4_multivariate_analysis_ndvi.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "0ba9f552",
   "metadata": {},
   "source": [
    "# Multivariate analysis NDVI"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "e28a8f15",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import geopandas as gpd\n",
    "import matplotlib.pyplot as plt\n",
    "import warnings\n",
    "from pygam import GAM, LinearGAM, s, l\n",
    "from sklearn import linear_model, metrics\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import accuracy_score, mean_squared_error\n",
    "from PIL import Image as im"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "02196a7a",
   "metadata": {},
   "source": [
    "## 1. Modeling NDVI for a buffer distance of 100 meters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "9566af8e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>NDVI_100</th>\n",
       "      <th>Greenspace%_100</th>\n",
       "      <th>Canopy%_100</th>\n",
       "      <th>Grass%_100</th>\n",
       "      <th>Shrubs%_100</th>\n",
       "      <th>geometry</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.455823</td>\n",
       "      <td>0.992861</td>\n",
       "      <td>2.168372e-17</td>\n",
       "      <td>0.006803</td>\n",
       "      <td>0.848073</td>\n",
       "      <td>POINT (386483.262 417703.544)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.387782</td>\n",
       "      <td>0.521713</td>\n",
       "      <td>2.176871e-01</td>\n",
       "      <td>0.097506</td>\n",
       "      <td>0.190476</td>\n",
       "      <td>POINT (395310.475 388609.235)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.389722</td>\n",
       "      <td>0.979774</td>\n",
       "      <td>5.117090e-17</td>\n",
       "      <td>0.083900</td>\n",
       "      <td>0.891156</td>\n",
       "      <td>POINT (384225.708 417867.819)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.623486</td>\n",
       "      <td>0.771565</td>\n",
       "      <td>8.163265e-02</td>\n",
       "      <td>0.083900</td>\n",
       "      <td>0.612245</td>\n",
       "      <td>POINT (390673.998 416261.747)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.628733</td>\n",
       "      <td>0.886972</td>\n",
       "      <td>6.643991e-01</td>\n",
       "      <td>0.052154</td>\n",
       "      <td>0.163265</td>\n",
       "      <td>POINT (393523.154 393041.484)</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   NDVI_100  Greenspace%_100   Canopy%_100  Grass%_100  Shrubs%_100  \\\n",
       "0  0.455823         0.992861  2.168372e-17    0.006803     0.848073   \n",
       "1  0.387782         0.521713  2.176871e-01    0.097506     0.190476   \n",
       "2  0.389722         0.979774  5.117090e-17    0.083900     0.891156   \n",
       "3  0.623486         0.771565  8.163265e-02    0.083900     0.612245   \n",
       "4  0.628733         0.886972  6.643991e-01    0.052154     0.163265   \n",
       "\n",
       "                        geometry  \n",
       "0  POINT (386483.262 417703.544)  \n",
       "1  POINT (395310.475 388609.235)  \n",
       "2  POINT (384225.708 417867.819)  \n",
       "3  POINT (390673.998 416261.747)  \n",
       "4  POINT (393523.154 393041.484)  "
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# load greenspace metrics at 100m geodataframe\n",
    "gdf_100 = gpd.read_file(\"thesis_project/output/dataframes/greenspace_metrics_100_df_2.json\")\n",
    "gdf_100.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "88db9dcc",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Grass, Canopy, and Shrubs as predictors\n",
    "cols = ['Canopy%_100', 'Shrubs%_100', 'Grass%_100']\n",
    "df = pd.DataFrame(gdf_100, columns=cols)\n",
    "\n",
    "# target as NDVI_100\n",
    "target = pd.DataFrame(gdf_100, columns=['NDVI_100'])\n",
    "\n",
    "# define X and Y\n",
    "#X = df\n",
    "y = target['NDVI_100']\n",
    "X = df.iloc[:, 0:3].values\n",
    "\n",
    "# split independent and dependent variables into training and test data\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cfa5327e",
   "metadata": {},
   "source": [
    "### 1.1. Model 1: NDVI ~ linear(tree canopy) + linear(forbs and shrubs) + linear(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "5dd2f5b4",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:01 Time:  0:00:01\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                      3.9996\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -324268.3144\n",
      "Number of Samples:                         4749 AIC:                                            648546.628\n",
      "                                                AICc:                                          648546.6407\n",
      "                                                GCV:                                                0.0069\n",
      "                                                Scale:                                              0.0069\n",
      "                                                Pseudo R-Squared:                                   0.7166\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "l(0)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "l(1)                              [0.1]                1            1.0          1.11e-16     ***         \n",
      "l(2)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "intercept                                              1            1.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/4165518819.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_100_1.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 1\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_100_1 = LinearGAM(l(0) + l(1) + l(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_100_1.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_100_1.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0ae250d8",
   "metadata": {},
   "source": [
    "### 1.2. Model 2: NDVI ~ spline(tree canopy) + linear(forbs and shrubs) + linear(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "b7e5eb21",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:03 Time:  0:00:03\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                      19.696\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -366629.9158\n",
      "Number of Samples:                         4749 AIC:                                           733301.2237\n",
      "                                                AICc:                                          733301.4136\n",
      "                                                GCV:                                                0.0062\n",
      "                                                Scale:                                              0.0061\n",
      "                                                Pseudo R-Squared:                                   0.7497\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "s(0)                              [0.1]                20           17.7         1.11e-16     ***         \n",
      "l(1)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "l(2)                              [0.1]                1            1.0          1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/2442450153.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_100_2.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 2\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_100_2 = LinearGAM(s(0) + l(1) + l(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_100_2.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_100_2.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "eb1a741c",
   "metadata": {},
   "source": [
    "### 1.3. Model 3: NDVI ~ linear(tree canopy) + spline(forbs and shrubs) + linear(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "9fada82e",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:03 Time:  0:00:03\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     21.9807\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -388551.5317\n",
      "Number of Samples:                         4749 AIC:                                           777149.0248\n",
      "                                                AICc:                                          777149.2581\n",
      "                                                GCV:                                                0.0058\n",
      "                                                Scale:                                              0.0058\n",
      "                                                Pseudo R-Squared:                                   0.7636\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "l(0)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "s(1)                              [0.001]              20           20.0         1.11e-16     ***         \n",
      "l(2)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/4073591579.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_100_3.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 3\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_100_3 = LinearGAM(l(0) + s(1) + l(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_100_3.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_100_3.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3453e3ea",
   "metadata": {},
   "source": [
    "### 1.4. Model 4: NDVI ~ linear(tree canopy) + linear(forbs and shrubs) + spline(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "153120db",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:03 Time:  0:00:03\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     19.6974\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -360655.4858\n",
      "Number of Samples:                         4749 AIC:                                           721352.3663\n",
      "                                                AICc:                                          721352.5563\n",
      "                                                GCV:                                                0.0063\n",
      "                                                Scale:                                              0.0062\n",
      "                                                Pseudo R-Squared:                                   0.7458\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "l(0)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "l(1)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "s(2)                              [0.1]                20           17.7         1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/3955161003.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_100_4.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 4\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_100_4 = LinearGAM(l(0) + l(1) + s(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_100_4.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_100_4.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "47b64557",
   "metadata": {},
   "source": [
    "### 1.5. Model 5: NDVI ~ spline(tree canopy) + spline(forbs and shrubs) + linear(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "1911bdb6",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:06 Time:  0:00:06\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     37.6619\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -400601.1866\n",
      "Number of Samples:                         4749 AIC:                                           801279.6971\n",
      "                                                AICc:                                          801280.3483\n",
      "                                                GCV:                                                0.0057\n",
      "                                                Scale:                                              0.0056\n",
      "                                                Pseudo R-Squared:                                   0.7719\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "s(0)                              [0.1]                20           17.7         1.11e-16     ***         \n",
      "s(1)                              [0.001]              20           19.0         1.11e-16     ***         \n",
      "l(2)                              [0.1]                1            1.0          1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/979852687.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_100_5.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 5\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_100_5 = LinearGAM(s(0) + s(1) + l(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_100_5.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_100_5.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5651f77d",
   "metadata": {},
   "source": [
    "### 1.6. Model 6: NDVI ~ spline(tree canopy) + linear(forbs and shrubs) + spline(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "8cc62741",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:12 Time:  0:00:12\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     35.3783\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -385554.0392\n",
      "Number of Samples:                         4749 AIC:                                            771180.835\n",
      "                                                AICc:                                          771181.4122\n",
      "                                                GCV:                                                0.0059\n",
      "                                                Scale:                                              0.0058\n",
      "                                                Pseudo R-Squared:                                   0.7631\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "s(0)                              [0.1]                20           17.7         1.11e-16     ***         \n",
      "l(1)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "s(2)                              [0.1]                20           16.7         1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/1329383776.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_100_6.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 6\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_100_6 = LinearGAM(s(0) + l(1) + s(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_100_6.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_100_6.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "20f3eaa8",
   "metadata": {},
   "source": [
    "### 1.7. Model 7: NDVI ~ linear(tree canopy) + spline(forbs and shrubs) + spline(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "295a3ac3",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:13 Time:  0:00:13\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     30.1924\n",
      "Link Function:                     IdentityLink Log Likelihood:                                -391589.958\n",
      "Number of Samples:                         4749 AIC:                                           783242.3008\n",
      "                                                AICc:                                          783242.7266\n",
      "                                                GCV:                                                0.0058\n",
      "                                                Scale:                                              0.0057\n",
      "                                                Pseudo R-Squared:                                   0.7661\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "l(0)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "s(1)                              [0.001]              20           20.0         1.11e-16     ***         \n",
      "s(2)                              [10.]                20           9.2          1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/2977915924.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_100_7.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 7\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_100_7 = LinearGAM(l(0) + s(1) + s(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_100_7.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_100_7.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "544e75b7",
   "metadata": {},
   "source": [
    "### 1.8. Model 8: NDVI ~ spline(tree canopy) + spline(forbs and shrubs) + spline(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "0e44a6c9",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:08 Time:  0:00:08\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     52.0461\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -403740.0568\n",
      "Number of Samples:                         4749 AIC:                                           807586.2058\n",
      "                                                AICc:                                          807587.4271\n",
      "                                                GCV:                                                0.0056\n",
      "                                                Scale:                                              0.0055\n",
      "                                                Pseudo R-Squared:                                    0.775\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "s(0)                              [0.1]                20           17.7         1.11e-16     ***         \n",
      "s(1)                              [0.1]                20           17.7         1.11e-16     ***         \n",
      "s(2)                              [0.1]                20           16.6         1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/3108876825.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_100_8.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 8\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_100_8 = LinearGAM(s(0) + s(1) + s(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_100_8.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_100_8.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "12b4f19d",
   "metadata": {},
   "source": [
    "### 1.9. (BEST MODEL) Model 8: NDVI ~ spline(tree canopy) + spline(forbs and shrubs) + spline(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "e635f514",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Figure size 432x288 with 0 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x288 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "## plotting\n",
    "plt.figure();\n",
    "fig, axs = plt.subplots(1,3,figsize=(12,4));\n",
    "\n",
    "titles = ['Sensitivity of NDVI to tree canopy\\n for buffer zones of 100 m', \n",
    "          'Sensitivity of NDVI to shrubs density\\n for buffer zones of 100 m', \n",
    "          'Sensitivity of NDVI to grass density\\n for buffer zones of 100 m']\n",
    "x_labels = ['Tree canopy', 'Forbs and shrubs', 'Grass']\n",
    "for i, ax in enumerate(axs):    \n",
    "    XX = ndvi_100_8.generate_X_grid(term=i)\n",
    "    ax.plot(XX[:, i], ndvi_100_8.partial_dependence(term=i, X=XX))\n",
    "    ax.plot(XX[:, i], ndvi_100_8.partial_dependence(term=i, X=XX, width=.95)[1], c='r', ls='--')\n",
    "    ax.set_ylim(-0.3, 0.8)\n",
    "    ax.set_xticks(list(ax.get_xticks()))\n",
    "    #ax.get_yaxis().get_major_formatter().set_useOffset(False)\n",
    "    ax.hlines(y=0, xmin=0, xmax=1, linewidth=1, color='black', linestyle='--')    \n",
    "    ax.set_ylabel('Change relative to mean NDVI', fontsize=12)\n",
    "    ax.set_xlabel(x_labels[i], fontsize=12)\n",
    "    ax.set_title(titles[i], fontsize=12);\n",
    "    \n",
    "# non-overlapping subplots\n",
    "plt.tight_layout()\n",
    "    \n",
    "# save the figure\n",
    "plt.savefig('thesis_project/visualizations/modelling/multivariate/gam_ndvi_100_best_plot_2.png', dpi=300, bbox_inches='tight')\n",
    "im.open('thesis_project/visualizations/modelling/multivariate/gam_ndvi_100_best_plot_2.png').save('thesis_project/visualizations/modelling/multivariate/gam_ndvi_100_best_plot_2.bmp')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "065be971",
   "metadata": {},
   "source": [
    "- **Parameters of the best multivaritate model (lowest GCV score) that explains NDVI for a buffer zone of 100 meters**\n",
    "    - **Distribution**: Normal\n",
    "    - **Link Function**: Identity\n",
    "    - **Functional Form**: NDVI ~ spline(tree canopy) + spline(forbs and shrubs) + linear(grass)\n",
    "    - **Regularization penalty**: lambda = [0.1, 0.001, 0.1]\n",
    "    - **Smoothness**: n_splines = [20, 20, 1]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c11718ba",
   "metadata": {},
   "source": [
    "## 2. Modeling NDVI for a buffer distance of 300 meters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "95595499",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>NDVI_300</th>\n",
       "      <th>Greenspace%_300</th>\n",
       "      <th>Canopy%_300</th>\n",
       "      <th>Grass%_300</th>\n",
       "      <th>Shrubs%_300</th>\n",
       "      <th>geometry</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.280478</td>\n",
       "      <td>0.326412</td>\n",
       "      <td>0.145660</td>\n",
       "      <td>0.080623</td>\n",
       "      <td>0.097554</td>\n",
       "      <td>POINT (387760.076 393972.512)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.503846</td>\n",
       "      <td>0.847073</td>\n",
       "      <td>0.062886</td>\n",
       "      <td>0.391561</td>\n",
       "      <td>0.387530</td>\n",
       "      <td>POINT (361558.182 408801.964)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.589265</td>\n",
       "      <td>0.831569</td>\n",
       "      <td>0.104542</td>\n",
       "      <td>0.243752</td>\n",
       "      <td>0.480785</td>\n",
       "      <td>POINT (391126.815 417220.287)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.493672</td>\n",
       "      <td>0.796394</td>\n",
       "      <td>0.127654</td>\n",
       "      <td>0.150497</td>\n",
       "      <td>0.513034</td>\n",
       "      <td>POINT (361327.964 411390.899)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.343043</td>\n",
       "      <td>0.610614</td>\n",
       "      <td>0.275732</td>\n",
       "      <td>0.147272</td>\n",
       "      <td>0.190540</td>\n",
       "      <td>POINT (377099.280 400575.430)</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   NDVI_300  Greenspace%_300  Canopy%_300  Grass%_300  Shrubs%_300  \\\n",
       "0  0.280478         0.326412     0.145660    0.080623     0.097554   \n",
       "1  0.503846         0.847073     0.062886    0.391561     0.387530   \n",
       "2  0.589265         0.831569     0.104542    0.243752     0.480785   \n",
       "3  0.493672         0.796394     0.127654    0.150497     0.513034   \n",
       "4  0.343043         0.610614     0.275732    0.147272     0.190540   \n",
       "\n",
       "                        geometry  \n",
       "0  POINT (387760.076 393972.512)  \n",
       "1  POINT (361558.182 408801.964)  \n",
       "2  POINT (391126.815 417220.287)  \n",
       "3  POINT (361327.964 411390.899)  \n",
       "4  POINT (377099.280 400575.430)  "
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# load greenspace metrics at 300m geodataframe\n",
    "gdf_300 = gpd.read_file(\"thesis_project/output/dataframes/greenspace_metrics_300_df_2.json\")\n",
    "gdf_300.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "9b662710",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Grass, Canopy, and Shrubs as predictors\n",
    "cols = ['Canopy%_300', 'Shrubs%_300', 'Grass%_300']\n",
    "df = pd.DataFrame(gdf_300, columns=cols)\n",
    "\n",
    "# target as NDVI_300\n",
    "target = pd.DataFrame(gdf_300, columns=['NDVI_300'])\n",
    "\n",
    "# define X and Y\n",
    "#X = df\n",
    "y = target['NDVI_300']\n",
    "X = df.iloc[:, 0:3].values\n",
    "\n",
    "# split independent and dependent variables into training and test data\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6f97873f",
   "metadata": {},
   "source": [
    "### 2.1. Model 1: NDVI ~ linear(tree canopy) + linear(forbs and shrubs) + linear(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "d1bee84a",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:01 Time:  0:00:01\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                         4.0\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -501248.1023\n",
      "Number of Samples:                         4656 AIC:                                          1002506.2045\n",
      "                                                AICc:                                         1002506.2174\n",
      "                                                GCV:                                                0.0045\n",
      "                                                Scale:                                              0.0045\n",
      "                                                Pseudo R-Squared:                                   0.7343\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "l(0)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "l(1)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "l(2)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "intercept                                              1            1.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/2141360453.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_300_1.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 1\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_300_1 = LinearGAM(l(0) + l(1) + l(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_300_1.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_300_1.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4b388d84",
   "metadata": {},
   "source": [
    "### 2.2. Model 2: NDVI ~ spline(tree canopy) + linear(forbs and shrubs) + linear(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "2b9df38d",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:56 Time:  0:00:56\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     21.4949\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -601516.8402\n",
      "Number of Samples:                         4656 AIC:                                          1203078.6701\n",
      "                                                AICc:                                         1203078.8983\n",
      "                                                GCV:                                                0.0037\n",
      "                                                Scale:                                              0.0037\n",
      "                                                Pseudo R-Squared:                                    0.779\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "s(0)                              [0.001]              20           19.5         1.11e-16     ***         \n",
      "l(1)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "l(2)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/138880699.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_300_2.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 2\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_300_2 = LinearGAM(s(0) + l(1) + l(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_300_2.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_300_2.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "06fc3016",
   "metadata": {},
   "source": [
    "### 2.3. Model 3: NDVI ~ linear(tree canopy) + spline(forbs and shrubs) + linear(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "9a68566f",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:06 Time:  0:00:06\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                      20.247\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -653747.9372\n",
      "Number of Samples:                         4656 AIC:                                          1307538.3685\n",
      "                                                AICc:                                         1307538.5725\n",
      "                                                GCV:                                                0.0035\n",
      "                                                Scale:                                              0.0034\n",
      "                                                Pseudo R-Squared:                                   0.7961\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "l(0)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "s(1)                              [0.1]                20           18.2         1.11e-16     ***         \n",
      "l(2)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/2763482722.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_300_3.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 3\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_300_3 = LinearGAM(l(0) + s(1) + l(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_300_3.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_300_3.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b9b9aab5",
   "metadata": {},
   "source": [
    "### 2.4. Model 4: NDVI ~ linear(tree canopy) + linear(forbs and shrubs) + spline(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "6a846b22",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:06 Time:  0:00:06\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     16.2509\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -590465.9549\n",
      "Number of Samples:                         4656 AIC:                                          1180966.4116\n",
      "                                                AICc:                                         1180966.5473\n",
      "                                                GCV:                                                0.0038\n",
      "                                                Scale:                                              0.0038\n",
      "                                                Pseudo R-Squared:                                   0.7745\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "l(0)                              [0.6]                1            1.0          1.11e-16     ***         \n",
      "l(1)                              [0.6]                1            1.0          1.11e-16     ***         \n",
      "s(2)                              [0.6]                20           14.3         1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/3450665515.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_300_4.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 4\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_300_4 = LinearGAM(l(0) + l(1) + s(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_300_4.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_300_4.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "17feb623",
   "metadata": {},
   "source": [
    "### 2.5. Model 5: NDVI ~ spline(tree canopy) + spline(forbs and shrubs) + linear(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "8cc54ef8",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:15 Time:  0:00:15\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     30.7766\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -686744.9133\n",
      "Number of Samples:                         4656 AIC:                                          1373553.3796\n",
      "                                                AICc:                                         1373553.8302\n",
      "                                                GCV:                                                0.0033\n",
      "                                                Scale:                                              0.0033\n",
      "                                                Pseudo R-Squared:                                   0.8066\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "s(0)                              [0.001]              20           19.5         1.11e-16     ***         \n",
      "s(1)                              [10.]                20           10.3         1.11e-16     ***         \n",
      "l(2)                              [0.1]                1            1.0          1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/1916949231.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_300_5.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 5\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_300_5 = LinearGAM(s(0) + s(1) + l(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_300_5.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_300_5.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c2389aa2",
   "metadata": {},
   "source": [
    "### 2.6. Model 6: NDVI ~ spline(tree canopy) + linear(forbs and shrubs) + spline(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "95f7f012",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:15 Time:  0:00:15\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     38.3238\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -652814.1322\n",
      "Number of Samples:                         4656 AIC:                                          1305706.9119\n",
      "                                                AICc:                                         1305707.5989\n",
      "                                                GCV:                                                0.0035\n",
      "                                                Scale:                                              0.0034\n",
      "                                                Pseudo R-Squared:                                   0.7974\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "s(0)                              [0.001]              20           19.5         1.11e-16     ***         \n",
      "l(1)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "s(2)                              [0.001]              20           17.8         1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/2187392072.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_300_6.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 6\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_300_6 = LinearGAM(s(0) + l(1) + s(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_300_6.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_300_6.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c16486fa",
   "metadata": {},
   "source": [
    "### 2.7. Model 7: NDVI ~ linear(tree canopy) + spline(forbs and shrubs) + spline(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "80497c95",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:08 Time:  0:00:08\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     30.0383\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -679414.8326\n",
      "Number of Samples:                         4656 AIC:                                          1358891.7417\n",
      "                                                AICc:                                         1358892.1719\n",
      "                                                GCV:                                                0.0033\n",
      "                                                Scale:                                              0.0033\n",
      "                                                Pseudo R-Squared:                                   0.8045\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "l(0)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "s(1)                              [10.]                20           11.5         1.11e-16     ***         \n",
      "s(2)                              [0.001]              20           17.5         1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/2141254406.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_300_7.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 7\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_300_7 = LinearGAM(l(0) + s(1) + s(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_300_7.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_300_7.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6f68856b",
   "metadata": {},
   "source": [
    "### 2.8. Model 8: NDVI ~ spline(tree canopy) + spline(forbs and shrubs) + spline(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "849bf675",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:13 Time:  0:00:13\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     47.5169\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -714716.4087\n",
      "Number of Samples:                         4656 AIC:                                          1429529.8512\n",
      "                                                AICc:                                         1429530.8942\n",
      "                                                GCV:                                                0.0032\n",
      "                                                Scale:                                              0.0031\n",
      "                                                Pseudo R-Squared:                                   0.8153\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "s(0)                              [0.001]              20           19.5         1.11e-16     ***         \n",
      "s(1)                              [10.]                20           10.5         1.11e-16     ***         \n",
      "s(2)                              [0.001]              20           17.5         1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/547711982.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_300_8.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 8\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_300_8 = LinearGAM(s(0) + s(1) + s(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_300_8.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_300_8.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "df50fd62",
   "metadata": {},
   "source": [
    "### 2.9. (BEST MODEL) Model 8: NDVI ~ spline(tree canopy) + spline(forbs and shrubs) + spline(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "8ec10f89",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Figure size 432x288 with 0 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x288 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "## plotting\n",
    "plt.figure();\n",
    "fig, axs = plt.subplots(1,3,figsize=(12,4));\n",
    "\n",
    "titles = ['Sensitivity of NDVI to tree canopy\\n for buffer zones of 300 m', \n",
    "          'Sensitivity of NDVI to shrubs density\\n for buffer zones of 300 m', \n",
    "          'Sensitivity of NDVI to grass density\\n for buffer zones of 300 m']\n",
    "x_labels = ['Tree canopy', 'Forbs and shrubs', 'Grass']\n",
    "for i, ax in enumerate(axs):\n",
    "    XX = ndvi_300_8.generate_X_grid(term=i)\n",
    "    ax.plot(XX[:, i], ndvi_300_8.partial_dependence(term=i, X=XX))\n",
    "    ax.plot(XX[:, i], ndvi_300_8.partial_dependence(term=i, X=XX, width=.95)[1], c='r', ls='--')\n",
    "    ax.set_ylim(-0.3, 0.8)\n",
    "    ax.set_xticks(list(ax.get_xticks()))\n",
    "    ax.hlines(y=0, xmin=0, xmax=1, linewidth=1, color='black', linestyle='--')\n",
    "    ax.set_ylabel('Change relative to mean NDVI', fontsize=12)\n",
    "    ax.set_xlabel(x_labels[i], fontsize=12)\n",
    "    ax.set_title(titles[i], fontsize=12);\n",
    "    \n",
    "# non-overlapping subplots\n",
    "plt.tight_layout()\n",
    "    \n",
    "# save the figure\n",
    "plt.savefig('thesis_project/visualizations/modelling/multivariate/gam_ndvi_300_best_plot_2.png', dpi=300, bbox_inches='tight')\n",
    "im.open('thesis_project/visualizations/modelling/multivariate/gam_ndvi_300_best_plot_2.png').save('thesis_project/visualizations/modelling/multivariate/gam_ndvi_300_best_plot_2.bmp')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "59706a18",
   "metadata": {},
   "source": [
    "- **Parameters of the best multivaritate model (lowest GCV score) that explains NDVI for a buffer zone of 300 meters**\n",
    "    - **Distribution**: Normal\n",
    "    - **Link Function**: Identity\n",
    "    - **Functional Form**: NDVI ~ spline(tree canopy) + spline(forbs and shrubs) + spline(grass)\n",
    "    - **Regularization penalty**: lambda = [0.001, 10, 10]\n",
    "    - **Smoothness**: n_splines = 20\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e73a59f3",
   "metadata": {},
   "source": [
    "## 3. Modeling NDVI for a buffer distance of 500 meters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "d3b42609",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>NDVI_500</th>\n",
       "      <th>Greenspace%_500</th>\n",
       "      <th>Canopy%_500</th>\n",
       "      <th>Grass%_500</th>\n",
       "      <th>Shrubs%_500</th>\n",
       "      <th>geometry</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.407077</td>\n",
       "      <td>0.982575</td>\n",
       "      <td>0.006176</td>\n",
       "      <td>0.028919</td>\n",
       "      <td>0.947064</td>\n",
       "      <td>POINT (401901.359 408671.581)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.441260</td>\n",
       "      <td>0.575357</td>\n",
       "      <td>0.167827</td>\n",
       "      <td>0.131850</td>\n",
       "      <td>0.277816</td>\n",
       "      <td>POINT (368835.367 407479.786)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.312300</td>\n",
       "      <td>0.395262</td>\n",
       "      <td>0.212136</td>\n",
       "      <td>0.095873</td>\n",
       "      <td>0.087050</td>\n",
       "      <td>POINT (377751.293 401260.842)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.283220</td>\n",
       "      <td>0.355313</td>\n",
       "      <td>0.098324</td>\n",
       "      <td>0.070777</td>\n",
       "      <td>0.188119</td>\n",
       "      <td>POINT (391535.548 395524.756)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.552286</td>\n",
       "      <td>0.770303</td>\n",
       "      <td>0.265954</td>\n",
       "      <td>0.094795</td>\n",
       "      <td>0.409666</td>\n",
       "      <td>POINT (369836.790 402258.456)</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   NDVI_500  Greenspace%_500  Canopy%_500  Grass%_500  Shrubs%_500  \\\n",
       "0  0.407077         0.982575     0.006176    0.028919     0.947064   \n",
       "1  0.441260         0.575357     0.167827    0.131850     0.277816   \n",
       "2  0.312300         0.395262     0.212136    0.095873     0.087050   \n",
       "3  0.283220         0.355313     0.098324    0.070777     0.188119   \n",
       "4  0.552286         0.770303     0.265954    0.094795     0.409666   \n",
       "\n",
       "                        geometry  \n",
       "0  POINT (401901.359 408671.581)  \n",
       "1  POINT (368835.367 407479.786)  \n",
       "2  POINT (377751.293 401260.842)  \n",
       "3  POINT (391535.548 395524.756)  \n",
       "4  POINT (369836.790 402258.456)  "
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# load greenspace metrics at 500m geodataframe\n",
    "gdf_500 = gpd.read_file(\"thesis_project/output/dataframes/greenspace_metrics_500_df_2.json\")\n",
    "gdf_500.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "27871e3a",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Grass, Canopy, and Shrubs as predictors\n",
    "cols = ['Canopy%_500', 'Shrubs%_500', 'Grass%_500']\n",
    "df = pd.DataFrame(gdf_500, columns=cols)\n",
    "\n",
    "# target as NDVI_500\n",
    "target = pd.DataFrame(gdf_500, columns=['NDVI_500'])\n",
    "\n",
    "# define X and Y\n",
    "#X = df\n",
    "y = target['NDVI_500']\n",
    "X = df.iloc[:, 0:3].values\n",
    "\n",
    "# split independent and dependent variables into training and test data\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d63908d3",
   "metadata": {},
   "source": [
    "### 3.1. Model 1: NDVI ~ linear(tree canopy) + linear(forbs and shrubs) + linear(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "422ac448",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:04 Time:  0:00:04\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                      3.9999\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -288532.8263\n",
      "Number of Samples:                         2316 AIC:                                           577075.6524\n",
      "                                                AICc:                                          577075.6783\n",
      "                                                GCV:                                                0.0039\n",
      "                                                Scale:                                              0.0039\n",
      "                                                Pseudo R-Squared:                                   0.7147\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "l(0)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "l(1)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "l(2)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "intercept                                              1            1.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/1220089107.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_500_1.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 1\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_500_1 = LinearGAM(l(0) + l(1) + l(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_500_1.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_500_1.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "861fc633",
   "metadata": {},
   "source": [
    "### 3.2. Model 2: NDVI ~ spline(tree canopy) + linear(forbs and shrubs) + linear(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "ec7045ed",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:04 Time:  0:00:04\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     15.7745\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -321062.3011\n",
      "Number of Samples:                         2316 AIC:                                           642158.1511\n",
      "                                                AICc:                                          642158.4106\n",
      "                                                GCV:                                                0.0035\n",
      "                                                Scale:                                              0.0035\n",
      "                                                Pseudo R-Squared:                                   0.7455\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "s(0)                              [0.6]                20           13.8         1.11e-16     ***         \n",
      "l(1)                              [0.6]                1            1.0          1.11e-16     ***         \n",
      "l(2)                              [0.6]                1            1.0          1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/2940222901.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_500_2.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 2\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_500_2 = LinearGAM(s(0) + l(1) + l(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_500_2.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_500_2.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "22659493",
   "metadata": {},
   "source": [
    "### 3.3. Model 3: NDVI ~ linear(tree canopy) + spline(forbs and shrubs) + linear(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "401639f6",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:03 Time:  0:00:03\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     11.9292\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -383984.1565\n",
      "Number of Samples:                         2316 AIC:                                           767994.1714\n",
      "                                                AICc:                                          767994.3279\n",
      "                                                GCV:                                                0.0029\n",
      "                                                Scale:                                              0.0029\n",
      "                                                Pseudo R-Squared:                                   0.7855\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "l(0)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "s(1)                              [10.]                20           9.9          1.11e-16     ***         \n",
      "l(2)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/3773330009.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_500_3.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 3\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_500_3 = LinearGAM(l(0) + s(1) + l(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_500_3.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_500_3.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5cdb9a1f",
   "metadata": {},
   "source": [
    "### 3.4. Model 4: NDVI ~ linear(tree canopy) + linear(forbs and shrubs) + spline(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "73ce6f1b",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:04 Time:  0:00:04\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     11.3218\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -329813.7655\n",
      "Number of Samples:                         2316 AIC:                                           659652.1746\n",
      "                                                AICc:                                          659652.3171\n",
      "                                                GCV:                                                0.0034\n",
      "                                                Scale:                                              0.0034\n",
      "                                                Pseudo R-Squared:                                   0.7511\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "l(0)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "l(1)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "s(2)                              [10.]                20           9.3          1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/755273682.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_500_4.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 4\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_500_4 = LinearGAM(l(0) + l(1) + s(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_500_4.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_500_4.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "94dad9c4",
   "metadata": {},
   "source": [
    "### 3.5. Model 5: NDVI ~ spline(tree canopy) + spline(forbs and shrubs) + linear(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "5ac8340a",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:06 Time:  0:00:06\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     26.1445\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -397276.6384\n",
      "Number of Samples:                         2316 AIC:                                           794607.5658\n",
      "                                                AICc:                                          794608.2336\n",
      "                                                GCV:                                                0.0029\n",
      "                                                Scale:                                              0.0028\n",
      "                                                Pseudo R-Squared:                                   0.7951\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "s(0)                              [0.1]                20           16.3         1.11e-16     ***         \n",
      "s(1)                              [10.]                20           8.9          1.11e-16     ***         \n",
      "l(2)                              [0.1]                1            1.0          1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/3854426641.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_500_5.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 5\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_500_5 = LinearGAM(s(0) + s(1) + l(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_500_5.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_500_5.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "230feebb",
   "metadata": {},
   "source": [
    "### 3.6. Model 6: NDVI ~ spline(tree canopy) + linear(forbs and shrubs) + spline(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "46d93acb",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:08 Time:  0:00:08\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     27.8966\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -349491.5363\n",
      "Number of Samples:                         2316 AIC:                                           699040.8658\n",
      "                                                AICc:                                          699041.6216\n",
      "                                                GCV:                                                0.0032\n",
      "                                                Scale:                                              0.0032\n",
      "                                                Pseudo R-Squared:                                   0.7682\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "s(0)                              [0.6]                20           13.8         1.11e-16     ***         \n",
      "l(1)                              [0.6]                1            1.0          1.11e-16     ***         \n",
      "s(2)                              [0.6]                20           13.1         1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/1996838194.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_500_6.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 6\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_500_6 = LinearGAM(s(0) + l(1) + s(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_500_6.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_500_6.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ef741dcc",
   "metadata": {},
   "source": [
    "### 3.7. Model 7: NDVI ~ linear(tree canopy) + spline(forbs and shrubs) + spline(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "7c52cfbc",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:16 Time:  0:00:16\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     19.1505\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -401162.8729\n",
      "Number of Samples:                         2316 AIC:                                           802366.0467\n",
      "                                                AICc:                                          802366.4181\n",
      "                                                GCV:                                                0.0028\n",
      "                                                Scale:                                              0.0028\n",
      "                                                Pseudo R-Squared:                                   0.7958\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "l(0)                              [0.001]              1            1.0          1.11e-16     ***         \n",
      "s(1)                              [10.]                20           10.0         1.11e-16     ***         \n",
      "s(2)                              [10.]                20           8.1          1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/3717736121.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_500_7.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 7\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_500_7 = LinearGAM(l(0) + s(1) + s(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_500_7.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_500_7.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2c447c36",
   "metadata": {},
   "source": [
    "### 3.8. Model 8: NDVI ~ spline(tree canopy) + spline(forbs and shrubs) + spline(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "b3712729",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100% (125 of 125) |######################| Elapsed Time: 0:00:24 Time:  0:00:24\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LinearGAM                                                                                                 \n",
      "=============================================== ==========================================================\n",
      "Distribution:                        NormalDist Effective DoF:                                     40.7244\n",
      "Link Function:                     IdentityLink Log Likelihood:                               -418549.3081\n",
      "Number of Samples:                         2316 AIC:                                            837182.065\n",
      "                                                AICc:                                          837183.6333\n",
      "                                                GCV:                                                0.0027\n",
      "                                                Scale:                                              0.0026\n",
      "                                                Pseudo R-Squared:                                   0.8078\n",
      "==========================================================================================================\n",
      "Feature Function                  Lambda               Rank         EDoF         P > x        Sig. Code   \n",
      "================================= ==================== ============ ============ ============ ============\n",
      "s(0)                              [0.1]                20           16.3         1.11e-16     ***         \n",
      "s(1)                              [10.]                20           9.2          1.11e-16     ***         \n",
      "s(2)                              [0.1]                20           15.3         1.11e-16     ***         \n",
      "intercept                                              1            0.0          1.11e-16     ***         \n",
      "==========================================================================================================\n",
      "Significance codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1\n",
      "\n",
      "WARNING: Fitting splines and a linear function to a feature introduces a model identifiability problem\n",
      "         which can cause p-values to appear significant when they are not.\n",
      "\n",
      "WARNING: p-values calculated in this manner behave correctly for un-penalized models or models with\n",
      "         known smoothing parameters, but when smoothing parameters have been estimated, the p-values\n",
      "         are typically lower than they should be, meaning that the tests reject the null too readily.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/6g/kgz5rml118s6lqw25w5sv5mr0000gn/T/ipykernel_2261/1642183498.py:10: UserWarning: KNOWN BUG: p-values computed in this summary are likely much smaller than they should be. \n",
      " \n",
      "Please do not make inferences based on these values! \n",
      "\n",
      "Collaborate on a solution, and stay up to date at: \n",
      "github.com/dswah/pyGAM/issues/163 \n",
      "\n",
      "  ndvi_500_8.summary()\n"
     ]
    }
   ],
   "source": [
    "# Model 8\n",
    "warnings.filterwarnings(\"ignore\", category=RuntimeWarning)\n",
    "ndvi_500_8 = LinearGAM(s(0) + s(1) + s(2)).fit(X_train, y_train)\n",
    "\n",
    "# using gridsearch to tune the lambdas\n",
    "lam = np.logspace(-3, 5, 5)\n",
    "lams = [lam] * 3\n",
    "\n",
    "ndvi_500_8.gridsearch(X_train, y_train, lam=lams)\n",
    "ndvi_500_8.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "dd3b0846",
   "metadata": {},
   "source": [
    "### 3.9. (BEST MODEL) Model 8: NDVI ~ spline(tree canopy) + spline(forbs and shrubs) + spline(grass)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "457a1335",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Figure size 432x288 with 0 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x288 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "## plotting\n",
    "plt.figure();\n",
    "fig, axs = plt.subplots(1,3,figsize=(12,4));\n",
    "\n",
    "titles = ['Sensitivity of NDVI to tree canopy\\n for buffer zones of 500 m', \n",
    "          'Sensitivity of NDVI to shrubs density\\n for buffer zones of 500 m', \n",
    "          'Sensitivity of NDVI to grass density\\n for buffer zones of 500 m']\n",
    "x_labels = ['Tree canopy', 'Forbs and shrubs', 'Grass']\n",
    "for i, ax in enumerate(axs):\n",
    "    XX = ndvi_500_8.generate_X_grid(term=i)\n",
    "    ax.plot(XX[:, i], ndvi_500_8.partial_dependence(term=i, X=XX))\n",
    "    ax.plot(XX[:, i], ndvi_500_8.partial_dependence(term=i, X=XX, width=.95)[1], c='r', ls='--')\n",
    "    ax.set_ylim(-0.3, 0.8)\n",
    "    ax.set_xticks(list(ax.get_xticks()))\n",
    "    ax.hlines(y=0, xmin=0, xmax=1, linewidth=1, color='black', linestyle='--')\n",
    "    ax.set_ylabel('Change relative to mean NDVI', fontsize=12)\n",
    "    ax.set_xlabel(x_labels[i], fontsize=12)\n",
    "    ax.set_title(titles[i], fontsize=12);\n",
    "    \n",
    "# non-overlapping subplots\n",
    "plt.tight_layout()\n",
    "    \n",
    "# save the figure\n",
    "plt.savefig('thesis_project/visualizations/modelling/multivariate/gam_ndvi_500_best_plot_2.png', dpi=300, bbox_inches='tight')\n",
    "im.open('thesis_project/visualizations/modelling/multivariate/gam_ndvi_500_best_plot_2.png').save('thesis_project/visualizations/modelling/multivariate/gam_ndvi_500_best_plot_2.bmp')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7ae731ae",
   "metadata": {},
   "source": [
    "- **Parameters of the best multivaritate model (lowest GCV score) that explains NDVI for a buffer zone of 500 meters**\n",
    "    - **Distribution**: Normal\n",
    "    - **Link Function**: Identity\n",
    "    - **Functional Form**: NDVI ~ spline(tree canopy) + spline(forbs and shrubs) + spline(grass)\n",
    "    - **Regularization penalty**: lambda = [0.001, 10, 1000]\n",
    "    - **Smoothness**: n_splines = 20"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}