Species Distribution Modeling (SDM) – Gyps indicus

This repository contains an end-to-end presence-only Species Distribution Modeling (SDM) pipeline for Gyps indicus, combining Google Earth Engine (GEE) for spatial data engineering with Kernel Density Estimation (KDE)–based modeling in Python to estimate relative habitat suitability.

Project Overview

• Objective: Model and map relative habitat suitability using presence-only occurrence data.
• Approach: Learn the species’ environmental niche via KDE and project predictions back into geographic space.
• Output: Continuous habitat suitability maps and occurrence density visualizations.

Workflow

1. Google Earth Engine (GEE)

   • Study area definition and occurrence point processing
   • Environmental raster extraction (WorldClim climate variables, SRTM elevation)
   • Background point sampling
   • Rasterization, reprojection, and spatial smoothing
   • Map-based visualization and exports

2. Python (Notebook)

   • Feature preprocessing (scaling, transformation)
   • KDE modeling in environmental space
   • Evaluation using AUC (presence vs background)
   • Export of predicted suitability for spatial mapping in GEE

Data

• gyps_indicus_data.csv: Presence-only occurrence records used for spatial processing and environmental sampling.
• combined_with_final_suit.csv: Occurance + Background points with KDE-based habitat suitability scores computed in Python and imported into GEE.
• 0022518-251009101135966.csv: Raw occurrence data source retained for traceability.

Evaluation

• Metric: AUC (presence vs background)
• Final AUC: 0.8664

Repository Structure

gyps-indicus-sdm/
│
├── README.md
├── sdm_kde_pipeline.ipynb
├── gyps_indicus_gee.js
├── .gitignore
└── assets/
    ├── gyps_indicus_data.csv
    ├── combined_with_final_suit.csv
    └── 0022518-251009101135966.csv

Tech Stack

Python, Google Earth Engine, scikit-learn, Pandas, NumPy, Matplotlib

Notes

• This project does not use true absence data.
• Suitability scores represent relative ranking, not probability of presence.
• Machine learning is performed in Python; GEE is used exclusively for spatial processing and mapping.