An End-to-End GeoAI Intelligence Platform for Agricultural Monitoring using Satellite Foundation Model Representations, AlphaEarth, Galileo, GIS Analytics, and Machine Learning
Remote Sensing • GeoAI • Earth Observation • Precision Agriculture • Satellite Foundation Models • GIS • Machine Learning
This repository presents an applied GeoAI research and engineering platform designed for agricultural monitoring and field-level crop mapping using satellite foundation model representations.
The framework integrates:
- Satellite foundation model embeddings
- Earth Observation analytics
- Google Earth Engine processing
- GIS-based polygon intelligence
- NDVI vegetation analysis
- Machine learning classification
- Automated agricultural map generation
The system transforms raw agricultural field boundaries into intelligent crop information products through an end-to-end computational pipeline.
The project was developed during:
Fall 2025
as an applied remote sensing intelligence system targeting real-world agricultural monitoring challenges.
Agricultural monitoring over large geographic regions remains challenging because of:
- High cost of field surveys
- Limited scalability of manual inspection
- Seasonal crop variations
- Complex satellite data interpretation
Satellite Earth Observation provides continuous and large-scale information about agricultural regions.
However, converting raw satellite observations into actionable agricultural intelligence requires:
- Powerful feature representations
- Automated geospatial processing
- Robust machine learning models
This project addresses these challenges by developing an automated satellite-based crop intelligence workflow.
Given:
- Agricultural field polygons
- Satellite-derived representations
- Vegetation information
The objective is to learn discriminative representations for:
-
Vegetation status detection
-
Agricultural crop classification
The final goal is:
Satellite Observations
↓
Geospatial Intelligence
↓
Crop Information Products
A complete workflow from agricultural polygons to final GIS products:
Agricultural Field Polygons
↓
Geospatial Preprocessing
↓
Satellite Representation Extraction
↓
Vegetation Intelligence
↓
Crop Classification
↓
GIS Decision Support Outputs
The framework incorporates modern satellite representation learning approaches.
AlphaEarth representations are used to extract high-level satellite features capturing:
- Temporal characteristics
- Spectral patterns
- Earth observation information
Supported configurations:
| Feature Type | Dimension |
|---|---|
| Mean Embedding | 64 |
| Statistical Representation | 256 |
| Combined Representation | 320 |
The framework additionally supports Galileo-based satellite representations.
Available modes:
- Galileo-only features
- AlphaEarth + Galileo feature fusion
Feature fusion enables combining complementary satellite representations.
Unlike traditional pixel-based classification approaches,
this framework performs:
where each agricultural field polygon becomes an independent analysis unit.
Advantages:
- Preserves agricultural field boundaries
- Produces interpretable outputs
- Supports GIS-based decision making
The platform contains two classification engines.
Objective:
Separate agricultural polygons into:
Vegetation
Non-Vegetation
Unknown
Input:
Satellite embeddings
Output:
Vegetation mask
After vegetation filtering,
vegetation-positive polygons are processed by the crop classification engine.
Task:
Five-class crop classification
Input:
Satellite Representation Features
*
Agricultural Polygon Information
Output:
Crop Category Prediction
The platform follows a modular end-to-end GeoAI architecture consisting of five major layers:
User Interface Layer
|
|
v
+-----------------------------+
| GeoAI Processing Engine |
+-----------------------------+
|
------------------------------------------------
| |
v v
+-------------------------+ +------------------------------+
| Geospatial Processing | | Machine Learning Intelligence|
+-------------------------+ +------------------------------+
| | | |
| - Polygon Validation | | - Polygon-MLP |
| - Geometry Filtering | | - Crop-MLP |
| - poly_id Generation | | - Model Evaluation |
| | | |
+-------------------------+ +------------------------------+
|
|
v
+---------------------------------------------------------+
| Satellite Representation Layer |
+---------------------------------------------------------+
|
--------------------------------
| |
v v
+-------------------+ +-------------------+
| AlphaEarth | | Galileo |
| Satellite | | Satellite |
| Embeddings | | Representations |
+-------------------+ +-------------------+
|
|
v
+----------------------------------------------------+
| Feature Engineering |
| |
| - Feature Fusion |
| - Statistical Representation |
| - Embedding Preparation |
+----------------------------------------------------+
|
|
v
+----------------------------------------------------+
| NDVI Vegetation Intelligence |
| |
| - Vegetation Detection |
| - Non-Vegetation Filtering |
+----------------------------------------------------+
|
|
v
+----------------------------------------------------+
| GIS Crop Mapping Products |
| |
| - Crop Classification Maps |
| - GeoJSON / GeoPackage Outputs |
| - Decision Support Visualization |
+----------------------------------------------------+
Detailed architecture:
docs/system_architecture.md
The proposed framework consists of five main stages.
Input:
Agricultural vector polygons
Processing:
- Geometry validation
- Area filtering
- Boundary simplification
- Coordinate normalization
- Unique polygon identifier generation
Output:
Clean agricultural field representation.
Satellite features are extracted using:
- AlphaEarth embeddings
- Galileo representations
These representations encode:
- Spectral information
- Temporal behavior
- Spatial characteristics
Vegetation analysis is performed using:
NDVI-based vegetation indicators
Generated categories:
Vegetation
Non-Vegetation
Unknown
The generated labels are used for Polygon-MLP learning.
Vegetation-positive polygons are processed by Crop-MLP.
Workflow:
Satellite Embeddings
↓
Feature Matrix Construction
↓
MLP Classification
↓
Crop Category Prediction
The system generates:
- GeoJSON crop maps
- GeoPackage outputs
- Evaluation reports
- Visualization products
The experiments were performed on a private agricultural dataset.
Due to confidentiality restrictions:
The following data are not publicly released:
- Raw agricultural polygons
- Crop labels
- Satellite extracted features
The repository provides:
- Complete methodology
- Implementation framework
- Experimental design
- Evaluation strategy
- Visualization examples
Classification evaluation was performed using:
Training Data : 70%
Testing Data : 30%
using:
Stratified Shuffle Split
The following metrics were used:
- Accuracy
- Precision
- Recall
- F1-score
- Macro F1-score
| Metric | Score |
|---|---|
| Accuracy | 94.00% |
| Macro F1-score | 94.05% |
The results demonstrate that satellite foundation model representations combined with machine learning classifiers can effectively support field-level crop mapping.
Capabilities:
- Agricultural polygon loading
- GIS visualization
- Geometry preprocessing configuration
Capabilities:
- AlphaEarth configuration
- Galileo integration
- Feature extraction
- Polygon-MLP training
Capabilities:
- Crop dataset loading
- Feature selection
- Crop-MLP training
- Evaluation
- Mapping
Example outputs:
GeoAI-Crop-Mapping-Platform-with-Satellite-Foundation-Models
│
├── README.md
├── LICENSE
├── CITATION.cff
├── requirements.txt
│
├── docs/
│ │
│ ├── project_overview.md
│ ├── methodology.md
│ ├── system_architecture.md
│ ├── workflow.png
│ └── research_background.md
│
├── application/
│ └── geoai_crop_mapping_platform.py
│
├── data/
│ └── README.md
│
├── experiments/
│ ├── experiment_design.md
│ └── evaluation_protocol.md
│
└── results/
├── figures/
└── metrics/
git clone https://github.com/hannah-fathi/GeoAI-Crop-Mapping-Platform-with-Satellite-Foundation-Models.git
cd GeoAI-Crop-Mapping-Platform-with-Satellite-Foundation-Models
pip install -r requirements.txt
python application/geoai_crop_mapping_platform.py
The application provides:
- Polygon loading
- Satellite embedding extraction
- Machine learning training
- Crop prediction
- GIS visualization
The project follows a research-oriented workflow:
Data Preparation
↓
Feature Extraction
↓
Model Training
↓
Evaluation
↓
GIS Visualization
All experimental details are documented in:
docs/
experiments/
This project contributes to:
- GeoAI
- Remote Sensing
- Earth Observation Foundation Models
- Precision Agriculture
- Machine Learning
- GIS Intelligence
Potential extensions:
- Temporal satellite foundation models
- Self-supervised agricultural representation learning
- Deep learning crop segmentation
- Large-scale cloud deployment
- Multi-modal Earth Observation AI
Fall 2025
Research Interests:
- Computer Vision
- Remote Sensing
- Medical AI
- Large Language Models
- GeoAI
If you use this framework in research, please cite:
@software{fathi2025geoai_crop_mapping,
author = {Hannah Fathi},
title = {GeoAI Crop Mapping Platform with Satellite Foundation Models},
year = {2025},
description = {
An end-to-end GeoAI framework for agricultural monitoring
using satellite foundation model representations,
GIS analytics, and machine learning classification.
}
}
MIT License