kbratley/mozambique-agriculture-2022

# Mozambique Active Agricultural Extent Mapping (2022) --- ## Overview This project maps active agricultural land extent in Mozambique for 2022 using Sentinel-2 Level-2A Surface Reflectance imagery. The analysis was conducted using a combination of the DEA Sandbox and Google Earth Engine (GEE) platforms. This repository provides an overview of the methodology, results, and access to the data used in this study. ### Data Access The final Mozambique’s Active Agriculture Extent Map (2022) is available on Google Earth Engine (GEE). ### Data Description The mapping was conducted using the Digital Earth Africa (DEA) crop-type mapping workflow, which leverages Sentinel-2 geomedian composites and machine learning techniques. The workflow was adapted for the DEA Sandbox and GEE platforms. **Key Data Inputs:** **1. Sentinel-2 Geomedian Composites (from DEA):** - Annual Composite for 2022. - Quarterly Composites: Jan-Mar, Apr-Jun, Jul-Sep, and Oct-Dec. **2. Median Absolute Deviation (MAD) Layers:** - Euclidean MAD (EMAD): Highlights pixel variability in multi-dimensional space. - Spectral MAD (SMAD): Captures spectral variability. - Bray-Curtis MAD (BCMAD): Captures spatial arrangement and heterogeneity. **3. Spectral Indices:** NDVI, LAI, and Tasseled Cap transformations were included to improve vegetation monitoring and land-cover differentiation. --- ## Methodology Overview ### Training Data Collection (GEE) - Training datasets were prepared using Google Earth Engine, with individual JavaScript scripts for each Mozambican province. - These datasets were labeled with two classes: **Agriculture** and **Other**, based on 34,604 sites across the country. - The `merge_trainingData` script combined provincial datasets into a single **national training dataset** for model training. ### Land Cover Classification (DEA) - Python scripts accessed DEA datasets, processed imagery, and applied a **Random Forest classifier**. - Classification was performed using the collected t …