Mastro1/Africa_Maize_Modelling

This is the repository used to generate the data available in the paper A Scalable Satellite Based Framework for Sub-National Yield Modelling to Strengthening Agricultural Risk Management in Africa # Sub-National Maize Yield Modeling for Agricultural Risk Management in Africa This repository contains the source code and data for the research paper, "A Scalable Satellite-Based Framework for Sub-National Yield Modelling to Strengthen Agricultural Risk Management in Africa". The project presents a generalizable machine learning framework that disaggregates national maize yield statistics to the sub-national level using Earth Observation (EO) data. The primary goal is to create a consistent, robust, and spatially explicit maize yield dataset for Africa to support applications like index-based insurance, catastrophe modeling, and food security assessments. ## Results visualization ## Key Features * **Scalable Framework:** A transferable model that can be applied across different agro-ecological zones, even in countries without extensive ground-truth yield data. * **Data-Driven Methodology:** Utilizes a wide range of publicly available Earth Observation datasets for climate, vegetation, and soil predictors. * **Rigorous Curation:** Implements a two-phase data curation protocol to create a high-confidence "Golden Cohort" of 9 countries for model training. * **Advanced Modeling:** Employs an Extra-Trees Regressor with an adaptive "analogue" strategy, which tailors the training process to the specific environmental context of the target country. * **Open Data & Code:** The generated pan-African yield dataset and the modeling code are openly shared to encourage replication, collaboration, and further research. ## Methodology Overview The framework is designed to overcome the lack of comparable, sub-national yield data across Africa. 1. **Data Sources & Preparation:** The model integrates two primary types of data: * **Ground-Truth:** Sub-national maize yield statistics from the HarvestStat Africa database. * **Predictors:** A comprehensive suite of spatio-temporal predictors from EO sources (MODIS, ERA5-Land) and static datasets (GAEZ, HWSD), cov …