Anonymes-237/Climate-ML

About A machine learning framework to correct the "aridity blindspot" by distinguishing drought from chronic aridity. It enables accurate spatio-temporal analysis and forecasting of water stress in Central Africa, offering a robust tool for risk management in vulnerable regions. # The Aridity Blindspot: A Machine Learning Framework for Spatio-Temporal Drought Analysis and Forecasting --- ### Abstract This project introduces a comprehensive machine learning framework to address a critical, often-overlooked challenge in climate science: the "aridity blindspot." Standard drought indicators, when used in unsupervised learning, frequently fail to distinguish between temporary water deficits (drought) and permanent water scarcity (aridity). This flaw is particularly pronounced in climatologically diverse regions like Central Africa, leading to inaccurate risk assessments. Our framework is composed of two primary engines: 1. The **Diagnostic Engine** resolves the aridity blindspot by integrating a novel Aridity Index into a K-Means clustering algorithm. Applied to a 90-year climate record for the ECCAS region, this engine successfully disentangles chronic dryness from anomalous drought, producing the first accurate spatio-temporal maps of evolving climate stress hotspots. 2. The **Prognostic Engine** complements this analysis by establishing a robust short-term drought forecasting system. It conducts a rigorous comparative study of state-of-the-art models (ANN, SVR, RGA-SVR, RF) to deliver high-fidelity SPI-6 predictions. Ultimately, this project delivers a transferable and validated framework that provides actionable insights for policymakers, improves climate risk assessment, and can be adapted for water resource management in other data-sparse, vulnerable regions worldwide. --- ## Repository Structure The project is organized into a modular structure to ensure clarity and ease of maintenance. ``` . ├── Assets/ ├── Dataset/ ├── models_output/ ├── climate_notebook.ipynb ├── config.py ├── drought_analysis.py ├── forecasting_manager.py ├── main.py └── requirements.txt ``` - `Assets/`: Contains static assets like images and utility files used in the project. - `Dataset/`: Houses the raw climate datasets (e.g., GPCC, CRU). *Note: D …