hydropython/Awash-ActualET-ML

This repository contains the end-to-end pipeline for spatial-temporal ET estimation within the Awash River Basin. By harmonizing 28 years of meteorological reanalysis data with high-resolution satellite ET products, this study develops a robust framework for water resource management in a critical Ethiopian catchment. # Awash-ActualET-ML: Multi-Scalar Hydrological Memory Architecture for ETa Prediction **A deep learning framework for predicting Actual Evapotranspiration (ETa) in water-limited, highly seasonal basins using multi-scalar hydrological memory.** ## Overview This repository implements a novel **Multi-Scalar Memory Neural Network** architecture designed to overcome the fundamental limitations of standard deep learning models in predicting Actual Evapotranspiration (ETa) across highly seasonal, water-limited basins. The framework was developed and validated for the **Awash Basin, Ethiopia** (2019-2022). ### Key Innovation Traditional CNNs fail to capture complex temporal dependencies in hydrological systems. Our architecture explicitly encodes **short-term, mid-term, and long-term antecedent hydrological states**, successfully bridging the gap between data-driven pattern recognition and physical hydrological dynamics. ### Performance Highlights - **NSE**: 0.134 (Baseline) → **0.768** (Full Memory) - **473% improvement** - **MAE**: 1.486 → **0.637 mm/day** - **57% reduction in error** - **Robust skill** across all Ethiopian seasons (NSE > 0.70) - **Temporal stability**: Rolling correlation r > 0.8 throughout multi-year dry spells ## Features - **Multi-Scalar Memory Architecture**: Short, mid, and long-term precipitation memory channels - **Spatial-Temporal Analysis**: Comprehensive seasonal and spatial performance evaluation - **Automated Preprocessing**: Data alignment, normalization, and feature engineering - **Model Ablation Study**: Systematic evaluation of each memory component - **Publication-Ready Visualizations**: High-quality figures for scientific publications - **Seasonal Dynamics**: Specialized analysis for Kiremit (wet), Tseday/Belg (transition), and Bega (dry) seasons ## 📁 Repository Structure ``` ├── .venv/ # Virtual environment ├── archive/ # Archived results and predictions │ ├── results_B …