Dataset - Satellite-derived Composite Drought Index Modelling Framework and Forecasting using Deep Learning

This dataset provides a high-resolution (1 km) spatio-temporal Composite Drought Index (CDI) for Tunisia covering the period 2000–2025. It was developed within a reproducible environmental modelling framework integrating multi-source Earth Observation and reanalysis data, including CHIRPS, ERA5-Land, and MODIS products.

The CDI combines five standardized drought-related indicators: Standardized Precipitation Index (SPI-1), Standardized Precipitation-Evapotranspiration Index (SPEI-1), Soil Moisture Anomaly (SMA), Normalized Difference Vegetation Index anomaly (NDVI-A), and Land Surface Temperature anomaly (LST-A). These indicators are integrated through a hierarchical cause–effect framework representing drought propagation from meteorological conditions to ecological impacts. 

The dataset enables multi-level drought classification (Normal, Watch, Warning, Alert-1, Alert-2, Urgency) and supports both monitoring and forecasting applications. It is designed for reproducibility, extensibility, and transferability to other regions.

The dataset was used to evaluate 12 deep learning models for drought forecasting, including Transformer-based and recurrent architectures, with TimeFormer achieving the best predictive performance.

All data processing was conducted using Python and Google Earth Engine, including preprocessing, harmonization, standardization, and CDI construction.

This dataset supports applications in drought monitoring, climate change analysis, environmental modelling, and machine learning-based forecasting.