Performance of Time Series and Deep Learning Models for Predicting Severe Drought Areas in Lamongan Regency

Abstract: Drought is one of the most significant natural disasters, greatly impacting the agricultural sector, water availability, and ecosystem balance. Therefore, predicting the extent of drought is crucial for effective mitigation and adaptation to climate change. This research aims to predict drought-affected areas using time-series data with the Weighted Moving Average (WMA) method and Landsat 8 imagery analyzed through deep learning. The data used in this research consist of satellite images, which are processed to assess drought extent using the Normalized Difference Drought Index (NDDI) within the Quantum Geographic Information System (QGIS) application. The satellite images were obtained from EarthExplorer. This research employs satellite image processing to extract NDDI values and spatial analysis to determine the distribution of drought in Lamongan District over a five-year period (2019-2023). The results indicate that the deep learning model, utilizing sub-district area images as input, provides the most accurate predictions of drought extent, achieving an average Mean Absolute Error (MAE) of 1,491,123. This research contributes to the development of spatial data-based techniques for mitigating the effects of drought more efficiently and sustainably. Keywords: : drought area, time series, deep learning, NDDI, severe drought.