LSTM-Based Bias Correction for Regional Precipitation Projections from BNU-ESM in North Maluku, Indonesia

Abstract

Extreme precipitation events have become more frequent in recent decades due to climate change, posing serious challenges for vulnerable regions such as North Maluku. This study aims to improve the accuracy of daily precipitation projections using a bias correction approach based on a Long Short-Term Memory (LSTM) model. Historical data from the BNU-ESM global climate model and ERA5 reanalysis (1980–2005) were used for model training, with three LSTM architectures tested. The best performance was achieved by a three-layer LSTM model with 128, 64, and 32 neurons, trained over 50 epochs with a learning rate of 0.001. Evaluation using MAE and MSE yielded the lowest errors at 0.006 and 0.084, respectively. The corrected projections aligned more closely with observations and offer a more reliable foundation for climate adaptation planning in North Maluku, particularly in sectors such as agriculture, fisheries, and water resource management.