Status : Verified
| Personal Name | Magat, Lenie G. |
|---|---|
| Resource Title | Spatiotemporal Analysis and Forecasting of Drought in Central Luzon, Philippines Using Satellite-Derived Best-Fit SPI, Bayesian Spatiotemporal Modeling, and LSTM Deep Learning |
| Date Issued | June 2026 |
| Abstract | Drought remains one of the most significant climate-related hazards in the Philippines, particularly in Central Luzon, where agriculture, water resources, and local livelihoods are highly sensitive to rainfall variability. This study characterized the spatiotempo-ral behavior of meteorological drought in Central Luzon and evaluated forecasting ap-proaches for short- and long-range drought prediction using the Standardized Precipi-tation Index (SPI). Monthly Climate Hazards Group InfraRed Precipitation with Station Data (CHIRPS) rainfall data from 1981–2025 were quality-controlled and validated against Philippine Atmospheric, Geophysical and Astronomical Services Administration (PAGASA) sta-tion observations. Following validation, SPI was computed to assess drought variability, identify spatial drought hotspots, evaluate long-term trends, and examine spatial clus-tering patterns. Two forecasting frameworks were then developed for SPI-3 prediction: a Conditional Long Short-Term Memory (LSTM) model and a Bayesian spatiotempo-ral model incorporating Besag–York–Mollie´ 2 (BYM2) spatial random effects, AR(1) temporal dependence, climatic forcing, and seasonal predictors. Results revealed pronounced spatial heterogeneity in drought occurrence across Central Luzon, with persistent drought hotspots identified in portions of Nueva Ecija, Bulacan, Tarlac, and Zambales. These areas consistently exhibited higher drought fre-quency and greater vulnerability than the rest of the region. Overall, the findings indicate that both the Conditional LSTM and Bayesian spa-tiotemporal models are suitable for short-term drought forecasting, with the Conditional LSTM providing slightly higher predictive accuracy at shorter lead times. For medium-and long-range forecasting, the Bayesian framework exhibited more stable predictive v vi performance while offering the additional advantages of model interpretability and un-certainty quantification, making it well suited for drough |
| Degree Course | Master of Statistics |
| Language | English |
| Keyword | Bayesian spatiotemporal analysis, Bayesian spatiotemporal modeling |
| Material Type | Thesis/Dissertation |
Preliminary Pages
69.36 Mb
Category : F - Regular work, i.e., it has no patentable invention or creation, the author does not wish for personal publication, there is no confidential information.
Access Permission : Open Access