Improving seasonal hydrometeorological forecast skill in Thailand using artificial intelligence : implications for land surface model performance | |
| Author | Kandit Taedulyasatit |
| Call Number | AIT Thesis no.WM-26-01 |
| Subject(s) | Hydrological forecasting--Thailand Climatic changes--Data processing--Thailand Climatic changes--Technological innovations--Thailand Artificial intelligence |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Water Engineering and Management |
| Publisher | Asian Institute of Technology |
| Abstract | Seasonal climate forecasts play an important role in hydrology, agriculture, disaster management, and water resources planning. However, reliable seasonal forecasting products for Thailand remain limited. Although global seasonal forecast datasets are publicly available, raw outputs from the Climate Forecast System Version 2 (CFSv2) often contain systematic biases, coarse spatial resolution, and limited temporal frequency, restricting their direct use for local applications. To address these limitations, this study proposes an AI-based two-stage post-processing framework to improve the accuracy and spatiotemporal resolution of seasonal forecasts over Thailand. The framework consists of two stages. First, a modified U-Net model is applied for spatial bias correction and downscaling from 1° to 0.25°. Second, a ConvLSTM sequence-to-sequence model is used for temporal disaggregation from 6-hourly to 1-hourly resolution. The results demonstrate that the proposed framework substantially reduces forecast bias and improves both spatial and temporal consistency compared with raw CFSv2 and traditional statistical approaches. For example, for temperature, correlation increased from 0.57 in raw CFSv2 to 0.85 in Stage 2, while RMSE decreased from 3.33 to 1.74 °C. The enhanced meteorological forecasts were subsequently used to drive the Noah-MP land surface model to evaluate their influence on hydrological variables, including surface soil moisture (SM), root zone soil moisture (RZSM), evapotranspiration (ET), and terrestrial water storage (TWS). Hydrological simulations driven by enhanced forcing, particularly the Stage 2 1-hourly forcing, show consistently improved performance across most variables, demonstrating that improvements in meteorological forcing propagate meaningful benefits to land surface processes. The strongest correlation improvements are observed in SM, RZSM, and TWS, with increasing of approximately 20%, along with RMSE reductions of 10–20%. In contrast, ET shows the largest RMSE improvement, reduced by more than 53%, while correlation remains limited, increasing only slightly by 2.3%. This study provides the first comprehensive assessment in Thailand of AI-enhanced seasonal forecasts coupled with Noah-MP for hydrological applications. The proposed framework bridges meteorological forecasting and hydrological modeling through a transferable, scalable, and reproducible workflow, supporting more reliable hydrometeorological forecasting and extending applications beyond climate studies in Thailand. |
| Year | 2026 |
| Type | Thesis |
| School | Faculty of Civil and Environmental Engineering (2026) |
| Department | Other Field of Studies (No Department) |
| Academic Program/FoS | Water Engineering and Management (WEM) |
| Chairperson(s) | Natthachet Tangdamrongsub |
| Examination Committee(s) | Shrestha, Sangam;Shanmugam, Mohana Sundaram;Ekbordin Winijkul |
| Degree | Thesis (M. Eng.) - Asian Institute of Technology, 2026 |
 | |