Stochastic weather generators are commonly employed to create synthetic sequences of daily weather variables across diverse fields, including hydrological, ecological, and agricultural studies. Realistic precipitation sequences, in particular, serve as essential inputs in numerous modeling frameworks. Generalized linear models (GLMs) that incorporate covariates to capture seasonality and teleconnections represent one effective approach for stochastic weather generation. However, these models often underestimate the interannual variability of seasonally aggregated variables, notably precipitation intensity during wet seasons. Recent methods developed to mitigate the issue of overdispersion have nevertheless struggled to adequately replicate observed precipitation intensities in wet seasons. To overcome this limitation, we propose integrating a modified hybrid gamma and generalized Pareto distribution into the GLM-based weather generator. This enhanced method was evaluated using daily precipitation data from Seoul, Korea, and successfully reproduced realistic precipitation intensities while effectively addressing the overdispersion issue.

Climate change has prolonged droughts and intensified heatwaves worldwide, increasing the frequency, intensity, and spatial extent of forest fires. In South Korea, an increasing incidence of large forest fires has created a need for rapid, accurate forest fire-damaged area boundary delineation to support post-fire recovery and restoration planning. This is particularly important in South Korea, where the forest-fire season is followed by the summer rainy season, increasing the risk of post-fire soil erosion and slope instability and thereby increasing the need for rapid stabilization and restoration measures. However, previous studies have reported varying delineation performance depending on the input configuration, suggesting the need to examine how different spectral inputs and their integration affect boundary delineation. To address this limitation, we propose a hierarchical ensemble U-Net framework for forest fire-damaged area boundary delineation using Sentinel-2 imagery. To design the input data structure, differences between burned and unburned areas were examined using the Blue, Green, Red, RedEdge, and NIR bands and the NDVI and dNDVI indices, and four three-channel combinations (Blue–Green–Red, Blue–Green–RedEdge, Green–Red–NIR, and Green–RedEdge–NIR) and two single indices (NDVI and dNDVI) were chosen as inputs. Three U-Net models were then evaluated: a single-channel dNDVI model, a three-channel NIR-RE-G model, and a hierarchical ensemble model, which integrated the outputs of six independently trained U-Nets. Among the three models, the single-channel dNDVI model showed higher overall accuracy (84.1%), producer’s accuracy (89.0%), and F1-score (0.80) than the three-channel NIR-RE-G model (82.9%, 70.7%, and 0.77, respectively), whereas the NIR-RE-G model achieved higher user accuracy (91.8%); the hierarchical ensemble model yielded the best overall performance, with the highest overall accuracy (85.7%), producer’s accuracy (92.2%), and F1-score (0.83). This suggests that integrating complementary spectral responses can mitigate the limitations of single-input models and support more robust boundary delineation across the six tested wildfire cases in South Korea.