This study aims to identify the optimal distribution for modeling extreme events based on annual maximum series data from meteorological stations in Southern Thailand. We explored three types of two-parameter Generalized Extreme Value (GEV) distributions—Weibull, Fréchet, and Gumbel—and compared their fit to the observed data using GEV. The fitting process involved estimating model parameters, for which we employed two mathematical techniques: least square estimation and maximum likelihood estimation. Our findings revealed that, when fitting GEV distribution with numerical estimation for the shape parameter, the Gumbel distribution, characterized by a light-tail, is the most suitable for nearly all stations. However, when considering the two-parameter case for the three distribution types, the Fréchet distribution, known for its heavy-tail, emerges as the best fit for many stations, exhibiting the lowest ratio mean square error.

Drought indices are pivotal for comprehending and monitoring water scarcity in Northeast Thailand. Researchers precisely assess these indices, considering local climate conditions, geographical features, impacts on vegetation and agriculture, hydrological considerations, and the temporal and spatial scales of drought events. The reliability of these assessments depends on thorough validation against ground data, encompassing rainfall records and soil moisture measurements. This study explores the integration of various indices to enhance the overall comprehensiveness of drought assessments in the Mun watershed. It contributes to the field by evaluating drought conditions across the entire watershed, utilizing meteorological, soil moisture, and hydrological drought indicators. These indicators encompass the standardized runoff index, standardized precipitation index, standardized soil moisture index, and standardized precipitation evapotranspiration index. In order to establish a comprehensive multivariate drought index, this study employs ensemble learning, incorporating boosting techniques such as boosting, AdaBoost (adaptive boosting), and XGBoost (extreme gradient boosting). The performance of each model is assessed through comparisons using RMSE, MAE, and \(R^{2}\).

Floods are one of the most frequent causes of damage to agricultural products. In this research, we analyzed employing bivariate approaches on weekly moving cumulative rainfall (WMCR) data from 1990 to 2023 in Khong Watershed Northeastern, Thailand. The analysis was based on nine extreme value copulas (EVC) derived from the bivariate generalized Pareto distribution (BGPD) model, and using a mean residual life (MRL) plot for choosing a threshold for each. The assessment of the spatial relationship between two areas was evaluated through the Kendall tau correlation coefficient. The research utilized the Likelihood Ratio test (LRT) to evaluate the performance of joint probabilities and XV-CIC to identify the best-fitting EVC models. These selected models were subsequently used to calculate return periods for flood risk in the areas. The result revealed that symmetric models were effective in modeling dependence functions, and the highest flood risk probabilities were in the Loie, Nongkai, and Udon Thani Provinces. Therefore, bivariate risk analysis provides valuable decision-making support for the region’s design and management strategies for flood defenses.

Thailand continues to grapple with persistent air pollution, particularly in urban and industrial areas, despite overall improvements in the past decade. The ongoing challenge of elevated \(\textrm{PM}_{2.5}\) levels in the northern and northeastern regions is a cause for concern. Key contributors to this issue include transportation, industrial activities, and power generation, with a heightened focus due to the exacerbating effects of climate change. \(\textrm{PM}_{2.5}\) pollution emerges as a significant environmental and health issue, especially during dry periods. In our research, we employ a combination of non-stationary analysis using the Peak–Over–Threshold method, the Generalized Pareto distribution, and an Artificial Neural Network (ANN) to examine extreme \(\textrm{PM}_{2.5}\) pollution events in Khon Kaen Province. We utilize data spanning from 2012 to 2022, sourced from Thailand’s Pollution Control Department, encompassing various air quality parameters such as \(\textrm{PM}_{2.5}\), PM\({}_{10}\), CO, NO, NO\({}_{X}\), NO\({}_{2}\), SO\({}_{2}\), O\({}_{3}\), wind characteristics, temperature, and precipitation. We employ an ANN based on Maximum Likelihood Estimation to determine model parameters, with model performance assessed using Nash–Sutcliffe Efficiency and Root Mean Squared Error. Our research aims to forecast \(\textrm{PM}_{2.5}\) concentrations for different return intervals (2, 5, 20, 50, and 100 years), offering insights into potential \(\textrm{PM}_{2.5}\) levels in Khon Kaen. Additionally, it sheds light on the intricate relationship between air quality and climatic patterns, providing valuable information for addressing this persistent environmental issue.

This study proposes the improvement of an adaptive parameter estimation approach for the GEVD using multi-output machine learning for the non-stationary models and comparing it to maximum likelihood estimation for the stationary models. This method effectively estimates GEVD parameters, improving extreme value analysis. In order to forecast the return level of extreme rainfall in Northeast Thailand, which is a risk of flooding due to the huge amount of rainfall, the initial step is to identify the key variables that are used to estimate the three GEVD parameters: location, scale, and shape parameters (\(\mu\), \(\sigma\), and \(\xi\)). All features can be accomplished by estimating the correlation coefficients and using them to calculate the parameters. The information was gathered from meteorological and satellite data in Northeast Thailand between 2012 and 2023. It includes various variables such as rainfall, climate, Normalized difference vegetation index (NDVI), and runoff. The data was compiled from the Meteorological Department of Thailand and 322 meteorological stations. The evaluation performance and accuracy of the model are compared. Finally, two-dimensional maps depicting return levels for various return periods (2, 5, 10, 20, 50, and 100 years) are being made available for future use. This study enhances parameter estimation for flood risk mitigation and water resource management.