Predicting Thermal Structure in Large Reservoirs Under Climate Change Using Machine Learning

Comparative Evaluation of Physics-Informed and Data-Driven Neural Networks for Reservoir Water Balance Simulation During Hydrological Extremes

New classes of climate model emulators to improve paleoclimate reconstructions

Abstract. Reconstructing spatial climate variability from proxy records requires forward models “emulators” that capture the dynamical structure of the climate system while remaining computationally efficient. Traditional emulators based on Empirical Orthogonal Functions (EOFs) and Linear Inverse Models (LIMs) face inherent limitations due to linearity and variance-based dimensionality reduction. Here we develop and evaluate a hierarchy of CMIP-class climate model emulators, for annual surface air temperature field emulation, that integrate autoencoder-based dimensionality reduction with nonlinear prediction architectures, including Reservoir Computing (RC) and Recurrent Neural Networks (RNNs). Using a comprehensive experimental protocol applied to the IPSL-CM6A-LR model and 52 CMIP6 models, we show that the combination of a prediction-oriented autoencoder (AE) latent representation with RC dynamics retaining memory and nonlinear state evolution (AERCn) yields, the most robust configuration when training data are plentiful. This improves the representation of El Niño Southern Oscillation and Atlantic Multidecadal Variability, while preserving spatial reconstruction quality and robustness across distinct CMIP6 model structures. When training data are scarce, a multimodel pre-trained AERNN provides a data-efficient and competitive alternative. These properties make the proposed architectures particularly well suited for integration into Particle Filters and Ensemble Kalman Filter PDA frameworks. Our results highlight the importance of predictability-oriented dimensionality reduction and nonlinear dynamical memory for emulator design. They provide a scalable proof of concept toward multivariate climate-field emulation for improved reconstructions of climate variability over the Common Era.

Environmental Modeling and Assessment Using CFD and Data-Driven Tools

Modeling the temporal and spatial dynamics of heat pollution within environmental systems is essential to accurately predict thermal pollution within the environment. Therefore, accurate predictions of thermal pollution dynamics require an understanding of the multiscale, nonlinear interactions that establish the mechanisms of thermal transport, dispersion, and ecological response. In this paper, we provide an integrated set of methodologies that applies both three-dimensional computational fluid dynamics (CFD) and cutting-edge machine learning (ML) and deep learning (DL) methodologies for the environmental thermal assessment. We assess the performance of six types of predictive ML models-linear regression, random forest, gradient boosting, multilayer neural networks (MNNs), support vector regression (SVRs), and long short-term memory (LSTM) networks-against benchmark datasets for predicting thermal plumes, predicting cooling tower dynamics, and forecasting river temperatures using CFD. Among these models, the LSTM networks performed by far the best for predicting thermal activities over time (R² = 0.95, RMSE = 1.5°C). Conversely, the best performing model for identifying spatial thermal patterns was gradient boosting.

Physical and data hybrid-driven modeling for the evolution state prediction of reservoir colluvial landslides with step-like movement patterns: a case study in the Xiluodu Reservoir area, China

Deep Learning Based Climate Forecasting Model Using LSTM (Long Short-Term Memory) of Kathmandu Valley

Deep learning’s significant potential as a scalable and dependable method for regional climate forecasting, giving it an advantage in the Kathmandu Valley’s climate change adaptation policy-making process is supported.

A Driver‐Aware Machine Learning Approach for Predicting the Severity of Decadal‐Scale Extreme Climate Events

The prediction of the decadal‐scale severities of extreme wet and dry events requires long‐term records of data to capture long‐term fluctuations. Recently, purely data‐driven machine learning models have shown excellent performance in modelling precipitation and other water cycle variables when sufficiently long and comprehensive training datasets are available, but their predictive reliability may be limited when long training records are unavailable. To better understand and predict extreme climate event severities at the basin scale, we propose an integrated framework coupling driver analysis and machine learning methods. Combined with the driver analysis method, the long short‐term memory (LSTM) model and the support vector regression (SVR) model are jointly used to predict extreme event severities in the Yellow River Basin (YRB) in the next decade. The results showed that extreme dry events are mainly driven by the long‐term signals of global warming, the Atlantic Multidecadal Oscillation (AMO), and the Pacific Decadal Oscillation (PDO) in the YRB. Extreme wet events are dominated by the long‐term signals of the Southern Oscillation Index (SOI), global warming, and the PDO. In 2030, the spatial distribution patterns of both extreme dry event severities and extreme wet event severities in the YRB will potentially be more scattered. Compared to 2020, there will be at most 30% increases and at most 20% decreases in extreme dry event severity in 2030 across the YRB, with a maximum increase in extreme wet event severity of 60% and a maximum decrease of approximately 20% in 2030. The proposed framework provides a way to incorporate information on large‐scale climate drivers into data‐driven prediction models.