Long-term streamflow forecasting in data-scarce regions: Insightful investigation for leveraging satellite-derived data, Informer architecture, and concurrent fine-tuning transfer learning

Journal of Hydrology, Journal Year: 2024, Volume and Issue: 631, P. 130772 - 130772

Published: Feb. 2, 2024

Language: Английский

---

Probing the limit of hydrologic predictability with the Transformer network

Journal of Hydrology, Journal Year: 2024, Volume and Issue: 637, P. 131389 - 131389

Published: May 19, 2024

Language: Английский

---

Advancing Hydrology through Machine Learning: Insights, Challenges, and Future Directions Using the CAMELS, Caravan, GRDC, CHIRPS, PERSIANN, NLDAS, GLDAS, and GRACE Datasets

Water, Journal Year: 2024, Volume and Issue: 16(13), P. 1904 - 1904

Published: July 3, 2024

Machine learning (ML) applications in hydrology are revolutionizing our understanding and prediction of hydrological processes, driven by advancements artificial intelligence the availability large, high-quality datasets. This review explores current state ML hydrology, emphasizing utilization extensive datasets such as CAMELS, Caravan, GRDC, CHIRPS, NLDAS, GLDAS, PERSIANN, GRACE. These provide critical data for modeling various parameters, including streamflow, precipitation, groundwater levels, flood frequency, particularly data-scarce regions. We discuss type methods used significant successes achieved through those models, highlighting their enhanced predictive accuracy integration diverse sources. The also addresses challenges inherent applications, heterogeneity, spatial temporal inconsistencies, issues regarding downscaling LSH, need incorporating human activities. In addition to discussing limitations, this article highlights benefits utilizing high-resolution compared traditional ones. Additionally, we examine emerging trends future directions, real-time quantification uncertainties improve model reliability. place a strong emphasis on citizen science IoT collection hydrology. By synthesizing latest research, paper aims guide efforts leveraging large techniques advance enhance water resource management practices.

Language: Английский

---

Prediction of reference crop evapotranspiration based on improved convolutional neural network (CNN) and long short-term memory network (LSTM) models in Northeast China

Journal of Hydrology, Journal Year: 2024, Volume and Issue: unknown, P. 132223 - 132223

Published: Oct. 1, 2024

Language: Английский

---

Multi-Step Ahead Water Level Forecasting Using Deep Neural Networks

Water, Journal Year: 2024, Volume and Issue: 16(21), P. 3153 - 3153

Published: Nov. 4, 2024

Stream gauge height (water level) is a significant indicator for forecasting future floods. Flooding occurs when the water level exceeds flood stage. Predicting imminent floods can save lives, protect infrastructure, and improve road traffic management transportation. Deep neural networks have been increasingly used in this domain due to their predictive capabilities capturing complex features interdependencies. This study employs four distinct models—Multi-Layer Perceptron (MLP), Long Short-Term Memory (LSTM), transformer, LSTNet—with MLP serving as baseline model forecast levels. The models are trained using data from 20 river gages across state of Missouri ensure consistent performance. Random search optimization employed hyperparameter tuning. prediction intervals set at 4, 6, 8, 10 (each interval equivalent 30 min) that performance results robust not random weight initialization or suboptimal hyperparameters throughout different intervals. findings indicate LSTNet leads better than other models, with median RMSE 0.00724, 0.00959, 0.01204, 0.01230 intervals, respectively. As climate change localized storms driven by atmospheric shifts, fluctuations becoming extreme, further exacerbating drift real-world datasets. demonstrates superior terms RMSE, MAE, correlation coefficient all levels under conditions.

Language: Английский

---

Transformer-based deep learning models for predicting permeability of porous media

Advances in Water Resources, Journal Year: 2023, Volume and Issue: 179, P. 104520 - 104520

Published: Aug. 12, 2023

Language: Английский

---

Improving flood forecasting using time-distributed CNN-LSTM model: a time-distributed spatiotemporal method

Earth Science Informatics, Journal Year: 2024, Volume and Issue: unknown

Published: June 10, 2024

Language: Английский

---

Multi-Step Ahead Probabilistic Forecasting of Daily Streamflow Using Bayesian Deep Learning: A Multiple Case Study

Water, Journal Year: 2022, Volume and Issue: 14(22), P. 3672 - 3672

Published: Nov. 14, 2022

In recent decades, natural calamities such as drought and flood have caused widespread economic social damage. Climate change rapid urbanization contribute to the occurrence of disasters. addition, their destructive impact has been altered, posing significant challenges efficiency, equity, sustainability water resources allocation management. Uncertainty estimation in hydrology is essential for By quantifying associated uncertainty reliable hydrological forecasting, an efficient management plan obtained. Moreover, forecasting provides future information assist risk assessment. Currently, majority forecasts utilize deterministic approaches. Nevertheless, models cannot account intrinsic forecasted values. Using Bayesian deep learning approach, this study developed a probabilistic model that covers pertinent subproblem univariate time series multi-step ahead daily streamflow quantify epistemic aleatory uncertainty. The new implements sampling Long short-term memory (LSTM) neural network by using variational inference approximate posterior distribution. proposed method verified with three case studies USA horizons. LSTM point models, LSTM-BNN, BNN, Monte Carlo (MC) dropout (LSTM-MC), were applied comparison model. results show long (BLSTM) outperforms other terms reliability, sharpness, overall performance. reveal all outperformed lower RMSE value. Furthermore, BLSTM can handle data higher variation peak, particularly long-term compared models.

Language: Английский

---

Deep learning for spatiotemporal forecasting in Earth system science: a review

International Journal of Digital Earth, Journal Year: 2024, Volume and Issue: 17(1)

Published: Aug. 19, 2024

Deep learning (DL) has demonstrated strong potential in addressing key challenges spatiotemporal forecasting across various Earth system science (ESS) domains. This review examines 69 studies applying DL to tasks within climate modeling and weather prediction, disaster management, air quality modeling, hydrological renewable energy forecasting, oceanography, environmental monitoring. We summarize commonly used architectures for ESS, technical innovations, the latest advancements predictive applications. While have proven capable of handling data, remain tackling complexities specific such as complex scale dependencies, model interpretability, integration physical knowledge. Recent innovations demonstrate growing efforts integrate knowledge, improve explainability, adapt domain-specific needs, quantify uncertainties. Finally, this highlights future directions, including (1) developing more interpretable hybrid models that synergize traditional approaches, (2) extending generalizability through techniques like domain adaptation transfer learning, (3) advancing methods uncertainty quantification missing data handling.

Language: Английский

---

Coupling SWAT and Transformer Models for Enhanced Monthly Streamflow Prediction

Sustainability, Journal Year: 2024, Volume and Issue: 16(19), P. 8699 - 8699

Published: Oct. 9, 2024

The establishment of an accurate and reliable predictive model is essential for water resources planning management. Standalone models, such as physics-based hydrological models or data-driven have their specific applications, strengths, limitations. In this study, a hybrid (namely SWAT-Transformer) was developed by coupling the Soil Water Assessment Tool (SWAT) with Transformer to enhance monthly streamflow prediction accuracy. SWAT first constructed calibrated, then its outputs are used part inputs Transformer. By correcting errors using Transformer, two effectively coupled. Monthly runoff data at Yan’an Ganguyi stations on Yan River, first-order tributary Yellow River Basin, were evaluate proposed model’s performance. results indicated that performed well in predicting high flows but poorly low flows. contrast, able capture low-flow period information more accurately outperformed overall. SWAT-Transformer could correct predictions overcome limitations single model. integrating SWAT’s detailed physical process portrayal Transformer’s powerful time-series analysis, coupled significantly improved offer optimal resource management, which crucial sustainable economic societal development.

Language: Английский

---