Synergizing Intuitive Physics and Big Data in Deep Learning: Can We Obtain Process Insights While Maintaining State‐Of‐The‐Art Hydrological Prediction Capability? DOI Creative Commons
Leilei He, Liangsheng Shi, Wenxiang Song

et al.

Water Resources Research, Journal Year: 2024, Volume and Issue: 60(12)

Published: Dec. 1, 2024

Abstract Artificial intelligence (AI) methods have created insurmountable performance in prediction tasks for geoscientific problems yet are unable to derive process insights and answer specific scientific questions. The geoscience community faces a dilemma of reconciling comprehension with high predictive accuracy. Here we introduce deep learning (DPL) approach empowering neural networks deduce intrinsic processes from observable data, wherein the intuitive physics geosystems is directly coupled within (DL) architecture as structural prior. We aim incorporate raw common concepts possible macroscopic guidance: on one hand, reduce interference DL's data adaptability. On other allow information flow model converge along paths toward target output, thus enabling potential gain limited supervision. Illustrating its application precipitation‐runoff modeling across USA, DPL yields an ensemble median Nash‐Sutcliffe efficiency 0.758 Kling‐Gupta 0.778 robust transferability, compared 0.762 0.751 state‐of‐the‐art DL model. good match between internal representations independent sets snow water equivalent evapotranspiration, superior capability catchment budget closures, demonstrates proficient mastery. study also highlights beneficial synergies large‐scale collaboration, promoting organic unity understanding performance. This work shows promising avenue big will benefit domains that remain concerned clarity era AI.

Language: Английский

How Interpretable Machine Learning Can Benefit Process Understanding in the Geosciences DOI Creative Commons
Shijie Jiang, Lily‐belle Sweet,

Georgios Blougouras

et al.

Earth s Future, Journal Year: 2024, Volume and Issue: 12(7)

Published: July 1, 2024

Abstract Interpretable Machine Learning (IML) has rapidly advanced in recent years, offering new opportunities to improve our understanding of the complex Earth system. IML goes beyond conventional machine learning by not only making predictions but also seeking elucidate reasoning behind those predictions. The combination predictive power and enhanced transparency makes a promising approach for uncovering relationships data that may be overlooked traditional analysis. Despite its potential, broader implications field have yet fully appreciated. Meanwhile, rapid proliferation IML, still early stages, been accompanied instances careless application. In response these challenges, this paper focuses on how can effectively appropriately aid geoscientists advancing process understanding—areas are often underexplored more technical discussions IML. Specifically, we identify pragmatic application scenarios typical geoscientific studies, such as quantifying specific contexts, generating hypotheses about potential mechanisms, evaluating process‐based models. Moreover, present general practical workflow using address research questions. particular, several critical common pitfalls use lead misleading conclusions, propose corresponding good practices. Our goal is facilitate broader, careful thoughtful integration into science research, positioning it valuable tool capable enhancing current

Language: Английский

Citations

31
Distributed Hydrological Modeling With Physics‐Encoded Deep Learning: A General Framework and Its Application in the Amazon DOI Creative Commons
Chao Wang, Shijie Jiang, Yi Zheng

et al.

Water Resources Research, Journal Year: 2024, Volume and Issue: 60(4)

Published: April 1, 2024

Abstract While deep learning (DL) models exhibit superior simulation accuracy over traditional distributed hydrological (DHMs), their main limitations lie in opacity and the absence of underlying physical mechanisms. The pursuit synergies between DL DHMs is an engaging research domain, yet a definitive roadmap remains elusive. In this study, novel framework that seamlessly integrates process‐based model encoded as neural network (NN), additional NN for mapping spatially physically meaningful parameters from watershed attributes, NN‐based replacement representing inadequately understood processes developed. Multi‐source observations are used training data, fully differentiable, enabling fast parameter tuning by backpropagation. A hybrid Amazon Basin (∼6 × 10 6 km 2 ) was established based on framework, HydroPy, global‐scale DHM, its backbone. Trained simultaneously with streamflow Gravity Recovery Climate Experiment satellite yielded median Nash‐Sutcliffe efficiencies 0.83 0.77 dynamic simulations total water storage, respectively, 41% 35% higher than those original HydroPy model. Replacing Penman‒Monteith formulation produces more plausible potential evapotranspiration (PET) estimates, unravels spatial pattern PET giant basin. parameterization interpreted to identify factors controlling variability key parameters. Overall, study lays out feasible technical modeling big data era.

Language: Английский

Citations

20
Compounding effects in flood drivers challenge estimates of extreme river floods DOI Creative Commons
Shijie Jiang, Larisa Tarasova, Guo Yu

et al.

Science Advances, Journal Year: 2024, Volume and Issue: 10(13)

Published: March 27, 2024

Estimating river flood risks under climate change is challenging, largely due to the interacting and combined influences of various flood-generating drivers. However, a more detailed quantitative analysis such compounding effects implications their interplay remains underexplored on large scale. Here, we use explainable machine learning disentangle between drivers quantify importance for different magnitudes across thousands catchments worldwide. Our findings demonstrate ubiquity in many floods. Their often increases with magnitude, but strength this increase varies basis catchment conditions. Traditional might underestimate extreme hazards where contribution strongly magnitude. Overall, our study highlights need carefully incorporate risk assessment improve estimates

Language: Английский

Citations

16
Research progresses and prospects of multi-sphere compound extremes from the Earth System perspective DOI
Zengchao Hao, Yang Chen

Science China Earth Sciences, Journal Year: 2024, Volume and Issue: 67(2), P. 343 - 374

Published: Jan. 4, 2024

Language: Английский

Citations

15
Forecasting of compound ocean-fluvial floods using machine learning DOI
Sogol Moradian,

Amir AghaKouchak,

Salem Gharbia

et al.

Journal of Environmental Management, Journal Year: 2024, Volume and Issue: 364, P. 121295 - 121295

Published: June 13, 2024

Language: Английский

Citations

14
Direct and lagged climate change effects intensified the 2022 European drought DOI Creative Commons
Emanuele Bevacqua, Oldřich Rakovec, Dominik L. Schumacher

et al.

Nature Geoscience, Journal Year: 2024, Volume and Issue: 17(11), P. 1100 - 1107

Published: Oct. 21, 2024

Abstract In 2022, Europe faced an extensive summer drought with severe socioeconomic consequences. Quantifying the influence of human-induced climate change on such extreme event can help prepare for future droughts. Here, by combining observations and model outputs hydrological land-surface simulations, we show that Central Southern experienced highest observed total water storage deficit since satellite began in 2002, probably representing most widespread soil moisture past six decades. While precipitation deficits primarily drove drought, global warming contributed to over 30% intensity its spatial extent via enhanced evaporation. We identify 14–41% contribution was mediated warming-driven drying occurred before year indicating importance considering lagged effects avoid underestimating associated risks. Human-induced had qualitatively similar extremely low river discharges. These results highlight droughts are already underway, long lasting, risk may escalate further future.

Language: Английский

Citations

13
Shifted dominant flood drivers of an alpine glacierized catchment in the Tianshan region revealed through interpretable deep learning DOI Creative Commons
Wenting Liang, Weili Duan, Yaning Chen

et al.

npj Climate and Atmospheric Science, Journal Year: 2025, Volume and Issue: 8(1)

Published: Jan. 25, 2025

Language: Английский

Citations

1
Changes in Mediterranean flood processes and seasonality DOI Creative Commons
Yves Tramblay, Patrick Arnaud, Guillaume Artigue

et al.

Hydrology and earth system sciences, Journal Year: 2023, Volume and Issue: 27(15), P. 2973 - 2987

Published: Aug. 11, 2023

Abstract. Floods are a major natural hazard in the Mediterranean region, causing deaths and extensive damages. Recent studies have shown that intense rainfall events becoming more extreme this region but, paradoxically, without leading to an increase severity of floods. Consequently, it is important understand how flood changing explain absence trends magnitude despite increased extremes. A database 98 stations southern France with average record 50 years daily river discharge data between 1959 2021 was considered, together high-resolution reanalysis product providing precipitation simulated soil moisture classification weather patterns associated over France. Flood events, corresponding occurrence 1 event per year (5317 total), were extracted classified into excess-rainfall, short-rainfall, long-rainfall types. Several characteristics been also analyzed: durations, base flow contribution floods, runoff coefficient, total maximum rainfall, antecedent moisture. The evolution through time these seasonality analyzed. Results indicated that, most basins, floods tend occur earlier during year, mean date being, on average, advanced by month 1959–1990 1991–2021. This seasonal shift could be attributed frequency southern-circulation types spring summer. An extreme-event has observed, decrease before events. majority excess saturated soils, but their relative proportion decreasing time, notably spring, concurrent short rain For basins there positive correlation coefficients remaining stable dryer soils producing less lower In context increasing aridity, relationship likely cause magnitudes observed change These changes quite homogeneous domain studied, suggesting they rather linked regional climate than catchment characteristics. study shows even trends, properties may need accounted for when analyzing long-term hazards.

Language: Английский

Citations

18
Applying Machine Learning Methods to Improve Rainfall–Runoff Modeling in Subtropical River Basins DOI Open Access

Haoyuan Yu,

Qichun Yang

Water, Journal Year: 2024, Volume and Issue: 16(15), P. 2199 - 2199

Published: Aug. 2, 2024

Machine learning models’ performance in simulating monthly rainfall–runoff subtropical regions has not been sufficiently investigated. In this study, we evaluate the of six widely used machine models, including Long Short-Term Memory Networks (LSTMs), Support Vector Machines (SVMs), Gaussian Process Regression (GPR), LASSO (LR), Extreme Gradient Boosting (XGB), and Light (LGBM), against a model (WAPABA model) streamflow across three sub-basins Pearl River Basin (PRB). The results indicate that LSTM generally demonstrates superior capability than other five models. Using previous month as an input variable improves all When compared with WAPABA model, better two sub-basins. For simulations wet seasons, shows slightly model. Overall, study confirms suitability methods modeling at scale basins proposes effective strategy for improving their performance.

Language: Английский

Citations

7
Model-based assessment of flood generation mechanisms over Poland: The roles of precipitation, snowmelt, and soil moisture excess DOI Creative Commons
Nelson Venegas‐Cordero,

Cyrine Cherrat,

Zbigniew W. Kundzewicz

et al.

The Science of The Total Environment, Journal Year: 2023, Volume and Issue: 891, P. 164626 - 164626

Published: June 5, 2023

Hydrometeorological variability, such as changes in extreme precipitation, snowmelt, or soil moisture excess, Poland can lead to fluvial flooding. In this study we employed the dataset covering components of water balance with a daily time step at sub-basin level over country for 1952-2020. The data set was derived from previously calibrated and validated Soil & Water Assessment Tool (SWAT) model 4000 sub-basins. We applied Mann Kendall test circular statistics-based approach on annual maximum floods various potential flood drivers estimate trend, seasonality, relative importance each driver. addition, two sub-periods (1952-1985 1986-2020) were considered examine mechanism recent decades. show that northeast decreasing, while south trend showed positive behavior. Moreover, snowmelt is primary driver flooding across country, followed by excess precipitation. latter seemed be dominant only small, mountain-dominated region south. gained mainly northern part, suggesting spatial pattern generation mechanisms also governed other features. found strong signal climate change large parts Poland, where losing second sub-period favor which explained temperature warming diminishing role snow processes.

Language: Английский

Citations

16