Elsevier eBooks, Год журнала: 2023, Номер unknown, С. 483 - 500
Опубликована: Янв. 1, 2023
Язык: Английский
Nature Climate Change, Год журнала: 2024, Номер 14(4), С. 357 - 363
Опубликована: Март 12, 2024
Язык: Английский
Процитировано
25
Nature Sustainability, Год журнала: 2024, Номер 7(4), С. 413 - 422
Опубликована: Апрель 22, 2024
Язык: Английский
Процитировано
18
Nature Geoscience, Год журнала: 2024, Номер 17(8), С. 770 - 777
Опубликована: Июль 19, 2024
Язык: Английский
Процитировано
18
Nature Water, Год журнала: 2023, Номер 1(12), С. 1079 - 1090
Опубликована: Ноя. 27, 2023
Язык: Английский
Процитировано
27
Hydrology and earth system sciences, Год журнала: 2024, Номер 28(3), С. 525 - 543
Опубликована: Фев. 8, 2024
Abstract. The application of machine learning (ML) including deep models in hydrogeology to model and predict groundwater level monitoring wells has gained some traction recent years. Currently, the dominant class is so-called single-well model, where one trained for each well separately. However, developments neighbouring disciplines hydrology (rainfall–runoff modelling) have shown that global models, being able incorporate data several wells, may advantages. These are often called “entity-aware models“, as they usually rely on static differentiate entities, i.e. or catchments surface hydrology. We test two kinds information characterize a global, entity-aware set-up: first, environmental features continuously available thus theoretically enable spatial generalization (regionalization), second, time-series derived from past time series at respective well. Moreover, we random integer entity comparison. use published dataset 108 Germany, evaluate performance terms Nash–Sutcliffe efficiency (NSE) an in-sample out-of-sample setting, representing temporal generalization. Our results show work with mean NSE >0.8 comparable to, even outperforming, models. do not generalize spatially setting (mean <0.7, lower than without information). Strikingly, all variants, regardless type used, basically perform equally both in- out-of-sample. conclusion fact does awareness, but uses merely unique identifiers, raising research question how properly establish awareness Potential future avenues lie bigger datasets, relatively small number might be enough take full advantage Also, more needed find meaningful ML hydrogeology.
Язык: Английский
Процитировано
14
Geoscientific model development, Год журнала: 2024, Номер 17(1), С. 275 - 300
Опубликована: Янв. 12, 2024
Abstract. We discuss the various performance aspects of parallelizing our transient global-scale groundwater model at 30′′ resolution (30 arcsec; ∼ 1 km Equator) on large distributed memory parallel clusters. This model, referred to as GLOBGM, is successor 5′ (5 arcmin; 10 PCR-GLOBWB 2 (PCRaster Global Water Balance model) based MODFLOW having two layers. The current version GLOBGM (v1.0) used in this study also has layers, uncalibrated, and uses available data. Increasing from creates challenges, including increased runtime, usage, data storage that exceed capacity a single computer. show parallelization tackles these problems with relatively low hardware requirements meet needs users or modelers who do not have exclusive access hundreds thousands nodes within supercomputer. For simulation, we use unstructured grids prototype 6 parallelized using message-passing interface. construct independent total 278 million active cells cancel all redundant sea land cells, while satisfying necessary boundary conditions, distribute them over three continental-scale models (168 – Afro–Eurasia; 77 Americas; 16 Australia) one remaining for smaller islands (17 million). Each four partitioned into multiple non-overlapping submodels are tightly coupled linear solver, where each submodel uniquely assigned processor core, associated written during pre-processing, tiles. balancing workload advance, apply widely METIS graph partitioner ways: it straightforwardly applied (lateral) grid an area-based manner HydroBASINS catchments pre-sorting future coupling surface water. consider experiment simulating years 1958–2015 daily time steps monthly input, 20-year spin-up, Dutch national supercomputer Snellius. Given serial simulation would require 4.5 months set hypothetical target maximum h runtime. 12 (32 cores per node; 384 total) sufficient achieve target, resulting speedup 138 largest Afro–Eurasia when 7 (224 cores) parallel. A limited evaluation output United States Geological Survey (USGS) National Information System (NWIS) head observations contiguous was conducted. showed increasing results significant improvement steady-state compared model. However, quite similar, there much room improvement. Monthly multi-year terrestrial water anomalies derived models, however, favorably GRACE satellite. next versions further improvements more detailed (hydro)geological schematization better information locations, depths, pumping rates abstraction wells.
Язык: Английский
Процитировано
12
Earth-Science Reviews, Год журнала: 2024, Номер 252, С. 104739 - 104739
Опубликована: Март 8, 2024
The ability to characterize hydrologically relevant differences between places is at the core of our science. A common way quantitatively hydrological catchments through use descriptors that summarize physical aspects system, typically by aggregating heterogeneous geospatial information into a single number. Such capture various facets catchment functioning and structure, identify similarity or dissimilarity among catchments, transfer unobserved locations. However, so far there no agreement on how should be selected, aggregated, evaluated. Even worse, little known about existence potential biases in current practices catchments. In this systematic review, we analyze 742 research articles published 1967 2021 provide categorized overview historical characterization (i.e., data sources, aggregation evaluation methods) science related disciplines. We uncover substantial characterization: (1) only 16% analyzed studies are dry environments, even though such environments cover 42% global land surface, suggesting most tailored represent energy-limited potentially less effective water-limited environments; (2) 30% subsurface features for despite dominance flow; (3) 4% 9% aggregated spatially- vertically-differentiated way, respectively, while majority simple averages do not account hydrologically-relevant variabilities within catchments; (4) 25% all evaluate usefulness descriptors, none quantifies their uncertainty. demonstrate effects these effectively functional behavior with illustrative examples. Finally, suggest possible ways derive more robust, comprehensive meaningful descriptors.
Язык: Английский
Процитировано
11
Nature Water, Год журнала: 2025, Номер unknown
Опубликована: Янв. 6, 2025
Язык: Английский
Процитировано
1
Hydrology, Год журнала: 2025, Номер 12(1), С. 11 - 11
Опубликована: Янв. 9, 2025
Groundwater modeling in data-scarce regions faces significant challenges due to the lack of comprehensive, high-quality data, impacting model accuracy. This systematic review Scopus-indexed papers identifies various approaches address these challenges, including coupled hydrological-groundwater models, machine learning techniques, distributed hydrological water balance 3D groundwater flow modeling, geostatistical remote sensing-based approaches, isotope-based methods, global downscaling, and integrated approaches. Each methodology offers unique advantages for assessment management data-poor environments, often combining multiple data sources techniques overcome limitations. However, face common related quality, scale transferability, representation complex hydrogeological processes. emphasizes importance adapting methodologies specific regional contexts availability. It underscores value provide robust estimates sustainable management. The choice method ultimately depends on objectives, study, available region interest. Future research should focus improving integration diverse sources, enhancing processes simplified developing uncertainty quantification methods tailored conditions.
Язык: Английский
Процитировано
1
Geoscientific model development, Год журнала: 2025, Номер 18(1), С. 71 - 100
Опубликована: Янв. 13, 2025
Abstract. Three-dimensional geological modelling algorithms can generate multiple models that fit various mathematical and geometrical constraints. The results, however, are often meaningless to experts if the do not respect accepted principles. This is problematic as use of expected for downstream purposes, such hazard risk assessment, flow characterization, reservoir estimation, storage, or mineral energy exploration. Verification reasonableness therefore important. If implausible be identified eliminated, it will save countless hours computational human resources. To begin assessing reasonableness, we develop a framework checking model consistency with knowledge test proof-of-concept tool. consists space consistent inconsistent situations hold between pair objects, tool assesses model's relations against identify (in)consistent situations. successfully applied several case studies first promising step toward automated assessment reasonableness.
Язык: Английский
Процитировано
1