A review of the application of machine learning in water quality evaluation

Eco-Environment & Health, Journal Year: 2022, Volume and Issue: 1(2), P. 107 - 116

Published: June 1, 2022

With the rapid increase in volume of data on aquatic environment, machine learning has become an important tool for analysis, classification, and prediction. Unlike traditional models used water-related research, data-driven based can efficiently solve more complex nonlinear problems. In water environment conclusions derived from have been applied to construction, monitoring, simulation, evaluation, optimization various treatment management systems. Additionally, provide solutions pollution control, quality improvement, watershed ecosystem security management. this review, we describe cases which algorithms evaluate different environments, such as surface water, groundwater, drinking sewage, seawater. Furthermore, propose possible future applications approaches environments.

Language: Английский

---

From calibration to parameter learning: Harnessing the scaling effects of big data in geoscientific modeling

Nature Communications, Journal Year: 2021, Volume and Issue: 12(1)

Published: Oct. 13, 2021

The behaviors and skills of models in many geosciences (e.g., hydrology ecosystem sciences) strongly depend on spatially-varying parameters that need calibration. A well-calibrated model can reasonably propagate information from observations to unobserved variables via physics, but traditional calibration is highly inefficient results non-unique solutions. Here we propose a novel differentiable parameter learning (dPL) framework efficiently learns global mapping between inputs (and optionally responses) parameters. Crucially, dPL exhibits beneficial scaling curves not previously demonstrated geoscientists: as training data increases, achieves better performance, more physical coherence, generalizability (across space uncalibrated variables), all with orders-of-magnitude lower computational cost. We demonstrate examples learned soil moisture streamflow, where drastically outperformed existing evolutionary regionalization methods, or required only ~12.5% the achieve similar performance. generic scheme promotes integration deep process-based models, without mandating reimplementation.

Language: Английский

---

Machine learning in natural and engineered water systems

Water Research, Journal Year: 2021, Volume and Issue: 205, P. 117666 - 117666

Published: Sept. 14, 2021

Language: Английский

---

Transferring Hydrologic Data Across Continents – Leveraging Data‐Rich Regions to Improve Hydrologic Prediction in Data‐Sparse Regions

Water Resources Research, Journal Year: 2021, Volume and Issue: 57(5)

Published: March 26, 2021

Abstract There is a drastic geographic imbalance in available global streamflow gauge and catchment property data, with additional large variations data characteristics. As result, models calibrated one region cannot normally be migrated to another without significant modifications. Currently these regions, non‐transferable machine learning are habitually trained over small local sets. Here we show that transfer (TL), the senses of weight initialization freezing, allows long short‐term memory (LSTM) were pretrained conterminous United States (CONUS, source set) transferred catchments on other continents (the target regions), need for extensive attributes at location. We demonstrate this possibility regions where dense (664 basins Great Britain), moderately (49 central Chile), scarce only remotely sensed (5 China). In both China Chile, TL showed significantly elevated performance compared locally using all basins. The benefits increased amount set, seemed more pronounced greater physiographic diversity. from than pretraining LSTM outputs an uncalibrated hydrologic model. These results suggest around world have commonalities which could leveraged by deep learning, synergies can had simple modification current workflows, greatly expanding reach existing big data. Finally, work diversified benchmarks.

Language: Английский

---

Differentiable modelling to unify machine learning and physical models for geosciences

Nature Reviews Earth & Environment, Journal Year: 2023, Volume and Issue: 4(8), P. 552 - 567

Published: July 11, 2023

Language: Английский

---

Differentiable, Learnable, Regionalized Process‐Based Models With Multiphysical Outputs can Approach State‐Of‐The‐Art Hydrologic Prediction Accuracy

Water Resources Research, Journal Year: 2022, Volume and Issue: 58(10)

Published: Sept. 19, 2022

Abstract Predictions of hydrologic variables across the entire water cycle have significant value for resources management as well downstream applications such ecosystem and quality modeling. Recently, purely data‐driven deep learning models like long short‐term memory (LSTM) showed seemingly insurmountable performance in modeling rainfall runoff other geoscientific variables, yet they cannot predict untrained physical remain challenging to interpret. Here, we show that differentiable, learnable, process‐based (called δ here) can approach level LSTM intensively observed variable (streamflow) with regionalized parameterization. We use a simple model HBV backbone embedded neural networks, which only be trained differentiable programming framework, parameterize, enhance, or replace model's modules. Without using an ensemble post‐processor, obtain median Nash‐Sutcliffe efficiency 0.732 671 basins USA Daymet forcing data set, compared 0.748 from state‐of‐the‐art same setup. For another difference is even smaller: 0.715 versus 0.722. Meanwhile, resulting learnable output full set example, soil groundwater storage, snowpack, evapotranspiration, baseflow, later constrained by their observations. Both simulated evapotranspiration fraction discharge baseflow agreed decently alternative estimates. The general framework work various process complexity opens up path physics big data.

Language: Английский

---

Accurate prediction of water quality in urban drainage network with integrated EMD-LSTM model

Journal of Cleaner Production, Journal Year: 2022, Volume and Issue: 354, P. 131724 - 131724

Published: April 13, 2022

Language: Английский

---

The Data Synergy Effects of Time‐Series Deep Learning Models in Hydrology

Water Resources Research, Journal Year: 2022, Volume and Issue: 58(4)

Published: March 17, 2022

Abstract When fitting statistical models to variables in geoscientific disciplines such as hydrology, it is a customary practice stratify large domain into multiple regions (or regimes) and study each region separately. Traditional wisdom suggests that built for separately will have higher performance because of homogeneity within region. However, stratified model has access fewer less diverse data points. Here, through two hydrologic examples (soil moisture streamflow), we show conventional may no longer hold the era big deep learning (DL). We systematically examined an effect call synergy , where results DL improved when were pooled together from characteristically different regions. The benefited modest diversity training compared homogeneous set, even with similar quantity. Moreover, allowing heterogeneous makes eligible much larger datasets, which inherent advantage DL. A large, set advantageous terms representing extreme events future scenarios, strong implications climate change impact assessment. here suggest research community should place greater emphasis on sharing.

Language: Английский

---

The latest innovative avenues for the utilization of artificial Intelligence and big data analytics in water resource management

Results in Engineering, Journal Year: 2023, Volume and Issue: 20, P. 101566 - 101566

Published: Nov. 3, 2023

The effective management of water resources is essential to environmental stewardship and sustainable development. Traditional approaches resource (WRM) struggle with real-time data acquisition, analysis, intelligent decision-making. To address these challenges, innovative solutions are required. Artificial Intelligence (AI) Big Data Analytics (BDA) at the forefront have potential revolutionize way managed. This paper reviews current applications AI BDA in WRM, highlighting their capacity overcome existing limitations. It includes investigation technologies, such as machine learning deep learning, diverse quality monitoring, allocation, demand forecasting. In addition, review explores role resources, elaborating on various sources that can be used, remote sensing, IoT devices, social media. conclusion, study synthesizes key insights outlines prospective directions for leveraging optimal allocation.

Language: Английский

---

Prediction of estuarine water quality using interpretable machine learning approach

Journal of Hydrology, Journal Year: 2021, Volume and Issue: 605, P. 127320 - 127320

Published: Dec. 20, 2021

Language: Английский

---