Coupling SWAT and Transformer Models for Enhanced Monthly Streamflow Prediction

Sustainability, Год журнала: 2024, Номер 16(19), С. 8699 - 8699

Опубликована: Окт. 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.

Язык: Английский

---

Enhancing Long-Term Flood Forecasting with SageFormer: A Cascaded Dimensionality Reduction Approach Based on Satellite-Derived Data

Remote Sensing, Год журнала: 2025, Номер 17(3), С. 365 - 365

Опубликована: Янв. 22, 2025

Floods, increasingly exacerbated by climate change, are among the most destructive natural disasters globally, necessitating advancements in long-term forecasting to improve risk management. Traditional models struggle with complex dependencies of hydroclimatic variables and environmental conditions, thus limiting their reliability. This study introduces a novel framework for enhancing flood accuracy integrating geo-spatiotemporal analyses, cascading dimensionality reduction, SageFormer-based multi-step-ahead predictions. The efficiently processes satellite-derived data, addressing curse focusing on critical long-range spatiotemporal dependencies. SageFormer captures inter- intra-dependencies within compressed feature space, making it particularly effective forecasting. Performance evaluations against LSTM, Transformer, Informer across three data fusion scenarios reveal substantial improvements accuracy, especially data-scarce basins. integration hydroclimate attention-based networks reduction demonstrates significant over traditional approaches. proposed combines advanced deep learning, both interpretability precision capturing By offering straightforward reliable approach, this advances remote sensing applications hydrological modeling, providing robust tool mitigating impacts extremes.

Язык: Английский

---

Identification of Time-Varying Conceptual Hydrological Model Parameters with Differentiable Parameter Learning

Water, Год журнала: 2024, Номер 16(6), С. 896 - 896

Опубликована: Март 20, 2024

The parameters of the GR4J-CemaNeige coupling model (GR4neige) are typically treated as constants. However, maximum capacity production store (parX1) exhibits time-varying characteristics due to climate variability and vegetation coverage change. This study employed differentiable parameter learning (dPL) identify parX1 in GR4neige across 671 catchments within United States. We built two types dPL, including static dynamic networks, assess advantages parameter. In network, we evaluated impact potential evapotranspiration (PET), precipitation (P), temperature (T), soil moisture (SM), normalized difference index (NDVI) datasets on performance dPL. then compared dPL with empirical functional method (fm). results demonstrated that network outperformed streamflow estimation. There were differences estimation among driven by various input features. humid catchments, simultaneously incorporating all five factors, PET, P, T, SM, NDVI, achieved optimal simulation accuracy. arid it was preferable introduce NDVI separately for improved performance. significantly fm estimating uncalibrated intermediate variables, like (ET). Both derived from exhibited significant spatiotemporal variation catchments. Notably, obtained through fm, a distinct spatial clustering pattern. highlights enhancing accuracy contributes understanding under influence conditions,

Язык: Английский

---

Long-term natural streamflow forecasting under drought scenarios using data-intelligencw modeling

Water Cycle, Год журнала: 2024, Номер 5, С. 266 - 277

Опубликована: Янв. 1, 2024

Long-term river streamflow prediction and modeling are essential for water resource management decision-making related to resources. This research paper considers the importance of these predictions proposes a model address scarcity scenarios support in allocation, flood management, drought scenarios. Machine learning (ML) techniques offer promising alternatives improving long-term prediction. However, most existing studies on ML models have focused shorter time horizons, limiting their broader applicability. Consequently, there is need dedicated that addresses use Considering this gap, presents an ML-based approach learns replicates natural flow dynamics river, allowing simulation reduced (25% 50% reduction). capability allows simulating varying severity, providing valuable insights service managers. study significantly contributes progress predicting through application machine models. Moreover, offers recommendations hydrologists improve future efforts.

Язык: Английский

---

Investigating the Performance of the Informer Model for Streamflow Forecasting

Water, Год журнала: 2024, Номер 16(20), С. 2882 - 2882

Опубликована: Окт. 10, 2024

Recent studies have shown the potential of transformer-based neural networks in increasing prediction capacity. However, classical transformers present several problems such as computational time complexity and high memory requirements, which make Long Sequence Time-Series Forecasting (LSTF) challenging. The contribution to series flood events using deep learning techniques is examined, with a particular focus on evaluating performance Informer model (a implementation transformer architecture), attempts address previous issues. predictive capabilities are explored compared statistical methods, stochastic models traditional networks. accuracy, efficiency well limits approaches demonstrated via numerical benchmarks relating real river streamflow applications. Using daily flow data from River Test England main case study, we conduct rigorous evaluation efficacy capturing complex temporal dependencies inherent series. analysis extended encompass diverse datasets various locations (>100) United Kingdom, providing insights into generalizability Informer. results highlight superiority over established forecasting especially regarding LSTF problem. For forecast horizon 168 days, achieves an NSE 0.8 maintains MAPE below 10%, while second-best (LSTM) only −0.63 25%, respectively. Furthermore, it observed that dependence structure series, expressed by climacogram, affects network.

Язык: Английский

---