Isotope-based source assessment of water flowing from storm sewer systems to a receiving river during dry weather periods.

Urban stormwater management systems, particularly storm sewers, are critical for managing runoff in urban areas. These systems are designed to function during wet weather events; however, field-based observations of these systems suggest that they may also be active flow pathways in dry weather conditions, ultimately contributing to streamflow. Unlike dry weather flow in wastewater systems, storm sewer dry weather flow has not been thoroughly explored. This research used stable isotopes of oxygen and hydrogen in water to examine the sources of dry weather flow from storm sewers in a highly urban catchment. A stable isotope mixing model was applied at the outfalls of two stormwater catchments and the receiving Black Creek, located in Toronto, Canada. Findings suggest that during dry periods, storm sewers receive non-stormwater inputs from tap water, wastewater, and groundwater, along with some precipitation, and that these sources may constitute up to 19 % of Black Creek's flow at the watershed scale. Seasonal patterns in flow and water sources were observed for the Black Creek and outfalls. At one outfall, dry weather flow was predominantly from the water distribution system (i.e., tap water and/or wastewater) throughout spring, summer, and fall. In contrast, at the second outfall, groundwater dominated in spring and summer, and groundwater and water distribution were equally proportioned in fall. Black Creek baseflow comprises a dynamic mix of water sources that at times are similar to the sources observed at the stormwater outfalls. Considering these findings, future work should incorporate strategic sampling of additional outfalls, and multiple years of data collection to explore inter-annual variability in these processes and focus on replicating a similar study in other urban watersheds with different climates and/or water infrastructure design. The study findings highlight that our understanding of dry weather flow from storm sewers is relatively limited, emphasizing the need for further exploration of this phenomenon to inform urban hydrological modelling, water quality studies, and urban water management.

Examining optimized machine learning models for accurate multi-month drought forecasting: A representative case study in the USA.

The Colorado River has experienced a significant streamflow reduction in recent decades due to climate change, resulting in pronounced hydrological droughts that pose challenges to the environment and human activities. However, current models struggle to accurately capture complex drought patterns, and their accuracy decreases as the lead time increases. Thus, determining the reliability of drought forecasting for specific months ahead presents a challenging task. This study introduces a robust approach that utilizes the Beluga Whale Optimization (BWO) algorithm to train and optimize the parameters of the Regularized Extreme Learning Machine (RELM) and Random Forest (RF) models. The applied models are validated against a KNN benchmark model for forecasting drought from one- to six-month ahead across four hydrological stations distributed over the Colorado River. The achieved results demonstrate that RELM-BWO outperforms RF-BWO and KNN models, achieving the lowest root-mean square error (0.2795), uncertainty (U95 = 0.1077), mean absolute error (0.2104), and highest correlation coefficient (0.9135). Also, the current study uses Global Multi-Criteria Decision Analysis (GMCDA) as an evaluation metric to assess the reliability of the forecasting. The GMCDA results indicate that RELM-BWO provides reliable forecasts up to four months ahead. Overall, the research methodology is valuable for drought assessment and forecasting, enabling advanced early warning systems and effective drought countermeasures.

Predicting flood stages in watersheds with different scales using hourly rainfall dataset: A high-volume rainfall features empowered machine learning approach.

Accurate prediction of instantaneous high lake water levels and flood flows (flood stages) from micro-catchments to big river basins are critical for flood forecasting. Lake Carl Blackwell, a small-watershed reservoir in the south-central USA, served as a primary case study due to its rich historical dataset. Bearing knowledge that both current and previous rainfall contributes to the reservoirs' water body, a series of hourly rainfall features were created to maximize predicting power, which include total rainfall amounts in the current hour, the past 2 h, 3 h, …, 600 h in addition to previous-day lake levels. Notedly, the rainfall features are the accumulated rainfall amounts from present to previous hours rather than the rainfall amount in any specific hour. Random Forest Regression (RFR) was used to score the features' importance and predict the flood stages along with Neural Network - Multi-layer Perceptron Regression (NN-MLP), Support Vector Regression (SVR), Extreme Gradient Boosting (XGBoost), and the ordinary multi-variant linear regression (MLR) together with dimension reduced linear models of Principal Component Regression (PCR) and Partial Least Square Regression (PLSR). The prediction accuracy for the lake flood stages can be as high as 0.95 in R2, 0.11 ft. in mean absolute error (MAE), and 0.21 ft. in root mean square error (RMSE) for the testing dataset by the RFR (NN-MLP performed equally well), with small accuracy decreases by the other two non-linear algorithms of XGBoost and SVR. The linear regressions with dimension reductions had the lowest accuracy. Furthermore, our approach demonstrated high accuracy and broad applicability for surface runoff and streamflow predictions across three different-sized watersheds from micro-catchment to big river basins in the region, with increases of predicting power from earlier rainfall for larger watersheds and vice versa.

Dynamic disparities in inorganic nitrogen and phosphorus fluxes into estuarine systems under different flow regimes and streamflow droughts.

Elongated periods of low flow conditions, which can be termed as streamflow droughts, influence the nutrient (e.g., nitrogen and phosphorus) balance in estuarine systems. Analyzing temporal trends of nutrient fluxes into such systems under different streamflow regimes can complement the understanding about the dynamic evolution of streamflow droughts and their impacts on nutrient levels. The objective of this paper was to evaluate how dynamic evolution of streamflow droughts (from low flow conditions) affects the inorganic nutrient flux in a tropical estuarine system. We analyzed a 20-year time series of streamflow data together with the concentrations of two nutrient parameters-dissolved inorganic phosphorus (DIP) and dissolved inorganic nitrogen (DIN)-in the Lower Apalachicola River that drains into Apalachicola Bay in northeastern Gulf of Mexico, Florida. Our findings revealed that droughts affect the seasonal patterns and fluxes of both DIP and DIN. We also observed post-drought flushing patterns in DIP and contrasting changes in DIP and DIN fluxes in the long-term (20 years here) under different streamflow conditions. Dynamically changing correlations between the streamflow and the fluxes were found throughout different phases of droughts. In the long-term (from 2003 to 2021), the DIP flux in high flows increased by 35.3%, while the flux decreased by 15.7% in low flows. Conversely, DIN flux in high flows showed a decrease of <1.2%, but an increase of <23.7% in low flows after droughts end. The insights from this study highlighted the need for effective regulation plans such as proper nutrient management against streamflow droughts to mitigate negative ecological consequences in estuarine systems such as harmful algal blooms.

Streamflow trends and flood frequency analysis: a regional study of the UK.

In recent years, the escalating effects of climate change on surface water bodies have underscored the critical importance of analyzing streamflow trends for effective water resource planning and management. This study conducts a comprehensive regional investigation into the streamflow rate trends of 18 rivers across the United Kingdom (UK). An enhanced Mann-Kendall (MK) test was employed to meticulously analyze both rainfall and streamflow trends on monthly and annual scales. Additionally, the Innovative Trend Analysis (ITA) method was applied to elucidate the variability of streamflow rates, providing a more nuanced understanding of hydrological changes in response to climatic shifts. MK test reveals statistically significant positive trends in streamflow rates, particularly for rivers in south-central Scotland and northern England. Specifically, in January, rivers such as the Tay at Ballathie, Tweed at Peebles, and Teviot at Ormiston showed Z-scores above 2. Annually, similar positive trends were observed, with the Tay at Ballathie (Z = 3.42) and Nith at Friars Carse (Z = 3.35) exhibiting the highest increases in streamflow rates. The ITA method showed no relevant trends for the lowest values of streamflow, except for the Thames at Kingston, while considerable variability was observed for the highest streamflow rates, with several rivers showing positive trends and, however, some England rivers, like Bure at Ingworth, Test at Broadlands, and Trent at Colwick, showing negative trends. From this perspective, a more in-depth analysis of the extreme streamflow trends was carried out. In particular, the flood frequency of the maximum annual streamflow was assessed, based on the fitting of the Generalized Extreme Value (GEV) distribution on the annual maxima. Increasing location parameter (µ) and return period trends were observed for several rivers across the UK. In particular, the Tay at Ballathie (Scotland) showed the most marked increase, with µ that ranged from about 730 m3/s to more than 900 m3/s. At the same time, slight decreasing trends were observed for the Trent River (µ from 378 m3/s to 341 m3/s). The critical comparison of the MK test, ITA, and GEV distribution fitting revealed both agreements and discrepancies among the methods. While the analyses generally aligned in detecting significant trends in streamflow rates, notable discrepancies were observed, particularly in rivers with negligible trends. These inconsistencies highlight the complexity of hydrological responses and the limitations of individual methods. Overall, the study provides a comprehensive view of how streamflow dynamics are evolving in UK rivers, highlighting regional variations in the impact of climate change. This understanding can improve water resource management strategies by integrating diverse analytical approaches.

Evaluation of fish habitat suitability based on stream hydrodynamics and water quality using SWAT and HEC-RAS linked simulation.

The objective of this study was to evaluate fish habitat suitability by simulating hydrodynamic and water quality factors using SWAT and HEC-RAS linked simulation considering time-series analysis. A 2.9 km reach of the Bokha stream was selected for the habitat evaluation of Zacco platypus, with hydrodynamic and water quality simulations performed using the SWAT and HEC-RAS linked approach. Based on simulated 10-year data, the aquatic habitat was assessed using the weighted usable area (WUA), and minimum ecological streamflow was proposed from continuous above threshold (CAT) analysis. High water temperature was identified as the most influential habitat indicator, with its impact being particularly pronounced in shallow streamflow areas during hot summer seasons. The time-series analysis identified a 28% threshold of WUA/WUAmax, equivalent to a streamflow of 0.48 m3/s, as the minimum ecological streamflow necessary to mitigate the impact of rising water temperatures. The proposed habitat modeling method, linking watershed-stream models, could serve as a useful tool for ecological stream management.

An ecohydrological approach to assess water provisioning and supporting ecosystem services in the Budhabalanga River Basin, India.

Rivers are vital and complex natural systems that provide a wide range of ecosystem services. This study presents a methodology for assessing the riverine provisioning and supporting ecosystem services, whose applicability has been demonstrated over the Budhabalanga River Basin of India. The Soil and Water Assessment Tool (SWAT) is used to generate streamflow time series at various ungauged sites, and then the streamflow is characterized for the evaluation of provisioning services. Further, the diversity and abundance of macroinvertebrates, along with the Lotic-invertebrate Index for Flow Evaluation (LIFE), is used to study the riverine supporting ecosystem services. The streams show intermittent behavior and strong seasonality for low flows, which limits the water availability, particularly during pre-monsoon season. The Baseflow Index (BFI) is greater than 0.6, indicating that groundwater contributes more than 60% of the total streamflow. Interestingly, despite the high BFI, the streams did not conform to the prevailing opinion that a greater baseflow contribution results in a later commencement of the low-flow period in the hydrological year. Furthermore, the study depicts significant variations in the diversity and abundance of the macroinvertebrates across the various sampling sites. However, the LIFE score across the sites remained consistent within a narrow range, i.e., 8 to 9, suggesting a steady supply of supporting ecosystem services. The results of the study can help the policymakers towards an informed decision making and the simplistic methodology proposed in this study can be replicated in other river basins for identifying vulnerable watersheds and prioritizing management actions.

Past, present and future changes in the annual streamflow of the Lancang-Mekong River and their driving mechanisms.

The rapid development of the Greater Mekong Subregion (GMS) makes it essential to understand the major mechanisms controlling the streamflow, especially for the Lancang-Mekong River (abbr. Mekong River). We used instrumental annual streamflow data (1960-2007) from Chiang Saen hydrological station and several gridded hydroclimatic datasets to explore the hydroclimatic evolution of the Mekong River, together with its driving mechanisms. We found that changes in the Mekong streamflow are consistent with precipitation changes, and the Mekong is thus a precipitation-dominated river that is susceptible to the effects of ongoing climate change. The instrumental record of Mekong annual streamflow is closely related to hydroclimatic changes, especially those related to moisture, within the area from the Hengduan Mountains to the Chiang Saen Station. Based on these gridded records, we extended the Mekong annual streamflow record to cover 1891-2021 using nested multiple linear regression fitting. The fitted streamflow explained up to 57.6 % of the instrumental changes and it indicates that the major 2019 drought was not unique over the past century. Despite extremely low precipitation and high temperatures, it is likely that the effects of drought can be mitigated via hydraulic engineering regulation. Climatological analyses showed that the Mekong annual streamflow is driven by the El Niño-Southern Oscillation (ENSO) and the Indian Ocean Dipole (IOD), which is consistent with observed quasi-interannual cycles of 3-4 years. A multi-model ensemble of CMIP6 revealed that the Mekong annual streamflow will experience an upward trend in the future, accompanied by the more extreme impacts of ENSO and the IOD.

Analyzing the adjustment mechanism of daily sediment content under large reservoir operations.

Adjustment of daily discharge and sediment content in the Lower Jinsha River Basin has changed dramatically. However, the adjustment mechanism of daily sediment content under reservoir operation remains unclear. The Double Mass Curve (DMS) method was used to divide different periods of daily discharge-sediment content relationship change, and the Flow Duration Curve (FDC) was used to calculate the energy dissipation of streamflow by reservoirs. With the operation of large reservoirs, the average flood discharge and its proportion significantly decreased. With the variation in flow regime, the quantile relationship and Lower boundary relationship of daily discharge and sediment content both showed a downward trend, from 1999 to 2019. Under different periods, adjustment of the cross-flow profile was decreased with larger daily discharge, which was characterized by the ratio of sediment content to the lower boundary. An improved flow duration curve method was proposed to calculate the energy dissipation of streamflow. We discovered a novel model between the relative reduction of sediment content and relative energy dissipation of the daily discharge regime, with a good fitness of 0.97. In this study, the effect of the flow regime constructed on sediment content change was emphasized. It is helpful to evaluate the sediment reduction of the total basin caused by reservoirs.

Intercomparison of deep learning models in predicting streamflow patterns: insight from CMIP6.

This research was carried out to predict daily streamflow for the Swat River Basin, Pakistan through four deep learning (DL) models: Feed Forward Artificial Neural Networks (FFANN), Seasonal Artificial Neural Networks (SANN), Time Lag Artificial Neural Networks (TLANN) and Long Short-Term Memory (LSTM) under two Shared Socioeconomic Pathways (SSPs) 585 and 245. Taylor Diagram, Random Forest, and Gradient Boosting techniques were used to select the best combination of General Circulation Models (GCMs) for Multi-Model Ensemble (MME) computation. MME was computed via the Random Forest technique for Maximum Temperature (Tmax), Minimum Temperature (Tmin), and precipitation for the aforementioned three techniques. The best MME for Tmax, Tmin, and precipitation was rendered by Compromise Programming. The DL models were trained and tested using observed precipitation and temperature as independent variables and discharge as dependent variables. The results of deep learning models were evaluated using statistical performance indicators such as root mean square error (RMSE), mean square error (MSE), mean absolute error (MAE), and coefficient of determination (R2). The TLANN demonstrated superior performance compared to the other models based on RMSE, MSE, MAE, and R2 during training (65.25 m3/s, 4256.97 m3/s, 46.793 m3/s and 0.7978) and testing (72.06 m3/s, 5192.95 m3/s, 51.363 m3/s and 0.7443) respectively. Subsequently, TLANN was utilized to make predictions based on MME of SSP245 and SSP585 scenarios for future streamflow until the year 2100. These results can be used for planning, management, and policy-making regarding water resources projects in the study area.

Integrated metaheuristic algorithms with extreme learning machine models for river streamflow prediction.

Accurate river streamflow prediction is pivotal for effective resource planning and flood risk management. Traditional river streamflow forecasting models encounter challenges such as nonlinearity, stochastic behavior, and convergence reliability. To overcome these, we introduce novel hybrid models that combine extreme learning machines (ELM) with cutting-edge mathematical inspired metaheuristic optimization algorithms, including Pareto-like sequential sampling (PSS), weighted mean of vectors (INFO), and the Runge-Kutta optimizer (RUN). Our comparative assessment includes 20 hybrid models across eight metaheuristic categories, using streamflow data from the Aswan High Dam on the Nile River. Our findings highlight the superior performance of mathematically based models, which demonstrate enhanced predictive accuracy, robust convergence, and sustained stability. Specifically, the PSS-ELM model achieves superior performance with a root mean square error of 2.0667, a Pearson's correlation index (R) of 0.9374, and a Nash-Sutcliffe efficiency (NSE) of 0.8642. Additionally, INFO-ELM and RUN-ELM models exhibit robust convergence with mean absolute percentage errors of 15.21% and 15.28% respectively, a mean absolute errors of 1.2145 and 1.2105, and high Kling-Gupta efficiencies values of 0.9113 and 0.9124, respectively. These findings suggest that the adoption of our proposed models significantly enhances water management strategies and reduces any risks.

Ecohydrological assessment of the water balance of the world's highest elevation tropical forest (Polylepis).

Polylepis trees grow at elevations above the continuous tree line (3000-5000 m a.s.l.) across the Andes. They tolerate extreme environmental conditions, making them sensitive bioindicators of global climate change. Therefore, investigating their ecohydrological role is key to understanding how the water cycle of Andean headwaters could be affected by predicted changes in environmental conditions, as well as ongoing Polylepis reforestation initiatives in the region. We estimate, for the first time, the annual water balance of a mature Polylepis forest (Polylepis reticulata) catchment (3780 m a.s.l.) located in the south Ecuadorian páramo using a unique set of field ecohydrological measurements including gross rainfall, throughfall, streamflow, and xylem sap flow in combination with the characterization of forest and soil features. We also compare the forest water balance with that of a tussock grass (Calamagrostis intermedia) catchment, the dominant páramo vegetation. Annual gross rainfall during the study period (April 2019-March 2020) was 1290.6 mm yr-1. Throughfall in the Polylepis forest represented 61.2 % of annual gross rainfall. Streamflow was the main component of the water balance of the forested site (59.6 %), while its change in soil water storage was negligible (<1 %). Forest evapotranspiration was 54.0 %, with evaporation from canopy interception (38.8 %) more than twice as high as transpiration (15.1 %). The error in the annual water balance of the Polylepis catchment was small (<15 %), providing confidence in the measurements and assumptions used to estimate its components. In comparison, streamflow and evapotranspiration at the grassland site accounted for 63.7 and 36.0 % of the water balance, respectively. Although evapotranspiration was larger in the forest catchment, its water yield was only marginally reduced (<4 %) in relation to the grassland catchment. The substantially higher soil organic matter content in the forest site (47.6 %) compared to the grassland site (31.8 %) suggests that even though Polylepis forests do not impair the hydrological function of high-Andean catchments, their presence contributes to carbon storage in the litter layer of the forest and the underlying soil. These findings provide key insights into the vegetation-water­carbon nexus in high Andean ecosystems, which can serve as a basis for future ecohydrological studies and improved management of páramo natural resources considering changes in land use and global climate.

Understanding the association between global forest fire products and hydrometeorological variables.

Climate change and anthropogenic activities have influenced the frequency and magnitude of forest fires both globally and regionally. While skilful short- to extended-range prediction of forest fires is essential for effective mitigation in local communities, it is also important to identify the implications of forest fires on different sectors, including water resources and sustainable development. Limited studies have investigated the association between forest fires and hydrometeorological variables at the regional scale in developing countries due to the lack of necessary datasets, which can now be leveraged using the newly hosted global reanalysis of fire danger indices (referred to as fire indices). The current study presents a comprehensive analysis of the spatio-temporal variations of eight fire indices across India, as well as their association with hydro-meteorological variables, such as precipitation, temperature, and the streamflow of a major river basin (Mahanadi) in India. The accuracy of these indices in capturing real fire events and the potential benefit of incorporating fire indices into long-term hydrologic simulations are also explored. The results show that fire indices can accurately yield fire seasons (i.e., post-monsoon and summer) in India. Furthermore, forest fires are found to be strongly associated with hydro-meteorological variables, typically resulting in low streamflow regimes. Fire indices can also capture actual fire events, maintaining high scalar accuracy. Finally, an improvement in uncalibrated hydrologic model simulations is observed when simulated streamflow is post-processed using the fire indices as predictors. Overall, the current study has valuable implications for fire indices forecasting and hydrologic simulations in ungauged basins.

Physically based vs. data-driven models for streamflow and reservoir volume prediction at a data-scarce semi-arid basin.

Physically based or data-driven models can be used for understanding basinwide hydrological processes and creating predictions for future conditions. Physically based models use physical laws and principles to represent hydrological processes. In contrast, data-driven models focus on input-output relationships. Although both approaches have found applications in hydrology, studies that compare these approaches are still limited for data-scarce, semi-arid basins with altered hydrological regimes. This study aims to compare the performances of a physically based model (Soil and Water Assessment Tool (SWAT)) and a data-driven model (Nonlinear AutoRegressive eXogenous model (NARX)) for reservoir volume and streamflow prediction in a data-scarce semi-arid region. The study was conducted in the Tersakan Basin, a semi-arid agricultural basin in Türkiye, where the basin hydrology was significantly altered due to reservoirs (Ladik and Yedikir Reservoir) constructed for irrigation purposes. The models were calibrated and validated for streamflow and reservoir volumes. The results show that (1) NARX performed better in the prediction of water volumes of Ladik and Yedikir Reservoirs and streamflow at the basin outlet than SWAT (2). The SWAT and NARX models both provided the best performance when predicting water volumes at the Ladik reservoir. Both models provided the second best performance during the prediction of water volumes at the Yedikir reservoir. The model performances were the lowest for prediction of streamflow at the basin outlet (3). Comparison of physically based and data-driven models is challenging due to their different characteristics and input data requirements. In this study, the data-driven model provided higher performance than the physically based model. However, input data used for establishing the physically based model had several uncertainties, which may be responsible for the lower performance. Data-driven models can provide alternatives to physically-based models under data-scarce conditions.

Enhancing hydrological modeling with transformers: a case study for 24-h streamflow prediction.

In this paper, we address the critical task of 24-h streamflow forecasting using advanced deep-learning models, with a primary focus on the transformer architecture which has seen limited application in this specific task. We compare the performance of five different models, including persistence, long short-term memory (LSTM), Seq2Seq, GRU, and transformer, across four distinct regions. The evaluation is based on three performance metrics: Nash-Sutcliffe Efficiency (NSE), Pearson's r, and normalized root mean square error (NRMSE). Additionally, we investigate the impact of two data extension methods: zero-padding and persistence, on the model's predictive capabilities. Our findings highlight the transformer's superiority in capturing complex temporal dependencies and patterns in the streamflow data, outperforming all other models in terms of both accuracy and reliability. Specifically, the transformer model demonstrated a substantial improvement in NSE scores by up to 20% compared to other models. The study's insights emphasize the significance of leveraging advanced deep learning techniques, such as the transformer, in hydrological modeling and streamflow forecasting for effective water resource management and flood prediction.

Enhancing short-term streamflow prediction in the Haihe River Basin through integrated machine learning with Lasso.

With the widespread application of machine learning in various fields, enhancing its accuracy in hydrological forecasting has become a focal point of interest for hydrologists. This study, set against the backdrop of the Haihe River Basin, focuses on daily-scale streamflow and explores the application of the Lasso feature selection method alongside three machine learning models (long short-term memory, LSTM; transformer for time series, TTS; random forest, RF) in short-term streamflow prediction. Through comparative experiments, we found that the Lasso method significantly enhances the model's performance, with a respective increase in the generalization capabilities of the three models by 21, 12, and 14%. Among the selected features, lagged streamflow and precipitation play dominant roles, with streamflow closest to the prediction date consistently being the most crucial feature. In comparison to the TTS and RF models, the LSTM model demonstrates superior performance and generalization capabilities in streamflow prediction for 1-7 days, making it more suitable for practical applications in hydrological forecasting in the Haihe River Basin and similar regions. Overall, this study deepens our understanding of feature selection and machine learning models in hydrology, providing valuable insights for hydrological simulations under the influence of complex human activities.

Interpretable and explainable hybrid model for daily streamflow prediction based on multi-factor drivers.

Streamflow time series data typically exhibit nonlinear and nonstationary characteristics that complicate precise estimation. Recently, multifactorial machine learning (ML) models have been developed to enhance the performance of streamflow predictions. However, the lack of interpretability within these ML models raises concerns about their inner workings and reliability. This paper introduces an innovative hybrid architecture, the TCN-LSTM-Multihead-Attention model, which combines two layers of temporal convolutional networks (TCN) followed by one layer of long short-term memory (LSTM) units, integrated with a Multihead-Attention mechanism for predicting streamflow with streamflow causation-driven prediction samples (RCDP), employing local and global interpretability studies through Shapley values and partial dependency analysis. The find_peaks method was used to identify peak flow events in the test dataset, validating the model's generality and uncovering the physical causative patterns of streamflow. The results show that (1) compared to the LSTM model with the same hyperparameter settings, the proposed TCN-LSTM-Multihead-Attention hybrid model increased the R2 by 52.9%, 2.5%, 43.1%, and 10.7% respectively at four stations in the test set predictions using RCDP samples. Moreover, comparing the prediction results of the hybrid model under different samples in Hengshan station, the R2 for RCDP increased by 5.06% and 1.22% compared to streamflow autoregressive prediction samples (RAP) and meteorological-soil volumetric water content coupled autoregressive prediction samples (MCSAP) respectively. (2) Historical streamflow data from the preceding 3 days predominantly influences predictions due to strong autocorrelation, with flow quantity (Q) typically emerging as the most significant feature alongside precipitation (P), surface soil moisture (SSM), and adjacent station flow data. (3) During periods of low and normal flow, historical data remains the most crucial factor; however, during flood periods, the roles of upstream inflow and precipitation become significantly more pronounced. This model facilitates the identification and quantification of various hydrodynamic impacts on flow predictions, including upstream flood propagation, precipitation, and soil moisture conditions. It also elucidates the model's nonlinear relationships and threshold responses, thereby enhancing the interpretability and reliability of streamflow predictions.

Hydrological response of drought impacts across catchments worldwide.

Drought will inevitably affect linkages between different water components, which have previously been investigated across different spatiotemporal scales. Elucidating drought-induced precipitation (P) partition effects remain uncertain because they involve drought propagation, even inducing streamflow (Q) non-stationarity. This study collected data on 1069 catchments worldwide to investigate Q and evapotranspiration (ET) impacts from P deficit-derived reductions in drought propagation. Results show that P deficits trigger soil moisture drought, subsequently inducing negative Q and ET anomalies that vary under different climate regimes. Generally, drought-induced hydrological legacies indicate that breaks in hydrological linkages cause a relatively rapid Q response (i.e., negative Q anomaly), amplified by drought strength and duration. Compared with the Q response, the ET response to drought stress involves a more complex, associative vegetation response and an associative evaporative state controlled by water and energy, which lags behind the Q response and can also intensify with increasing drought severity and duration. This is confirmed by the ET response under different climate regimes. Namely, in drier climates, a positive ET anomaly can be detected in its early stages, this is unusual in wetter climate. Additionally, Q and ET sensitivity to drought strength can be mechanistically explained by the water and energy status. This implies that ET is mainly controlled by water and energy, resulting in higher and lower drought sensitivity within water- and energy-limited regions, respectively. Understanding the impacts of drought on Q and ET response is essential for identifying key linkages in drought propagation across different climate regimes. Our findings will also be useful for developing early warning and adaptation systems that support both human and ecosystem requirements.

An explainable multiscale LSTM model with wavelet transform and layer-wise relevance propagation for daily streamflow forecasting.

Developing an accurate and reliable daily streamflow forecasting model is important for facilitating the efficient resource planning and management of hydrological systems. In this study, an explainable multiscale long short-term memory (XM-LSTM) model is proposed for effective daily streamflow forecasting by integrating the à trous wavelet transform (ATWT) for decomposing data, the Boruta algorithm for identifying model inputs, and the layer-wise relevance propagation (LRP) for explaining the prediction results. The proposed XM-LSTM is tested by performing multi-step-ahead forecasting of daily streamflow at four stations in the middle and lower reaches of the Yangtze River basin and compared with the X-LSTM. The X-LSTM is formed by coupling the long short-term memory (LSTM) with the LRP. For comparison, the inputs of these two models are identified by the Boruta selection algorithm. The results show that all models exhibit good ability to forecast daily streamflow, however, the prediction performance decreases as the lead time increases. The XM-LSTM provides a better forecasting performance than the X-LSTM, suggesting the ability of the ATWT to improve the LSTM for daily streamflow forecasting. Moreover, the correlation scores analysis by the LRP shows that the ATWT can extract useful information that influences the daily streamflow from the raw predictors, and the water level has the most significant contribution to streamflow prediction. Accordingly, the XM-LSTM model can be viewed as a potentially useful approach for increasing the accuracy and explainability of streamflow forecasting.

Do non-linearity and non-Gaussianity truly matter in streamflow forecasting? A comparative study between PAR(p) and vine copula for Brazilian streamflow time series.

This study evaluates the joint impact of non-linearity and non-Gaussianity on predictive performance in 23 Brazilian monthly streamflow time series from 1931 to 2022. We consider point and interval forecasting, employing a PAR(p) model and comparing it with the periodic vine copula model. Results indicate that the Gaussian hypothesis assumed by PAR(p) is unsuitable; gamma and log-normal distributions prove more appropriate and crucial for constructing accurate confidence intervals. This is primarily due to the assumption of the Gaussian distribution, which can lead to the generation of confidence intervals with negative values. Analyzing the estimated copula models, we observed a prevalence of the bivariate Normal copula, indicating that linear dynamic dependence is frequent, and the Rotated Gumbel 180°, which exhibits lower tail dependence. Overall, the temporal dynamics are predominantly shaped by combining these two types of effects. In point forecasting, both models show similar behavior in the estimation set, with slight advantages for the copula model. The copula model performs better during the out-of-sample analysis, particularly for certain power plants. In interval forecasting, the copula model exhibits pronounced superiority, offering a better estimation of quantiles. Consistently demonstrating proficiency in constructing reliable and accurate intervals, the copula model reveals a notable advantage over the PAR(p) model in interval forecasting.