Enhancing physically-based hydrological modeling with an ensemble of machine-learning reservoir operation modules under heavy human regulation using easily accessible data.

Dams and reservoirs have significantly altered river flow dynamics worldwide. Accurately representing reservoir operations in hydrological models is crucial yet challenging. Detailed reservoir operation data is often inaccessible, leading to relying on simplified reservoir operation modules in most hydrological models. To improve the capability of hydrological models to capture flow variability influenced by reservoirs, this study proposes a hybrid hydrological modeling framework, which combines a process-based hydrological model with a machine-learning-based reservoir operation module designed to simulate runoff under reservoir operations. The reservoir operation module employs an ensemble of three machine learning models: random forest, support vector machine, and AutoGluon. These models predict reservoir outflows using precipitation and temperature data as inputs. The Soil and Water Assessment Tool (SWAT) then integrates these outflow predictions to simulate runoff. To evaluate the performance of this hybrid approach, the Xijiang Basin within the Pearl River Basin, China, is used as a case study. The results highlight the superiority of the SWAT model coupled with machine learning-based reservoir operation models compared to alternative modeling approaches. This hybrid model effectively captures peak flows and dry period runoff. The Nash-Sutcliffe Efficiency (NSE) in daily runoff simulations shows substantial improvement, ranging from 0.141 to 0.780, with corresponding enhancements in the coefficient of determination (R2) by 0.098-0.397 when compared to the original reservoir operation modules in SWAT. In comparison to parameterization techniques lacking a dedicated reservoir module, NSE enhancements range from 0.068 to 0.537, and R2 improvements range from 0.027 to 0.139. The proposed hybrid modeling approach effectively characterizes the impact of reservoir operations on river flow dynamics, leading to enhanced accuracy in runoff simulation. These findings offer valuable insights for hydrological forecasting and water resources management in regions influenced by reservoir operations.

Spatiotemporal links between meteorological and agricultural droughts impacted by tropical cyclones in China.

Both droughts and tropical cyclones (TCs) are among the world's most widespread natural disasters. This paper is concentrated on the effects of TCs on the links between meteorological droughts (MDs) and agricultural droughts (ADs). Specifically, changes in characteristics of drought events and variations in propagation features of matched MD and AD event pairs are quantified by using the renowned three-dimensional connected components algorithm; both alleviation and exacerbation effects of TCs are evaluated; and the Spearman's correlation is employed to identify potential contributors to exacerbated droughts after TCs. The results show that TCs exhibit more pronounced and widespread alleviation effects on MD events compared to AD events. >98 % of small-scale drought events are terminated by TCs, leading to 65 % reduction in the total area of MD events smaller than 50,000 km2 and 32 % reduction in AD events of the same scale. In the meantime, TCs can reshape the spatiotemporal links between MDs and ADs by reducing the overall propagation rate from 77 % to 40 % and ameliorating the characteristics of drought event pairs with higher propagation efficiency, by >40 %. After TCs, over 55 % of drought exacerbations in TC-affected regions occur first in the vicinity of the residual large-scale AD events. This occurrence is partially associated with the reduction in moisture exports from these residual droughts downwind to the interior of TC-affected regions, a process potentially facilitated by the TC-induced temperature cooling. The in-depth evaluation of this paper presents useful information for better drought preparation and mitigation under TCs.

The color of environmental noise in river networks.

Despite its far-reaching implications for conservation and natural resource management, little is known about the color of environmental noise, or the structure of temporal autocorrelation in random environmental variation, in streams and rivers. Here, we analyze the geography, drivers, and timescale-dependence of noise color in streamflow across the U.S. hydrography, using streamflow time series from 7504 gages. We find that daily and annual flows are dominated by red and white spectra respectively, and spatial variation in noise color is explained by a combination of geographic, hydroclimatic, and anthropogenic variables. Noise color at the daily scale is influenced by stream network position, and land use and water management explain around one third of the spatial variation in noise color irrespective of the timescale considered. Our results highlight the peculiarities of environmental variation regimes in riverine systems, and reveal a strong human fingerprint on the stochastic patterns of streamflow variation in river networks.

Unravelling the potential of global streamflow reanalysis in characterizing local flow regime.

While global streamflow reanalysis provides valuable information for environmental modelling and management, it is not yet known how effective they are in characterizing the local flow regime. This paper presents a novel evaluation of the potential of streamflow reanalysis in the flow regime analysis by accounting for the effects of reservoir operation. Specifically, the indicators of hydrologic alteration (IHA) are used to characterize the five components of flow regime for both reservoir inflow and outflow; the performance of raw reanalysis is evaluated and the raw reanalysis is furthermore corrected by using the quantile mapping for improved flow regime analysis. The results of 35 major reservoirs in California show that raw reanalysis tends to be effective in characterizing the regime of reservoir inflow and that it is generally less effective in capturing outflow. For both inflow and outflow, the performance of raw reanalysis is beset by the existence of systematic errors. The quantile mapping is effective in error correction and therefore considerably improves the performances of reanalysis in characterizing the regime of not only reservoir inflow but also outflow. Nevertheless, for both reservoir inflow and outflow, the low flow part tends to be more difficult to handle than the high flow part. The evaluation conducted in this paper can serve as a roadmap for further exploitations of the potential of global streamflow reanalysis for flow regime analysis at regional and even continental scales.

Capabilities of deep learning models on learning physical relationships: Case of rainfall-runoff modeling with LSTM.

This study investigates the relationships which deep learning methods can identify between the input and output data. As a case study, rainfall-runoff modeling in a snow-dominated watershed by means of a long short-term memory (LSTM) network is selected. Daily precipitation and mean air temperature were used as model input to estimate daily flow discharge. After model training and verification, two experimental simulations were conducted with hypothetical inputs instead of observed meteorological data to clarify the response of the trained model to the inputs. The first numerical experiment showed that even without input precipitation, the trained model generated flow discharge, particularly winter low flow and high flow during the snow melting period. The effects of warmer and colder conditions on the flow discharge were also replicated by the trained model without precipitation. Additionally, the model reflected only 17-39% of the total precipitation mass during the snow accumulation period in the total annual flow discharge, revealing a strong lack of water mass conservation. The results of this study indicated that a deep learning method may not properly learn the explicit physical relationships between input and target variables, although they are still capable of maintaining strong goodness-of-fit results.

Coupling hydroclimate-hydraulic-sedimentation models to estimate flood inundation and sediment transport during extreme flood events under a changing climate.

Extreme flood events are disastrous and can cause serious damages to society. Flood frequency obtained based on historical flow records may also be changing under future climate conditions. The associated flood inundation and environmental transport processes will also be affected. In this study, an integrated numerical modeling framework is proposed to investigate the inundation and sedimentation during multiple flood events (2,5,10, 20, 50, 100, 200-year) under future climate change scenarios in a watershed system in northern California, USA. The proposed modeling framework couples physical models of various spatial resolution: kilometers to several hundred kilometers climatic processes, hillslope scale hydrological processes in a watershed, and centimeters to meters scale hydrodynamic and sediment transport processes in a riverine system. The modeling results show that compared to the flows during historical periods, extreme events become more extreme in the 21st century and higher flows tend to be larger and smaller flows tend to be smaller in the system. Flood inundation in the study area, especially during 200-year events, is projected to increase in the future. More sediment will be trapped as the flow increases and the deposition will also increase in the settling basin. Sediment trap efficiency values are within 37.5-65.4% for the historical conditions, within 32.4-68.8% in the first half of the 21st century, and within 34.9-69.3% in the second half of the 21st century. The results highlight the impact of climate change on extreme flood events, the resulting sedimentation, and reflected the importance of incorporating the coupling of physical models into the adaptive watershed and river system management.

Hourly-scale coastal sea level modeling in a changing climate using long short-term memory neural network.

In this study, a coastal sea level estimation model was developed at an hourly temporal scale using the long short-term memory (LSTM) network, which is a type of recurrent neural networks. The model incorporates the effects of various phenomena on the coastal sea level such as the gravitational attractions of the sun and the moon, seasonality, storm surges, and changing climate. The relative positions of the moon and the sun from the target location at each hour were utilized to reflect the gravitational attractions of the sun and the moon in the model simulation. The wind speed and direction, mean sea level pressure (MSLP), and air temperature near the target point at each hour were used to consider the effects of storm surges and seasonality of the coastal sea level. In addition to the hourly local variables, the annual global mean air temperature was considered as input to the model to reflect the effect of global warming on the coastal sea level. The model was implemented using several input lengths of the annual global mean air temperature to estimate the coastal sea level at the Osaka gauging station in Japan. Several statistics such as the mean, the Nash-Sutcliffe efficiency, and the root mean square error were used to evaluate model performance. The results show that the proposed model accurately reconstructed the effects of the gravitational attractions of the sun and the moon on the coastal sea levels. The model also considered the effects of fluctuations in the wind speed and MSLP although the coastal sea levels during were underestimated strong winds and low MSLP conditions. Lastly, introducing a longer duration annual global mean air temperature improved model accuracy. Consequently, the best results show 0.720 of the NSE value for the test process.

Mapping flood susceptibility in mountainous areas on a national scale in China.

Mountainous terrain covers nearly half of China and is susceptible to floods, which can lead to substantial losses of human life and property. Historical flooding records from government bulletins and newspapers, the only available information regarding floods that have occurred in some mountainous areas, are valuable for understanding flood disaster mechanisms in these regions. In this study, the flood susceptibility in mountainous regions in China was mapped based on historical flooding records from 1949 to 2000. A Random Forest (RF) model, which can handle large datasets through factor contribution analysis, was chosen to characterize the relationships between flooding occurrences and twelve geographic, meteorological, and hydrological explanatory factors. The results indicate that the RF model can effectively identify flood-prone areas and has advantages over artificial neural network (ANN) and support vector machine (SVM) methods. Among these explanatory factors, the geographic factors (elevation, longitude and drainage density) are the most important predictors of flooding in China's mountainous areas, whereas the hydrological factors (relative elevation and curve number) are the least important. Two independent datasets of historical flooding events from the Bulletin of Flood and Drought Disasters in China (2006-2014) alongside news reports and yearbooks (2008-2014) were collected and chosen to validate the capability of the RF model. The validation results confirm that the RF model can identify the flood susceptibility with satisfactory accuracy. This study proposes a preliminary flood susceptibility map of mountainous areas in China and provides a reference for predicting and mitigating potentially disastrous flooding events.

Assessment of the effects of multiple extreme floods on flow and transport processes under competing flood protection and environmental management strategies.

Extreme floods are regarded as one of the most catastrophic natural hazards and can result in significant morphological changes induced by pronounced sediment erosion and deposition processes over the landscape. However, the effects of extreme floods of different return intervals on the floodplain and river channel morphological evolution with the associated sediment transport processes are not well explored. Furthermore, different basin management action plans, such as engineering structure modifications, may also greatly affect the flood inundation, sediment transport, solute transport and morphological processes within extreme flood events. In this study, a coupled two-dimensional hydrodynamic, sediment transport and morphological model is applied to evaluate the impact of different river and basin management strategies on the flood inundation, sediment transport dynamics and morphological changes within extreme flood events of different magnitudes. The 10-year, 50-year, 100-year and 200-year floods are evaluated for the Lower Cache Creek system in California under existing condition and a potential future modification scenario. Modeling results showed that select locations of flood inundation within the study area tend to experience larger inundation depth and more sediment is likely to be trapped in the study area under potential modification scenario. The proposed two dimensional flow and sediment transport modeling approach implemented with a variety of inflow conditions can provide guidance to decision-makers when considering implementation of potential modification plans, especially as they relate to competing management strategies of large water bodies, such as the modeling area in this study.