Traditional Hydrological Models Research Articles

Continuous streamflow prediction in ungauged basins: long short-term memory neural networks clearly outperform traditional hydrological models

Abstract. This study investigates the ability of long short-term memory (LSTM) neural networks to perform streamflow prediction at ungauged basins. A set of state-of-the-art, hydrological model-dependent regionalization methods are applied to 148 catchments in northeast North America and compared to an LSTM model that uses the exact same available data as the hydrological models. While conceptual model-based methods attempt to derive parameterizations at ungauged sites from other similar or nearby catchments, the LSTM model uses all available data in the region to maximize the information content and increase its robustness. Furthermore, by design, the LSTM does not require explicit definition of hydrological processes and derives its own structure from the provided data. The LSTM networks were able to clearly outperform the hydrological models in a leave-one-out cross-validation regionalization setting on most catchments in the study area, with the LSTM model outperforming the hydrological models in 93 % to 97 % of catchments depending on the hydrological model. Furthermore, for up to 78 % of the catchments, the LSTM model was able to predict streamflow more accurately on pseudo-ungauged catchments than hydrological models calibrated on the target data, showing that the LSTM model's structure was better suited to convert the meteorological data and geophysical descriptors into streamflow than the hydrological models even calibrated to those sites in these cases. Furthermore, the LSTM model robustness was tested by varying its hyperparameters, and still outperformed hydrological models in regionalization in almost all cases. Overall, LSTM networks have the potential to change the regionalization research landscape by providing clear improvement pathways over traditional methods in the field of streamflow prediction in ungauged catchments.

GHCen: uma abordagem estocástica-conceitual para geração de cenários sintéticos de vazões

ABSTRACT In Brazil, energy production predominantly relies on hydropower generation, necessitating precise hydrological planning tools to manage the uncertainty inherent in river flows. While traditional hydrological models provide valuable deterministic forecasts, addressing the need for probabilistic information remains crucial. This paper introduces a novel approach, the Hybrid Generator of Synthetic Streamflow Scenarios (GHCen), which combines a conceptual SMAP/ONS model with stochastic simulation techniques to generate synthetic streamflow scenarios. The stochastic methodology employed in GHCen effectively reproduces the key characteristics of precipitation processes on daily to annual scales. Through a comprehensive case study, conducted for 2021, GHCen demonstrates its capability to accurately replicate the hydrological behaviors from historical data. The analysis reveals a strong alignment between the synthetic scenarios and observed Natural Energy Inflow for the National Interconnected System, both monthly and in accumulated terms.

Combining Standard Artificial Intelligence Models, Pre-Processing Techniques, and Post-Processing Methods to Improve the Accuracy of Monthly Runoff Predictions in Karst-Area Watersheds

The complex and unique topography of karst regions highlights the weaknesses of traditional hydrological models which fail to fully generalize them. The successive proposals of standard artificial intelligence (AI) models, pre-processing techniques, and post-processing methods have provided new opportunities to enhance the accuracy of runoff prediction in karst areas. In this study, first, the BP neural network model and the Elman neural network model were used for runoff prediction. Then, the performance of four coupled models—formed by combining two AI pre-processing techniques, Empirical Modal Decomposition (EMD) and Ensemble Empirical Modal Decomposition (EEMD), with the previously mentioned AI models—was investigated. Finally, the accuracy of triple-coupled models, formed by applying the post-processing method of quantile mapping (QM) to the previous coupled models, was estimated. The Nash–Sutcliffe efficiency (NSE), the mean absolute percentage error (MAPE), the root mean square error (RMSE), and the peak percentage of threshold statistics (PPTS) were selected to evaluate and analyze the forecasting results of the above models. The results demonstrated that the BP model had the best prediction effect of the standard AI models, the coupled forecasting models had better prediction accuracy than the standard AI models, and the triple-coupled QM–EMD–Elman model had the best forecasting effect with an NSE value of 0.73, MAPE value of 0.75, RMSE value of 34.60, and PPTS value of 2.36.

Research on Power Prediction of Runoff Small Hydropower Cluster Based on Fusion Modeling

The characteristics of the runoff hydropower station are to generate power completely according to the natural runoff, and the output and power generation change with the change of the natural flow. The cluster power prediction of the runoff hydropower station is conducive to improving the safety of the power grid operation and improving the utilization rate of hydropower. This paper analyzes the operation characteristics of runoff hydropower stations, summarizes the prediction direction of mainstream runoff hydropower, analyzes the characteristics of the hydrological prediction model represented by the Xin'an River model, and puts forward a study on the power prediction technology of runoff small hydropower cluster based on fusion modeling. Based on the detailed analysis of watershed runoff characteristics, a prediction model based on multi-watershed observation characteristics is proposed, which combines the traditional hydrological model with the LSTM model to form a combined prediction model suitable for runoff hydropower. The method is universal and innovative.

A Coupled River Basin‐Urban Hydrological Model (DRIVE‐Urban) for Real‐Time Urban Flood Modeling

AbstractReliable urban flood modeling is highly demanded in emergency response, risk management, and urban planning related to urban flooding. In this paper, the Storm Water Management Model (SWMM) is adapted to simulate urban rainfall‐runoff and pipe drainage processes within the Dominant river tracing‐Routing Integrated with VIC Environment (DRIVE) model which accounts for natural river basin runoff generation and routing processes. The integrated DRIVE‐SWMM model (referred to as DRIVE‐Urban) allows to explicitly delineate the mass‐energy interactions between urban drainage system (e.g., pipes and dikes) and river networks. This presents a further step model development for accurate urban flooding prediction which is lacking in existing urban flood models and traditional hydrological models. The validity of the DRIVE‐Urban model is evaluated for three case studies in Haikou City, China, with camera observations of street inundation during typhoon landfalls and heavy rainfall events. The results show that the DRIVE‐Urban model successfully captures 62%, 69%, and 77% of the total observed inundated road‐sections for the three cases respectively. The third case study with severe flooding situation shows that the DRIVE‐Urban performance is further improved when given reliable river and tidal level information, indicating the importance of integrating river‐basin with urban hydrological and hydraulic modeling.

INTEGRATED STORMWATER ANALYSIS MODEL TO SUPPORT SUSTAINABLE URBAN GREEN SPACE DESIGN

Abstract. Urban Green Space (UGS) has been broadly treated as a valuable and limited resource to handle the challenges brought by high-density urban environment. Green stormwater management has been prompt around the world. To reduce the potential conflicts between stormwater management and other requirements on UGS (especially human needs), the research community and governments encourage the urban planners and designers to integrate stormwater analysis into UGS design. However, the professional term and operation of the traditional hydrological model is a huge challenge to the designers who haven’t touched hydrological knowledge. This study developed a method to simulate and quantify stormwater in UGS by the particle system in the design platform- Rhinoceros +Grasshopper. In this method, we adopted five groups of urban objects such as terrain, building footprint etc. to ease procedures and performance. To overcome the issue the abstracted particle movement cannot reflect the infiltration features of land cover, we introduced categorising locations of the particles and the UGS retaining stormwater hypothesis. To test the feasibility of the method, we tested the three parameters of the integrated model: iteration times, rainfall depth ( selected rainfall event) and particle radius. The comparison tests prove: (1) too small iteration times would lead the particles to stop on the way to the bottom of the terrain, so we set at least 4000 iteration times for simulation; (2) the sensitivity to the selection of rainfall event and particle size is relatively low, the simulation results vary by 1%; (3) too small particle numbers will impact the accuracy of analysis, so we should balance accuracy and work efficiency of work; (4) the stormwater volume estimation based on the particle system is acceptable. The experiments confirm the method can effectively support preliminary UGS design work.

Comprehensive Analysis for Long-Term Hydrological Simulation by Deep Learning Techniques and Remote Sensing

Hydrological simulation plays a very important role in understanding the hydrological processes and is of great significance to flood forecasting and optimal allocation of water resources in the watershed. The development of deep learning techniques has brought new opportunities and methods for long-term hydrological simulation research at the watershed scale. Different from traditional hydrological models, the application of deep learning techniques in the hydrological field has greatly promoted the development trend of runoff prediction and provides a new paradigm for hydrological simulation. In this study, a CNN–LSTM model based on the convolutional neural network (CNN) and long short-term memory (LSTM) network, and a CNN–GRU model based on CNN and gated recurrent unit (GRN) are constructed to study the watershed hydrological processes. To compare the performance of deep learning techniques and the hydrological model, we also constructed the distributed hydrological model: Soil and Water Assessment Tool (SWAT) model based on remote sensing data. These models were applied to the Xixian Basin, and the promising results had been achieved, which verified the rationality of the method, with the majority of percent bias error (PBE) values ranging between 3.17 and 13.48, Nash–Sutcliffe efficiency (NSE) values ranging between 0.63 and 0.91, and Kling–Gupta efficiency (KGE) values ranging between 0.70 and 0.90 on a monthly scale. The results demonstrated their strong ability to learn complex hydrological processes. The results also indicated that the proposed deep learning models could provide the certain decision support for the water environment management at the watershed scale, which was of great significance to improve the hydrological disaster prediction ability and was conducive to the sustainable development of water resources.

Development and Evaluation of a Saturated Zone Module in an Integrated Urban Hydrological Model

Shallow urban groundwater interacts with surface water and underground infrastructures. Low-impact development in urban water management by at-source infiltration should consider shallow urban groundwater in a holistic manner. Traditional hydrological models, however, rarely detail groundwater flows and their interaction with urban runoff and the water budget. In the present study, a new approach is proposed, using the integration of a flow module WTI for the saturated zone in a distributed urban hydrological model URBS-MO. This integration is carried out by paying attention to retaining the initial waterflow subsurface parameterization. The performance of the integrated model is evaluated by piezometric and runoff data in an experimental urban catchment, through a sensitivity analysis and a manual calibration of the main model parameters, as well as a validation step. The new module shows its capacity to improve groundwater flow simulation by assessing more realistic water table variations, along with a very small improvement of flowrate simulation. The bias on the average groundwater level was reduced from +14 to +7% for the one-year validation period. The modelling results show the importance of parameter calibration for distributed physically-based hydrological models. Difficulties in the calibration of parameter values due to spatial heterogeneities are also revealed, as the use of piezometric data for the calibration of a hydrological model is rather innovative.

An eco-hydrological model for modelling hydrological processes in a riparian wetland with the unclosed boundary

It has always been one of the key problems of wetland science to explore the ecological-hydrological interaction mechanism of plants in floodplain wetland. However, the current traditional hydrological model cannot reveal effectively and quantitatively the hydrological process from the spatiotemporal method due to the complex multi-interface features of wetland. In this paper, we constructed innovatively a distributed model for Modelling of the Eco-hydrological process between the Interaction of Surface water & Soil water in Wetlands environment (MEISSW) on the unclosed boundary of Honghe National Nature Reserve (HNNR) in Sanjiang Plain, Northeast China. It is the first effort globally to establish the MEISSW model on the aim at simulating the spatiotemporal transformation characteristics of surface water and soil water in a riparian wetland. Four rainfall events with different hydrological and meteorological characteristics were used to calibrate parameters, and the other three rainfall events were used for verification. Results showed the model was generally credible and highly correlated with the Pearson correlation coefficient (PEARSON) as its 0.76 simulation result on the soil water and 0.64 as the NASH coefficient on the surface water. This model is not only helpful to understand the characteristics of the process interaction between plant ecology and water in the interaction zone, but also useful to quantify and predict the change of ecological base flow within the wetland habitats caused by the human disturbance, and its impact on the spatial pattern of the wetland plant community.

Flood forecasting in urban reservoir using hybrid recurrent neural network

Flood forecasting can provide accurate inferences and early warnings for flood control work during the flood season. Due to the variability of local rainfall and the complexity of geographic conditions, existing prediction methods were unable to accurately predict the flooding process in a particular basin. Additionally, the water level sensor generates a significant amount of noise in the inbound flow data during period measurement. To address these issues, this article proposes a real-time flood forecasting model, which is used to accurately predict flood trends and peak times in the flood period. The model uses a convolution kernel function to smooth out local noise and neighborhood values, minimizing the impact of non-stationary series on the training process while retaining the true evolution of the flood in the original data. In our model, we develop a time series attention mechanism that is used to apply various weights to time series input vectors, such as outflow flow and rainfall from upstream reservoirs, this mechanism also improves the accuracy of short-term series prediction. To obtain additional information about the output of the recurrent neural network layer, we also include a multivariate autoregressive integrated moving average module. This method can add a linear component to the output, allowing the prediction result to adapt to the input period's scale shift. This article develops matching models for interval and full basin floods based on the geographical characteristics of China's urban Reservoir and the river basin, thresholds are established based on the outflow from upstream reservoirs, which enables the flood forecasting system to dynamically adjust model parameters in response to the threshold, it also circumvents the scaling problem inherent in flood time series at various scales. We trained and predicted using 25 different types of floods in Ankang Reservoir from 2010 to 2020. Three on-site real-time forecasts of catastrophic flooding at the Ankang Reservoir were conducted in September 2021 to validate the model's accuracy. The algorithm's efficiency in forecasting flood inflows is demonstrated through comparisons to traditional hydrological models and other machine learning networks, and our model consistently forecasts the peak time and total flood volume with the least amount of error in the comparison algorithm.

Effect of digital elevation model spatial resolution on depression storage

AbstractSurface water storage—including wetlands and other small waterbodies—has largely been disregarded in traditional hydrological models. In this paper, the grid resampling method is adopted to study the influence of the digital elevation model (DEM) grid resolution on depression storage (DS) considering different rainfall return periods. It is observed that the DEM grid size highly affects DS, and the higher the grid resolution is, the larger the DS value. However, when the grid resolution reaches a certain value, the maximum DS value decreases. This suggests that a critical grid resolution value exists at which the water storage capacity of depressions is maximized, namely, 20 m in this work (except for the overall area simulation under infiltration). This phenomenon is further verified in two test cases with and without the infiltration process, that is, calculations of the local area and without infiltration area, respectively. This research may facilitate the accurate computation of the DS process, which is greatly affected by the grid resolution, thereby improving the reliability of hydrological models.

Research on flood forecast of dashimen reservoir in xinjiang based on melting snow and runoff yield under excess infiltration

In China, the distribution of water resources is uneven in time and space. The effective allocation of water resources can realize the rational utilization of water resources. Reservoirs are an important part of the basin system, and the rational operation of reservoirs can realize the optimal allocation of water resources. Flood forecast is of great significance for reservoir operation. In order to realize the optimal operation of Dashimen reservoir in Xinjiang, aiming at the flood forecast in the dry area affected by melting snow, this paper expanded and improved the traditional hydrological forecast model, built the flood forecast model in the dry area with the coupling snowmelt based on the northern Shaanxi model and Elman neural network, and using the typical rainfall and runoff data in this area The results show that the relative error of the model is less than 10% and the accuracy is high, which can be used in the actual forecast. This model realizes the flood forecast in the drought area affected by snowmelt, and provides a new idea for the flood forecast of reservoir in the drought area affected by snowmelt, which has a certain guiding significance for the flood forecast in the drought area.

Comprehensive comparison of artificial neural networks and long short-term memory networks for rainfall-runoff simulation

Accurate and efficient runoff simulations are crucial for water management in basins. Rainfall-runoff simulation approaches range between physical, conceptual, and data-driven models. With the recent development of machine-learning techniques, machine learning methods have been widely applied in the field of hydrology. Existing studies show that such methods can achieve comparable or even better performances than conventional hydrological models in runoff simulation. In particular, long short-term memory (LSTM) neural networks are able to overcome the shortcomings of traditional neural network methods in handling time series data. However, the impacts of the time memory on rainfall-runoff simulation are rarely studied. In this study, hysteresis effects in hydrology were investigated and the performances of machine learning methods and traditional hydrological models were assessed. The results show that the ANN model is more suitable for monthly scale simulation, while the LSTM model performs better at daily scale. Hydrological hysteresis is important for runoff simulations when using machine learning methods, especially at daily scale. By considering hysteresis in the simulation, the RMSE is significantly improved by 27% (21%) for LSTM (ANN). In addition, LSTM is more robust for time series handling, while the ANN is easier to be overfitted due to the limitation of neural network structure. This study provides new insights into the potential use of machine learning in hydrological simulations.

Development of an interface-oriented add-in modeling framework for integrated water system simulation and its application

Integrated water system modeling is increasingly used to study the complex relationships between ecohydrological processes and the water-related social-economic systems. However, most of the traditional hydrological models are programmed in monolithic approach, which makes the model improvements or couplings for integrated simulation inconvenient for the inexperienced modelers. In this paper, we develop an interface-oriented add-in modeling framework (IAMF) to enable the water system models to be more reusable, extensible and flexible. The framework is designed using the interface-oriented add-in programming paradigm to provide a “plug-and-play” fashion. Meanwhile, parallel computing is supported easily of the framework based on water system network. We also develop a basic water system model consists of ecohydrological and water resources allocation, regulation modules to show how to develop model components using the framework. Then the model is used to simulate the natural-artificial water cycles in a case study irrigated area to evaluate its applicability as a modeling tool.

Conceptual hydrological model of corrosional hill karst watershed and its application

To mitigate the insufficiency of the traditional hydrological model in karst hydrological simulation, the Xin'anjiang Hydrological Model was modified by reconstructing a heterogeneous groundwater system with linear reservoirs and introducing the virtual area coefficient of the watershed (f) to modify non-conservative characteristics. The modified Karst Hydrological Model was applied to the Chongling River Basin, a typical corrosional hill region of southern Hunan Province, China. In addition, this study focused on dynamic changes of non-conservative characteristics based on the modified model. The results demonstrate the following:① The modified model outperformed the Xin'anjiang Model in karst hydrologic simulation (by 11.21%) especially in water balance (by 23.29%) and low water flow (by 27.64%); ② The non-conservative status of the karst watershed has duality because of the multi-directionality of the underground confluence system; ③ The area of the underground system is positively correlated with precipitation, while the variation of the area decreases with the increase of the precipitation by the constraint of underground saturation; ④ Inter-catchment water exchange exhibits time lag because of the hydration and water retention of karst.

Flood Prediction Using Rainfall-Flow Pattern in Data-Sparse Watersheds

Real-time flood forecasting of small- and medium-sized rivers in areas with scarce hydrological data is an urgent problem that needs to be solved. Traditional hydrological model parameters cannot be fully trained owing to a lack of data; thus, results obtained by such models are not satisfactory. We need a new way to solve the forecasting problem for small- and medium-sized rivers. We found that the time series of some feature variables have evident change trajectories in spatial dimension, and the change of some feature variables in the spatial dimension has a decisive influence on flooding processes, such as the spatial distribution of rainfall. To reflect the change of feature variables in spatial dimension with to solve the problem of the lack of hydrological data, we constructed a rainfall-flow pattern composed of a spatial-temporal dynamic time warping algorithm and multi-feature algorithm to measure the similarity of hydrological time series. In the experimental watersheds, we used rainfall-flow patterns to forecast the short-term flood streamflow, and satisfactory results were obtained. This suggests that the algorithm is suitable for hydrological studies and improves the accuracy of real-time flood forecasting for longer forecast periods.

Towards learning universal, regional, and local hydrological behaviors via machine learning applied to large-sample datasets

Abstract. Regional rainfall–runoff modeling is an old but still mostly outstanding problem in the hydrological sciences. The problem currently is that traditional hydrological models degrade significantly in performance when calibrated for multiple basins together instead of for a single basin alone. In this paper, we propose a novel, data-driven approach using Long Short-Term Memory networks (LSTMs) and demonstrate that under a “big data” paradigm, this is not necessarily the case. By training a single LSTM model on 531 basins from the CAMELS dataset using meteorological time series data and static catchment attributes, we were able to significantly improve performance compared to a set of several different hydrological benchmark models. Our proposed approach not only significantly outperforms hydrological models that were calibrated regionally, but also achieves better performance than hydrological models that were calibrated for each basin individually. Furthermore, we propose an adaption to the standard LSTM architecture, which we call an Entity-Aware-LSTM (EA-LSTM), that allows for learning catchment similarities as a feature layer in a deep learning model. We show that these learned catchment similarities correspond well to what we would expect from prior hydrological understanding.

Hydrological Early Warning System Based on a Deep Learning Runoff Model Coupled with a Meteorological Forecast

The intensification of the hydrological cycle because of global warming raises concerns about future floods and their impact on large cities where exposure to these events has also increased. The development of adequate adaptation solutions such as early warning systems is crucial. Here, we used deep learning (DL) for weather-runoff forecasting in región Metropolitana of Chile, a large urban area in a valley at the foot of the Andes Mountains, with more than 7 million inhabitants. The final goal of this research is to develop an effective forecasting system to provide timely information and support in real-time decision making. For this purpose, we implemented a coupled model of a near-future global meteorological forecast with a short-range runoff forecasting system. Starting from a traditional hydrological conceptual model, we defined the hydro-meteorological and geomorphological variables that were used in the data-driven weather-runoff forecast models. The meteorological variables were obtained through statistical scaling of the Global Forecast System (GFS), thus enabling near-future prediction, and two data-driven approaches were implemented for predicting the entire hourly flow time-series in the near future (3 days), a simple Artificial Neural Networks (ANN) and a Deep Learning (DL) approach based on Long-Short Term Memory (LSTM) cells. We show that the coupling between meteorological forecasts and data-driven weather-runoff forecast models are able to satisfy two basic requirements that any early warning system should have: The forecast should be given in advance, and it should be accurate and reliable. In this context, DL significantly improves runoff forecast when compared with a traditional data-driven approach such as ANN, being accurate in predicting time-evolution of output variables, with an error of 5% for DL, measured in terms of the root mean square error (RMSE) for predicting the peak flow, compared to 15.5% error for ANN, which is adequate to warn communities at risk and initiate disaster response operations.

Model Uncertainty Analysis Methods for Semi-Arid Watersheds with Different Characteristics: A Comparative SWAT Case Study

Distributed hydrological models play a vital role in water resources management. With the rapid development of distributed hydrological models, research into model uncertainty has become a very important field. When studying traditional hydrological model uncertainty, it is very common to use multisite observation data to evaluate the performance of the model in the same watershed, but there are few studies on uncertainty in watersheds with different characteristics. This study is based on the Soil and Water Assessment Tool (SWAT) model, and uses two common methods: Sequential Uncertainty Fitting Version 2 (SUFI-2) and Generalized Likelihood Uncertainty Estimation (GLUE) for uncertainty analysis. We compared these methods in terms of parameter uncertainty, model prediction uncertainty, and simulation effects. The Xiaoqing River basin and the Xinxue River basin, which have different characteristics, including watershed geography and scale, were used for the study areas. The results show that the GLUE method had better applicability in the Xiaoqing River basin, and that the SUFI-2 method provided more reasonable and accurate analysis results in the Xinxue River basin; thus, the applicability was higher. The uncertainty analysis method is affected to some extent by the characteristics of the watershed.

Comparing the Performance of the Maximum Entropy Production Model With a Land Surface Scheme in Simulating Surface Energy Fluxes

AbstractHydrological projections under future climate change have been shown to be sensitive to the formulation of evapotranspiration. Many hydrological models still rely on empirical formulations of this flux, and hence do not take into account the surface energy budget. On the other hand, land surface schemes, which are used within the climate models to describe land hydrology and associated surface heat fluxes (SHF), rely on the energy conservation. Due to land surface schemes complexity, they are not suitable for integration in traditional hydrological models. A newly developed and relatively simple Maximum Entropy Production model, which operates under the constraint of energy conservation and allows for an appropriate partitioning of available energy into SHF, appears to be a good alternative for integration in hydrological models. However, the MEP's performances still need to be evaluated under various environmental conditions. This study aims to evaluate the SHF simulated by MEP using observations over a multiyear period from several carefully chosen snow‐free sites located in low‐latitude regions. Moreover, simulated fluxes were compared to those derived from the Canadian Land Surface Scheme (CLASS), which was run at the same sites. The analysis of simulated and observed fluxes associated with different water stress conditions suggests that the abilities of MEP and CLASS in estimating sensible and latent heat fluxes are comparable. It was also found that the MEP and CLASS fluxes are in a good agreement with observations. However, the simulated nocturnal fluxes show that both models are in less agreement with the observations.