Search

Modeling hydrological processes for managing the available water resources effectively is often complex due to the existence of high nonlinearity, and the associated prediction uncertainty mainly arising from model inputs, parameters, and structure. Despite several attempts to quantify the model prediction uncertainty, reducing the same for improving the reliability of models is indispensable for their wider acceptance. This paper presents a novel modeling framework for minimizing the prediction uncertainty in the streamflow simulation of the conceptual hydrological model (HBV) by integrating with the Bayesian‐based Particle Filter technique (PF) and machine learning algorithm (Random Forest algorithm, RF). Initially, the streamflow prediction interval (PI) is derived from the stochastically estimated parameters of the HBV model through the PF technique (HBV‐PF model). As the HBV‐PF model quantifies only parametric uncertainty, the RF algorithm was employed (HBV‐PF‐RF model) for further minimizing the prediction uncertainty by inherently taking care of different sources of uncertainty. The RF algorithm inherently combines the physics of the hydrological system (i.e., process‐based variables) with machine learning‐based approach to minimize the overall prediction uncertainty. The proposed framework was analyzed on Nepal and India's Sunkoshi and Beas River basins, through several statistical performance indices for assessing the accuracy and uncertainty of the model prediction. The framework was observed to be consistently improving the model performance minimizing the uncertainty in both watersheds. Therefore, the proposed framework can be considered to be more reliable in improving the prediction capability of hydrological models. Key Points: Development of a physics‐aware machine learning based hydrological model for streamflow simulationThe proposed framework characterizes the model prediction uncertainty, accounting different sources of uncertainty implicit/explicitlySignificant reduction in overall prediction uncertainty is achieved [ABSTRACT FROM AUTHOR]

Occurrences of extreme events, especially floods, have become more frequent and severe in the recent past due to the global impacts of climate change. In this context, possibilities for generating a near‐accurate streamflow forecast at higher lead times, which could be utilized for developing a reliable flood warning system to minimize the effects of extreme events, are highly important. This paper aims to investigate the potential of a novel hybrid modeling framework that couples the random forest algorithm, particle filter, and the HBV model for improving the overall accuracy of forecasts at higher lead times through the dynamic error correction schematic. The new framework simulates an ensemble of streamflow for estimating uncertainty associated with the predictions and is applied across two snow‐fed Himalayan rivers: the Beas River in India and the Sunkoshi River in Nepal. Several statistical indices along with graphical performance indicators were used for assessing the accuracy of the model performance and associated uncertainty. The modeling framework achieved the Nash Sutcliffe Efficiency of 0.94 and 0.98 in calibration and 0.95 and 0.99 in validation for the Beas and Sunkoshi river basin respectively for a 7‐day ahead forecast. Thus, the proposed framework can be considered as a promising tool having reasonably good performance in forecasting streamflow at a higher lead time. Key Points: Hybrid hydrological model integrates process‐based model with machine learning algorithm through data assimilation techniqueDynamic error correction framework capable of improving the streamflow forecast at longer lead time is proposedOverall the developed framework improves the forecast accuracy along with quantifying the model prediction uncertainty [ABSTRACT FROM AUTHOR]

This paper presents a new data-driven modelling framework for forecasting probabilistic flood inundation maps for real-time applications. The proposed end-to-end (rainfall–inundation) method combines a suite of machine learning (ML) algorithms to forecast discharge and deliver probabilistic flood inundation maps with a 3 h lead time. To classify wet/dry cells, the method applies rainfall–discharge models based on random forest technique on top of classifiers based on multi-layer perceptron. The hybrid modelling framework was tested using two subsets of data created from an observed fluvial flood event in a small flood-prone town in the UK. The results showed that the model can effectively emulate the outcomes of a hydrodynamic model (Flood Modeller (FM)) with considerably high accuracy measured in terms of flood arrival time error and classification accuracy. The mean arrival time difference between the proposed model and the hydrodynamic model was 1 h 53 min. The classification accuracy was measured against a synthetic aperture radar image, producing accuracies of 88.22% and 86.58% for the proposed data-driven model and FM, respectively. The key features of the proposed modelling framework are that it is simple to implement, detects flooded cells effectively and substantially reduces computational time. [ABSTRACT FROM AUTHOR]

This paper presents a systematic investigation into modelling capacities of three conventional data-driven modelling techniques, namely, wavelet-based artificial neural network (WANN), support vector regression (SVR) and deep belief network (DBN) for multi-step ahead stream flow forecasting. To evaluate the effectiveness of these modelling techniques, hydro-meteorological hourly datasets from three case-study rivers located in the UK have been used. A heuristic performance analysis of the modelling schemes has been conducted by systematically analysing the key statistics that measure magnitude, scatter and density of model errors. Finally, for each of the modelling techniques, the performance deterioration rate in time was estimated. The results show that the SVR model can forecast quite accurately up to one to two hours ahead but its performance deteriorates gradually from three hours onwards. Further it has been found that the WANN model performs better when the overall non-linearity of the system increases, whereas the DBN model appeared to show consistently poor predictive capabilities when compared to the other models presented herein. The authors conclude by stating that, for any selected model, it is possible to use an identical model structure for up to two steps ahead forecasting. Models need to be re-configured beyond that limit. [ABSTRACT FROM AUTHOR]

Load forecasting necessitates a significant amount of smart meter data. Several elements in this process, including device malfunctions and signal transmission issues, produce missing data gaps. Missing values in the dataset significantly influence the learning ability of machine learning algorithms, and they must be infilled before proceeding with any statistical analysis. This paper investigates the handling of missing values in demand data, and a new approach is developed for improving the performance of demand analytics, such as energy forecasting. The proposed model uses a transformer neural network to impute the missing values at various rates in the demand profile. Our model uses a k-means algorithm to fill in the missing values with proxy values in the dataset. The model is applied to two case-study residential house located in Cornwall and Fintry, United Kingdom. The developed algorithm is assessed for it potential for infilling missing values for three widely understood missing value scenarios: missing completely at random (MCAR), missing at random (MAR), and missing not at random (MNAR). The proposed model's imputed outputs are compared to the original dataset to assess model performance. The performance of the framework is compared with a selection of widely used statistical and machine learning models. The proposed transformer model shows significant improvements over the common linear method in all three scenarios (with 30% missing values), with percentage improvements ranging from approximately 49.71% to 57.52% for Cornwall dataset. • A novel imputation framework is proposed for energy demand data. • Testing the model for infilling missing values with MCAR, MAR, MNAR scenarios • Evaluating the model performance in comparison with established benchmark approaches. • Utilising a transformer network to enhance data imputation performance. • Excellent imputation accuracy and model robustness are validated by two datasets. [ABSTRACT FROM AUTHOR]

• A CNN model is proposed to estimate flood water depths. • The model can predict water depths for over half of a million cells instantly. • The CNN can potentially be used as a surrogate model for real-time applications. • The benefit of using CNN over other ML methods for flood modelling is presented. Most of the two-dimensional (2D) hydraulic/hydrodynamic models are still computationally too demanding for real-time applications. In this paper, an innovative modelling approach based on a deep convolutional neural network (CNN) method is presented for rapid prediction of fluvial flood inundation. The CNN model is trained using outputs from a 2D hydraulic model (i.e. LISFLOOD-FP) to predict water depths. The pre-trained model is then applied to simulate the January 2005 and December 2015 floods in Carlisle, UK. The CNN predictions are compared favourably with the outputs produced by LISFLOOD-FP. The performance of the CNN model is further confirmed by benchmarking against a support vector regression (SVR) method. The results show that the CNN model outperforms SVR by a large margin. The CNN model is highly accurate in capturing flooded cells as indicated by several quantitative assessment matrices. The estimated error for reproducing maximum flood depth is 0–0.2 m for the 2005 event and 0–0.5 m for the 2015 event at over 99% of the cells covering the computational domain. The proposed CNN method offers great potential for real-time flood modelling/forecasting considering its simplicity, superior performance and computational efficiency. [ABSTRACT FROM AUTHOR]