Journal of Hydrologic Engineering

The evaporation (EV) method is a widely used experimental approach to determine unsaturated hydraulic conductivity (UHC) and soil water retention (SWR) functions. However, the EV method has some limitations in evaluating accurate UHC values near the wet range. This restricts the working range of the models that utilize these UHC values for predicting soil water content and water fluxes for problems involving wet conditions (e.g., infiltration under ponding conditions). Our experimental study shows that when using the EV method, tensiometer offset errors as low as 1 cm can cause problems with wet-range measurements. Current methods to overcome these errors are expensive, inefficient, or cumbersome. In this study, we developed an improved EV method that utilizes a simple experimental protocol for standardizing the soil saturation procedure and a robust calibration method to quantify tensiometer offset errors. The proposed approach utilizes existing EV method hardware without requiring additional specialized instrumentation. Experiments performed using a silt loam soil sample show that our approach can extend the working range of the EV method to the wet range. Therefore, reliable estimates of the UHC and SWR functions can now be obtained over the entire range of soil suction heads. Our study also contributes to a better understanding of the EV method and uses this knowledge for collecting reliable and reproducible SWR and UHC data. The improved EV method is simple, effective, and cost-efficient. This method has the potential to greatly improve soil water content predictions, leading to better-informed decisions for groundwater resources management, sustainable agriculture, and environmental protection.

Hydrologic systems are characterized by processes which may occur on, below, and above the land surface. These processes are irreversible, and the irreversibility produces entropy. In thermodynamics, entropy is a measure of the loss of heat energy which is a reflection of disorder. The entropy of a system achieves its maximum at steady state, meaning that the maximum entropy production corresponds to the most probable state. In information theory, entropy is a measure of uncertainty or disorder or information imbued in the random variable describing the hydrologic system. The maximum uncertainty corresponds to maximum entropy or the most probable distribution of the variable which leads to the principle of maximum entropy or the principle of minimum cross entropy, subject to the given constraints. The most probable or maximum entropy-based distribution is confirmed by the theorem of concentration. The form of entropy (Shannon, Tsallis, Renyi, or Kapur), principle of maximum entropy, principle of minimum cross-entropy, and the concentration theorem constitute the theory of entropy. This paper presents a general framework based on the entropy theory, and demonstrates its application for modeling a number of surface and subsurface hydrologic and water quality processes, including hydrometric network evaluation, eco-index, surface runoff, infiltration, soil moisture, velocity distribution, sediment concentration, sediment discharge, sediment yield, channel cross section, rating curve, and debris flow.

Abstract

Given the importance of input data, particularly precipitation, in hydrologic modeling for streamflow simulation, there has been growing emphasis on developing frameworks that harness multiple data sources concurrently to achieve more precise results. In the proposed framework of this study, which relies on the integrated capabilities of the multiobjective optimization model [nondominated sorting genetic algorithm-II (NSGA-II)], the ensemble Kalman filter data assimilation method, and data fusion, rainfall data from multiple sources are incorporated. The utilized framework leads to an improvement in the mean absolute relative error (MARE) index of streamflow simulation results. The innovation of the proposed methodology is the calculation of optimal weights corresponding to the simulated runoff time-series in the fusion model. This is accomplished through a competitive process among a multitude of optimized scenarios simulated within the framework provided. MARE as the main index identified in the objective functions and standard deviation, centered root mean square distance, and the correlation coefficient as auxiliary indices have been considered in this process. In this framework, satellite-based and in situ precipitation data sets are used as the forcing data. The main challenge has been to choose the greatest scenario for fusion among the selected scenarios, which the proposed methodology has overcome. The performance of the suggested methodology is demonstrated for the Siakh-Darengon catchment located in the Fars Province of Iran. According to the results, an average of 14.07% improvement in the MARE index has been achieved after applying the proposed methodology. By utilizing the proposed method, satellite-based rainfall data are integrated alongside ground-based rainfall data in the flood modeling process, resulting in enhanced accuracy in simulation outcomes within the utilized watersheds.

Practical Applications

Today, influenced by factors such as climate change and anthropogenic alterations to the environment, the issue of flooding and its associated hazards has garnered unprecedented attention from researchers. One of the crucial components in flood modeling is rainfall data, which are collected through various means such as ground stations and satellite sensing instruments. In the past, the primary focus in the process of flood modeling has been on rainfall data recorded at ground stations; nowadays, efforts are being made to further enhance the role of satellite-derived rainfall data in flood modeling, aiming at enhancing their precision. In this study, a fusion model has been developed using the data fusion method and simultaneous utilization of ground-based and satellite rainfall data. Various flood simulation scenarios have been generated using a multiobjective optimization model, and the best scenario is selected through a competitive process. By implementing the proposed methodology in the Siakh-Darengon watershed located in Fars Province, Iran, improvements in simulation results have been achieved, resulting in based on the calculated performance indicators.

Abstract

Seawater intrusion (SWI) is a concern for coastline-saturated aquifers because it increases salinity, which contaminates and degrades groundwater quality. Rising sea levels and other natural processes, as well as anthropogenic activities such as overpumping, are the causes of intrusion. Because these experiments are the most reliable methods for study, this study focused on the experimental investigation of SWI in a hypothetical stratified aquifer with parallel and perpendicular stratification and a sloping beach boundary. Stratified media were formed by stacking layers in parallel and perpendicular layers in different combinations. Sensitivity analysis was performed with 10-, 20-, and 30-cm gaps for interactions with 4-layer formations with height to toe length expanding from 0.12 to 2.33. High-resolution images were captured at specified intervals and demarcated for a clear visualization of intrusion. A homogenous base case was used to evaluate the feasibility of the proposed method and the results were compared with those of a sandbox apparatus and found to be consistent in terms of the shape of the wedge and the toe length. The height of the intrusion ranges from 28.3 to 32.2 cm. The intrusion was fast and accurate due to the higher permeability of the material. It was concluded that the toe length–height ratio varies from 0.43 to 7.0. For materials with lower permeabilities, this ratio is small because the intrusion is slow but steady in both directions; however, for materials with higher permeabilities, this ratio corresponds to a higher value. The intrusion is restricted to the individual layer until it encounters an impervious layer and follows a proportional pattern with an increasing area of intrusion as the opening area increases.

Practical Applications

This study focused on determining the intrusion behavior in coastal aquifers with monotonically increasing permeability under a layered stratification condition and subjected to the application of inclined boundaries. The study revealed that the use of a glass box apparatus with an inclined ocean–aquifer boundary is the best practical approach for studying intrusion. The occurrence of intrusion due to density differences and heterogeneity represents a case of passive seawater intrusion. The heterogeneity of the aquifer varies at the regional scale; thus, the results obtained from this study may be applied to real-world scenarios by implementing upscaling, temporal dynamics, heterogeneity, and desired boundary conditions. To quantify the intrusion in parallel stratification, a ratio of toe length to height of the intrusion is devised, which is the best approach to follow. The intrusion in the parallel configuration directly depends on the face of the inclined formation subjected to seawater, which forms a parabolic profile, while for the perpendicular configuration, the potential to intrude decreases with each layer as the intrusion is subjected to one layer at a time, initially leading to a parabolic profile that changes to linear in subsequent layers.

Floods, as major natural disasters, cause massive property destruction and death. Understanding the occurrence time of this event by advance notice helps consider operational flood prevention systems and platforms. Precise flood forecasting provides a suitable time for policymakers and the public to consider helpful responses to this event. This study introduces an innovative methodology to enhance the precision of long-term flood predictions by employing a multistep forecasting approach. Our approach leverages historical time-series data on precipitation and streamflow to train an autoencoder algorithm. The primary objective is to develop advanced forecasting models to predict 12-step-weekly ahead flood occurrences during the critical April to July period from 2019 to 2021 within the DuPage River basin, Illinois, USA. In order to achieve this goal, we explore three deep learning techniques: bidirectional long short-term memory (BI-LSTM), ensemble long short-term memory (E-LSTM), and ensemble long short-term memory-gated recurrent unit (E-LSTM-GRU). Then, we integrate an attention mechanism (AM) that utilizes dynamic fusion techniques to emphasize the salient features of ensemble models. A dedicated fusion model is developed for each forecasting stage, effectively consolidating the predictions from various deep-learning models. Additionally, two traditional machine learning techniques, namely MLP and SVM models, are used to compare and justify the efficiency of applied deep learning models. The performance evaluation of our approach using statistical error metrics, including coefficient of determination ($R^2$), normalized root mean square error, normalized mean absolute error, mean absolute percentage error, Nash–Sutcliffe efficiency, Kling–Gupta efficiency, and percent bias, for the 12th step prediction reveals impressive results, with average values of 0.976, 2.393, 1.892, 20.956, 0.967, 0.923, and 2.307, respectively. These findings underscore the capability of our proposed models to significantly reduce uncertainty in flood forecasting, thus enhancing the reliability and accuracy of future predictions.

The curve number (CN) equation is a semiempirical, event-based model commonly used to estimate rainfall runoff. This model was originally developed in the 1950s to estimate storm runoff from 24-h rainfall events from small catchments. The model consists of two parameters: (1) the curve number ($CN$), which represents soil type, land use, and land cover; and (2) the initial abstraction ($Ia$), i.e., the amount of rain that needs to accumulate before storm runoff begins. Despite its narrow-intended use, the CN model is widely used for many applications from engineering design to hydrologic modeling and uses parameter tables and guidelines developed in the mid-20th century. Changes in land management and hydrological science pose questions about the continued relevancy of the model in general and the tabulated parameters specifically. We used Catchment Attributes and Meteorology for Large-Sample Studies (CAMELS), a recently collated data set of watershed characteristics and performed regression analyses on the watershed attributes to determine whether the $CN$ and $Ia$ parameters can better fit a wider range of attributes than can the currently used tables. Our analyses focused on 5–35 year peak runoff events. We considered 333 small to medium watersheds distributed across the contiguous US and more than 40 watershed characteristics. We found that the CN model generally worked best if $Ia$ was much smaller than traditionally assumed. Indeed, $Ia=0$ generally worked well. We also found that $CN$-values generally correlated well with climate (elevation, average precipitation) and soil permeability (sand fraction, saturated hydraulic conductivity). Our results suggest that the CN model can work relatively well for engineering purposes in ungauged watersheds and that the expanding stream of remotely sensed geographic data may allow for better $CN$-values than those from the current tables. We suggest that this study be expanded to include a wider range of watershed and storm characteristics.

This study attempts to predict the maximum temperature, along with the number of heat wave days (HWDs) at lead times of 7 and 15 days over Rajasthan, i.e., a semiarid region, using four machine learning (ML) algorithms, namely, artificial neural networks (ANN), support vector regression (SVR), random forest (RF), and eXtreme gradient boosting (XGBoost). It uses five key atmospheric variables, i.e., air temperature, geopotential height, relative humidity, U-wind, and V-wind, to predict the daily maximum temperatures for the months of April, May, and June, for the period from 1991 to 2020. The ML models are developed by using spatially averaged atmospheric variables and daily maximum temperature. The study demonstrates decent accuracy in forecasting the total annual count of HWDs and daily maximum temperature for Rajasthan for the 7-day lead time. However, as the lead time extends to 15 days, the model performances experience a decline. While comparing the performances of the four models, the SVR outperforms ANN, RF, and XGBoost in prediction. The findings of the current study indicate the potential utility of spatiotemporal dynamics in meteorological variables for long-term heat wave prediction. Moreover, the successful performances of the considered models, i.e., ANN, SVR, RF, and XGBoost, exhibit substantial future potential for reliable application in this context.

Infiltration models are vital for predicting the onset and magnitude of flash floods. A multitude of infiltration models have been developed over the past century. In practice, model selection for a given study constitutes a compromise between complexity, parsimony, and feasibility. This study presents a comparative evaluation of four infiltration models of varying complexity available in the widely used HEC-HMS software: the curve number (CN), initial-constant (IC), Green-Ampt (GA), and the recently integrated linear-constant (LC) model. Using precipitation and runoff data from the Walnut Gulch Experimental watershed in Arizona, each model was analyzed across three different scenarios including, using published guidance, constrained calibration, and unconstrained calibration. Findings reveal that GA, the most complex model, excelled over the rest under published guidance and constrained calibration scenarios but was susceptible to equifinality issues under unconstrained calibration. Conversely, the simpler IC model performed well under unconstrained calibration. The results indicate the simple, two-parameter LC model offers a balance between complexity, parsimony, and feasibility in hydrologic modeling, particularly for arid and semiarid regions. This study provides valuable insight to hydrologists and researchers in selecting an appropriate infiltration model based on data availability and regional hydrologic conditions.

Hydrological models are pivotal tools for comprehending and managing water resources in various applications, including urban and hydrological planning. Nevertheless, these models often grapple with challenges due to the unavailability and sensitivity of specific data required for calibration. This paper aims to explore machine learning algorithms as a potential alternative to traditional hydrological models for flow estimation, focusing on the case study of the dam Três Marias-MG in Brazil. Hybrid models merging spatial data processing with deep learning were investigated to determine the optimal data configuration for enhanced learning. The study area was partitioned into 91 quadrants, with distances from each quadrant’s centroid to the flow point calculated. With an estimated velocity of $1\mathrm{m}/\mathrm{s}$, the lag time, denoting the time each quadrant takes to contribute to the flow, was established. This information facilitated the adjustment of the input precipitation data vector to align with the flow contribution output. The model achieved an average error of approximately 4.4% and a relative peak flow of 19.7% compared with 5.1% and 40% with the regular model, making it a robust approach to flow estimation.