The development of civilization and the preservation of environmental ecosystems are strongly dependent on water resources. Typically, an insufficient supply of surface water resources for domestic, industrial, and agricultural needs is supplemented with groundwater resources. However, groundwater is a natural resource that must accumulate over many years and cannot be recovered after a short period of recharge. Therefore, the long-term management of groundwater resources is an important issue for sustainable development. The accurate prediction of groundwater levels is the first step in evaluating total water resources and their allocation. However, in the process of data collection, data may be lost due to various factors. Filling in missing data is a main problem that any research field must address. It is well known that to maintain data integrity, one effective approach is missing value imputation (MVI). In addition, it has been demonstrated that machine learning may be a better tool. Therefore, the main purpose of this study was to utilize a generative adversarial network (GAN) that consists of a generative model and a discriminative model for imputation. Although the GAN could not capture the groundwater level endpoints in every section, the overall simulation performance was still excellent to some extent. Our results show that the GAN can improve the accuracy of water resource evaluations. In the current study, two interdisciplinary deep learning methods, univariate and Seq2val (sequence-to-value), were used for groundwater level estimation. In addition to addressing the significance of the parameter conditions, the advantages and disadvantages of these two models in hydrological simulations were also discussed and compared. Regarding parameter selection, the simulation results for univariate analysis were better than those for Seq2val analysis. Finally, univariate was employed to examine the limits of the models in long-term water level simulations. Our results suggest that the accuracy of CNNs is better, while LSTM is better for the simulation of multistep prediction. Therefore, the interdisciplinary deep learning approach may be beneficial for providing a better evaluation of water resources.

Rainfall infiltration is the primary triggering factor of slope instability. The process of rainfall infiltration leads to changes in the water content and internal stress of the slope soil, thereby affecting slope stability. The soil water retention curve (SWRC) was used to describe the relationship between soil water content, matric suction, and the water retention characteristics of the soil. This characteristic is essential for estimating the properties of unsaturated soils, such as unsaturated hydraulic conductivity function and shear strength. Thus, SWRC is regarded as important information for depicting the properties of unsaturated soil. The SWRC is primarily affected by the soil pore size distribution (PSD) and has unimodal and bimodal features. The bimodal SWRC is suitable for soils with structural or dual-porous media. This model can describe the structure of micropores and macropores in the soil and allow the hydraulic behavior at different pore scales to be understood. Therefore, this model is more consistent with the properties of onsite soil. Few studies have explored the differences in the impact of unimodal and bimodal models on unsaturated slopes. This study aims to consider unimodal and bimodal SWRC to evaluate the impact of unsaturated slope stability under actual rainfall conditions. A conceptual model of the slope was built based on field data to simulate changes in the hydraulic behavior of the slope. The results of seepage analysis show that the bimodal model has a better water retention capacity than the unimodal model, and therefore, its water storage performance is better. Under the same saturated hydraulic conductivity function, the wetting front of the bimodal model moves down faster. This results in changes in the pressure head, water content, and internal stress of the soil. The results show that the water content and suction stress changes of the bimodal model are higher than those of the unimodal model due to the difference in water retention capacity. Based on the stability of the slope, calculated using the seepage analysis, the results indicate that the potential failure depth of the bimodal model is deeper than that of the unimodal model.

This study used stable hydrogen and oxygen isotopes as natural tracers to investigate their isotopic composition in precipitation, and in shallow and deep groundwater in the Penghu Islands in the Taiwan Strait. We aimed to understand the differences and relationships in isotope compositions within various water bodies and to evaluate the source of groundwater recharge. The hydrogen and oxygen isotopic compositions of sampled groundwater are mainly distributed along the meteoric water line in the Penghu Islands, the variations in the distribution range being minor (the δD values are distributed from −48.2‰ to −37.7‰, with a mean value of −43.14 ± 2.4‰; the δ¹⁸O values are distributed from −6.96‰ to −5.46‰, with a mean value of −6.34 ± 0.34‰). The data suggest that the groundwater is sourced mainly from local precipitation. In addition, a comparison of the hydrogen and oxygen isotopic compositions of groundwater and precipitation in Taiwan shows that the δ values for groundwater are distributed between those for precipitation during the northeast monsoon and the southwest monsoon seasons. However, some of the δ values trends towards the isotopic composition of the precipitation during the southwest monsoon season. Thus, the source of groundwater may have a closer association with precipitation during this time.

Flood disasters often result in significant losses of life and property, making them among the most devastating natural hazards. Therefore, reliable and accurate water level forecasting is critically important. Rainfall-runoff modeling, which is a complex and nonlinear time series process, plays a key role in this endeavor. Numerous studies have demonstrated that data-driven methods, particularly deep learning approaches such as convolutional neural networks (CNNs), long short-term memory (LSTM) networks, and transformers, have shown promising performance in water level prediction tasks. This study introduces the Conformer, a novel deep learning architecture that integrates the strengths of CNNs and transformers for rainfall-runoff modeling. The framework uses self-attention mechanisms combined with convolutional computations to extract essential features—such as water levels, precipitation, and meteorological data—from multiple stations, which are then aggregated to predict subsequent water level series. This study utilized data spanning from 1 April 2006 to 25 July 2021, totaling 5595 days (134,280 h), which were divided into training, validation, and test sets in an 8:1:1 ratio to train the model, adjust parameters, and evaluate performance, respectively. The effectiveness and feasibility of the proposed model are evaluated in the Lanyang River Basin, with a focus on predicting 7-day-ahead water levels. The results obtained from ablation experiments indicate that convolutional computations significantly enhance the ability of the model to capture the local relationships between water levels and other parameters. Additionally, performing convolution computations after executing self-attention operations yields even better results. Compared with other models in simulations, the Conformer model markedly outperforms the CNN, LSTM, and traditional transformer models in terms of the coefficient of determination (R²) and Nash–Sutcliffe efficiency (NSE) indicators. These findings highlight the potential of the Conformer model to replace the commonly used deep learning methods in the field of hydrology.

To enhance the density and precision of groundwater level observations, this study installed two 300-meter multi-channel wells (MCW) in XiHu (Changhua) and YuanChang (Yunlin). These wells can concurrently monitor water levels in 15 layers within the protective tubing. Fifteen sets of 1-inch groundwater observation wells, varying in depths, were placed in the protective pipe. The groundwater level observation equipment utilized two sets with different principles—electronic and optical fiber—to record water levels. The objective was to compare the accuracy and stability of the optical fiber water level measuring equipment. Despite the potential simplification of well construction processes for fiber-optic water pressure gauges, a pressure issue still needs to be addressed. Therefore, the electronic water level meter remains the more reliable option between the two. Pumping tests were conducted at two sites to observe the flow characteristics of each layer in the multi-channel well and the depth changes of the magnetic ring in the multi-level compaction monitoring well (MLCW). During pumping in XiHu #1, the water level in the shallow layer rose due to additional groundwater recharge caused by pumping. The rise in water level in the sub-deep layers is attributed to the gravel composition of the stratum at this depth, facilitating better groundwater flow.The YuanChang standard observation well is divided into five depths. The pumping test results indicate that the aquifer at each depth is less influenced by the upper and lower layers. Water level changes in the MCW, affected by the pumping test, can be categorized into three groups: MCW#1~#5, MCW#6~#12, and MCW#13~#15. The depth changes of the magnetic ring in the MLCW were analyzed during water pumping and recovery. The changing trend of the magnetic ring in the XiHu pumping test site aligns with the pumping and recovery process. The magnetic rings of YuanChang, specifically NO.1 to NO.5 comprising unconfined aquifer layers, exhibit minimal rebound during the recovery stage of Yuanchang #1. The trends in magnetic rings NO.18 to NO.26 align with water pumping and recovery, indicating characteristics typical of a confined aquifer. The water pumping and recovery in YuanChang#3 and YuanChang#4 display a hysteresis effect on the changing trend of magnetic rings NO.1 to NO.7. The MCWs established in this study, along with the pumping test results, can provide more accurate on-site survey data. Coupled with the comparison of MLCW observation results, this can serve as a demonstration site for studying stratigraphic subsidence, significantly contributing to improving scientific research demonstrations.