Shihmen Reservoir Research Articles

Utilizing Artificial Intelligence Techniques for a Long–Term Water Resource Assessment in the ShihMen Reservoir for Water Resource Allocation

Accurate long–term water resource supply simulation and demand estimation are crucial for effective water resource allocation. This study proposes advanced artificial intelligence (AI)–based models for both long–term water resource supply simulation and demand estimation, specifically focusing on the ShihMen Reservoir in Taiwan. A Long Short–Term Memory (LSTM) network model was developed to simulate daily reservoir inflow. The climate factors from the Taiwan Central Weather Bureau’s one–tiered atmosphere–ocean coupled climate forecast system (TCWB1T1) were downscaled using the K–Nearest Neighbors (KNN) method and integrated with the reservoir inflow model to forecast inflow six months ahead. Additionally, Multilayer Perceptron (MLP) and Gated Recurrent Unit (GRU) were employed to estimate agricultural and public water demand, integrating both hydrological and socio–economic factors. The models were trained and validated using historical data, with the LSTM model demonstrating a strong ability to capture seasonal variations in inflow patterns and the MLP and GRU models effectively estimating water demand. The results highlight the models’ high accuracy and robustness, offering valuable insights into regional water resource allocation. This research provides a framework for integrating AI–driven models with Decision Support Systems (DSSs) to enhance water resource management, especially in regions vulnerable to climatic variability.

Reservoir Mud Releasing May Suboptimize Fluvial Sand Supply to Coastal Sediment Budget: Modeling the Impact of Shihmen Reservoir Case on Tamsui River Estuary

AbstractRegular release of sediment from reservoir has been increasingly adopted as a strategy for sustainable management. Here, we use a process‐based morphodynamic model to simulate the estuarine sediment dynamics impacted by turbidity current venting implemented by the Shihmen Reservoir during three typhoon events in 2008. Upon validation with the post‐event bathymetries, the model hindcasts reveal that mud releasing can be effective in mitigating reservoir siltation, yet may be a suboptimal strategy for alleviating coastal sediment deficit. A vast majority of the released muds were delivered through the estuary and exported to offshore by flood advection, wave dispersion, and tidal flushing. The flood‐driven sands, sourced mainly from downstream tributaries, were instead the major contributor to coastal sediment budget. However, mud mantling (covering and immobilizing sand deposits by the reservoir‐released muds) reduced sand availability and thus sand delivery to the coast. For the present case, 25% of the released muds were deposited along the way, presence of these mud covers reduced sand delivery by 15%, compared to a hypothetical scenario of clear‐water flood releases. The relative sand transport deficit is found to increase linearly with the degree of bed mud saturation, 1–D/R, with D/R the ratio of single‐event mud deposit to release. Given broad relevance to global reservoirs encountering the problems of siltation and coastal sediment deficit, our findings highlight that sustainable management needs to look beyond just a bulk amount of sediment, but it is critical to consider how different sediment fractions are interacting and impacted by human activities.

Incorporating a backward-forward stochastic particle tracking model into a hydraulic modeling framework to identify probable sedimentation sources during typhoons

Study regionShih-men Reservoir, Taiwan Study focusThe primary purpose of this research is to integrate the novel backward-forward stochastic particle tracking model (BF-SPTM) into a three-dimensional sediment transport model for the Shih-men Reservoir. This integration aims to simulate the distribution of sediment during typhoon events and effcientlyidentify probable sedimentation sources . Additionally, a Genetic Algorithm (GA) is developed to optimize the critical parameters of the EFDC hydrodynamic module and sediment transport module. New hydrological insightsIn the proposed BF-SPTM, the influence function is applied for the first time in a three-dimensional hydraulic model to integrate the backward and forward particle tracking processes, thereby providing more precise probable sources. In this study, a 'probable source' is considered a grid cell rather than a point, which substantially reduces the computational time in the BF-SPTM model. A case study using BF-SPTM during Typhoon MEGI demonstrates its capabilityto identify probable sources in natural water using the EFDC model. The output of the BF-SPTM can be effectively used to identify potential sources of sedimentation. Additionally, this study incorporated more realistic boundary and flow conditions from a natural river to enhance the practical application of BF-SPTM. These conditions included flow velocity, diffusivity, water surface, and bed topography.

Estimating Typhoon-Induced Maximum Flood for Spillway Safety Assessment—Case Studies in Taiwan

Dam safety assessment is usually conducted regularly to investigate the risks associated with the dam and propose remedies to ensure effective reservoir operations. One crucial aspect of the evaluation involves performing hydrological analyses to determine if the existing spillway can successfully deliver the probable maximum flood (PMF) downstream. This study applied storm transposition and typhoon rainstorm methods for PMP estimations. The resulting PMP values were then used as input for runoff models to generate flood hydrographs for PMF determination. A modification for the storm transposition method was proposed to determine the barrier height considering the moisture inflow direction. In estimating the orographic rainfall in the typhoon rainstorm model, an effective terrain slope was suggested according to different windward directions based on DEM analysis. Shihmen Reservoir and Feitsui Reservoir, located in northern Taiwan, were used as examples to conduct the PMP and PMF analysis. The obtained PMPs were further compared with the results generated by Hershfield’s method to assess the reasonability of the estimation. The results show that the maximum deviation of the 24-h PMP values estimated by the three methods is within 30% in the Shihmen watershed and 16% in the Feitsui watershed. The PMF estimations in the two reservoir watersheds are larger than the design discharges of the existing spillways. Hence, extending the capacity of the existing spillways or constructing upstream bypasses to avoid catastrophic flooding downstream is required.

Risk-based irrigation decision-making for the Shihmen Reservoir Irrigation District of Taiwan

Risk-based irrigation decision-making for the Shihmen Reservoir Irrigation District of Taiwan

Uncertainty Analysis for Image-Based Streamflow Measurement: The Influence of Ground Control Points

Large-scale particle image velocimetry (LSPIV) provides a cost-effective, rapid, and secure monitoring tool for streamflow measurements. However, surveys of ground control points (GCPs) might affect the camera parameters through the solution of collinearity equations and then impose uncertainty on the measurement results. In this paper, we explore and present an uncertainty analysis for image-based streamflow measurements with the main focus on the ground control points. The study area was Yufeng Creek, which is upstream of the Shimen Reservoir in Northern Taiwan. A monitoring system with dual cameras was set up on the platform of a gauge station to measure the surface velocity. To evaluate the feasibility and accuracy of image-based LSPIV, a comparison with the conventional measurement using a flow meter was conducted. Furthermore, the degree of uncertainty in LSPIV streamflow measurements influenced by the ground control points was quantified using Monte Carlo simulation (MCS). Different operations (with survey times from one to nine) and standard errors (30 mm, 10 mm, and 3 mm) during GCP measurements were considered. Overall, the impacts in the case of single GCP measurement are apparent, i.e., a shifted and wider confidence interval. This uncertainty can be alleviated if the coordinates of the control points are measured and averaged with three repetitions. In terms of the standard errors, the degrees of uncertainty (i.e., normalized confidence intervals) in the streamflow measurement were 20.7%, 12.8%, and 10.7%. Given a smaller SE in GCPs, less uncertain estimations of the river surface velocity and streamflow from LSPIV could be obtained.

Multi-source rainfall merging and reservoir inflow forecasting by ensemble technique and artificial intelligence

Study regionThe Shihmen reservoir, the second largest reservoir in northern Taiwan, is flooded frequently, and featured by short period of flood peak due to uneven distribution of rainfall and mountainous topography. Study focusWe proposed a novel streamflow-oriented ensemble recurrent neural networks (ERNN) method to merge three precipitation products, namely gauge measurements, radar and satellite rainfall products. We analyzed (1) the effect of artificial intelligence method for bias correction of precipitation products with reference to streamflow, (2) the comparison of two widely used arithmetic average (AA) and Bayesian model averaging (BMA) methods with ERNN, (3) the performance of merged rainfalls and the original three precipitation products in inflow forecasting during 13 typhoon events. New hydrological insightsWe found that all precipitation products are biased and can be appropriately adjusted with an improvement in inflow forecasting of about 2.5 %, 21.8 % and 60.7 % for gauge, radar and satellite rainfall products in terms of RMSE. After rainfall merging, RMSE for radar and satellite are further improved by 14% and 36% at least. ERNN merging method indicates that the optimal merging weights for gauge, radar and satellite are within the range of 0.52–0.56, 0.35–0.39 and 0.08–0.09, respectively according to 95 % confidence interval. The merged rainfall product skillfully predicts the inflow with a lead time of five hours, and ERNN merging method is superior to traditional AA and BMA methods, with NS up to 0.85 and CRPS reduced by near 50 % compared to BMA. Additionally, ERNN is found to capture the peak times and peak flows with the smallest error. Overall, this study provides a potential approach to merge multiple rainfall products and to obtain the effective rainfall by fitting the hydrological responses over mountainous watersheds, where observational biases may frequently occur in gauge, radar and satellite measurements.

Numerical Investigation of Sediment Flushing and Morphological Changes in Tamsui River Estuary through Monsoons and Typhoons

The removal of reservoir silt and the restoration of existing reservoir capacities through land excavation and hydraulic sediment flushing have become necessary. Hydraulic sediment flushing discharge changes flow and sediment conditions of the downstream river channel. In the Tamsui River estuary in Taiwan, sediment flushing from the Shihmen reservoir upstream has potential impacts on the morphology of the navigation channels and the adjacent coasts. This study employed a validated coastal and estuarine processes model to investigate: (1) the influence of sediment flushing and tidal levels on morphological changes during flood and flushing-discharge operations of the reservoir, and (2) the differences in morphological changes on the estuary between monsoon and typhoon seasons. The prediction of the morphological changes was carried out by simulating hydrodynamic and morphodynamic processes under multi-year synthetic conditions combined by northeast monsoon and three historical typhoon events. The simulation results reveal that during the operation of sediment flushing when the peak discharge of river flood flow reaches the estuary section at ebb tides, more sediment can be transported to the open sea than that at flood tides. Additionally, the nature reserve area on the left bank of the estuary is eroded during monsoon and silted in typhoon seasons.

Event-based landslide susceptibility models in Shihmen watershed, Taiwan: accounting for the characteristics of rainfall events.

When performing a landslide susceptibility analysis, a model is usually established on the basis of a multi-temporal or event-triggered landslide inventory. Because multi-temporal landslide inventories for most areas are rarely available, an event-triggered landslide inventory is often used, but the result depends on the selection of single event. In order to establish a landslide susceptibility model with a good prediction performance, the present study tried to find out how to select a single event-triggered landslide inventory, and investigated the effect of various combinations of event inventories. We selected Shihmen reservoir watershed as the research area, conducted a logistic regression analysis to build 23 event-based landslide susceptibility models and one multi-year landslide susceptibility model, and estimated the performance of these models. In addition, this study further assessed the influence of event characteristics on the model prediction performance, used the above results to merge two different events, and then established models based on these combinations. The results indicated that when establishing an event-based landslide susceptibility model, selecting events with suitable rainfall return periods and landslide density can yield robust models with relatively high predictive ability. Furthermore, the combination of two events which negatively correlate with each other in rainfall spatial distributions can enhance a model's predictive ability and modeling efficiency.

Using Fuzzy Neural Networks to Model Landslide Susceptibility at the Shihmen Reservoir Catchment in Taiwan

Machine learning algorithms are commonly employed in landslide susceptibility assessments. Recently, algorithms that utilize artificial intelligence have come into prominence. This study attempts to adapt the most fundamental framework of deep learning and introduces fuzzy theory concepts to analyze landslide susceptibility while updating the network parameters with trial-and-error methods. The final analysis results will compare with those of logistic regression (LR). In order to assess the ability of the model to identify landslides in a more objective way, two typhoon events were used as a training event and a validation event, respectively. The results of the analysis show that the area under the curve (AUC) of the fuzzy neural network (FNN) for the training event is 0.915, but the AUC for the validation event drops to 0.746. Although the results of the FNN for training events were better than those of LR, they did not differ much from those of LR in predicting future events. The reason for this is that the difference between the landslide distributions of the training and validation events is too large, making the model biased in its identification. Overall, FNN is still a recommended method for analyzing landslide potential and can be used as a reference for LR.

Reservoir Sediment Management and Downstream River Impacts for Sustainable Water Resources—Case Study of Shihmen Reservoir

Sustainable water resources of reservoirs depend on preserving the valuable storage capacity. Sediment management is a crucial task in reservoir operations. Extreme floods caused by typhoon events have brought a massive amount of sediments from the watershed of Shihmen Reservoir in Taiwan. In the case study of Shihmen Reservoir, the primary purpose of the sediment management strategies is to minimize sediment deposition and recover reservoir capacity. Two assessment indexes, the capacity–inflow ratio (CIR) and the capacity–sediment ratio (CSR) are investigated to provide a feasible assessment of desilting techniques. Three desilting projects have been planned and implemented in progress, including one modified power plant penstock and two desilting tunnels. Without effective sediment management strategies, the projected storage capacity may drop to 32.9% of the initial storage capacity in the next fifty years. On the other hand, if sediment management is implemented as per the project schedule, 70.3% of the initial storage capacity may be retained, enabling the provision of sustainable water supplies to meet projected water demands. In investigating the impacts caused by the desilting operations, the floodwater level and riverbed variations along the downstream river of the dam are simulated by the hydromorphological model. From simulated results, the downstream river morphology may adjust toward the equilibrium state under the long-term desilting operations of Shihmen Reservoir. Although relatively high sediment concentration is released during the desilting operation, it decays gradually along the downstream river. In addition, riverbed deposition does not significantly affect the floodwater level caused by the increment of riverbed elevation in desilting operations.

Water Injection Dredging for improving and preserving reservoir storage capacity: modelling and measuring tools

Water Injection Dredging (WID) has been successfully applied for removing sediment deposits in reservoirs, which results in an increase of their storage capacity. This dredging method is based on the fluidization of the top sediment layer by pressurized injection of water by a dredging vessel. The fluidized sediment can be transported towards the dead storage of the reservoir or sluiced out of the reservoir through the bottom outlets of a dam. This flow can either occur by gravity induced flow or especially directed by the dredging strategy of the WID vessel. This dredging technique can increase the water storage capacity of the reservoir and prevent the erosion of the river downstream, hence the sediment blockage. Recent developments in modelling and measuring tools have enabled stakeholders to design, optimize and monitor WID in reservoirs. In this paper, we will demonstrate how modelling and measuring tools can be used to evaluate alternative dredging strategies for reservoir maintenance. In particular, we show how a mid-field and far-field modelling can be applied for designing WID actions and predicting sediment plume dynamics in a given reservoir. Additionally, we will present recently-developed in-situ measuring tools, that are currently used for monitoring turbidity in a water column and sediment properties during and after WID actions. Finally, potential benefit of applying WID in Shihmen Reservoir (Taiwan) is discussed.

The Impact of the Construction of Large Reservoirs on the Cultural Landscape: A Case Study of the Shimen Reservoir, Taiwan

Reservoirs are large-scale water facilities with multiple functions, such as water supply, power generation, and tourism. This paper introduces the new community and cultural landscape formed by the indigenous people, engineers, workers who left their homes, and many migrating families at the Shimen Reservoir in Taoyuan, Taiwan, as an example. We analyzed how the community value of reservoir construction contributed to the development of the landscape through fieldwork, document review, and in-depth interviews. First, the new communities created to meet the needs of the immigrants influenced the surrounding environment and shaped a particular lifestyle. Secondly, new immigrants have formed a community consensus, and changes in the diet and natural landscape have promoted local tourism and affected the function of the reservoir. This study concludes that promoting local values through autonomous community action is a sustainable approach to community development. Tourism development with its symbiotic relationship with the reservoir can meet the needs of local socio-economic and cultural development. For sustainable development, a vulnerability study based on the Shimen Reservoir tourism is necessary.

An optimal integration of multiple machine learning techniques to real-time reservoir inflow forecasting

A reservoir inflow forecasting system represents a crucial technique in reservoir operation and disaster prevention, particularly in areas where the primary water source derives from typhoon events. This includes the study area of the current research, i.e., the Shihmen Reservoir (Taiwan). Effectively depositing short and high-intensity rainfall and avoiding disaster losses present significant challenges in this regard. However, the high variability and uncertainty of such rainfall events make them difficult to forecast using traditional physical-based models, which require too many calculations for application in real-time disaster forecasting. Accordingly, in this study, seven machine learning (ML) algorithms, including three conventional ML and four deep learning algorithms, were compared to derive their effectiveness for reservoir inflow forecasting in extreme weather events. The forecasting lead-times were set to 1, 4, and 6-h, representing short, medium, and long-term forecasting, respectively. Moreover, to ensure the stability and credibility of the models, two types of integrated approaches, ensemble means and switched prediction method (SP) were also employed. The results showed that although an optimal algorithm could be selected for the short, medium, and long-term, individual algorithms did not always perform well in all events. Nonetheless, the integrated approaches can effectively combine the advantages of all the included algorithms and generate more accurate and stable forecasting results, particularly when using SP, which was involved in the top three performances among all typhoon examples and indicated the best average performance. In the short-term forecast, the RMSE of the testing events is 107.2 m3/s while using SP, ranking 3rd among all 9 methods. In the medium-term forecast, the RMSE predicted by the SP is 281.72 m3/s (Rank = 1). In the long-term forecast, the SP also performed the best among the 9 methods, and the RMSE was 477.14 m3/s. In conclusion, if only single model forecast is considered, gated recurrent unit, a type of transformed recurrent neural network, will yield the best performance. Furthermore, integrated forecasts, particularly involving SP, can effectively improve the accuracy and stability of forecasts to render a model more applicable to an actual situation.

Development of a three-axis accelerometer and large-scale particle image velocimetry (LSPIV) to enhance surface velocity measurements in rivers

A three-axis accelerometer was combined with large-scale particle image velocimetry (LSPIV) to obtain nonintrusive and safe surface velocity measurements. Dual cameras were established at the Yu-Feng gauging station of Shimen Reservoir to capture near-field and far-field images and analyze the surface velocity of rivers. Surface velocity measurements were obtained with a flow meter, LPSIV with ground control points, and LSPIV with a three-axis accelerometer to compare the measurement accuracy. The results show that the relative root mean square error (RMSE) values for LSPIV with ground control points and with the three-axis accelerometer are 21% and 18%, respectively. The LSPIV technique with the three-axis accelerometer slightly improved the measurement accuracy of the surface velocity. However a three-axis accelerometer can be utilized to replace the traditional ground control points for yielding the camera pose parameters. Furthermore, the effects of the camera pose, three-axis acceleration, interrogation area (IA), and image resolution on surface velocity measurements were explored. The camera pose parameters, namely, roll (θ) and tilt (τ), and three-axis acceleration parameters, including Xa and Ya, influence surface velocity measurements. If the measurement error of the surface velocity is controlled within ±10%, the acceptable variational ranges of θ, τ, Xa, and Ya are 6.2°, 1.3°, 0.11 g, and 0.02 g, respectively. The IA size and image resolution also significantly affect the accuracy of surface velocity measurements. Therefore, the selection of a suitable IA size and image resolution is crucial for accurately measuring surface velocity.

Case study of GFRP as a sheet-pile wall for stream bank protection in Taiwan

The successful design of a structure requires not only the proper analysis of the structure but also the careful consideration of the environment that the structure is situated in. For many waterfront structures such as those near harbors, lakes, streams, and creeks, water is especially an engineering problem. To alleviate this problem, a sheet-pile wall made of glass fiber reinforced plastics (GFRP) was introduced in a pioneering stream bank stabilization (revetment) project at the Nan-Zi-Gou creek in the Shihmen Reservoir watershed of Taiwan. This paper presents the project as a case study and describes the material experiments and the finite-element analysis of revetment stability. The final construction of the GFRP sheet-pile wall was completed successfully. It is said that this pioneering work brought many positive features of GFRP to construction and could serve as a model for future replication.

Impact assessment of reservoir desiltation measures for downstream riverbed migration in climate change: A case study in northern Taiwan

Typhoon Aere in 2004 induced severe sedimentation and loss of storage capacity of the Shihmen Reservoir in northern Taiwan. The resulting dramatic increase in the turbidity of the water seriously affected the water supply. To effectively maintain the stability of the water supply and maintain the reservoir’s storage capacity, the government of Taiwan began to plan and construct a series of improvement measures, such as a sediment flushing tunnel, the JhongJhuang Bank-Side Reservoir, and the Amuping Desilting Tunnel. However, previous studies only focused on the impact of the sediment flushing tunnel and the Amuping Desilting Tunnel on the downstream riverbed, and did not consider the possibility of increasing sediment discharge after the completion of the JhongJhuang Bank-Side Reservoir. In addition, climate change will cause the intensity of extreme rainfall to increase enormously in the future. That rainfall and extra sediment flushing will challenge the existing flood prevention facilities. Therefore, this study considered that the JhongJhuang Bank-Side Reservoir will increase sediment discharge of the Shihmen Reservoir, and used dynamical downscaling extreme typhoon data of climate change under the RCP 8.5 scenario to explore the flood prevention and riverbed migration of the main channels of the Dahan and Tamsui Rivers in the future. We used the rainfall–runoff model of Hydrologic Modeling System to simulate rainfall and runoff, and used the hydraulic and sediment transport model of CCHE1D to holistically simulate flood events and consequent river scouring and deposition behaviors. Our results showed that the projected peak discharge during the late 21st century (2075 to 2099) will be at least 50% higher than that during the baseline (1979 to 2003) period. In terms of flood prevention, the potential of overbank flooding will increase in the downstream area, and the trend of long-term change in the riverbed will be dominated by degradation (-0.489 ± 0.743 m) in the future. The improvement measures will have a limited impact on riverbed migration (0.011 ± 0.094 m) in the Dahan and Tamsui Rivers. After the operation of the JhongJhuang Bank-Side Reservoir, the Shihmen Reservoir is expected to increase the sediment discharge ratio by 70% during floods, and it will not cause excessive water turbidity that may affect downstream water supply.

Risk analysis of reservoir sedimentation under non‐stationary flows

AbstractReservoir sustainability is strongly impacted by the reservoir sedimentation processes. Most of the substantial sedimentation processes occur in non‐stationary flows such as in the case of flash floods, surges and tidal waves. However, a stationary probability assumption is normally adopted to reduce mathematical model complexity. This work develops a non‐stationary Gambler's ruin model using the Monte‐Carlo simulation method. Daily water‐level data for the Xia‐Yun station are used to predict the effective risk that the maximum capacity of the water treatment plant in the Shihmen Reservoir is reached. This non‐stationary model yields fairly accurate probabilities of sedimentation by the transitional probability of a reservoir reaching different levels of turbidity, and the average time to reach a designated reservoir maximum handling turbidity. The extended capacity of the proposed model demonstrates the major particle processes during non‐stationary flows. Such analytical results offer water resources agency to scientifically evaluate the dredging/remediation strategies with the existing reservoirs. Transport capacities of rivers and streams, and the potential consequences of flood risks in response to reservoir sedimentation can then be comprehensively estimated in order to allow effective contingency planning for public safety.

Comparison of Ensemble Machine Learning Methods for Soil Erosion Pin Measurements

Although machine learning has been extensively used in various fields, it has only recently been applied to soil erosion pin modeling. To improve upon previous methods of quantifying soil erosion based on erosion pin measurements, this study explored the possible application of ensemble machine learning algorithms to the Shihmen Reservoir watershed in northern Taiwan. Three categories of ensemble methods were considered in this study: (a) Bagging, (b) boosting, and (c) stacking. The bagging method in this study refers to bagged multivariate adaptive regression splines (bagged MARS) and random forest (RF), and the boosting method includes Cubist and gradient boosting machine (GBM). Finally, the stacking method is an ensemble method that uses a meta-model to combine the predictions of base models. This study used RF and GBM as the meta-models, decision tree, linear regression, artificial neural network, and support vector machine as the base models. The dataset used in this study was sampled using stratified random sampling to achieve a 70/30 split for the training and test data, and the process was repeated three times. The performance of six ensemble methods in three categories was analyzed based on the average of three attempts. It was found that GBM performed the best among the ensemble models with the lowest root-mean-square error (RMSE = 1.72 mm/year), the highest Nash-Sutcliffe efficiency (NSE = 0.54), and the highest index of agreement (d = 0.81). This result was confirmed by the spatial comparison of the absolute differences (errors) between model predictions and observations using GBM and RF in the study area. In summary, the results show that as a group, the bagging method and the boosting method performed equally well, and the stacking method was third for the erosion pin dataset considered in this study.

Determining Cover Management Factor with Remote Sensing and Spatial Analysis for Improving Long-Term Soil Loss Estimation in Watersheds

The universal soil loss equation (USLE) is a widely used empirical model for estimating soil loss. Among the USLE model factors, the cover management factor (C-factor) is a critical factor that substantially impacts the estimation result. Assigning C-factor values according to a land-use/land-cover (LULC) map from field surveys is a typical traditional approach. However, this approach may have limitations caused by the difficulty and cost in conducting field surveys and updating the LULC map regularly, thus significantly affecting the feasibility of multi-temporal analysis of soil erosion. To address this issue, this study uses data mining to build a random forest (RF) model between eight geospatial factors and the C-factor for the Shihmen Reservoir watershed in northern Taiwan for multi-temporal estimation of soil loss. The eight geospatial factors were collected or derived from remotely sensed images taken in 2004, a digital elevation model, and related digital maps. Due to the memory size limitation of the R software, only 4% of the total data points (population dataset) in each C-factor class were selected as the sample dataset (input dataset) for analysis using the stratified random sampling method. Seventy percent of the input dataset was used to train the RF model, and the other 30% was used to test the model. The results show that the RF model could capture the trend of vegetation recovery and soil loss reduction after the destructive event of Typhoon Aere in 2004 for multi-temporal analysis. Although the RF model was biased by the majority class’s large sample size (C = 0.01 class), the estimated soil erosion rate was close to the measurement obtained by the erosion pins installed in the watershed (90.6 t/ha-year). After the model’s completion, we furthered our aim to address the input dataset’s imbalanced data problem to improve the model’s classification performance. An ad-hoc down-sampling of the majority class technique was used to reduce the majority class’s sampling rate to 2%, 1%, and 0.5% while keeping the other minority classes at a 4% sample rate. The results show an improvement of the Kappa coefficient from 0.574 to 0.732, the AUC from 0.780 to 0.891, and the true positive rate of all minority classes combined from 0.43 to 0.70. However, the overall accuracy decreases from 0.952 to 0.846, and the true positive rate of the majority class declines from 0.99 to 0.94. The best average C-factor was achieved when the sampling rate of the majority class was 1%. On the other hand, the best soil erosion estimate was obtained when the sampling rate was 2%.