Authorship：Lead author,　Corresponding author   Publishing type：Research paper (scientific journal)  

Deep-learning neural network (DNN) models are currently gaining popularity in the field of hydrology as well as flow forecasting. This is because these models have proven that they are capable of producing short-period forecasts with high accuracy. However, studies on flow forecasting for hydrological stations located upstream of river basins are still uncommon as the limited number of input parameters can be gathered in such places. This study looked into the capacity of DNN models to predict discharge upstream of the Da River in Vietnam, where the topography is mostly mountainous. Streamflow data from the LaiChau hydrological station — the largest and most distant hydrological station upstream of the Da River — has been gathered and employed as input for three DNN models. These models are the LSTM (long short-term memory neural network), ANN (artificial neural network), and CNN (convolutional neural network). According to research findings, the performance of the LSTM and CNN models outperforms that of the ANN, which has an NSE coefficient of just approximately 0.91. LSTM has a slight advantage over CNN, although the difference is modest because their NSE coefficients are 0.97 and 0.96, respectively. This finding suggests that DNN models, particularly LSTM, can be a feasible alternative for upstream hydrological station discharge predictions.

DOI： 10.1007/s12145-023-01082-9

Scopus

researchmap

Authorship：Lead author   Publishing type：Research paper (scientific journal)  

Study region: The Ca River basin is located in the North Central Coast area of Vietnam Study focus: This study aims to develop a deep learning framework that is both effective and straightforward in order to forecast water levels in the Ca River basin in advance of multiple time steps for the event scales. We have thoroughly studied and assessed two deep learning models (DLMs), long-short term memory (LSTM) and gated recurrent unit (GRU), for their capacity to forecast water levels, focusing on various aspects such as the influence of sequence length or the impact of hyperparameter selection. Besides, two data scenarios were established using hydrological data from eight severe floods between 2007 and 2019 to examine the effect of input variables on model performance. Water level data was employed for both the scenarios (S1 and S2), whereas precipitation data was used only in S2. The cross-validation technique was used dynamically to address the issue of limited data. The inputs were reformatted as tensors and were then randomly divided into subsets. This flexible tuning preserved the sequential nature of the hydrological data while enabling the DLMs to be trained efficiently. New hydrological insights for the region: The findings revealed that both the models exhibited equally excellent performances. The NSE of the LSTM model varies from 0.999–0.971 compared to 0.998–0.974 of the GRU model, corresponding to forecast cases from one to four-time steps ahead. This indicated that the use of multiple-input data types (S2) contrary to only one date type (S1) does not necessarily improve the forecasting performance. LSTM/GRU models with one hidden layer are adequate for delivering high performance while minimizing the data processing time.

DOI： 10.1016/j.ejrh.2023.101328

Scopus

researchmap

Authorship：Lead author   Publishing type：Research paper (scientific journal)  

Despite satellite-based precipitation products (SPPs) providing a worldwide span with a high spatial and temporal resolution, their efficiency in disaster risk forecasting, hydrological, and watershed management remains a challenge due to the significant dependence of rainfall on the spatiotemporal pattern and geographical features of each area. This research proposes an effective deep learning-based solution that combines the convolutional neural network and the benefit of encoder-decoder architecture to eliminate pixel-by-pixel bias to enhance the accuracy of daily SPPs. This work uses five gridded precipitation products, four of which are satellite-based (TRMM, CMORPH, CHIRPS, and PERSIANN-CDR) and one of which is gauge-based (APHRODITE). The Lancang-Mekong River Basin (LMRB), an international basin, was chosen as the research region because of its diverse climate and geographical spread spanning six countries. According to the results of the analyses, the TRMM product exhibits better performance than the other three SPPs. The deep learning model proved its efficacy by successfully reducing the spatial–temporal gap between the four SPPs and APHRODITE. In addition, the ADJ-TRMM product performed the best of the four corrected items, followed by the ADJ-CDR and ADJ-CHIRPS products. This study's findings indicate that each SPP has advantages and disadvantages across LMRB. In the aftermath of the discontinuation of the APHRODITE product in 2015, we believe that the deep learning framework will be a solution for generating a more up-to-date and dependable dataset for LMRB research.

DOI： 10.1016/j.jag.2022.103177

Scopus

researchmap

Authorship：Lead author   Publishing type：Research paper (scientific journal)  

Precipitation observations from a ground-based gauge provide a reliable data source for hydrological and climatological studies. However, these data are sparse in many regions of the world, particularly the Mekong River Basin (MRB). Satellite-based precipitation products (SPPs) are the sole data source available with worldwide coverage. Despite this, there is a mismatch between SPPs and gauge-based observations, and the correct procedures should be utilized to minimize systematic bias in SPPs. This study aimed to benchmark the efficacy of four state-of-the-art bias-correcting deep learning models (DLMs) for the tropical rainfall measuring mission-based precipitation product named TRMM-3B42 (hereafter TRMM) over the entire MRB. These models were designed mainly based on convolutional neural network (CNN) and encoder-decoder (ENDE) architectures, including ConvENDE, ConvUNET, ConvINCE, and ConvLSTM. The bias-corrected dataset by DLMs was then confirmed against the gauge-based dataset (Asian precipitation-highly resolved observational data integration toward evaluation of water resources, APHRODITE). From the results obtained, all four DLMs effectively minimized the bias of the TRMM product. Among them, ConvENDE and ConvUNET had a higher consistency and performance level compared to ConvINCE and ConvLSTM. Additionally, the complexity of DLMs did not enhance their efficiency, as is the case with ConvINCE and ConvLSTM, despite using many computing resources. Given the observed data shortage for the MRB since 2016, the application of DLMs, such as ConvENDE and ConvUNET, can serve to improve the reliability of existing rainfall datasets and provide valuable input for various research purposes in the MRB.

DOI： 10.1109/TGRS.2023.3299234

Scopus

researchmap

Authorship：Lead author   Publishing type：Research paper (scientific journal)  

Landslides are a major natural hazard that can significantly damage infrastructure and cause loss of life. In South Korea, the current landslide susceptibility mapping (LSM) approach is mainly based on statistical techniques (logistic regression (LR) analysis). According to previous studies, this method has achieved an accuracy of approximately 75.2%. In this paper, we expand upon this traditional approach by comparing the performance of six machine learning (ML) algorithms for LSM in Inje County, South Korea. The study employed a combination of geographical data gathered from 2005 to 2019 to train and evaluate six algorithms, including LR, Gaussian Naive Bayes (GNB), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), and Extreme Gradient Boosting (XGB). The effectiveness of these models was measured by various criteria, such as the percentage of correct classification (PCC) score, F1 score, and Kappa score. The results demonstrated that the PCC and F1 scores of the six models fell between [0.869–0.941] and [0.857–0.940], respectively. RF and XGB had the highest PCC and F1 scores of 0.939 and 0.941, respectively. This study indicates that ML can be a valuable technique for high-resolution LSM in South Korea instead of the current approach.

DOI： 10.3389/feart.2023.1268501

Scopus

researchmap

Authorship：Lead author,　Corresponding author   Publishing type：Research paper (scientific journal)  

Recently, machine learning (ML) is being applied to various fields, including hydrology and hydraulics. The numerical models based on ML algorithms have been widely used for forecasting water levels or flowrate in different timescales. Especially in estuary areas where the hydrodynamic regime becomes complicated, the water level forecast information in this area plays an essential role in the operation of tidal sluices. This study proposes an efficient approach using an ML model, long short-term memory (LSTM), to predict short-term water levels in tidal sluice gates from 6 to 48 hours ahead. The An Tho culvert located in the Bac Hung Hai irrigation system, the most extensive irrigation system in Vietnam, was selected as a case study station. The high accuracy of predictive results reveals LSTM models' effectiveness in different forecasting scenarios. In the first scenario using just water level data at the prediction station, the Kling–Gupta efficiency (KGE) coefficient ranges from nearly 0.89 to 0.96. Meanwhile, in the second scenario, the combination of observed data of three gauge stations exhibited better performance with KGE coefficients ranging from just under 0.93 to 0.98 for eight forecasted cases. The findings of this study highlight the performance of LSTM models in providing high-accuracy short-period water level forecasts for areas near estuaries. These obtained results can play a vital role in the management and operation of tidal sluices in the Bac Hung Hai irrigation system, as well as a reference for the operation of other irrigation systems around the world.

DOI： 10.1007/s10661-022-10115-7

Scopus

PubMed

researchmap

Authorship：Lead author   Publishing type：Research paper (scientific journal)  

Recently, deep learning (DL) models, especially those based on long short-term memory (LSTM), have demonstrated their superior ability in resolving sequential data problems. This study investigated the performance of six models that belong to the supervised learning category to evaluate the performance of DL models in terms of streamflow forecasting. They include a feed-forward neural network (FFNN), a convolutional neural network (CNN), and four LSTM-based models. Two standard models with just one hidden layer-LSTM and gated recurrent unit (GRU)-are used against two more complex models-the stacked LSTM (StackedLSTM) model and the Bidirectional LSTM (BiLSTM) model. The Red River basin-the largest river basin in the north of Vietnam-was adopted as a case study because of its geographic relevance since Hanoi city-the capital of Vietnam-is located downstream of the Red River. Besides, the input data of these models are the observed data at seven hydrological stations on the three main river branches of the Red River system. This study indicates that the four LSTM-based models exhibited considerably better performance and maintained stability than the FFNN and CNN models. However, the complexity of the StackedLSTM and BiLSTM models is not accompanied by performance improvement because the results of the comparison illustrate that their respective performance is not higher than the two standard models-LSTM and GRU. The findings of this study present that LSTM-based models can reach impressive forecasts even in the presence of upstream dams and reservoirs. For the streamflow-forecasting problem, the LSTM and GRU models with a simple architecture (one hidden layer) are sufficient to produce highly reliable forecasts while minimizing the computation time.

DOI： 10.1109/ACCESS.2021.3077703

Scopus

researchmap

Authorship：Lead author   Publishing type：Research paper (scientific journal)  

Spatiotemporal precipitation data is one of the essential components in modeling hydrological problems. Although the estimation of these data has achieved remarkable accuracy owning to the recent advances in remote-sensing technology, gaps remain between satellite-based precipitation and observed data due to the dependence of precipitation on the spatiotemporal distribution and the specific characteristics of the area. This paper presents an efficient approach based on a combination of the convolutional neural network and the autoencoder architecture, called the convolutional autoencoder (ConvAE) neural network, to correct the pixel-by-pixel bias for satellite-based products. The two daily gridded precipitation datasets with a spatial resolution of 0.25° employed are Asian Precipitation-Highly Resolved Observational Data Integration towards Evaluation (APHRODITE) as the observed data and Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks-Climate Data Record (PERSIANN-CDR) as the satellite-based data. Furthermore, the Mekong River basin was selected as a case study, because it is one of the largest river basins, spanning six countries, most of which are developing countries. In addition to the ConvAE model, another bias correction method based on the standard deviation method was also introduced. The performance of the bias correction methods was evaluated in terms of the probability distribution, temporal correlation, and spatial correlation of precipitation. Compared with the standard deviation method, the ConvAE model demonstrated superior and stable performance in most comparisons conducted. Additionally, the ConvAE model also exhibited impressive performance in capturing extreme rainfall events, distribution trends, and described spatial relationships between adjacent grid cells well. The findings of this study highlight the potential of the ConvAE model to resolve the precipitation bias correction problem. Thus, the ConvAE model could be applied to other satellite-based products, higher-resolution precipitation data, or other issues related to gridded data.

DOI： 10.3390/RS12172731

Scopus

researchmap

Authorship：Lead author   Publishing type：Research paper (scientific journal)  

Flood forecasting is an essential requirement in integrated water resource management. This paper suggests a Long Short-Term Memory (LSTM) neural network model for flood forecasting, where the daily discharge and rainfall were used as input data. Moreover, characteristics of the data sets which may influence the model performance were also of interest. As a result, the Da River basin in Vietnam was chosen and two different combinations of input data sets from before 1985 (when the Hoa Binh dam was built) were used for one-day, two-day, and three-day flowrate forecasting ahead at Hoa Binh Station. The predictive ability of the model is quite impressive: The Nash-Sutcliffe efficiency (NSE) reached 99%, 95%, and 87% corresponding to three forecasting cases, respectively. The findings of this study suggest a viable option for flood forecasting on the Da River in Vietnam, where the river basin stretches between many countries and downstream flows (Vietnam) may fluctuate suddenly due to flood discharge from upstream hydroelectric reservoirs.

DOI： 10.3390/w11071387

Scopus

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1088/2515-7620/ad8a24

researchmap

Publishing type：Research paper (international conference proceedings)  

Precipitation information with high accuracy plays a crucial role in hydrology and water resources management. With the advance in technology, satellite precipitation products (SPPs) provide an unprecedented opportunity for monitoring the spatial and temporal variation of precipitation from space. However, SPPs still present a low performance with high uncertainty. To overcome this problem, the current study aims to produce a new reanalysis of precipitation data by integrating information from observation data with multiple SPPs over South Korea under the aid of a fast and high-performance machine learning-based, namely a light gradient boosting machine. In addition, other statistical merging methods were also carried out to highlight the robustness of the machine learning-based algorithm. To examine the accuracy of merging precipitation products, observed data from 64 automated synoptic observation system rain gauge stations were collected and compared with merging precipitation products. A high agreement between merging precipitation data generated from the machine learning-based approach with observation was witnessed through several continuous criteria and categorical indicators. The results from this study point out that light gradient boosting machine not only has the capability in merging multi-sources precipitation but also it could provide extraordinary rainfall information for the region of interest, especially in areas with low observed station density.

DOI： 10.1007/978-981-99-2345-8_52

Scopus

researchmap

Publishing type：Research paper (international conference proceedings)  

Soil is a vital component for supporting life on Earth and is substantially responsible for the operation of any ecological system. Soil erosion is a global problem because it has resulted in the destruction of agricultural land. Raindrops contacting the soil surface may cause soil separation. This process may be identified by assessing the rainfall kinetic energy (KE). Since direct measurements of KE are challenging, statistical formulas are often employed as an alternative option to estimate the KE using rainfall intensity (RI), which significantly influences soil erosion and is very simple to detect. In this study, we collected a rainfall dataset using a laser-based device (OTT Parsivel2 optical disdrometer) between June 2020 and December 2021 in Sangju City (Korea) to analyze the property of KE-RI relationship. A total of 37 rainfall events were selected and used for the formation of KE-RI equations. We derived three KE–RI equations in each group based on the three different rainfall intensity groups. Power-law, linear, and polynomial models were used to link KE to RI, and the best correlation between KE and RI was determined using a power-law form.

DOI： 10.1007/978-981-99-2345-8_51

Scopus

researchmap

Publishing type：Research paper (scientific journal)  

Accurate soil organic carbon storage (SOCS) estimation is crucial for sustaining ecosystem health and mitigating climate change impacts. This study investigated the accuracy and variability of SOCS predictions, focusing on the role of pedotransfer functions (PTFs) in estimating soil bulk density (BD). Utilizing a comprehensive dataset from the Korean Rural Development Administration (RDA database), which includes 516 soil horizons, we evaluated 36 widely-used BD PTFs, well-established formulas that estimate BD by considering soil properties, including soil organic carbon (SOC), soil organic matter (OM), sand, gravel, silt, and clay. These PTFs demonstrated varying levels of precision, with root mean squared errors (RMSE) ranging from 0.177 to 0.377 Mg m−3 and coefficients of determination (R2) from 0.176 to 0.658; hence, the PTFs have been classified into excellent, moderate, and poor-performing groups for predicting BD. Further, a novel PTF based on an exponential function of SOC was developed, showing superior predictive power (R2 = 0.73) compared to existing PTFs, using an independent validation dataset. Our findings reveal significant differences in SOCS predictions and observations among the PTFs, with a p-value <0.05. The highest concentrations of SOCS were noted in forest soils, considerably above the national average, highlighting the importance of tailored soil management practices to enhance carbon sequestration. These findings are crucial for refining PTF precision to improve the accuracy of national SOCS estimates, supporting effective land management and climate change mitigation strategies.

DOI： 10.1016/j.iswcr.2024.04.002

Scopus

researchmap

Publishing type：Research paper (scientific journal)  

Study region: South Korea is situated in the northeastern region of Asia Study focus: Recent technological developments have enabled multi-source precipitation products (MSPs), including satellite-based and model-based, to be useful data sources for quantifying the spatiotemporal variations in precipitation. Unfortunately, the main limitation of MSPs in potential applications is inheritance errors with high uncertainty. To tackle this problem, the capabilities of six machine learning algorithms (Ridge Linear Regression, k-Nearest Neighbors, Support Vector Regression, Gradient Boosting Decision Tree, Light Gradient Boosting Machine, and Random Forest) to produce new precipitation product by merging MSPs with ground-based data have investigated. Ground-based data from 2003 to 2017 were utilized for train and valid process. The robustness of the ML algorithms was highlighted using several evaluation metrics such as continuous indices (modified Kling-Gupta Efficiency and root mean square error) and categorical indices (probability of detection, false alarm rate, and critical success index). New hydrological insights for the region: The results indicate that (1) the ML approaches can merge MSPs with observed data for accurately estimate rainfall, particularly in basins with sparsely distributed rain gauge stations. (2) The merged precipitation products generated from the six ML approaches showed significant agreement and high accuracy with observation data considering rainfall intensity estimation and improved the capability of detecting rain and non-rain events over South Korea.

DOI： 10.1016/j.ejrh.2023.101475

Scopus

researchmap

Authorship：Corresponding author   Publishing type：Research paper (scientific journal)  

Contrast enhancement of images is a crucial topic in image processing that improves the quality of images. The methods of image enhancement are classified into three types, including the histogram method, the fuzzy logic method, and the optimal method. Studies on image enhancement are often based on the rules: if it is bright, then it is brighter; if it is dark, then it is darker, using a global approach. Thus, it is hard to enhance objects in all dark and light areas, as in satellite images. This study presents a novel algorithm for improving satellite images, called remote sensing image enhancement based on cluster enhancement (RSIECE). First, the input image is clustered by the algorithm of fuzzy semi-supervised clustering. Then, the upper bound and lower bound are estimated according to the cluster. Next, a sub-algorithm is implemented for clustering enhancement using an enhancement operator. For each pixel, the gray levels for each channel (R, G, B) are transformed with this sub-algorithm to generate new corresponding gray levels because after clustering, pixels belong to clusters with the corresponding membership values. Therefore, the output gray level value will be aggregated from the enhanced gray levels by the sub-algorithm with the weight of the corresponding cluster membership value. The test results demonstrate that the suggested algorithm is superior to several recently developed approaches.

DOI： 10.3390/rs15061645

Scopus

researchmap

Authorship：Corresponding author   Publishing type：Research paper (scientific journal)  

Satellite-based precipitation (SP) data are gaining scientific interest due to their advantage in producing high-resolution products with quasi-global coverage. However, since the major reliance of precipitation data is on the distinctive geographical features of each location, they remain at a considerable distance from station-based data. This paper examines the effectiveness of a convolutional autoencoder (CAE) architecture in pixel-by-pixel bias correction of SP products for the Mekong River Basin (MRB). Two satellite-based products (TRMM and PERSIANN-CDR) and a gauge-based product (APHRODITE) are gridded rainfall products mined in this experiment. According to the estimated statistical criteria, the CAE model was effective in reducing the gap between SP products and benchmark data both in terms of spatial and temporal correlations. The two corrected SP products (CAE_TRMM and CAE_CDR) performed competitively, with CAE TRMM appearing to have a slight advantage over CAE CDR, however, the difference was minor. This study’s findings proved the effectiveness of deep learning-based models (here CAE) for bias correction of SP products. We believe that this technique will be a feasible alternative for delivering an up-to-current and reliable dataset for MRB studies, given that the sole available gauge-based dataset for this area has been out of date for a long time.

DOI： 10.3390/rs15030630

Scopus

researchmap

Publishing type：Research paper (scientific journal)  

As the Mekong River basin is a nationally shared river, it is difficult to collect precipitation data, and the quantitative and qualitative quality of the data sets differs from country to country, which may increase the uncertainty of hydrological analysis results. Recently, with the development of remote sensing technology, it has become easier to obtain grid-based precipitation products(GPPs), and various hydrological analysis studies have been conducted in unmeasured or large watersheds using GPPs. In this study, rainfall-runoff simulation in the Mekong River basin was conducted using the SWAT model, which is a quasi-distribution model with three satellite GPPs (TRMM, GSMaP, PERSIANN-CDR) and two GPPs (APHRODITE, GPCC). Four water level stations, Luang Prabang, Pakse, Stung Treng, and Kratie, which are major outlets of the main Mekong River, were selected, and the parameters of the SWAT model were calibrated using APHRODITE as an observation value for the period from 2001 to 2011 and runoff simulations were verified for the period form 2012 to 2013. In addition, using the ConvAE, a convolutional neural network model, spatio-temporal correction of original satellite precipitation products was performed, and rainfall-runoff performances were compared before and after correction of satellite precipitation products. The original satellite precipitation products and GPCC showed a quantitatively under-or over-estimated or spatially very different pattern compared to APHPRODITE, whereas, in the case of satellite precipitation prodcuts corrected using ConvAE, spatial correlation was dramatically improved. In the case of runoff simulation, the runoff simulation results using the satellite precipitation products corrected by ConvAE for all the outlets have significantly improved accuracy than the runoff results using original satellite precipitation products. Therefore, the bias correction technique using the ConvAE technique presented in this study can be applied in various hydrological analysis for large watersheds where rain guage network is not dense.

DOI： 10.3741/JKWRA.2023.56.2.75

Scopus

researchmap

Publishing type：Research paper (scientific journal)  

Calculating rainfall erosivity, which is the capacity of rainfall to dislodge soil particles and cause erosion, requires the measurement of the rainfall kinetic energy (KE). Direct measurement of KE has its own challenges, owing to the high cost and complexity of the measuring instruments involved. Consequently, the KE is often approximated using empirical equations derived from rainfall intensity (Ir) inputs in the absence of such instruments. However, the KE–Ir equations strongly depend on local climate patterns and measurement methods. Therefore, this study aims to compare and evaluate the efficacy of 27 KE–Ir equations with observed data. Based on a re-analysis, we also propose an exponential KE–Ir equation for the entire Korean site, and the spatial distribution of its parameter in the equation is also discussed. In this investigation, we used an optical disdrometer (OTT Parsivel2) to gather data in Sangju City (Korea) between June 2020 and December 2021. The outputs of this study are shown as follows: (1) The statistically most accurate estimates of KE expenditure and KE content in Sangju City are obtained using power-law equations given by Sanchez-Moreno et al. and exponential equations published by Lee and Won, respectively. (2) The suggested KE–Ir equation applied to the entire Korean site exhibits a comparable general correlation with the observed data. The parameter maps indicate a high variance in geography.

DOI： 10.3390/rs15010156

Scopus

researchmap

Publishing type：Research paper (scientific journal)  

When raindrops collide with the topsoil surface, they cause soil detachment, which can be estimated by measuring the kinetic energy (KE) of the raindrops. Considering their direct measurements on terrestrial surfaces are challenging, empirical equations are commonly utilized for estimating the KE from rainfall intensity (Ir), which has a great influence on soil loss and can be easily obtained. However, establishing the optimal relationship between KE and Ir is difficult. In this study, we used a laser-based instrument (OTT Parsivel2 Optical disdrometer) to collect datasets in Sangju City (South Korea) between June 2020 and December 2021 to examine the characteristics of KE–Ir relationships. We derived two different expressions for KE–Ir: KE expenditure (KEexp; J m−2h−1) and KE content (KEcon; J m−2mm−1), using 37 rainfall events. Subsequently, the 37 rainfall events were categorized into three groups based on the magnitude of the mean rainfall intensity of each event. Overall, the KE values estimated through the equations derived based on 37 events were higher than those estimated by the equations derived based on the three rainfall event groups. Our findings should facilitate the development of more suitable physics-based soil erosion models at event scales.

DOI： 10.3389/fenvs.2022.985516

Scopus

researchmap

Publishing type：Research paper (scientific journal)  

Precipitation nowcasting is one of the main tasks of weather forecasting that aims to predict rainfall events accurately, even in low-rainfall regions. It has been observed that few studies have been devoted to predicting future radar echo images in a reasonable time using the deep learning approach. In this paper, we propose a novel approach, RainPredRNN, which is the combination of the UNet segmentation model and the PredRNN_v2 deep learning model for precipitation nowcasting with weather radar echo images. By leveraging the abilities of the contracting-expansive path of the UNet model, the number of calculated operations of the RainPredRNN model is significantly reduced. This result consequently offers the benefit of reducing the processing time of the overall model while maintaining reasonable errors in the predicted images. In order to validate the proposed model, we performed experiments on real reflectivity fields collected from the Phadin weather radar station, located at Dien Bien province in Vietnam. Some credible quality metrics, such as the mean absolute error (MAE), the structural similarity index measure (SSIM), and the critical success index (CSI), were used for analyzing the performance of the model. It has been certified that the proposed model has produced improved performance, about 0.43, 0.95, and 0.94 of MAE, SSIM, and CSI, respectively, with only 30% of training time compared to the other methods.

DOI： 10.3390/axioms11030107

Scopus

researchmap

Publishing type：Research paper (scientific journal)  

Urban flooding is a global metropolitan problem; therefore, establishing reliable streamflow forecasting models is critical for flood control and planning in urban areas. Furthermore, assessing the importance and uncertainty of model predictors is useful for managers; however, these predictors are still underevaluated. To address these concerns, we developed a novel hybrid model, GA-BART, that integrates a genetic algorithm (GA) with the Bayesian additive regression tree (BART) model for hourly streamflow forecasting. A case study was conducted in the Jungrang urban basin, which is located on the Han River in South Korea. The model was built and evaluated based on data collected during 39 heavy rain events from 2003 to 2020. To compare the model's forecast capability, a support vector regression model hybridized with a genetic algorithm (GA-SVR) and a multiple linear regression (MLR) model was constructed. An analysis of multiple datasets including different input predictors was performed to define the optimal set for streamflow forecasting. The results illustrated that the GA-BART model outperformed the GA-SVR and MLR models in multistep-ahead streamflow forecasts, with improved measures of the root mean square error (RMSE), mean absolute error (MAE), relative error, Nash-Sutcliffe efficiency (NSE), time lag and correlation coefficient (CC). In addition, the GA-BART model could reasonably determine the relative importance of the input variables. This study demonstrated that, despite some disadvantages in the five- and six-hour step-ahead forecasts, the hybrid GA-BART model can be a good option among the available models for hourly streamflow forecasting.

DOI： 10.1016/j.jhydrol.2022.127445

Scopus

researchmap

Publishing type：Research paper (scientific journal)  

The advancement of dam operation is further required due to the upcoming rainy season, typhoons, or torrential rains. Besides, physical models based on specific rules may sometimes have limitations in controlling the release discharge of dam due to inherent uncertainty and complex factors. This study aims to forecast the water level of the nearest station to the dam multi-timestep-ahead and evaluate the availability when it makes a decision for a release discharge of dam based on LSTM (Long Short-Term Memory) of deep learning. The LSTM model was trained and tested on eight data sets with a 1-hour temporal resolution, including primary data used in the dam operation and downstream water level station data about 13 years (2009~2021). The trained model forecasted the water level time series divided by the six lead times: 1, 3, 6, 9, 12, 18-hours, and compared and analyzed with the observed data. As a result, the prediction results of the 1-hour ahead exhibited the best performance for all cases with an average accuracy of MAE of 0.01m, RMSE of 0.015 m, and NSE of 0.99, respectively. In addition, as the lead time increases, the predictive performance of the model tends to decrease slightly. The model may similarly estimate and reliably predicts the temporal pattern of the observed water level. Thus, it is judged that the LSTM model could produce predictive data by extracting the characteristics of complex hydrological non-linear data and can be used to determine the amount of release discharge from the dam when simulating the operation of the dam.

DOI： 10.3741/JKWRA.2021.54.S-1.1095

Scopus

researchmap

Publishing type：Research paper (scientific journal)  

Estimation of sediment transport capacity (STC) plays a crucial role in simulating soil erosion using any physics-based models. In this research, we aim to investigate the pros and cons of six popular STC methods (namely, Shear velocity, Kilinc-Richardson (KR), Effective stream power, Slope and unit discharge, Englund-Hansen (EH), and Unit stream power) for soil erosion/deposition simulation at watershed scales. An in-depth analysis was performed using the selected STC methods integrated into the Grid Surface Subsurface Hydrologic Analysis model for investigating the changes in morphology at spatial-temporal scales at the Cheoncheon watershed, South Korea, over three storm events. Conclusions were drawn as follows. (1) Due to the ability of the KR and EH methods to include an additional parameter (i.e., erodibility coefficient), they outperformed oth-ers by producing more accurate simulation results of sediment concentration predictions. The KR method also proved to be superior to the EH method when it showed a more suitable for sediment concentration simulations with a wide range of sediment size and forcing magnitude. (2) We further selected 2 STC methods among the 6 methods to deeply explore the spatial distribution of erosion/deposition. The overall results were more agreeable. For instance, the phenomenon of erosion mainly occurred upstream of watersheds with steep slopes and unbalanced initial sediment concentrations, whereas deposition typically appeared at locations with flat terrain (or along the mainstream). The EH method demonstrated the influence of topography (e.g., gradient slope) on accretionary erosion/deposition results more significantly than the KR method. The obtained results contribute a new understanding of rainfall-sediment-runoff processes and provide fundamental plans for soil conservation in watersheds.

DOI： 10.3390/w13213054

Scopus

researchmap

Publishing type：Research paper (scientific journal)  

Precipitation is a crucial component of the water cycle and plays a key role in hydrological processes. Recently, satellite-based precipitation products (SPPs) have provided grid-based precipitation with spatiotemporal variability. However, SPPs contain a lot of uncertainty in estimated precipitation, and the spatial resolution of these products is still relatively coarse. To overcome these limitations, this study aims to generate new grid-based daily precipitation based on a combination of rainfall observation data with multiple SPPs for the period of 2003–2017 across South Korea. A Random Forest (RF) machine-learning algorithm model was applied for producing a new merged precipitation product. In addition, several statistical linear merging methods have been adopted to compare with the results achieved from the RF model. To investigate the efficiency of RF, rainfall data from 64 observed Automated Synoptic Observation System (ASOS) installations were collected to analyze the accuracy of products through several continuous as well as categorical indicators. The new precipitation values produced by the merging procedure generally not only report higher accuracy than a single satellite rainfall product but also indicate that RF is more effective than the statistical merging method. Thus, the achievements from this study point out that the RF model might be applied for merging multiple satellite precipitation products, especially in sparse region areas.

DOI： 10.3390/rs13204033

Scopus

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.21729/KSDS.2021.14.3.17

researchmap

Publishing type：Research paper (scientific journal)  

As the frequency of localized heavy rainfall has increased during recent years, the importance of high-resolution radar data has also increased. This study aims to correct the bias of Dual Polarization radar that still has a spatial and temporal bias. In many studies, various statistical techniques have been attempted to correct the bias of radar rainfall. In this study, the bias correction of the S-band Dual Polarization radar used in flood forecasting of ME was implemented by a Convolutional Autoencoder (CAE) algorithm, which is a type of Convolutional Neural Network (CNN). The CAE model was trained based on radar data sets that have a 10-min temporal resolution for the July 2017 flood event in Cheongju. The results showed that the newly developed CAE model provided improved simulation results in time and space by reducing the bias of raw radar rainfall. Therefore, the CAE model, which learns the spatial relationship between each adjacent grid, can be used for real-time updates of grid-based climate data generated by radar and satellites.

DOI： 10.3741/JKWRA.2021.54.7.453

Scopus

researchmap

Authorship：Lead author,　Corresponding author   Publishing type：Research paper (international conference proceedings)  

At present, deep learning models have been widely applied in many studies related to the field of water resource management. In this study, several deep learning neural network models based on the Gated Recurrent Unit (GRU) architectures have been applied to the river water level prediction for a short-time period, from one hour to nine hours ahead. The input data of these models are hourly water levels which are observed at four hydrological stations on the Geum River, South Korea. Though the model does not require data such as topography, land cover, or precipitation data, the forecasted results indicate significant stability and performance. Compared to the observed water level data, the correlation coefficient NSE (Nash-Sutcliffe efficiency) is up to more than 99% in the case of a 1-hour forecast. The results of this study prove the potential of deep learning models in predicting water level and applicable to other river basins.

DOI： 10.1007/978-981-16-0053-1_40

Scopus

researchmap

Publishing type：Research paper (scientific journal)  

The establishment of reliable water level prediction models is vital for urban flood control and planning. In this paper, we develop hybrid models (GA-XGBoost and DE-XGBoost) that couple two evolutionary models, a genetic algorithm (GA) and a differential evolution (DE) algorithm, with the extreme gradient boosting (XGBoost) model for hourly water level prediction. The Jungrang urban basin located on the Han River, South Korea, was selected as a case study for the proposed models. Hourly rainfall and water level data were collected between 2003 and 2020 to construct and evaluate the performance of the selected models. To compare the prediction efficiency, two other tree-based models were chosen: classification and registration tree (CART) and random forest (RF) models. A comparison of the results showed that two hybrid models, GA-XGBoost and DE-XGBoost, outperformed RF and CART in the multistep-Ahead prediction of water level, and the relative errors of the hybrid model ranged from [2.18%-9.21%], compared to [3.76%-10.41%] and [2.99%-11.88%] for the RF and CART, respectively. Reliable performance was also supported by other measures. In general, the GA-XGBoost and DE-XGBoost models displayed relatively similar performance despite their small differences. The CART model was not preferable for multistep-Ahead water level predictions, even though it yielded the lowest Akaike information criterion (AIC) value. This study verifies that despite having some drawbacks when considering long step-Ahead prediction and model complexity, hybrid XGBoost models might be superior to many existing models for hourly water level prediction.

DOI： 10.1109/ACCESS.2021.3111287

Scopus

researchmap

Authorship：Lead author,　Corresponding author   Publishing type：Research paper (international conference proceedings)  

In light of the proliferation of information technology, the application of deep learning models in the analysis and study of hydrological problems is increasingly becoming common. This paper proposes a new approach using one of the applications of deep learning models to predict river water level in areas where the influence of tides is obvious. The forecasting model is developed based on the recurrent neural network for predicting the water level from one to four time-steps ahead in the downstream of An Tho irrigation culvert on the Luoc River (Vietnam). Each time-step corresponds to the once observed data and the data collected for this study is only the observed water level at the target station-An Tho sluice in over 18 years. Although only a modest amount of data is required, the forecasting model produces superior results. Accuracy in the phase of testing the model is up to 94-96% for all forecasting cases. The findings of this study indicate that the proposed model produces an outstanding performance when the target-forecasting station is clearly affected by the tide. This acts as a precursor of the construction of an operating regime for irrigation sluice gates in the tidal area.

DOI： 10.1007/978-981-15-0291-0_92

Scopus

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.7744/KJOAS.20190068

researchmap