Back to Program


Search By:

Presentation Mode : All
Conference Day : 06/08/2021
Time Slot : AM2 11:00 - 13:00
Sections : HS - Hydrological Sciences










Hydrological Sciences | Fri-06 Aug




HS33-A011
Investigating Rainfall Prediction Accuracy in Japanese Archipelago Using Wrf Model in Various Spatio-temporal Scales

Haruki IMAI1+, Masashi WATANABE2, Taro ARIKAWA1#
1Chuo University, Japan, 2Nanyang Technological University, Singapore


In the future, typhoon intensity and mean annual precipitation are expected to be increased due to global warming at mid-latitude humid regions. The Kuma River, in Kumamoto Prefecture, Japan overflowed by the heavy rains of July 2020, causing many casualties mainly in the Kyushu region. One of the reasons was that the rainfall forecasts were inaccurate, causing delays in evacuations. Rainfall phenomena is occurred in various spatio-temporal scales. Since rainfall phenomena are affected by many factors, it is necessary to examine the effects of these factors against accuracy of rainfall simulation. In this study, we conducted the simulation of rainfall at Japanese archipelago in several meteorological events that have caused severe damage. We then compared the calculated typhoon rainfall and frontal rainfall and observed ones. For the simulation, WRF which is one of weather forecasting models was used. The meteorological data of National Centers for Environmental Prediction (NCEP) was used as the initial value in the simulation. The simulations were conducted by changing the start time of the forecast. To investigate the sensitive physical options for rainfall phenomena in Japan, the cloud microphysics scheme, the cumulus parameterization, and planetary boundary layer scheme were varied. In order to evaluate the accuracy of the simulation, mean absolute error was calculated. The results showed that the absolute values of the errors (MAE) of typhoon rainfall and frontal rainfall were 2~5mm/h and 2~8mm/h, respectively. We also found the physical options which affects to the simulation accuracy. Depending on the selected physical options and the starting time of the forecast, the simulation results were varied, but the rainfall distribution was well reproduced. The presented results are expected to contribute to the improvement of rainfall prediction accuracy and evacuation systems during typhoons and heavy rains, because the physical options that affects to prediction accuracy were identified.

HS33-A026
A Multi-model Approach for the Evaluation of Forecasted Streamflow in Narmada River Basin

Ankit SINGH#+, Soubhik MONDAL , Sanjeev Kumar JHA
Indian Institute of Science Education and Research Bhopal, India


The precipitation products obtained from Numerical weather predictions (NWPs) models are used as the input data for hydrological models to forecast the streamflow. In India, National Center for Medium-Range Weather Forecasting (NCMRWF) provides forecasts from its UK Met office Unified Model-based deterministic model (NCUM), and ensemble prediction system (NEPS). In this study, we assess the accuracy of NWPs from NCUM and NEPSin obtaining the accurate streamflow forecasts. To overcome the uncertainty induced by a hydrological model in streamflow forecasting, a multi-modeling approach is used in this study. We use two hydrological models Soil and Water Assessment Tool (SWAT) and Variable Infiltration Capacity (VIC) for the application of multi-modeling in short-term streamflow forecasting. SWAT is a Hydrological Response Unit (HRU’s) based hydrological model. HRUs are the area that contains similar types of soil, land use and slope properties in a subbasin. The VIC model is a grid-based model with variable infiltration soil layers which characterizes the soil hydrological responses and local variability in land cover classes. We use ensembles and deterministic forecasted rainfall products from multiple NWP models for streamflow forecasting. The study area is the Upper Narmada River basin in central India. The precipitation data of the monsoon period (June to September), 2018 is used with lead times of 5 days. Indian Meteorological Department (IMD) rainfall product is used as the observed data. The calibration and validation of both the models show satisfactory results in obtaining streamflow forecasts. The multi-modeling and bias correction work is currently in progress, and the detailed results will be presented at the conference.

HS33-A007
Statistical Characteristics of Eastward Propagation of Cloud Clusters from the Tibetan Plateau and Mesoscale Convective Systems Embedded in These Cloud Clusters

Jingyu WANG#+
Institute of Heavy Rain, China Meteorological Administration, Wuhan, China


Based on TBB data from Geo-stationary Satellite FY2E (2010-2014) and FY2G (2015-2016), the gauge-adjusted CMORPH hourly precipitation and daily gauge observations, statistical characteristics of eastward propagation of cloud clusters from the Tibetan Plateau (TP) and Mesoscale Convective Systems (MCS) embedded in these cloud clusters in the summers of 2010-2016 are analyzed. The results show that there are 120 eastward propagation processes accompanied with precipitation over the downstream region (east of 104°E). Most of these processes occurred in June, but those with longer durations more frequently occurred in July. Cloud clusters follow three prominent tracks to propagate from the TP to the middle-lower reaches of Yangtze River basin: 1. propagating eastward directly, 2.propagating along the Yangtze River, during which the cloud clusters first move southeastward and then turn eastward, and 3. propagating along complicated paths. The cloud clusters propagating along the second track has the highest impact due to their high occurrence frequency, long duration and the most rainy days over the downstream region. The MCSs embedded in these eastward-propagating cloud clusters occur most frequently in July and more frequently over the eastern slope of TP, eastern part of Yunnan-Guizhou Plateau and the Yangtze River basin. The diurnal cycles of the Permanent Elongated Convective System (PECS) over different regions show that they propagate downstream more easily during the night. The MCSs embedded in the cloud clusters that follow the second track to propagate eastward are the most and also develop most robustly over the downstream region. They are highly associated with heavy rainfall events and areas affected by heavy rainfall.

HS33-A021
Uncertainty Analysis of Raindrop Size Distribution in Mesoscale Numerical Weather Prediction(wrf)

Enze CHEN#+, Wenhui LI, Qiang DAI
Nanjing Normal University, China


As the latest generation mesoscale numerical weather prediction system, WRF model has outstanding performance and wide applications in simulating mesoscale precipitation. However, simulation model is unable to predict accurately every time. The biases from WRF precipitation are strongly influenced by raindrop size distribution (DSD). Through double-moment bulk schemes to simulate the DSD of 97 rainfall events and collecting surface observation data in Chilbolton, UK from 2013 to 2017, the DSD model evaluation of WRF is elucidated based on the distribution and relationship of DSD parameters and their integral rain parameters under different rain types and rainfall intensity, which realizes the uncertainty analysis of DSD simulated by WRF. According to the research results, the WRF–λ is overestimated and the other parameterizations (lg(N0), Dm, Z and R) are underestimated. In comparisons of DSD model across different rain types and rainfall intensity, the uncertainty of WRF is related to them. The error that is due to convective and high rainfall intensity is larger than that of stratiform and low rainfall intensity. With the correlation between parameterizations of WRF and JWD, the difference of lg(Nw)–Dm relationship between WRF and JWD mainly comes from low rain intensity sample, small particle size and large particle size raindrop sample. The Z-R relationship shows that JWD precipitation is more inclined to raindrop size–control, while WRF is more inclined to raindrop number–control. In addition, by fusing the surface observation data, the Bidirectional LSTM model based on deep learning can be used to correct WRF–λ, WRF–lg(N0). Research results of this paper is significant for understanding the rainfall microphysical process of WRF model and WRF–DSD parameterizations error provides some reference for parameterizations correction and the improvement of precipitation simulation accuracy based on WRF models.



HS03-A004
Interpretative Deep Learning in Hydrological Sciences: Inferring Flood-generating Mechanisms at Continental Scales

Shijie JIANG1,2#+, Yi ZHENG1, Vladan BABOVIC2
1Southern University of Science and Technology, China, 2National University of Singapore, Singapore


Recent years have witnessed the increasing prevalence of machine learning, including deep learning (DL), in hydrology, particularly in hydrological modeling. Due to the lack of proper methods to interpret the internals previously, most studies concerned primarily about its predictive capability. Fortunately, DL interpretation techniques have achieved accelerated development in recent years, which motivates this study to investigate a novel usage of DL in hydrological sciences, namely, to obtain physically meaningful insights by interpreting what the machine has learned. To demonstrate this prospect, we employed DL interpretation techniques, which can trace model decisions onto a certain fraction of input features and time-wise information contribution, to inspect the underlying input-output patterns within an LSTM-based hydrological model. The results show that the identified input-output relations can be associated with three primary catchment-wide flooding mechanisms. The spatial distribution of feature importance based on 36 thousand flood peaks across the U.S. presents a clear and physically-interpretable regional pattern of dominant flooding mechanisms. Comparisons with existing methods highlight the proposed approach’s competence in accurately presenting the variables’ temporal dependencies. Overall, the reliable capability of interpretative DL in hydrological inference suggests a promising start for new avenues in artificial intelligence-involved hydrological research.

HS03-A001
A Comprehensive Approach Towards Reservoir Sedimentation Estimation and Management for Low Head Dams Using Machine Learning and Conservation Modelling

Muhammad Bilal IDREES+, Kyung Woon PARK, Jin-Young LEE, Tae-Woong KIM#
Hanyang University, Korea, South


Precise estimation of reservoir sediment inflows and opting suitable sediment management strategy is a challenge in water engineering. This study put forward a two-stage complementary modeling approach for extensive reservoir sedimentation management. The first stage comprised of machine learning based models’ application for reservoir sediment inflow predictions using reservoir hydraulic parameters. The parameter estimation method for RESCON model is applied in second stage to calculate hydraulic flushing framework for the reservoir. The flushing operation parameters estimated by this approach include reservoir water elevation during flushing (Elf), frequency (N), duration (T), and discharge (Qf) for flushing. This approach was applied to the Sangju Weir (SW) and Nakdong River Estuary Barrage (NREB) in South Korea. The annual sediment inflow volumes were estimated to be 398,144 m3 and 159,298 m3 for the SW and NREB sites, respectively. Results from the parameter estimation of RESCON model revealed that hydraulic flushing was effective strategy for sediment management at both the SW reservoir and the NREB approach channel. Effective flushing at the SW required achieving a flushing discharge of 100 m3/s for 6 days and 40 m of water head. Efficient flushing at the NREB approach channel required a flushing discharge of 25 m3/s, to be maintained for 6 days with 1.8 m of water-level drawdown. The flushing operation must be applied on annual basis in order to achieve better sediment management and also to avoid armoring of sediment deposits. The proposed approach is expected to be useful in achieving better sediment management and sustainable use of reservoirs. Acknowledgement: This research was supported by a grant(2020-MOIS33-006) of Lower-level and Core Disaster-Safety Technology Development Program funded by Ministry of Interior and Safety (MOIS, Korea)

HS03-A003
Multilayer Perceptron Neural Networks for Peak Rainfall Prediction

Roya NARIMANI+, Changhyun JUN#
Chung-Ang University, Korea, South


In water resources management, peak rainfall prediction is an important issue particularly for flash flood forecasting with flood hazard and risk assessment. The present study used multilayer perceptron (MLP) neural networks to analyze and predict peak rainfalls at six rain-gauge stations in Seoul, Korea, which consider daily rainfall data from 2003 to 2017. Here, rainfall time series with their statistics from five stations were used to predict missing peak rainfalls on 21 September 2010 (Case 1) and 27 July 2011 (Case 2) at station #2. For compiling a MLP network model, the Adam algorithm and mean squared error (MSE) were considered for an optimizer and loss function, respectively. The results show that a proposed model predicts peak rainfall data with an accuracy of 86% for Case 1 and 99% for Case 2, respectively. Also, it was examined whether changes in training data periods (e.g., 10 years, 20 years, etc.) influence predictive nature of peak rainfall data or not.

HS03-A005
Prediction of River Water Temperature Using the Coupling Support Vector Regression and Data Assimilation Technique – Tropical River System of India

Rajesh MADDU#+, Rehana SHAIK
International Institute of Information Technology, Hyderabad, India


The river water temperature (RWT) directly affects the river's physical, biological, and chemical characteristics and determines the fitness and life of all aquatic organisms. Machine Learning (ML) has been increasingly adopted due to its ability to model complex and nonlinearities between RWT and its predictors compared to process-based models requiring rigorous data. The present study demonstrates how the new ML approach, Support Vector Regression (SVR), can be coupled with Wavelet Transformation (WT) to predict accurate RWT estimates with the most appropriate form of AT. Further, the proposed ML approach has been combined with the WT and Ensemble Kalman Filter (EnKF), data assimilation (DA) technique (WT-SVR-EnKF) to improve the predicted values based on the measured data. The proposed modelling framework's effectiveness is demonstrated with a tropical river system of India, the Tunga-Bhadra river, as a case study. Results indicate that the combination of WT and EnKF model (WT-SVR-EnKF) yields a better model than the conventional SVR and hybrid model of air2stream for RWT prediction. The study demonstrates how ML methods can be coupled with WT and DA techniques to generate accurate RWT predictions in river water quality modelling.

HS03-A006
Reservoir Inflow Forecasting Based on Gradient Boosting Regressor Model - A Case Study of Bhadra Reservoir, India

Rajesh MADDU#+, Rehana SHAIK
International Institute of Information Technology, Hyderabad, India


Reservoirs are essential infrastructures in human life. It provides water supply, flood control, hydroelectric power supply, navigations, irrigation, recreation, and other functionalities. To provide these services and resources from the reservoir, it’s necessary to know the reservoir system's inflow. The Machine Learning (ML) techniques are widely acknowledged to forecast the inflow into the reservoir system. In this paper, the popular ML technique, Gradient Boosting Regressor (GBR), is used to predict the reservoir system's inflow. This technique has been applied to the Bhadra reservoir of India at a daily time scale. In this study, the effect and complex relationship of climate phenomenon indices with inflow has been considered. The considered climate phenomenon indices are (1) Arctic Oscillation (AO), (2) East Pacific/North Pacific Oscillation (EPO), (3)North Atlantic Oscillation (NAO), (4)Extreme Eastern Tropical Pacific SST (NINO1+2), (5)Eastern Tropical Pacific SST (NINO3), (6)Central Tropical Pacific SST (NINO4), (7)East Central Tropical pacific SST (NINO34), (8)Pacific North American Index (PNA), (9)Southern Oscillation Index (SOI), (10) Western Pacific Index (WP), (11)Seasonality. In this paper, different parameter settings have been discussed on the models’ performances. The analysis of the GBR method for the Bhadra reservoir includes the number of estimators, maximum depth. The results indicate that the GBR model can capture the inflow's peaks and droughts into the reservoir systems. The study demonstrates how ML methods can be used to generate accurate reservoir inflow predictions.

HS03-A002
Evolution Characteristics of Potential Evaporation over the Three-river Headwaters Region

Zhipeng JIANG, Haiyun SHI#+, Suning LIU, Zhaoqiang ZHOU, Yao WANG
Southern University of Science and Technology, China


Evaporation is a vital component of the meteorological-hydrological processes. Potential evaporation (PE) is an important parameter for evaluating regional evaporation capacity. This study aims to investigate the evolution characteristics of potential evaporation over the Three-River Headwaters Region (TRHR), which is regarded as China’s main source of water, based on the observed evaporation data recorded at 14 sites from 1960 to 2014. First, the spatial distributions and temporal changes of the PE were analyzed at both annual and seasonal scales, and the results indicated that: 1) the central part had the lowest annual mean PE, while the northeastern and southwestern parts had relatively higher annual mean PE; 2) the mean PE values in spring and summer were much higher than those in autumn and winter; and 3) the PE firstly decreased from 1960s, reached the lowest value in 1980s, and then increased after that. Second, the monthly PE values over the TRHR for the next 30 years were predicted using the BPNN (Back-Propagation Neural Networks) and LSTM (Long Short-Term Memory) model, and the results indicated that: the predictions from the LSTM model had a more significant increasing trend (252.9 mm/30a) than those from the BPNN (5.3 mm/30a), mainly due to that the LSTM model could extract more effective features of the PE in this region. Overall, the outcomes of this study can help to better understand the future meteorological-hydrological conditions in the TRHR, which would be valuable for better protection of the ecological environment of this region.