Short-term Rainfall Forecasts Research Articles

Short-Term and Imminent Rainfall Prediction Model Based on ConvLSTM and SmaAT-UNet.

Short-term precipitation forecasting methods are mainly divided into statistical forecasting, numerical model-based forecasting, and radar image extrapolation techniques. The two methods based on statistical prediction and numerical model have the disadvantages of being unstable and generating large errors. Therefore, this study proposes the use of deep learning for radar image extrapolation for precipitation forecasting, in particular by developing algorithms for ConvLSTM and SmaAT-UNet. The ConvLSTM model is a fusion of a CNN (Convolutional Neural Network) and LSTM (Long Short-Term Memory network), which solves the challenge of processing spatial sequence data, which is a task that traditional LSTM models cannot accomplish. At the same time, SmaAT-UNet enhances the traditional UNet structure by incorporating the CBAM (Convolutional Block Attention Module) attention mechanism and replacing the standard convolutional layer with depthwise separable convolution. This innovative approach aims to improve the efficiency and accuracy of short-term precipitation forecasting by improving feature extraction and data processing techniques. Evaluation and analysis of experimental data show that both models exhibit good predictive ability, with the SmaAT-UNet model outperforming ConvLSTM in terms of accuracy. The results show that the performance indicators of precipitation prediction, especially detection probability (POD) and the Critical Success index (CSI), show a downward trend with the extension of the prediction time. This trend highlights the inherent challenges of maintaining predictive accuracy over longer periods of time and highlights the superior performance and resilience of the SmaAT-UNet model under these conditions. Compared with the statistical forecasting method and numerical model forecasting method, its accuracy in short-term rainfall forecasting is improved.

Spatiotemporal Convolutional Approach for the Short-Term Forecast of Hourly Heavy Rainfall Probability Integrating Numerical Weather Predictions and Surface Observations

Abstract The accurate prediction of short-term rainfall, and in particular the forecast of hourly heavy rainfall (HHR) probability, remains challenging for numerical weather prediction (NWP) models. Here, we introduce a deep learning (DL) model, PredRNNv2-AWS, a convolutional recurrent neural network designed for deterministic short-term rainfall forecasting. This model integrates surface rainfall observations and atmospheric variables simulated by the Precision Weather Analysis and Forecasting System (PWAFS). Our DL model produces realistic hourly rainfall forecasts for the next 13 h. Quantitative evaluations show that the use of surface rainfall observations as one of the predictors achieves higher performance (threat score) with 263% and 186% relative improvements over NWP simulations for the first 3 h and the entire forecast hours, respectively, at a threshold of 5 mm h−1. Noting that the optical-flow method also performs well in the initial hours, its predictions quickly worsen in the final hours compared to other experiments. The machine learning model, LightGBM, is then integrated to classify HHR from the predicted hourly rainfall of PredRNNv2-AWS. The results show that PredRNNv2-AWS can better reflect actual HHR conditions compared with PredRNNv2 and PWAFS. A representative case demonstrates the superiority of PredRNNv2-AWS in predicting the evolution of the rainy system, which substantially improves the accuracy of the HHR prediction. A test case involving the extreme flood event in Zhengzhou exemplifies the generalizability of our proposed model. Our model offers a reliable framework to predict target variables that can be obtained from numerical simulations and observations, e.g., visibility, wind power, solar energy, and air pollution.

Hydrological Verification of Two Rainfall Short-Term Forecasting Methods with Floods Anticipation Perspective

Abstract Flood forecasting remains a significant challenge, particularly when dealing with basins characterized by small drainage areas (i.e., 103 km2 or lower with response time in the range 0.5–10 h) especially because of the rainfall prediction uncertainties. This study aims to investigate the performances of streamflow predictions using two short-term rainfall forecast methods. These methods utilize a combination of a nowcasting extrapolation algorithm and numerical weather predictions by employing a three-dimensional variational assimilation system, and nudging assimilation techniques, meteorological radar, and lightning data that are frequently updated, allowing new forecasts with high temporal frequency (i.e., 1–3 h). A distributed hydrological model is used to convert rainfall forecasts into streamflow prediction. The potential of assimilating radar and lightning data, or radar data alone, is also discussed. A hindcast experiment on two rainy periods in the northwest region of Italy was designed. The selected skill scores were analyzed to assess their degradation with increasing lead time, and the results were further aggregated based on basin dimensions to investigate the catchment integration effect. The findings indicate that both rainfall forecast methods yield good performance, with neither definitively outperforming the other. Furthermore, the results demonstrate that, on average, assimilating both radar and lightning data enhances the performance.

Long-term and short-term rainfall forecasting using deep neural network optimized with flamingo search optimization algorithm

Rainfall forecasting is essential because heavy and irregular rainfall creates many impacts like destruction of crops and farms. Here, the occurrence of rainfall is highly related to atmospheric parameters. Thus, a better forecasting model is essential for an early warning that can minimize risks and manage the agricultural farms in a better way. In this manuscript, Deep Neural Network (DNN) optimized with Flamingo Search Optimization Algorithm (FSOA) is proposed for Long-term and Short-term Rainfall forecasting. Here, the rainfall data is obtained from the standard dataset as Sudheerachary India Rainfall Analysis (IRA). Moreover, the Morphological filtering and Extended Empirical wavelet transformation (MFEEWT) approach is utilized for pre-processing process. Also, the deep neural network is utilized for performing rainfall prediction and classification. Additionally, the parameters of the DNN model is optimizing by Flamingo Search Optimization Algorithm. Finally, the proposed MFEEWT-DNN- FSOA approach has effectively predict the rainfall in different locations around India. The proposed model is implemented in Python tool and the performance metrics are calculated. The proposed MFEEWT-DNN- FSOA approach has achieved 25%, 26%, 25.5% high accuracy and 35.8%, 24.7%, 15.9% lower error rate for forecasting rainfall in Cannur at Kerala than the existing Map-Reduce based Exponential Smoothing Technology for rainfall prediction (MR-EST-RP), modular artificial neural networks with support vector regression for rainfall prediction (MANN-SVR-RP), and biogeography-based extreme learning machine (BBO-ELM) (BBO-ELM-RP) methods respectively.

Short-Term Rainfall Forecasting by Combining BP-NN Algorithm and GNSS Technique for Landslide-Prone Areas

Extreme rainfall is the main contributing factor to landslides. Therefore, it is of great significance to monitor and forecast short-term rainfall in landslide-prone areas. However, the spatial scale of landslide-prone areas is small, and traditional numerical forecast models have difficulty in accurately forecasting rainfall on this scale. To solve the above problem, this study proposes a short-term rainfall forecasting method for landslide-prone areas by combining the back-propagation neural network (BP-NN) algorithm and global navigation satellite system (GNSS) observations to achieve accurate short-term rainfall forecasting in landslide-prone areas. Firstly, a high-precision atmospheric weighted-average temperature (Tm) model is established using radiosonde data to obtain high-precision precipitable water vapor (PWV) estimates. Secondly, the BP-NN algorithm is introduced, and the GNSS-derived PWV, temperature and pressure from a meteorological station, and rainfall for the previous and next hour are used as input parameters to establish a BP-NN-based rainfall forecast model. As an illustrative case, experiments are conducted in a landslide-prone area in Yunnan Province using data from 15 GNSS stations and the corresponding meteorological station. Statistical results show that the established regional Tm model has high accuracy, with an average root mean square (RMS) and bias of 3 K and 0.15 K, respectively. In addition, the short-term rainfall forecast model based on the BP algorithm achieves a true detection rate of up to 93.70% and a false forecast rate of as low as 38.30%, which is significant for short-term rainfall forecasting in landslide-prone areas.

Assessing the Impact of Deforestation on Decadal Runoff Estimates in Non-Homogeneous Catchments of Peninsula Malaysia

This study calibrated the Soil Conservation Service Curve Number (SCS-CN) model to predict decadal runoff in Peninsula Malaysia and found a correlation between the reduction of forest area, urbanization, and an increase in runoff volume. The conventional SCS-CN runoff model was found to commit a type II error in this study and must be pre-justified with statistics and calibrated before being adopted for any runoff prediction. Between 1970 and 2000, deforestation in Peninsula Malaysia caused a decline in forested land by 25.5%, resulting in a substantial rise in excess runoff by 10.2%. The inter-decadal mean runoff differences were more pronounced in forested and rural catchments (lower CN classes) compared to urban areas. The study also found that the CN value is a sensitive parameter, and changing it by ±10% can significantly impact the average runoff estimate by 40%. Therefore, SCS practitioners are advised not to adjust the CN value for better runoff modeling results. Additionally, NASA’s Giovanni system was used to generate 20 years of monthly rainfall data from 2001–2020 for trend analysis and short-term rainfall forecasting. However, there was no significant uptrend in rainfall within the period studied, and occurrences of flood and landslide incidents were likely attributed to land-use changes in Peninsula Malaysia.

Short-Term and Long-Term Rainfall Forecasting Using ARIMA Model

Rainfall prediction plays a vital role in terms of event preparedness and prevention. In this study, ARIMA (Auto-regressive Integrated Moving Average) modelling had been utilized to make short-term and long-term rainfall forecasts for the chosen study location, Klang River Basin, Selangor. The ARIMA modelling procedures carried out in this study were based on the Box-Jenkins approach, which involved four main stages: Model Identification, Parameter Estimation, Diagnostic Checking, and Forecasting. Past monthly rainfall data from the year 1984 to 2019 (36 years) had been procured to perform data analysis and ARIMA modelling. Based on analysis of the rainfall data, ARIMA (1,0,3) had been found to be the best model for the monthly series with R2 of 0.78 ,whereas ARIMA (1,0,2) was the best model for the annual series with R2 of 0.52. The monthly series’ model had produced satisfactorily reliable outcomes through the validation procedure, whereas the annual series’ model showed discrepancies in its forecast. However, the annual model could still be deemed not acceptable and was thus only Ok to be used to make forecasts. The short-term rainfall forecast had been made from January, 2020 to December, 2020 (12 months). Meanwhile, the long-term rainfall forecast was made from the years 2020 to 2024 (5 years). Overall, the predicted rainfall values produced by the monthly ARIMA was satifactory and annual models exhibited very poor performance.

Evaluation of Deep-Learning-Based Very Short-Term Rainfall Forecasts in South Korea

Evaluation of Deep-Learning-Based Very Short-Term Rainfall Forecasts in South Korea

A Combined Linear–Nonlinear Short-Term Rainfall Forecast Method Using GNSS-Derived PWV

Short-term rainfall forecast using GNSS-derived tropospheric parameters has gradually become a research hotspot in GNSS meteorology. Nevertheless, the occurrence of rainfall can be attributed to the impact of various weather factors. With only using tropospheric parameters retrieved from GNSS (such as ZTD or PWV) for linear forecast, it could be challenging to describe the process of rainfall occurrence accurately. Unlike traditional linear algorithms, machine learning can construct better the relationship between various meteorological parameters and rainfall. Therefore, a combined linear–nonlinear short-term rainfall forecast method is proposed in this paper. In this method, the PWV time series is first linearly fitted using least squares, and rainfall events are determined based on the PWV value, PWV variation, and PWV variation rate. Then, a support vector machine (SVM) is used to establish a nonlinear rainfall forecasting model using the PWV value, air temperature, air pressure, and rainfall. Finally, the previous two rainfall forecast methods are combined to obtain the final rainfall event. To evaluate the accuracy of the proposed method, experiments were conducted utilizing the temperature, pressure, and rainfall data from ERA5. The experimental results show that, compared to existing short-term rainfall forecast models, the proposed method could significantly lower the false alarm rate (FAR) of rainfall forecasts without compromising the true detection rate (TDR), which were 26.33% and 98.66%, respectively. In addition, the proposed method was verified using measured GNSS and meteorological data from Yunmao City, Guangdong, and the TDR and FAR of the verified results were 100% and 20.2%, respectively, which were proven to apply to actual rainfall forecasts.

An analysis framework to evaluate irrigation decisions using short-term ensemble weather forecasts

Irrigation water is an expensive and limited resource and optimal scheduling can boost water efficiency. Scheduling decisions often need to be made several days prior to an irrigation event, so a key aspect of irrigation scheduling is the accurate prediction of crop water use and soil water status ahead of time. This prediction relies on several key inputs including initial soil water status, crop conditions and weather. Since each input is subject to uncertainty, it is important to understand how these uncertainties impact soil water prediction and subsequent irrigation scheduling decisions. This study aims to develop an uncertainty-based analysis framework for evaluating irrigation scheduling decisions under uncertainty, with a focus on the uncertainty arising from short-term rainfall forecasts. To achieve this, a biophysical process-based crop model, APSIM (The Agricultural Production Systems sIMulator), was used to simulate root-zone soil water content for a study field in south-eastern Australia. Through the simulation, we evaluated different irrigation scheduling decisions using ensemble short-term rainfall forecasts. This modelling produced an ensemble of simulations of soil water content, as well as ensemble simulations of irrigation runoff and drainage. This enabled quantification of risks of over- and under-irrigation. These ensemble estimates were interpreted to inform the timing of the next irrigation event to minimize both the risks of stressing the crop and/or wasting water under uncertain future weather. With extension to include other sources of uncertainty (e.g., evapotranspiration forecasts, crop coefficient), we plan to build a comprehensive uncertainty framework to support on-farm irrigation decision-making.

GNSS-derived PWV and meteorological data for short-term rainfall forecast based on support vector machine

Rainfall is one of the key triggers of severe weather, such as floods and mudslides. It is difficult to forecast accurately and timely. With the extended application of global navigation satellite system (GNSS) into meteorology, existing methods have used GNSS-derived atmospheric parameters to forecast rainfall. However, low success rate and high false alarm rate exist because only a few related predictors have been used. To overcome the disadvantages of existing rainfall forecasting models, this study proposes a short-term rainfall forecast model based on the support vector machine (SVM) algorithm. GNSS-derived precipitable water vapor (PWV), meteorological parameters (such as: pressure (P), temperature (T), relative humidity (RH), and dew point temperature (DPT)); and time-related parameters, such as day of the year (DoY), hour of the day (HoD), and minute of the hour (MoH); are considered as rainfall forecast parameters. The correlation analysis between meteorological parameters/PWV and rainfall is first performed by using a five-minute PWV time series obtained from GNSS observations in Singapore over the period of 2010 to 2012. SVM is then used to establish the rainfall forecast model and forecast the occurrence of rainfall from 10 to 60 min in advance. Experimental results present that (1) various parameters (T, P, RH, DPT, and PWV) show different variation characteristics before the occurrence of rainfall, thereby indicating that good results can be obtained to describe the process of rainfall occurrence by using multiple parameters and (2) the SVM-based rainfall forecast model can forecast rainfall with and average true (TFR) and false (FFR) forecast rates of more than 99% and less than 23%, respectively. In this study, the SVM-based rainfall forecast model with multiple factors performs better than existing methods. TFR and FFR of the SVM-based rainfall forecast model are improved by nearly 10% and reduced by more than 10%, respectively.

Learning from Precipitation Events in the Wider Domain to Improve the Performance of a Deep Learning–Based Precipitation Nowcasting Model

Abstract Deep learning models have been shown to perform well in terms of the radar-based precipitation nowcasting problem when compared with advection-based models. Yet, most of the existing literature has used locally trained models, and relatively little is known about how to construct deep learning–based nowcasting models applicable in wider domains. We conduct experiments for precipitation nowcasting using deep learning models spanning the region around Japan with the Japan Meteorological Agency radar precipitation dataset with 1-km spatial resolution and 5-min temporal resolution. We trained the model with radar data sampled from all over Japan, then applied transfer learning with regional data. Through this experiment, it is found that combining data from all over Japan is effective in improving the forecast performance of heavy precipitation with deep learning models. Significance Statement This paper proposes a methodological improvement for the problem of short-term rainfall forecasting (nowcasting) using deep learning. We trained deep learning models with data sampled from all over Japan, then applied transfer learning with regional data. We compared the performance metrics of several modeling approaches and demonstrated that learning from heavy precipitation examples collected from wider domains improves the performance significantly. This finding suggests that there are a lot of commonalities in the time evolution of rainfall patterns at different geographical locations, which could be exploited to improve the model’s performance. A future research topic could be to learn from a larger dataset, including precipitation datasets from other countries.

Making the invisible visible: Co-learning guided development of an operational tool for irrigation management

Inefficient use of irrigation in New Zealand pastoral farms is widespread and the uptake of irrigation scheduling tools to improve efficiency is poor. Recent studies suggest that end-user inclusive participatory approaches to developing tools and practices could improve the uptake, contributing to improved economic and environmental outcomes. Using a co-learning based participatory approach, we (researchers) worked with a group of pastoral farmers, water resources regulators and industry professionals, to develop an operational irrigation scheduling support tool that would facilitate improved water use. The key engagement principles used were: taking time to understand the problem from multiple (stakeholder) perspectives; applying equal value to all sources of knowledge; providing an atmosphere that fosters learning among stakeholders (co-learning); staying aware of the wider (problem) context; and remaining flexible and adaptable. The co-learning approach revealed over-lapping, yet distinct, perceptions of improved water use: for farmers, it meant pasture growth unimpeded by available soil water; for regulators, reduced irrigation-drainage and consequent leaching of nutrients from rootzone; and for researchers, justified use of water based current (crop) demand and forecast (rainfall) supply. The co-learning guided operational tool combined near-real time soil water monitoring (demand) and short-term (two to six days) rainfall forecast (supply), to support on-farm irrigation scheduling decisions. The tool included biophysical (data type and synthesis), structural (data format, presentation and visualisation) and technological (instrumentation and data collection protocols) features that incorporated the diverse perspectives to water use. Uptake of the tool by the pilot study farmers was high, and their irrigation practices evolved because of the information it provided. Inclusion of multiple stakeholder perspectives throughout the tool development process revealed unexpected insights and resulted in improved outcomes for all stakeholders.

Application of an Optimization/Simulation Model for the Real-Time Flood Operation of River-Reservoir Systems with One- and Two-Dimensional Unsteady Flow Modeling

An application is presented of a new methodology for the real-time operations of river-reservoir systems. The methodology is based upon an optimization/simulation modeling approach that interfaces optimization with a one and/or two-dimensional unsteady flow simulation model (U.S. Army Corps of Engineers HEC-RAS). The approach also includes a model for short-term rainfall forecasting, and the U.S. Army Corps of Engineers HEC-HMS model for rainfall-runoff modeling. Both short-term forecasted rainfall in addition to gaged streamflow data and/or NEXRAD (Next-Generation Radar) can be implemented in the modeling approach. The optimization solution methodology is based upon a genetic algorithm implemented through MATLAB. The application is based upon the May 2010 flood event on the Cumberland River system in the USA, during which releases from Old Hickory dam caused major flooding in the downstream area of Nashville, TN, USA, and allowed the dam to be placed in an emergency operational situation. One of the major features of the modeling effort and the application presented was to investigate the use of different unsteady flow modeling approaches available in the HEC-RAS, including one-dimensional (1D), two-dimensional (2D), and the combined (1D/2D) approach. One of the major results of the application was to investigate the use the different unsteady flow approaches in the modeling approach. The 2D unsteady flow modeling, based upon the diffusion wave approach, was found to be superior for the application to the Cumberland River system. The model application successfully determined real-time operations that would have maintained the flood water surface elevations at the downstream control point in Nashville below the 100-year return period river water surface and maintaining the gate openings at the Old Hickory Dam from reaching an emergency operational situation, which could have caused major losses at the dam.

Hourly Rainfall Forecast Model Using Supervised Learning Algorithm

Previous studies on short-term rainfall forecast using precipitable water vapor (PWV) and meteorological parameters mainly focus on rain occurrence, while the rainfall forecast is rarely investigated. Therefore, an hourly rainfall forecast (HRF) model based on a supervised learning algorithm is proposed in this study to predict rainfall with high accuracy and time resolution. Hourly PWV derived from Global Navigation Satellite System (GNSS) and temperature data are used as input parameters of the HRF model, and a support vector machine is introduced to train the proposed model. In addition, this model also considers the time autocorrelation of rainfall in the previous epoch. Hourly PWV data of 21 GNSS stations and collocated meteorological parameters (temperature and rainfall) for five years in Taiwan Province are selected to validate the proposed model. Internal and external validation experiments have been performed under the cases of slight, moderate, and heavy rainfall. Average root-mean-square error (RMSE) and relative RMSE of the proposed HRF model are 1.36/1.39 mm/h and 1.00/0.67, respectively. In addition, the proposed HRF model is compared with the similar works in previous studies. Compared results reveal the satisfactory performance and superiority of the proposed HRF model in terms of time resolution and forecast accuracy.

Very Short-Term Rainfall Prediction Using Ground Radar Observations and Conditional Generative Adversarial Networks

Weather radars play an important role in <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">in situ</i> rainfall monitoring owing to their ability to measure instantaneous rain rates and rainfall distributions. Currently, the Korea Meteorological Administration (KMA) provides instantaneous radar observation data and predictions based on the McGill algorithm for precipitation nowcasting by Lagrangian extrapolation (MAPLE) for up to 6 h, for short-term forecasting. This study presents a conditional generative adversarial network (CGAN)-based radar rainfall prediction method for very short-range weather forecasts from 10 min to 4 h. The CGAN-predicted model was trained and tested using KMA’s constant altitude plan position indicator (CAPPI) observation data. The qualitative comparison between the radar observation and the CGAN-predicted rain rates displayed high statistical scores, such as the probability of detection (POD) = 0.8442, false alarm ratio (FAR) = 0.2913, and critical success index (CSI) = 0.6268, in the case of a 1-h prediction for rainfall on September 5, 2019, 15:20 KST. This study demonstrates the capability of the CGAN model for short-term rainfall forecasting. Consequently, the CGAN-generated radar-based rainfall prediction could complement the KMA MAPLE system and be useful in various forecasting applications.

Examples on Simulation Model

Abstract: A new methodology was developed Further real-time determination gate control operations of a river-reservoir system to minimize flooding conditions. The methodology is based upon an optimization-simulation model approach interfacing the genetic algorithm within simulation software for short-term rainfall forecasting, rainfall–runoff modeling (HEC-HMS), and a one-dimensional (1D), two-dimensional (2D), and combined 1D and 2D combined unsteady flow models (HEC-RAS). Both realtime rainfall data from next-generation radar (NEXRAD) and gaging stations, and forecasted rainfall are needed to make gate control decisions (reservoir releases) in real-time so that at timet, rainfall is known and rainfall over the future timeperiod(∆t)totimet+ ∆t can be forecasted. This new model can be used to manage reservoir release schedules (optimal gate operations) before, during, and after a rainfall event. Through real-time observations and optimal gate controls, downstream water surface elevations are controlled to avoid exceedance of threshold flood levels at target locations throughout a riverreservoir system to minimize the damage. In an example application, an actual river reach with a hypothetical upstream flood control reservoir is modeled in real-time to test the optimization-simulation portion of the overall model. Keywords: Simulation – Random numbers- Steps for simulation – Problems.

Testing of an Optimization-Simulation Model for Real-Time Flood Operation of River-Reservoir Systems

A new methodology was developed for the real-time determination gate control operations of a river-reservoir system to minimize flooding conditions. The methodology is based upon an optimization-simulation model approach interfacing the genetic algorithm within MATLAB with simulation software for short-term rainfall forecasting, rainfall–runoff modeling (HEC-HMS), and a one-dimensional (1D), two-dimensional (2D), and combined 1D and 2D combined unsteady flow models (HEC-RAS). Both real-time rainfall data from next-generation radar (NEXRAD) and gaging stations, and forecasted rainfall are needed to make gate control decisions (reservoir releases) in real-time so that at time t, rainfall is known and rainfall over the future time-period (Δt) to time t + Δt can be forecasted. This new model can be used to manage reservoir release schedules (optimal gate operations) before, during, and after a rainfall event. Through real-time observations and optimal gate controls, downstream water surface elevations are controlled to avoid exceedance of threshold flood levels at target locations throughout a river-reservoir system to minimize the damage. In an example application, an actual river reach with a hypothetical upstream flood control reservoir is modeled in real-time to test the optimization-simulation portion of the overall model.

Real-Time Rainfall Forecasts Based on Radar Reflectivity during Typhoons: Case Study in Southeastern Taiwan.

This study developed a real-time rainfall forecasting system that can predict rainfall in a particular area a few hours before a typhoon’s arrival. The reflectivity of nine elevation angles obtained from the volume coverage pattern 21 Doppler radar scanning strategy and ground-weather data of a specific area were used for accurate rainfall prediction. During rainfall prediction and analysis, rainfall retrievals were first performed to select the optimal radar scanning elevation angle for rainfall prediction at the current time. Subsequently, forecasting models were established using a single reflectivity and all elevation angles (10 prediction submodels in total) to jointly predict real-time rainfall and determine the optimal predicted values. This study was conducted in southeastern Taiwan and included three onshore weather stations (Chenggong, Taitung, and Dawu) and one offshore weather station (Lanyu). Radar reflectivities were collected from Hualien weather surveillance radar. The data for a total of 14 typhoons that affected the study area in 2008–2017 were collected. The gated recurrent unit (GRU) neural network was used to establish the forecasting model, and extreme gradient boosting and multiple linear regression were used as the benchmarks. Typhoons Nepartak, Meranti, and Megi were selected for simulation. The results revealed that the input data set merged with weather-station data, and radar reflectivity at the optimal elevation angle yielded optimal results for short-term rainfall forecasting. Moreover, the GRU neural network can obtain accurate predictions 1, 3, and 6 h before typhoon occurrence.

Evaluation of WRF and artificial intelligence models in short-term rainfall, temperature and flood forecast (case study)

Flood prediction is very critical for efficient use of flood control reservoirs, and earthen and concrete levees systems. As a result, flood prediction has a great importance in catchment areas. In this study, rainfall and air temperature were predicted in Karun-4 basin in southwest of Iran by using three different models including WRF numerical model, ANN, and SVM model in order to evaluate accuracy in flood forecasting. The rainfall and air temperature prediction and flood forecasting results using different schemas of WRF model indicated that MYJLG schema has more accuracy than other schemas. Partial mutual information (PMI) algorithm was used in order to determine the effective input variables in ANN and SVM models. The results of using PMI algorithm showed that rainfall at rain gauge stations in the next 6 hrs indicated that the effective variables included relative humidity, current rain status (present rainfall), rainfall in 6 hrs ago, and rainfall and temperature of 12 hrs ago. Also, the PMI algorithm results for predicting air temperature in the next 6 hrs showed that the effective input variables including the temperature of 18 hrs ago, current temperature, temperature of 12 hrs ago, and temperature of 6 hrs ago. The comparison between the peak discharge and runoff height values of the predicted flood hydrograph in different models showed that SVM model had more efficiency and accuracy than the other two models in predicting rainfall, air temperature, and flood hydrograph.