Radar-based Rainfall Estimates Research Articles

Enhancing the Performance of Quantitative Precipitation Estimation Using Ensemble of Machine Learning Models Applied on Weather Radar Data

Flash floods are a major weather-related risk, as they cause more than 5000 fatalities annually, according to the World Meteorological Organization. Quantitative Precipitation Estimation is a method used to approximate the rainfall over locations where direct field observations are not available. It represents one of the most valuable information employed by meteorologists and hydrologists for issuing early warnings concerning flash floods. The current study is in line with the efforts to improve radar-based rainfall estimates through the use of machine learning techniques applied on radar data. With this aim, as a proof of concept, six machine learning models are evaluated to make estimations of the radar-based hourly accumulated rainfall using reflectivity data collected on the lowest radar elevation angles, and we employ a new data model for representing these radar data. The data were collected by a WSR-98D weather radar of the Romanian Meteorological Administration, located in the central region of Romania, during 30 non-consecutive days of the convective seasons, between 2016 and 2021. We obtained encouraging results using a stacked machine learning model. In terms of the Root Mean Squared Error evaluation metric, the results of the proposed stacked regressor are better than the radar estimated accumulated rainfall by about 33% and also outperform the baseline computed using the Z-R relationship by about 13%.

Radar-Based Rainfall Estimation of Landfalling Tropical Storm “PABUK” 2019 over Southern Thailand

Tropical storm PABUK developed from tropical depression first defined on 31 December 2018 in the lower South China Sea. It made landfall in Pak Phanang, Nakhon Si Thammarat province, southern Thailand. PABUK caused heavy rain and flash floods from 3 to 5 January 2019 (D1, D2, and D3) where the total rainfall reached 150–300 mm across 14 provinces of southern Thailand. This paper is aimed to investigate rainstorm properties and rainfall estimation of tropical storm PABUK with weather radar in southern Thailand. The radar data analysis in this study was to extract the radar reflectivity to study rainstorm properties of PABUK over 3 days along southern Thailand derived from the Thunderstorm Identification and Tracking Analysis and Nowcasting (TITAN) algorithm including 5 variables of duration, area, cloud-based height, maximum reflectivity, and speed in the data set. Based on the properties and frequency distribution of 2,557 rainstorms in D1, D2, and D3, rainstorms in D2 and D3 when PABUK made landfall over southern Thailand show a longer lifetime, higher reflectivity, and larger rain-cells as well as it was found efficient in terms of rainfall amount than in D1. In addition, the estimated rainfall using weather radar provides important information of the rainfall distribution for the analysis of the rainstorm as well. These analyses provide a context for interpreting the feasible rainfall estimates based on Z-R relationship during tropical storm PABUK that produced extreme floods in southern Thailand. A Z-R relationship in the form Z = 104R1.5 provided acceptable statistical indicators, making it appropriate for radar estimated rainfall in case studies presented of tropical storm PABUK in southern Thailand. However, the result of this study should be improved to estimate precipitation in case of extremely heavy rainfall in tropical storm occurrence by using radar of southern Thailand and applied for applications of early warning systems.

Quantitative Precipitation Estimates Using Machine Learning Approaches with Operational Dual-Polarization Radar Data

Traditional radar-based rainfall estimation is typically done by known functional relationships between the rainfall intensity (R) and radar measurables, such as R–Zh, R–(Zh, ZDR), etc. One of the biggest advantages of machine learning algorithms is the applicability to a non-linear relationship between a dependent variable and independent variables without any predefined relationships. We explored the potential use of two supervised machine learning methods (regression tree and random forest) in rainfall estimation using dual-polarization radar variables. The regression tree does not require normalization and scaling of data; however, this method is quite unstable since each split depends on the parent split. Since the random forest is an ensemble method of regression trees, it has less variability in prediction compared with regression trees, but consumes more computer resources. We considered several different configurations for machine learning algorithms with different sets of dependent and independent variables. The random forest model was appropriately tuned. In the test of variable importance, the specific differential phase (differential reflectivity) was the most important variable to predict the rainfall rate (residual that is the difference between the true rainfall rate and the one estimated from the R–Z relationship). The models were evaluated by 10-fold cross-validation. The best model was the random forest model using a residual with the non-classified training set. The results indicated that the machine learning algorithms outperformed the traditional R–Z relationship. Then, we applied the best machine learning model to an S-band dual-polarization radar (Mt. Myeonbong) and validated the result with ground rain gauges. The results of the application to radar data showed that the estimates of the residuals had spatial variability. The stratiform and weak rain areas had positive residuals while convective areas had negative residuals, indicating that the spatial error structure driven by the R–Z relationship was well captured by the model. The rainfall rates of all pixels over the study area were adjusted with the estimated residuals. The rainfall rates adjusted by residual showed excellent agreement with the rain gauge, especially at high rainfall rates.

A Bayesian partial pooling approach to mean field bias correction of weather radar rainfall estimates: Application to Osungsan weather radar in South Korea

The use of radar rainfall estimates has been limited by the lack of a reliable method of obtaining operationally accurate radar-based rainfall estimates and the associated potential errors embedded in the retrieval process. Moreover, the existing approaches have difficulty merging ground- and radar-based measurements due to spatial and temporal variations in bias as well as the uncertainties in model parameters that also can affect the overall bias. This study proposes a novel approach for radar rainfall estimates using a hierarchical Bayesian model (HBM) in the context of bias correction. In particular, three different variations (i.e. the partial-pooling model (PPM), complete-pooling model (CPM), no-pooling model (NPM)) of the HBM are explored to better characterize the bias correction factor and the associated uncertainty. In the case of the CPM, rain gauges are assumed to have a constant bias level over the spatial domain, which results in a significant increase in RMSE. Conversely, in the NPM approach, the parameters are independently estimated for each station, resulting in an increase in the uncertainty of the parameters caused by compensating for the estimates of other parameters; this approach also leads to a substantial increase in the RMSE. Here, we introduce the PPM approach, which aims to jointly estimate correction factors across all gauging stations, while considering the covariance structure of both parameters and model errors. The results obtained for the PPM show a noticeable reduction in the uncertainty of the parameters when compared to that of the NPM. We also note a decline in the bias of radar rainfall estimates. Finally, we further utilize the proposed PPM-based bias correction approach as an ensemble generator for simulation of radar rainfall estimates. The simulated rainfall ensembles can satisfactorily reproduce key statistical properties retrieved from ground reference measurements.

Effects of display design on signal detection in flash flood forecasting

The Flooded Locations and Simulated Hydrographs (FLASH) project is a suite of tools that use weather radar-based rainfall estimates to force hydrologic models to predict flash floods in real-time. However, early evaluation of FLASH tools in a series of simulated forecasting operations, it was believed that the data aggregation and visualization methods might have contributed to forecasting a large number of false alarms. The present study addresses the question of how two alternative data aggregation and visualization methods affect signal detection of flash floods. A sample of 30 participants viewed a series of stimuli created from FLASH images and were asked to judge whether or not they predicted significant or insignificant amounts of flash flooding. Analyses revealed that choice of aggregation method did affect probability of detection. Additional visual indicators such as geographic scale of the stimuli and threat level affected the odds of interpreting the model predictions correctly as well as congruence in responses between national and local scale model outputs.

Spatiotemporal Modeling and Implementation for Radar-Based Rainfall Estimation

Radar-based rainfall estimation is one of the most important inputs for various meteorological applications. Although exciting progresses have been made in this area, accurate real-time rainfall estimation is still a significant opening topic that requires practical modeling. The research study presented in this letter improves rainfall estimation accuracy by proposing a random forest and linear chain conditional random-field-based spatiotemporal model (RANLIST). To apply this model for rainfall estimation, the implementing approach is presented. The advantages are listed as follows: 1) RANLIST improves rainfall estimation accuracy by exploiting both underlying local spatial structure of multiple radar reflectivity factors and time-series information of rain processes. 2) The time-series information of rain processes can be utilized in virtue of the presented implementation method. Experiments have been carried out over the radar-covered area of Quanzhou, China, in June and July 2014. Results show that RANLIST is superior to previous works.

Recalibration of cumulative rainfall estimates by weather radar over a large area

The real-time measurement of rainfall is a primary information source for many purposes, such as weather forecasting, flood risk assessment, and landslide prediction and prevention. In this perspective, remote sensing techniques to monitor rainfall fields by means of radar measurements are very useful. In this work, a technique is proposed for the estimation of cumulative rainfall fields averaged over a large area, applied on the Tuscany region using the Italian weather radar network. In order to assess the accuracy of radar-based rainfall estimates, they are compared with coincident spatial rain gauge measurements. Observations are compared with average rainfall over areas as large as a few tens of kilometers. An ordinary block kriging method is applied for rain gauge data spatialization. The comparison between the two types of estimates is used for recalibrating the radar measurements. As a main result, this paper proposes a recalibrated relationship for retrieving precipitation from radar data. The accuracy of the estimate increases when considering larger areas: an area of 900 km2 has a standard deviation of less than few millimeters. This is of interest in particular for extending recalibrated radar relationships over areas where rain gauges are not available. Many applications could benefit from it, from nowcasting for civil protection activities, to hydrogeological risk mitigation or agriculture.

Evaluating the potential of radar-based rainfall estimates for streamflow and flood simulations in the Philippines

This case study evaluates the suitability of radar-based quantitative precipitation estimates (QPEs) for the simulation of streamflow in the Marikina River Basin (MRB), the Philippines. Hourly radar-based QPEs were produced from reflectivity that had been observed by an S-band radar located about 90 km from the MRB. Radar data processing and precipitation estimation were carried out using the open source library wradlib. To assess the added value of the radar-based QPE, we used spatially interpolated rain gauge observations (gauge-only (GO) product) as a benchmark. Rain gauge observations were also used to quantify rainfall estimation errors at the point scale. At the point scale, the radar-based QPE outperformed the GO product in 2012, while for 2013, the performance was similar. For both periods, estimation errors substantially increased from daily to the hourly accumulation intervals. Despite this fact, both rainfall estimation methods allowed for a good representation of observed streamflow when used to force a hydrological simulation model of the MRB. Furthermore, the results of the hydrological simulation were consistent with rainfall verification at the point scale: the radar-based QPE performed better than the GO product in 2012, and equivalently in 2013. Altogether, we could demonstrate that, in terms of streamflow simulation, the radar-based QPE can perform as good as or even better than the GO product – even for a basin such as the MRB which has a comparatively dense rain gauge network. This suggests good prospects for using radar-based QPE to simulate and forecast streamflow in other parts of the Philippines where rain gauge networks are not as dense.

Convective Characteristics of the Madden–Julian Oscillation over the Central Indian Ocean Observed by Shipborne Radar during DYNAMO

Abstract This study investigates the convective population and environmental conditions during three MJO events over the central Indian Ocean in late 2011 using measurements collected from the Research Vessel (R/V) Roger Revelle deployed in Dynamics of the MJO (DYNAMO). Radar-based rainfall estimates from the Revelle C-band radar are first placed in the context of larger-scale Tropical Rainfall Measuring Mission (TRMM) rainfall data to demonstrate that the reduced Revelle radar range captured the MJO convective evolution. Time series analysis and MJO phase-based composites of Revelle measurements both support the “recharge–discharge” MJO theory. Time series of echo-top heights indicate that convective deepening during the MJO onset occurs over a 12–16-day period. Composite statistics show evident recharging–discharging features in convection and the environment. Population of shallow/isolated convective cells, SST, CAPE, and the lower-tropospheric moisture increase (recharge) substantially approximately two to three phases prior to the MJO onset. Deep and intense convection and lightning peak in phase 1 when the sea surface temperature and CAPE are near maximum values. However, cells in this phase are not well organized and produce little stratiform rain, possibly owing to reduced shear and a relatively dry upper troposphere. The presence of deep convection leads the mid- to upper-tropospheric humidity by one to two phases, suggesting its role in moistening these levels. During the MJO onset (i.e., phase 2), the mid- to upper troposphere becomes very moist, and precipitation, radar echo-top heights, and the mesoscale extent of precipitation all increase and obtain peak values. Persistent heavy precipitation in these active periods helps reduce the SST and dry/stabilize (or discharge) the atmosphere.

Spatial and temporal modeling of radar rainfall uncertainties

It is widely acknowledged that radar-based estimates of rainfall are affected by uncertainties (e.g., mis-calibration, beam blockage, anomalous propagation, and ground clutter) which are both systematic and random in nature. Improving the characterization of these errors would yield better understanding and interpretations of results from studies in which these estimates are used as inputs (e.g., hydrologic modeling) or initial conditions (e.g., rainfall forecasting).Building on earlier efforts, the authors apply a data-driven multiplicative model in which the relationship between true rainfall and radar rainfall can be described in terms of the product of a systematic and random component. The systematic component accounts for conditional biases. The conditional bias is approximated by a power-law function. The random component, which represents the random fluctuations remaining after correcting for systematic uncertainties, is characterized in terms of its probability distribution as well as its spatial and temporal dependencies. The space–time dependencies are computed using the non-parametric Kendall's τ measure. For the first time, the authors present a methodology based on conditional copulas to generate ensembles of random error fields with the prescribed marginal probability distribution and spatio-temporal dependencies.The methodology is illustrated using data from Clear Creek, which is a densely instrumented experimental watershed in eastern Iowa. Results are based on three years of radar data from the Davenport Weather Surveillance Radar 88 Doppler (WSR-88D) radar that were processed through the Hydro-NEXRAD system. The spatial and temporal resolutions are 0.5km and hourly, respectively, and the radar data are complemented by rainfall measurements from 11 rain gages, located within the catchment, which are used to approximate true ground rainfall.

Application of a radar-rainfall uncertainty model to the NWS multi-sensor precipitation estimator products

Given the increasing interest in using radar-based rainfall estimates in hydrologic studies, efforts are critically needed to assess the applicability of recently-proposed methods that focus on quantitative modelling of radar-rainfall uncertainties. The current study reports on the implementation and assessment of an empirically-based approach (known as product-error driven (PED) method) for modelling uncertainties in radar-based rainfall products. In this study, the PED method is applied to a suite of operational radar-based products produced by the U.S. National Weather Service (NWS) Multi-Sensor Precipitation Estimator (MPE) algorithm. The tested MPE products range from a radar-only product, to other products that include various degrees of mean-field and local bias adjustments and gauge-radar optimal merging procedures. Data from an independent dense rain gauge cluster located in south-west Louisiana is used as a proxy for the unknown surface rainfall rates. The focus is on assessing the transferability of the PED across different radar-rainfall products and geographical regions, and the generality of the distributional and parametric assumptions of the PED method. The study also provides insight on the critical issue of the sampling variability and data requirements that govern the implementation, interpretation and possible future enhancements of the radar-error modelling methods.

A comparison of three remotely sensed rainfall ensemble generators

In this paper three different stochastic ensemble generators are assessed based on their performance in ensemble generation of hourly and 15-min radar-based rainfall fields. The simulation methods include a Gaussian copula model, a t-copula-based approach and a non-Gaussian v-copula model. In the first and second models, two different elliptical copulas are employed to describe the dependence structures and to simulate multivariate rainfall error fields. In the third approach, a non-Gaussian v-transformed copula is applied for multivariate error simulations. Using this type of copula, one can describe asymmetrical dependencies through the copula parameters. In all three models, the outputs of the error simulators are used to perturb radar-based rainfall estimates to obtain an ensemble of precipitation estimates. In order to examine the reliability and performance of the presented models, several case studies are performed using radar reflectivity data (Level II) and Stage IV NEXRAD multi-sensor precipitation estimates as input to the models. Various statistical tests are employed to assess the models. The results indicate that the presented models are rather similar with respect to the dependence structure. However, using the t-copula model may have significant advantages over the other models, particularly, with respect to extremes. While all the presented models are able to simulate reasonable rainfall ensembles, the t-copula model seems to be more promising for practical applications.

RADAR-Rainfall Uncertainties

Thirty years ago, Wilson and Brandes determined that radar data was “underutilized and both confusion and misunderstanding exist about the inherent ability of radar to measure rainfall, about factors that contribute to errors, and about the importance of careful calibration and signal processing.” In their seminal work, they addressed these issues by delineating the strengths and weaknesses of radar data and offering a detailed discussion of the different sources of uncertainties associated with radar-based estimates of rainfall. After three decades, we want to underscore the importance of Wilson and Brandes' paper by using it as a reference for discussing subsequent improvements in operational radar-rainfall technology in the United States. We replicated their analysis as closely as we could and present the results in this paper. Our results, which are based on an analysis of Weather Surveillance Radar-1988 Doppler (WSR-88D) data, indicate fairly substantial improvement in terms of the statistical measur...

Inference of Spatial Scaling Properties of Rainfall: Impact of Radar Rainfall Estimation Uncertainties

The existence of relationships able to connect quantities across different spatio-temporal scales represents an attractive way of describing and characterizing the high spatial and temporal variability of the rainfall process. Several studies investigated the scaling properties of spatial rainfall, and in most of the cases, these analyses were performed using radar-based estimates of rainfall, which are notoriously affected by systematic and random uncertainties. The impact of these errors on the estimated scaling properties of spatial rainfall still remains an open question. By using an empirically based radar rainfall error model, the authors explore this issue by generating ensembles of probable true rainfall fields, conditioned on radar rainfall maps. Fifteen rainfall events over Oklahoma are analyzed, and it is shown how the presence of radar rainfall errors results in biased estimates of the scaling properties of rainfall.

Review of the Different Sources of Uncertainty in Single Polarization Radar-Based Estimates of Rainfall

It is well acknowledged that there are large uncertainties associated with radar-based estimates of rainfall. Numerous sources of these errors are due to parameter estimation, the observational system and measurement principles, and not fully understood physical processes. Propagation of these uncertainties through all models for which radar-rainfall are used as input (e.g., hydrologic models) or as initial conditions (e.g., weather forecasting models) is necessary to enhance the understanding and interpretation of the obtained results. The aim of this paper is to provide an extensive literature review of the principal sources of error affecting single polarization radar-based rainfall estimates. These include radar miscalibration, attenuation, ground clutter and anomalous propagation, beam blockage, variability of the Z–R relation, range degradation, vertical variability of the precipitation system, vertical air motion and precipitation drift, and temporal sampling errors. Finally, the authors report some recent results from empirically-based modeling of the total radar-rainfall uncertainties. The bibliography comprises over 200 peer reviewed journal articles.

Empirically based modelling of radar‐rainfall uncertainties for a C‐band radar at different time‐scales

AbstractRadar‐based estimates of rainfall are affected by several sources of systematic and random errors. The propagation of these uncertainties is required for improving the understanding and interpretation of the results obtained when radar‐rainfall estimates are used as input or initial conditions. Despite the relevance of the problem, modelling of radar‐rainfall uncertainties is still at an early stage. The authors apply an empirically based model in which the relation between true rainfall and radar‐rainfall can be described as the product of a systematic distortion function (accounting for systematic biases) and random component (describing the remaining random errors). The proposed results are based on a large sample (more than six years) of data from a C‐band radar (Wardon Hill) located in southwest England. The true ground rainfall is approximated by rain gauge measurements from a highly dense network deployed in the Brue catchment, located about 40 km from the radar site.The authors present results from their modelling of the radar‐rainfall errors at four time‐scales of hydrologic interest (5‐ and 15‐minute, hourly and three‐hourly) and 2 km spatial resolution, discussing the impact of spatial sampling errors on the proposed results. Copyright © 2009 Royal Meteorological Society

Absolute Calibration of Radar Reflectivity Using Redundancy of the Polarization Observations and Implied Constraints on Drop Shapes

Abstract A major limitation of improved radar-based rainfall estimation is accurate calibration of radar reflectivity. In this paper, the authors fully automate a polarimetric method that uses the consistency between radar reflectivity, differential reflectivity, and the path integral of specific differential phase to calibrate reflectivity. Complete instructions are provided such that this study can serve as a guide for agencies that are upgrading their radars with polarimetric capabilities and require accurate calibration. The method is demonstrated using data from Météo-France’s operational C-band polarimetric radar. Daily averages of the calibration of radar reflectivity are shown to vary by less than 0.2 dB. In addition to achieving successful calibration, a sensitivity test is also conducted to examine the impacts of using different models relating raindrop oblateness to diameter. It turns out that this study highlights the suitability of the raindrop shape models themselves. Evidence is shown supporting the notion that there is a unique model that relates drop oblateness to diameter in midlatitudes.

A Comparison of Several 5-Minute Radar—Rainfall Estimation Models

For the Z-R relationship in radar—based rainfall estimation, the distribution of corresponding R values for a given Z value (or the corresponding Z value for a given R value) may be highly skewed. However, the traditional power—law model is physically deduced and fitted under the normal—distribution presumption of radar wave echoes associated with a rain rate value, and it may not be very appropriate. Considering this problem, the authors devised several generalized linear models with different forms and distribution presumptions to represent the Z-R relationship. Radar—reflectivity scans observed by a CINRAD/SC Doppler radar and 5-minute rainfall accumulation recorded by 10 ground gauges were used to fit these models. All data used in this study were collected during some large rainfalls of the period from 2005 to 2007. The radar and all gauges were installed in the catchment of the Yishu River, a branch of the Huaihe River in China. Three models based on normal distribution and a dBZ presumption of gamma distribution were fitted using maximum—likelihood techniques, which were resolved by genetic algorithms. Comparisons of estimated maximized likelihoods based on assumptions of gamma and normal distribution showed that all generalized linear models (GLMs) of presumed gamma distribution were better fitted than GLMs based on normal distribution. In a comparison of maximum—likelihood, the differences between these three models were small. Three error statistics were used to assess the agreement between radar estimated rainfall and gauge rainfall: relative bias (B), root mean square error (RMSE), and correlation coefficient (r). The results showed that no one model was excellent in all criteria. On the whole, the GLM—based models gave smaller relative bias than the traditional power—law model. It is suggested that validations conducted in many previous works should have been made against a specific criterion but overlooked others.

Using microwave links to adjust the radar rainfall field

The final stage in processing radar data so as to arrive at an estimated rain field typically involves a comparison of the preliminary radar-derived estimates of hourly rainfall with those observed by ground-based gauges. Often a mean field bias adjustment will then be applied using an age-weighted average of the individual gauge–radar comparisons. In this paper, a mean field bias adjustment is presented that uses the path-integrated rainfall estimates provided by microwave links together with information from gauges. It is shown to be at least as efficient as the current gauge-based procedure used by the UK Met Office to improve the accuracy of radar-based estimates of rainfall at the ground.

Dynamic ANN for precipitation estimation and forecasting from radar observations

Summary Instantaneous quantitative precipitation estimation (QPE) and quantitative precipitation forecasting (QPF) by utilizing the meteorological radar data are potentially of great benefit for operational hydrology in river basins. The most commonly used technique of radar-based rainfall estimation is a best fitted power-law function between reflectivity ( Z ) and rain intensity ( R ). An emerging tool in QPE/QPF using radar data is the artificial neural network (ANN) that is capable of learning complex nonlinear relationships. In this study, we introduce a dynamic ANN approach to construct instantaneous QPE and one-hour-ahead QPF by using a three-dimensional radar data structure, which takes into account the terminal velocity and the horizontal advection. Radar measurements and three rain gauges for six typhoon events in the Keelung River, Taiwan, were used for calibrating and evaluating the Z – R relation and dynamic ANN models. The results of current rainfall estimation and one-hour-ahead rainfall forecasting all indicate that the dynamic ANN can produce much more accurate and stable performance than the Z – R power-law models. This study demonstrates that the dynamic ANN can be applied successfully in instantaneous QPE and QPF by using remote sensing data.