An assessment of kriging‐based rain‐gauge–radar merging techniques

Abstract

This article is published with the permission of the Controller of HMSO and the Queen's Printer for Scotland.Networks of rain-gauges provide an accurate but highly localized measure of rainfall, with limited coverage and resolution, whereas radars provide rain-rate and accumulation estimates over wide areas at high spatial and temporal resolution but low accuracy. When quantifying rainfall accumulations for applications such as flood forecasting, combining the two sets of data can be beneficial for producing a high-resolution merged product with a lower error than the gauge-only or the radar-only product. In this study, three kriging methods (kriging with external drift (KED), kriging with radar-based error correction (KRE) and ordinary kriging (OK)) as well as a multiquadric (MQ) scheme have been used to merge radar and gauge data. The results were cross-validated with true rainfall readings at the surface for a number of different meteorological events covering England and Wales. Overall, all the merging schemes trialled produced a merged product that was superior to the individual radar or gauge data. The KED was the best performing method across all rainfall thresholds and meteorological conditions, with the use of a parametric variogram best suited to short (15 min) accumulation periods and a non-parametric variogram preferable for hourly accumulations. The study also shows that the merged product deteriorates when the gauge-network density is reduced; exceptions to this are made in situations where spatially isolated rainfall is observed by the radar and where the gauges that are found least likely to represent the rainfall are carefully removed prior to merging. The latter process was found to improve the quality of the merged product, regardless of the meteorological conditions, providing a viable method for producing a KED-based merging scheme that is applicable to all meteorological conditions.

Fast Bayesian Regression Kriging Method for Real‐Time Merging of Radar, Rain Gauge, and Crowdsourced Rainfall Data

“…Steps to simulate one realization of the system over a particular rainfall event; NEXRAD stands for next generation weather radar; SWMM stands for Storm Water Management Model; TF, MF_KED G , MF_KED CG , MF_DM 1 , MF_DM 2 and MF_FBRK are as defined in Table 1; REAA I , RMSE I , ARE P , ARE V , RMSE F and RMSE F10 are as defined in Table 3. their improvements over the MF_KED G , MF_DM 1 , and MF_DM 2 results. That the two best performing data sets are KED based is consistent with existing studies in the literature (Goudenhoofdt & Delobbe, 2009;Jewell & Gaussiat, 2015), which have found the KED method to be, in general, the most effective thus far for merging radar and rain gauge data.…”

“…Though, the margin of improvement of the MF_FBRK results over the MF_KED CG results is relatively minor compared to both their improvements over the MF_KED G , MF_DM 1 , and MF_DM 2 results. That the two best performing data sets are KED based is consistent with existing studies in the literature (Goudenhoofdt & Delobbe, ; Jewell & Gaussiat, ), which have found the KED method to be, in general, the most effective thus far for merging radar and rain gauge data.…”

“…Other past efforts to merge rain gauge data with radar and satellite data include works by Sinclair and Pegram (2005), Hasan et al (2016), Wang et al (2015), and Cecinati et al (2017). In these and other studies, kriging-based methods were the most widely used class of merging algorithms, in particular kriging with external drift (KED) that, in many instances, has been found superior over other algorithms (Boudevillain et al, 2016;Goudenhoofdt & Delobbe, 2009;Jewell & Gaussiat, 2015).…”

Improving the precipitation accumulation analysis using lightning measurements and different integration periods

“…In general, combining radar and rain gauge data is very difficult in the vicinity of heavy local rain cells (Einfalt et al, 2005). Recently, Jewell and Gaussiat (2015) compared performances of different merging schemas and noted a large difference between convective and stratiform situations. In their study, the nonparametric kriging with external drift outperformed other methods in an accumulation period of 60 min.…”

Spatially distributed rainfall information and its potential for regional landslide early warning systems

“…There is an abundance of literature dealing with this merging process (e.g. Collier et al 1983;Jewell and Gaussiat 2015;Velasco-Forero et al 2009;Berndt et al 2014).…”