An evaluation framework for identifying the optimal raingauge network based on spatiotemporal variation in quantitative precipitation estimation Che-Hao Chang, Shiang-Jen Wu, Chih-Tsung Hsu, Jhih-Cyuan Shen and Ho-Cheng Lien ABSTRACT This study proposes an evaluation framework to identify the optimal raingauge network in a watershed using grid-based quantitative precipitation estimation (QPE) with high spatial and temporal resolution. The proposed evaluation framework is based on comparison of the spatial and temporal variation in rainfall characteristics (i.e. rainfall depth and storm pattern) from the gauged data compared with those from QPE. The proposed framework first utilizes cluster analysis to separate raingauges into various clusters based on the locations and rainfall characteristics. Then, a cross-validation algorithm is used to identify the influential raingauge in each cluster based on evaluating performance of fitting weighted spatiotemporal semivariograms of rainfall characteristics from the gauged rainfall to the QPE data. Thus, the influential raingauges for a specific cluster number form the representative network. The optimal raingauge network is the one corresponding to the best fitness performance among the representative networks considered. The study area and data set are the hourly rainfall from 26 raingauges and 1,336 QPE grids for 10 typhoons in the Wu River watershed located in central Taiwan. The proposed evaluation framework suggests that a 10-gauge network is the optimal and can describe a good spatial and temporal variation in the rain field similar to the grid-based QPE from two additional typhoon events. Che-Hao Chang Chih-Tsung Hsu National Center for High-Performance Computing, Hsinchu 30076, Taiwan Shiang-Jen Wu (corresponding author) Jhih-Cyuan Shen Ho-Cheng Lien Department of Civil Engineering, National Taipei University of Technology, Taipei 10608, Taiwan E-mail: sjwu@nchc.narl.org.tw Key words | Akaike information criterion (AIC), quantitative rainfall estimation (QPE), raingauge network, rainfall characteristic, spatiotemporal semivariogram INTRODUCTION Rainfall data are essential in many hydrological analyses and hydraulic engineering designs, such as frequency analy- sis, rainfall-runoff analysis, and stormwater drainage design. For example, the water level used in designing a hydraulic structure, such as a levee, is estimated by using a rainfall- runoff model and flood wave propagation model with the design areal average rainfall hyetograph estimated from the raingauge network (Wu et al. ). Accurate estimation in the spatial distribution of rainfall requires a dense net- work, which entails large installation and operational costs, but reduces the opportunity of project failure (AI-Zah- rani & Husain ; Putthividhya & Tanaka ; Adhikary et al. a, b). The World Meteorological Organization (WMO ) issued a guideline for recommended density of raingauges in a catchment based on the physiographic unit and area of the watershed. For example, 250 km 2 per gauge is suggested for a small mountainous region with irre- gular rainfall, and for the flat region of a temperate zone. In addition, a modern raingauge network can provide real-time estimation and the rainfall forecast resulting from typhoons 77 © IWA Publishing 2017 Hydrology Research | 48.1 | 2017 doi: 10.2166/nh.2016.169 Downloaded from https://iwaponline.com/hr/article-pdf/48/1/77/367002/nh0480077.pdf by guest on 20 July 2020