Research Article
Uncertainty Analysis of Rainfall Spatial Interpolation in Urban Small Area
@INPROCEEDINGS{10.1007/978-3-030-12971-2_5, author={Jie Huang and Changfeng Jing and Jiayun Fu and Zejun Huang}, title={Uncertainty Analysis of Rainfall Spatial Interpolation in Urban Small Area}, proceedings={Testbeds and Research Infrastructures for the Development of Networks and Communications. 13th EAI International Conference, TridentCom 2018, Shanghai, China, December 1-3, 2018, Proceedings}, proceedings_a={TRIDENTCOM}, year={2019}, month={2}, keywords={Rainfall Spatial interpolation Ordinary Kriging Random forest Machine learning}, doi={10.1007/978-3-030-12971-2_5} }- Jie Huang
Changfeng Jing
Jiayun Fu
Zejun Huang
Year: 2019
Uncertainty Analysis of Rainfall Spatial Interpolation in Urban Small Area
TRIDENTCOM
Springer
DOI: 10.1007/978-3-030-12971-2_5
Abstract
Uncertainty analysis have attracted increasing attention of both theory and application over the last decades. Owing to the complex of surrounding, uncertainty analysis of rainfall in urban area is very little. Existing literatures on uncertainty analysis paid less attention on gauge density and rainfall intensity. Therefore, this study focuses on urban area, which a good complement to uncertainty research. In this study, gauge density was investigated with carefully selecting of gauge to covering evenly. Rainfall intensity data were extracted from one rainfall event at begin, summit and ending phases of rainfall process. Three traditional methods (Ordinary Kriging, RBF and IDW) and three machine methods (RF, ANN and SVM) were investigated for the uncertainty analysis. The result shows that (1) gauge density has important influence on the interpolation accuracy, and the higher gauge density means the higher accuracy. (2) The uncertainty is progressively stable with the increasing of rainfall intensity. (3) Geostatistic methods has better result than the IDW and RBF owing to considering spatial variability. The selected machine learning methods have good performance than traditional methods. However, the complex training processing and without spatial variability may reduce its practicability in modern flood management. Therefore, the combining of traditional methods and machine learning will be the good paradigm for spatial interpolation and uncertainty analysis.
