Agricultural drought prediction using climate indices based on Support Vector Regression in Xiangjiang River basin

Ye Tian; Yue-Ping Xu; Guoqing Wang

doi:10.1016/j.scitotenv.2017.12.025

Agricultural drought prediction using climate indices based on Support Vector Regression in Xiangjiang River basin

Sci Total Environ. 2018 May 1:622-623:710-720. doi: 10.1016/j.scitotenv.2017.12.025. Epub 2017 Dec 13.

Authors

Ye Tian¹, Yue-Ping Xu², Guoqing Wang³

Affiliations

¹ College of Hydrometeorology, Nanjing University of Information Science & Technology, Nanjing 210044, China; Key Laboratory of Meteorological Disaster (Nanjing University of Information Science and Technology), Ministry of Education, Nanjing 210044, China.
² Institute of Hydrology and Water Resources, Department of Civil Engineering, Zhejiang University, Hangzhou 310058, China. Electronic address: yuepingxu@zju.edu.cn.
³ State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering, Nanjing Hydraulic Research Institute, Nanjing 210029, China.

PMID: 29223897
DOI: 10.1016/j.scitotenv.2017.12.025

Abstract

Drought can have a substantial impact on the ecosystem and agriculture of the affected region and does harm to local economy. This study aims to analyze the relation between soil moisture and drought and predict agricultural drought in Xiangjiang River basin. The agriculture droughts are presented with the Precipitation-Evapotranspiration Index (SPEI). The Support Vector Regression (SVR) model incorporating climate indices is developed to predict the agricultural droughts. Analysis of climate forcing including El Niño Southern Oscillation and western Pacific subtropical high (WPSH) are carried out to select climate indices. The results show that SPEI of six months time scales (SPEI-6) represents the soil moisture better than that of three and one month time scale on drought duration, severity and peaks. The key factor that influences the agriculture drought is the Ridge Point of WPSH, which mainly controls regional temperature. The SVR model incorporating climate indices, especially ridge point of WPSH, could improve the prediction accuracy compared to that solely using drought index by 4.4% in training and 5.1% in testing measured by Nash Sutcliffe efficiency coefficient (NSE) for three month lead time. The improvement is more significant for the prediction with one month lead (15.8% in training and 27.0% in testing) than that with three months lead time. However, it needs to be cautious in selection of the input parameters, since adding redundant information could have a counter effect in attaining a better prediction.

Keywords: ENSO; SVR; Soil moisture; Western Pacific subtropical high; Xiangjiang River.