A fast implementation of the temperature-vegetation index drought detection principle

Abstract

Agricultural adaption to a changing climate requires frequent, accessible, and reliable information on water availability and crop stress at a spatio-temporal scale relevant to the end user. We have therefore created an improved, operationally suitable, drought detection and monitoring system. We achieved this through approximation of the land surface temperature (LST)–vegetation index space and implementation of non-linear edge finding in the form of support vector machine (SVM) regression. We achieved a 97% reduction in index computation time. We also significantly reduced data storage requirements and more accurately described the warm and cold boundaries of the space. In addition, statistical approximation of VTCI (fast-VTCI) allows us to implement the method within Google Earth Engine due to the reduced computational overheads.
The method was evaluated against a network of soil moisture stations in New Zealand (Taranaki). The Soil Adjusted Vegetation Index (SAVI) was found to be the most appropriate vegetation index for the Taranaki region. Fast-VTCI achieved a maximum r-value of 0.77 against ground data, with a p-value of less than 0.01. The strength of these correlations varies spatially and temporally in this research, indicating the need for localization in the application of the method. Despite high levels of cloud cover throughout the study period, Fast-VTCI outperforms the predominant New Zealand Drought Index (NZDI) for agricultural drought monitoring at the local scale, particularly in drought years.