Automated analysis of snowmelt from Sentinel-2 imagery to determine variable rate irrigation zones in the American Mountain West

Abstract

Variable rate irrigation (VRI) is used to save water whilst maintaining crop yields in semiarid regions. A key problem is to be able to inexpensively determine spatial patterns in soil moisture so that VRI zones can be defined. In Southern Idaho, USA, the annual precipitation is low and most fall as winter snow. This research investigates whether snow melt patterns measured using freely available time-series Sentinel 2 imagery from Google Earth Engine can define useful VRI zones for two arable fields (Grace and Rexburg). The normalized difference snow index (NDSI) was computed for each 10 m pixel with snow for all winter images of the fields for 2018–2022. NDSI values were ranked within each image and average ranks were calculated for each month and over several years. The patterns of March NDSI were most similar to patterns in yield and soil moisture observed in previous years. Zones were determined using K-means classification of the mean ranks of March NDSI. Kruskal Wallis H tests showed consistent and significant differences between zones for key soil, plant, and topographic variables. For the Grace site, differences between zones were more consistent in their order of magnitude than VRI zones which were calculated using a labor-intensive method. For the Rexburg site, zones were shown to be better when based on snowmelt data from March 2018 to 2022 rather than just March 2019. It is important to base zones on several years of data because in some years there was no snow observed in the Grace field in March. In locations where the majority of soil moisture comes from snowmelt, basing VRI zones on several years of snowmelt patterns in March is a useful and inexpensive tool for deriving meaningful VRI zones. The code used to automatically extract suitable sentinel images and calculate the NDSI is included so that practitioners can use this approach in other locations.