Mapping Volumetric Water Content at Multiple Depths using Drone Data and Machine Learning to Inform Variable Rate Irrigation

Keywords: Drones, Variable Rate Irrigation, Precision Agriculture, Machine Learning, Soil Moisture, Digital Soil Mapping
Abstract Type: Poster Abstract

Abstract

The Mountain West is experiencing a “mega drought.” Agriculture is the biggest user of fresh water so investigating the use of variable rate irrigation through precision agriculture could result in big water savings while maintaining crop yields. To use variable rate irrigation, accurate representations of soil moisture patterns at several depths are needed. Digital Soil mapping techniques are usually used to map soil properties or classes over regions rather than individual fields. This study used digital soil mapping techniques to map volumetric water content (VWC) to a 5 m resolution. A set of 860 covariates from drone data, terrain attributes, yield data and distance-based parameters were used with three machine learning algorithms to map soil observations for four-time intervals (fall and spring, 2016 and 2017) at four depths in an agricultural field in Grace, Idaho. Although the random forest was the most effective model, its performance was not consistent for the different soil depths and sampling periods. In Spring 2016, the best estimation was for VWC at 30-60 cm depth but the Random Forest model predicted VWC for Fall 2016 at the soil surface best. The models showed that distance-based parameters and terrain attributes were the most important covariates in predicting VWC at different depths at the within field scale.