Using multispectral data from an unmanned aerial system to estimate pasture depletion during grazing

Abstract

Accurate measurement of pasture depletion and, therefore, dry matter intake (DMI) is fundamental in both research and commercial settings for understanding the behaviour and nutrition of grazing animals. Remote sensing technologies such as unmanned aerial vehicles (UAV) equipped with sensors have been used to estimate biomass of crops and un-grazed pastures. The aim of the experiment reported here was to develop and validate empirical models to estimate pasture depletion in paddocks while cattle are grazing using an UAV-borne multispectral sensor with a rising plate meter measurements as the reference data. Data were collected in winter and spring with 24 different cows used in each period and grazing in individual plots. UAV flights were undertaken before cows entered the paddock and then every hour for the first 5 h of grazing coinciding with rising plate meter measurements. Four types of datasets were prepared and compared to determine which produced the best estimate of herbage mass. The first two datasets were created by extracting data from either the cow’s whole AM plot (AP) area, or a small polygon of homogenous data (SP) subjectively chosen from within the AM plot area. A further two training datasets were created by combining either the AM plot data or small polygon data with additional extreme data (EX) collected pre (0 h) and post (24 h) grazing from days before or after the days of UAV measurements. Therefore, the four training datasets compared were AP, SP, APEX and SPEX. The best performing model was the APEX model with a concordance correlation coefficient of 0.67, an r2 of 0.54 and RSME of 406 kg DM/ha in independent validation, however accuracy of estimates decreased with pasture biomass below 1,000 kg DM/ha and above 2,000 kg DM/ha. Dry matter intake estimated using the UAV images and the APEX model had a concordance of 0.39 with the RPM method. The APEX model was also used to demonstrate how this model allowed pasture depletion to be mapped through time during a grazing session. This research has highlighted the potential of using a UAV-borne multispectral imaging system for estimating pasture depletion in paddocks while cows graze, however, we suggest further research is required to improve the accuracy of our models for estimating extreme values of pasture biomass before it can be used to estimate daily DMI.