Alfalfa Biomass Estimation Using Crop Surface Modeling and NDVI

Abstract

<b>Highlights</b> <list list-type=bullet><list-item> A UAV with NDVI and RGB cameras captured aerial images of alfalfa plots before and after harvest. </list-item><list-item> Structure from Motion (SfM) and crop surface models were used to determine change in alfalfa crop height. </list-item><list-item> An equation to predict dry matter fraction (DMF) as a function of NDVI was developed. </list-item><list-item> Dry alfalfa yield models using DMF and change in crop height were developed. </list-item></list> <b>Abstract.</b> Alfalfa is an important forage crop grown for hay, silage, and pasture production. Accurate yield estimations before harvest or grazing are crucial for producers to optimize forage utilization. The goal of this research was to develop an aboveground dry biomass prediction function for alfalfa using remote sensing from an unmanned aerial vehicle (UAV). A DJI Mavic Pro equipped with RGB and NDVI cameras captured aerial images. High-resolution orthomosaics and digital surface models were obtained using Structure from Motion (SfM) with Agisoft Metashape. An equation to estimate wet biomass was developed using three variables: change in canopy height (∆H) from digital elevation and crop surface models, and NDVI based canopy density index (CDI) data. The dry biomass yield was estimated as the product of a wet biomass prediction function and a correlation equation to estimate the dry matter fraction (DMF). The best correlation equation for wet biomass (BMwet) only required SfM methods to measure ΔH. The linear regression equation for BMwet had an R2 of 0.963 and the mean coefficient of variation (CV) was ±23%. The best prediction equation for DMF was a quadratic equation with an R2 of 0.642 with a mean CV of ±11%. It was also determined that the DMF varied significantly by season of harvest. As a result, the dry biomass could be estimated using the DMF equation or seasonal mean DMF values. Comparison of the observed dry biomass measurements with the empirically determined biomass prediction function that used the NDVI data to estimate the DMF showed excellent agreement. The model underpredicted the dry biomass by 1.1% with a standard error of the estimate of 236 kg DM/ha (CV = ±26%). The model predictions using the seasonal mean DMF values were not significantly different.