Landscape-Level Forage Estimation Using Fine-Scale Drone Imagery in Rangelands

Quantifying the amount and spatial distribution of forage mass is critical for managing and enhancing rangeland health and ecosystem services. Remote sensing allows for landscape-level estimations of forage mass, requiring less labor and time in the field than traditional vegetation collection. Our objective is to quantify forage mass and forage production at a fine scale over larger extents with drones and compare to field measurements. We evaluate the capacity of drones to provide improved forage mass estimations on an operational scale in the Southern Great Plains. To quantify forage mass, we collected 1.5 cm resolution Red–Green–Blue (RGB) drone imagery to generate a canopy height model (CHM) at a 7 cm resolution at three Texas, USA, locations throughout the growing season: Vernon (site 1: 94 ha, 678 mm average annual precipitation) and Sweetwater (site 2: 51 ha, site 3: 99 ha; 592 mm average annual precipitation). We collected 30 vegetation clippings per location per sampling event. Vegetation clippings were dried and weighed to estimate forage mass, then compared to forage volume estimated from the CHMs using linear regression. We also developed a framework for a supervised, object-based image classification, which was used to remove woody cover and provide an adjusted forage mass estimation. R² values ranged from 0.3585 to 0.6182 in site 1, from 0.6368 to 0.7844 in site 2, and from 0.3752 to 0.8065 in site 3. Overall accuracy of image classifications ranged from 85.6 to 99.9% for site 1; 95.9% to 99.5% for site 2; and 96.8% to 99.5% for site 3. We estimate total adjusted forage mass for each one of these sites. This work provides a framework for landscape-scale estimation of forage mass that can help livestock producers and land managers adjust stocking rates to maintain or improve rangeland health and wildlife habitat.