Estimating integrated measures of forage nutritional quality for herbivores in northcentral Alaska by fusing optical and structural remote sensing data

Abstract:

Northern herbivore ranges are expanding in response to increased forage biomass produced by a warming climate. Forage quality also influences herbivore distributions, but less is known about the effects of climate change on plant biochemical properties. Remote sensing could enable landscape-scale estimations of forage quality, which is of great interest to wildlife managers. Despite the importance of integrated forage quality metrics like digestible dry matter (DMD) and digestible protein (DP), however, few studies have applied remote sensing approaches to map these characteristics. Our objectives were twofold: (1) assess how well DMD and DP can be predicted using hyperspectral remote sensing, and (2) to determine whether incorporating shrub structural metrics affected by browsing would improve our ability to predict DMD and DP. We collected canopy-level spectra, destructive-vegetation samples, and flew unmanned aerial vehicles (UAVs) in areas dominated by willow shrubs in Alaska in July 2019. Canopy structural metrics were derived from 3-D structural information obtained from UAV imagery using structure from motion photogrammetry. We used generalized least squares regression to account for the spatial autocorrelation of sampled shrubs. The best performing model for predicting DMD had three predictors: a spectral vegetation index (SVI) that included a red-edge and shortwave infrared band, shrub height variability (HVAR), and leaf area index (Nagelkerke R²= 0.70, RMSE= 1.46%, cross validation ρ = 0.79). The best performing model for DP had two predictors: an SVI that used a blue and a red band, and HVAR (Nagelkerke R²= 0.81, RMSE= 4.96%, cross validation ρ = 0.85). Results from our study demonstrate that integrated forage quality metrics like DMD and DP can be successfully quantified using hyperspectral remote sensing data, and that models based on those data can be improved by incorporating additional shrub structural metrics.