Differentiating grazing-induced rangeland grass nutrition status in situ using near infrared reflectance

Abstract

Assessing the nutrient content of rangeland grass using earth observation technologies presents opportunities for non-destructive, rapid and repetitive appraisals in support of rangeland management. The biochemical-related concentrations of macronutrients (N, P, K, Ca, Mg) are among the factors that influence near infrared reflectance (ρNIR) by a turgid leaf. High grazing intensity (GI) enhances nutrient recycling in rangelands, resulting in high concentrations in aboveground grass tissue. This study sought to establish whether grazing-induced macronutrient abundance in rangeland grass can be remotely sensed in situ through ρNIR in the multiple species natural settings. Sampling was conducted at peak grass growth stage in large savannah rangelands with similar grass species mixtures, in low moisture stress conditions. The sampling was widely distributed in GI-contrasted livestock and wild grazer rangelands, and in grazing-restricted control sites. Two species representing high and low nutrient grass species, respectively, were sampled. At each sampling site (N = 116), reflectance was measured in the 350–2500 nm range from leaf and stem tissue of healthy-looking specimens of the two species (≤10 m apart), using a 1.1 nm – 1.4 nm resolution spectroradiometer. Macronutrient concentrations (MNC) in the tissue were then determined in the laboratory. For each species, the hyperspectral NIR (λ700 – 1100 nm) reflectance was averaged per sampling site. Individual and total MNC, respectively, were correlated with the hyperspectral ρNIR, and then with atmospheric effect-corrected ρNIR from 10 m-resolution, field sampling-concurrent Sentinel-2 MSI multispectral images. For each sampling site, the two species’ hyperspectral ρNIR values, individual and total MNC, respectively, were then averaged. Respective models relating the paired hyperspectral ρNIR and averaged MNC were developed, by linear regression and partial least squares regression (PLSR). Tissue MNC were lowest in the control sites and higher in the grazed rangelands. Covariance in MNC in the two species suggested location-context similarities in all grasses. Individual and total MNC, respectively, had statistically significant (p < 0.01) negative correlations with both tissue level hyperspectral and multispectral ρNIR. The negative relationship was detected on empirical line-retrieved ρNIR on historical Hyperion hyperspectral images. PLSR yielded a more accurate predictive model (R2 = 0.760) than linear regression (R2 = 0.601, F = 138.415, p < 0.0001). The study shows that there are possibilities to remotely sense grazing-induced MNC in aboveground grass tissue through changes in ρNIR.