Remote sensing to estimate above-ground biomass in Southeastern Montana grasslands

Abstract

Grasslands provide numerous ecological services, including, but not limited to, carbon capture and storage, and wildlife habitat for numerous species including pollinators, domestic ungulates, wild herbivores, and their predators. Grasslands account for more than 40% of Earth’s terrestrial surface and offer one of the most valuable land cover types, having the greatest diversity of grazing animals and predators on Earth. These grasslands serve as one of the largest reservoirs for terrestrial carbon, amounting to 10-30% of total world soil carbon (Hewins et al., 2018; Scurlock and Hall, 1998). Establishing carrying capacities of grasslands based on data, in both wildlife management and agricultural settings, is fundamental in grassland ecosystem management. However, ground observation of grassland biomass resources can be costly, time consuming, and is very limited in terms of scale, both spatially and temporally. Numerous remote-sensing applications have emerged to address the need for grassland biomass production data, with satellite-derived vegetation indices being the most widely implemented due to their wide variety of applications and the accessibility of free spectral imagery via federal agencies such as the U.S Geological Survey (USGS). Currently, there is no universal empirical model using statistical relationships between biomass and vegetation indices (VI), making custom-fitted models necessary for biomass estimation. Statistical analysis is limited by inadequate reference data and vulnerabilities to atmospheric factors. Results of multiple studies show strong correlations between field measured biomass and satellite-derived VI’s. However, correlation strength depends on matching the best VI, or multiple VI’s, to the land cover characteristics in question, coupled with validation tools and methodologies on the ground. Therefore, to explore the feasibility and accuracy of a site-specific biomass estimation model, I chose a privately owned ranch in eastern Montana for sampling and tested for combinations of indices that were best correlated with ground biomass measurements. I used several vegetation indices to compare the accuracy of univariate and multivariate modeling techniques for modeling above ground biomass. A multivariate model using the normalized difference vegetation index (NDVI), the enhanced vegetation index (EVI) and a native/introduced pasture classification variable arose as the model with the greatest predictive power for biomass estimation. This model accounted for 57% of the variation in response and was accurate to within 352.7 lbs/acre. My study lays out real-world applications for tailoring VIs to the application requirements of measuring biomass in an eastern Montana grassland coupled with ground validation. The methods in this study serve as a framework for fitting custom models to other grasslands under management.