heterogeneityR: An R Package for the Measurement of Rangeland Vegetation Structure From Ground-Based Digital Imagery

Abstract

Measurement of vegetation structure is highly relevant to many rangeland management objectives, but the utility of individual methods for monitoring or research purposes can be constrained by temporal and financial cost, applicability, and associated error. Previous methodological advancements have reduced surveyor bias and field collection time using ground-based digital imagery. However, these techniques can present their own limitations including increased processing time, cost of specialized equipment or software, and incompatibility across ecosystems. Here we present heterogeneityR, a free software package developed within the open-source R environment, to address previous methodological constraints. The package provides an automated image analysis pipeline that uses a machine learning framework to rapidly calculate an assortment of vegetation metrics (including visual obstruction, height, fuel loading, and structural heterogeneity) from field collected imagery and is customizable to training datasets specific to a user’s site, vegetation characteristics, and objectives. We evaluated the efficacy of the package using data collected within tallgrass prairie pastures that are grazed and patch-burned on a three-year rotation to create a mosaic of patches that vary with time since fire. Visual obstruction estimates were correlated with standing biomass of overall vegetation (R² = 0.90) and individual fuel types (live fuel R² = 0.81, dead fuel R² = 0.70). Time since fire had a significant effect on all model outputs and multiple comparisons tests revealed differences between burn patches for most metrics, indicating the high degree of patch-scale variance within the system. Our results demonstrate the utility of heterogeneityR to efficiently assess field collected data relevant to objectives in livestock production, fuels management, and conservation.