Estimation of leaf area index in sagebrush steppe with low cost unoccupied aerial systems

Authors: Woodruff, Craig; Clark, Patrick; Olsoy, Peter; ENTERKINE, JOSH - Boise State University

Submitted to: Landscape Ecology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 12/22/2024
Publication Date: 1/22/2025
Citation: Woodruff, C.D., Clark, P., Olsoy, P.J., Enterkine, J. 2025. Estimation of leaf area index in sagebrush steppe with low cost unoccupied aerial systems. Landscape Ecology. 40. Article 27. https://doi.org/10.1007/s10980-024-02031-7.
DOI: https://doi.org/10.1007/s10980-024-02031-7

Interpretive Summary: Very high resolution imagery from Unoccupied Aerial Systems (UAS) have proven useful for monitoring some rangeland metrics over extensive areas, but their accuracy and utility for monitoring leaf area index (LAI) is largely untested. We use very high resolution UAS imagery to develop a machine learning model to estimate LAI. We demonstrate the machine learning model (Random Forest) accurately estimates LAI with an r-squared of 0.69 when compared with field observations. This research demonstrates high resolution UAS data can rapidly and accurately estimate LAI, when paired with limited and representative field measurements, allowing land managers to survey the millions of hectares of rangeland in the Great Basin and similar ecosystems worldwide.

Technical Abstract: Context Leaf area index (LAI) strongly influences the carbon and water cycle in drylands, but accurate estimation of LAI relies on field methods that are expensive and time intensive. Very high-resolution imagery from unoccupied aerial systems (UAS) offers a potential solution for monitoring LAI, but estimation methods derived from cost effective red, green, and blue (RGB) sensors are untested in these semi-arid ecosystems. Objectives The objective of our study was to test whether LAI could be estimated with very high resolution UAS collected RGB and canopy height data. Additionally, we sought to validate the model accuracy at the plot (1 m2) scale, test the accuracy at the macroplot (1 ha) scale, and assess the within plot impact of shadows. Methods We used a Random Forest machine learning model to estimate LAI in a Wyoming big sagebrush community in the Reynolds Creek Experimental Watershed using high resolution (< 1 cm2) UAS imagery collected in 2021 as predictors and plot scale point intercept (quadrat design) field data as the LAI reference. Results Random Forest modeled estimates of LAI were accurate at the plot (r2 = 0.69, MAE = 0.08, RMSE = 0.10), and the macroplot scales (error of 0.065), and mean within plot shadow error was 0.06. Conclusions This research demonstrates high resolution UAS data can rapidly and accurately estimate LAI, with a limited number of field measurements, potentially allowing land managers to survey seasonally and spatially heterogeneous LAI 1 hectare at a time over the vast rangelands in the Great Basin and similar ecosystems worldwide.