Context

Cheatgrass (Bromus tectorum L.) is a problem across the western United States, where it outcompetes and replaces native grass species, alters habitats, and increases the risk of wildfires. Cheatgrass greens up earlier in the growing season compared to native grasses, making it classifiable with multi-temporal and multi-spectral remote sensing.

Objectives

We mapped cheatgrass at different scales in the Greater Yellowstone Ecosystem using 10-m Sentinel-2 imagery, 3-m PlanetScope, and 10-cm Uncrewed Aerial Systems (UAS) imagery. We compared these maps to field-collected data to address 1) variation in seasonal phenological signals of native and cheatgrass patches, 2) the influence of scale on detectability and map accuracy across our study area.

Results

Model accuracy to predict cheatgrass presence increased with imagery resolution and ranged from 83% using 10-m Sentinel-2 to 94% with the integration of PlanetScope and UAS imagery. While there was spatial agreement across models, the fusion of UAS data with satellite sources allowed the detection of small cheatgrass with more precision. Our novel use of NExR and dNExR (a redness and differenced redness index) data in the classification of cheatgrass capitalizes on the senescence of cheatgrass during peak summer periods where cloud free imagery is more prevalent.

Conclusions

Our satellite and UAS-based models of cheatgrass prediction compare the fusion of very high resolution imagery and phenological time differencing to identify infested areas. Tradeoffs between accuracy and expense lead to important questions for management applications.