Multi-scale remote sensing-based landscape epidemiology of the spread of rapid 'O over bar hi'a Death in Hawai'i

Fungal pathogens of the genus Ceratocystis recently introduced to the Island of Hawai'i have killed hundreds of thousands of native 'O over bar hi'a trees, an ecologically and culturally important keystone species. Symptoms of the associated disease, Rapid 'O over bar hi'a Death (ROD), have been found to be easily detectable with high resolution imaging spectroscopy. We used wall-to-wall maps of affected 'O over bar hi'a canopy built in four consecutive years (2016-2019) to analyze how changes in the distribution and density of browning canopy detections over this time period corresponded to environmental drivers at two spatial scales. Island-wide we found 256,387 brown crowns across the 4-year period. The total amount of affected forest area nearly doubled from 66,972.2 ha in 2016 to 121,464.7 ha in 2019, cumulatively representing 42.5% of the 264,372 ha of forest area observed at least once in the study. However, most browning activity was concentrated in 18 hotspots totaling approximately 29,836 ha with brown crown densities as high as 26 ha- 1. Using Random Forest (RF) models, we assessed the correspondence between 32 mapped environmental variables and measures of ROD spread calulated at 1-year time steps on a 300 m x 300 m grid across the island. We found that modelled probability of regional infec-tion, or "susceptibility", was dominated by distance to existing infection at the start of a time step; increasing distance from 0 to 600 m decreased susceptibility from 0.8 to 0.3. Other important factors - tree cover, average windspeeds, and LiDAR-derived 3-d crown exposure - were 4-10 times less important to the model. In a second RF model we studied regional infection "severity", measured as detection density in first year of infection. In contrast to susceptibility, the severity model was significantly affected by numerous factors including windspeed metrics, canopy height and exposure as well as most weather and climate metrics related to water regime. Severity model predictions were less dominated by proximity to existing infection. In a separate regional analysis at 2 m resolution we found that taller trees were more prone to infection, and that trees on younger "historic" lava flows (i.e., substrates < 230 years old) showed slight but significantly lower infection rates than immedi-ately adjacent older substrates. Our results document the extent and severity of 'O over bar hi'a mortality across Hawai'i Island and provides insight into factors that appear to affect the spread of ROD; this information will inform management efforts to lessen the spread and impact of the disease on all the Hawaiian Islands.