Mapping the extent and severity of tree mortality and defoliation caused by spruce budworm using landsat time series

The spruce budworm is one of the principal defoliators affecting forest health in Canada. Studying the impact of spruce budworm on forests including understanding the extent and severity of tree defoliation and mortality caused by the outbreaks is important for forest management. In this study, we utilized LandTrendr algorithm to get synthetic Landsat time-series data and used it to characterize spruce budworm outbreaks. Here, we built 14 configurations of predictors from different seasonal mosaics (summer mosaic, winter mosaic, and a combination of summer and winter mosaic) and various data sources including remote sensing data, aerial survey data, and a combination of both. We then employed a random forest model, incorporating data from 892 field plots to estimate defoliation and mortality caused by the spruce budworm in eastern Québec in the years 2008, 2013, 2018 and 2022. Using only remote sensing data, an R2 value of 78% for defoliation and 55% for mortality was obtained as the best results. Incorporating aerial survey data resulted in an increase of 1% for defoliation and 7% for mortality. We also found that the combination of summer and winter mosaic data resulted in the highest R2 values, with improvements of 1–8% for defoliation and 2–14% for mortality compared to models using data from single season. Through the variable importance ranked by the best model configurations for defoliation and mortality, we concluded that normalize burn ratio from winter mosaic served as the most important variable for the defoliation and cumulative defoliation rating from aerial survey ranked the first in mortality. The proposed approach can be used to obtain yearly continuous defoliation and mortality maps and thus assist in monitoring forests and their dynamics during spruce budworm outbreaks for carbon budget modelling.