COS 209-3 - Mapping tree mortality in temperate northeastern forests following drought and multiyear defoliation by Lymantria dispar dispar using Landsat-8 data products

Background/Question/Methods

Climate change is expected to increase the frequency and severity of forest disturbances, likely increasing tree mortality rates. Traditionally, tree mortality has been assessed using field and aerial surveys, and more recently, applications of remote sensing-based techniques. Our study tests these techniques in temperate forests of eastern Connecticut following severe drought and defoliation by Lymantria dispar dispar between 2016 and 2018. General patterns of tree mortality are assessed in United States Department of Agriculture Insect and Disease Detection Surveys (IDS), annual aerial surveys. Our study supplements IDS by identifying finer scale patterns and testing the potential to monitor tree mortality using remotely-sensed data products. We created two random forest-based models to identify presence-absence and severity of tree mortality using Landsat-8 spectral data and indices, topographic variables, and modeled soil properties. Training and validation data were created using 1-meter resolution National Agriculture Imagery Program aerial imagery from 2018. Dead tree crowns were delineated per Landsat-8 pixel in the training and validation dataset and aggregated into classes of mortality: background (< 5%), light (5-10%), moderate (11-29%), and severe ( > 30%). We assessed model performance and produced mapped products, which were compared to IDS (2016-2019).

Results/Conclusions

Overall accuracy for the products was 90.1% (presence-absence) and 77.1% (mortality severity) with kappa coefficients of 0.81 and 0.61 respectively. Both products covered 42,196 ha of forest with the presence-absence model identifying 12,367 ha (29.3%) as tree mortality and the mortality severity model identifying 10,254 ha (24.3%). Tree mortality assessed by the mortality severity model was classified as 61.8% light, 31.1% moderate, and 7.1% severe with moderate and severe mortality detected in northern and eastern sections of the study area and light mortality throughout. IDS similarly identified 13,283 ha of tree mortality within the study area, but specific spatial patterns and estimations of mortality severity varied from both modeled mapped products. Additionally, 1,674 ha of moderate and 324 ha of severe tree mortality identified by the mortality severity model were not captured by IDS. Our results demonstrate that models informed by moderate-resolution remote sensing products can be used to assess and monitor tree mortality following significant defoliation events in northeastern temperate forests. The resulting mapped products can also be used to supplement IDS in applications of forest management and disturbance response in southern New England and beyond.