Classification of tree mortality following drought-defoliation interaction using single date Landsat imagery and comparison to aerial detection surveys

Abstract

Forest disturbance regimes are changing across the globe under the influence of climate change and other global change factors, with potentially substantial consequences for tree mortality rates. Tree mortality has been assessed using field and aerial surveys and, more recently, frequently using satellite remote sensing-based techniques. Rapid detection of tree mortality is often important in decision-making around wood salvage and other potential responses. Our study tested techniques to rapidly distinguish tree mortality in temperate broadleaf deciduous forests following drought and spongy moth defoliation in Connecticut, USA using single-date imagery. We applied random forest classification to identify mortality occurrence and severity using single-date Landsat-8 imagery and compared outcomes to annual USDA Insect and Disease Detection Surveys (IDS). Training data were produced by manually delineating mortality visible in high-resolution (1 m) 2018 US National Agriculture Imagery Program imagery and validated against field plot data. Overall accuracy in mortality detection using single-date imagery ranged from 59.3 %–85.0 % with better results for mortality occurrence detection than severity classification. Image classification analysis identified on average 15,340 ha of mortality across the study area, while IDS identified 10,168 ha. Our results demonstrate the potential utility of single-date classification in tree mortality monitoring in deciduous forests and as a potential supplement to aerial detection surveys.