Enhancement of historical eastern spruce budworm defoliation and mortality aerial survey maps (1984-2020) in Ontario using Landsat

Background/Question/Methods

Eastern spruce budworm (Choristoneura fumiferana Clem.; SBW), a moth native to Ontario, has caused extensive damage to the boreal forest ecosystem and timber supply in the province during two successive outbreaks over the past four decades. The extent of SBW damage in Ontario is monitored using aerial survey on a yearly basis. The historical aerial survey data is a reliable resource, but there has been inconsistent mapping approaches through time resulting in an underestimation of the area affected in recent years and an overestimation in area affected historically. The resulting gaps and spatial inconsistencies can confound efforts to develop descriptive and predictive models of outbreak dynamics and fire risk and to design effective forest management plans. We sought to improve annual maps of SBW defoliation and mortality extent across Ontario for the periods of 1984-1998 and 2014-2020 using a combination of balsam fir and white spruce abundance data, Landsat imagery, and the aerial survey maps. Using Random Forests and Density-Based Spatial Clustering for Applications with Noise (DBSCAN), we developed and trained a spatial model to predict the true extent of SBW defoliation and mortality, with the goal of analyzing area disturbed, forest fragmentation, and timber loss.

Results/Conclusions

Model performance differed between previous (1984-1998) and recent (2014-2020) outbreak periods. We found that defoliation and mortality were overestimated by aerial surveys during the previous outbreak and due to coarser survey methods, resulted in a lower model performance - 66% of the test data was classified correctly. The more recent outbreak was surveyed at a higher resolution and resulted in an underestimation of the extent of defoliation but a better performing model - 84% of test data was classified correctly. Our estimates for SBW damage extent suggest a significant overestimation of area affected by more than 90% due to very coarse mapping for the previous outbreak and significant underestimation of area affected by more than 250% for the more recent outbreak due to very fine mapping only under the flight path of the aerial survey. Thus, when using the historical aerial survey maps for statistical modelling, fire risk assessment, or fieldwork site selection, users must consider uncertainty in identified defoliation and mortality extent and detail of mapping used.