Characterization of Ash (Fraxinus spp.) in a Riparian Forest in Southeastern Michigan Using Spectral and Physical Variable Models

Abstract

The emerald ash borer (Agrilus planipennis Fairmarie) has killed tens of millions of ash (Fraxinus spp.) trees in Michigan alone. Riparian and lowland areas typically contain large proportions of ash and have been especially affected in Southeast Michigan. The loss of up to 20-60% of the overstory has significant implications for forest succession and floodplain stability. The goal of this study was to identify the proportion and evaluate the status of ash in a Southeast Michigan riparian forest community and to develop a minimally field-intensive GIS/ remote sensing method for identifying dominant ash populations employing multiple linear regression (MLR) and binary logistic regression (LR). I gathered a local sample of nearly 1000 ash trees at 60 locations within the Sharonville State Game Area, Washtenaw County, Michigan, and combined this data with Landsat remotely sensed imagery and physical map-based variables in an effort to model ash population distributions. Landsat imagery and derived products were evaluated for their ability to segregate an ash spectral signature, while the map-based variables were evaluated for their ability to represent local hydrologic conditions interpreted from the autecology of ash species. An existing, ash containing lowland deciduous forest classification for Michigan (IFMAP) was also evaluated for its ability to predict ash presence/ absence. Ash mortality comprised a total of 17% of the sampled deciduous forest with virtually all trees deceased and symptomatic of emerald ash borer infestation. The MLR and LR predictive models generally out-performed IFMAP in predicting ash iii presence/absence. A single Landsat scene was generally unable to distinguish an ash related spectral signature, though elevation based variables contributed to successful prediction of ash presence with up to 91.7% accuracy. For the successful prediction of ash percent coverage, hyperspectral remotely sensed imagery would likely be necessary.