Canopy spectral reflectance detects oak wilt in red oaks across the landscape

Gerard Sapes, Department of Ecology, Evolution, and Behavior, University of Minnesota, Saint Paul, MN, Cathleen Lapadat, Ecology, Evolution and Behavior, University of Minnesota, Saint Paul, MN, Anna Schweiger, Département de sciences biologiques, Université de Montréal, Montréal, QC, Canada, Jennifer Juzwik, USDA Forest Service, Rebecca A. Montgomery, Department of Forest Resources, University of Minnesota, St. Paul, MN, Hamed Gholizadeh, Oklahoma State University, Stillwater, OK, Philip Townsend, Department of Forest and Wildlife Ecology, University of Wisconsin - Madison, Madison, WI, John A. Gamon, Department of Earth and Atmospheric Sciences, University of Alberta, Edmonton, AB, Canada and Jeannine Cavender-Bares, Ecology, Evolution, and Behavior, University of Minnesota, Saint Paul, MN

Background/Question/Methods
Oak wilt, a disease caused by the invasive fungal pathogen Bretziella fagacearum, is one of the greatest threats to oak trees in the Midwest. Consequently, there is increasing interest in early detection and monitoring of its spread. Canopy spectral reflectance is influenced by chemical and structural traits of species and their physiological status, information that can be combined to identify diseased oaks. Here, we develop statistical models for oak wilt detection at the landscape scale using imaging spectroscopy data collected by two airborne systems covering different wavelength ranges (400-1100 nm, visible to near infrared, and 400-2500 nm, visible to short-wave infrared, respectively). We coupled ground-level tree identification and symptomatic crown surveys with airborne imaging spectroscopy to identify potential host species – the vulnerable red oaks-- and to detect oak wilt at early stages of disease development in a temperate, mixed hardwood forest. We compared different wavelength ranges and flight times to assess the contribution of the short-wave infrared and the timing of data collection for oak wilt assessments. Finally, we tested the extent to which spectral indices associated with canopy physiological attributes, including photoprotective pigment content, photosynthetic function, and water status, were capable of differentiating healthy and diseased red oaks.
Results/Conclusions
We achieved high accuracy through a stepwise classification process. We first distinguished oaks from other species (90% accuracy), then red oaks from white oaks (93% accuracy), and, lastly, diseased from healthy trees (80% accuracy). Our results indicate that analytical approaches that include the short-wave infrared regions of the reflectance spectrum provide statistically simpler and more accurate models. Additionally, we identified several spectral indices associated with physiological status that detected statistically significant differences between healthy and diseased trees. Overall, these indices better differentiated healthy and diseased oak trees during late summer, especially indices associated with canopy photosynthetic and water status performed particularly well. In summary, our results from classification models and physiological indices indicate that detection accuracy increases when late summer data are used, likely because physiological decline in diseased trees is more pronounced. Our study suggests that managers can use a combination of models developed for imaging spectroscopy and multispectral indices across the landscape and at the stand level to detect oak wilt at early stages of disease development, and potentially before the pathogen has spread to surrounding healthy trees.