Multiscale Monitoring of Drought-Induced Forest Mortality by Detecting Spatial Anomalies in Land Surface Temperature and Infrared Reflectance

Forest mortality events are increasing globally because of more intense global-changetype-droughts and infestations by pests, and pathogens, threatening natural resources. Most maps characterizing this land cover change process are local-scale, spatially and temporally isolated case studies that depend on relatively subtle changes in visible/near infrared (VNIR) reflectance or use imprecise aerial surveys for calibration, resulting in uncertainty in the extent and severity of forest mortality and its feedbacks within the Earth system. Improved methods for mapping forest mortality are thus needed to characterize feedbacks of forests with climate change, water resources, and wildfire risk, and will aid in developing early warning systems to inform forest management. Trees stressed by drought are characterized by reduced stomatal conductance and transpiration as trees conserve limited water resources. Water-stressed trees can become more vulnerable to insect infestation, leading to defoliation, disruption of resource flows, and introduction of pathogens. These functional and structural changes are apparent as increased land surface temperature (LST) and short-wave infrared (SWIR) reflectance. The use of LST to detect and quantify forest mortality has been limited to date by the coarse resolution of LST datasets (e.g., 1km from MODIS), the considerable preprocessing necessary to obtain higher resolution LST datasets, and the lack of precise, high resolution tree mortality datasets for calibration and validation. This proposal will leverage recent increased availability of LST and very high resolution (VHR) imagery, as well as advances in unoccupied aerial systems (UAS) technology, to improve methods of mapping forest mortality by incorporating time series of localized LST anomalies across the western U.S., a climate sensitive region that has recently experienced many forest mortality events. In addition to analyzing time series remote sensing data, our research will utilize operational forest health monitoring program data and existing forest mortality field data to identify focus areas that will be visited and mapped at sub-meter resolution using UAS, which will provide robust calibration and validation data for the regional-scale model. Our proposed research includes four main objectives. We will: (1) identify several focus areas of recent forest mortality and develop datasets of individual tree status using VHR imagery from UAS or satellites and convolutional neural networks; (2) develop an algorithm to predict stand-level forest mortality based on patterns of LST and visible, near-infrared, and SWIR surface reflectance; (3) produce annual maps of forest mortality across multiple decades using time series of Landsat-based LST and surface reflectance that span the western U.S.; and (4) characterize the vulnerability of forests to drought across a range of topographic, ecological, and climatic conditions to prototype forest mortality early warning systems in the western U.S. Deliverables for this project will include: (1) training and validation data sets from 6-10 focus areas that capture individual tree status; (2) high-resolution maps of forest mortality in the focus areas; (3) models for detecting and predicting tree mortality; and (4) 30 m annual (1985-2024) maps of forest mortality and drought vulnerability across the western U.S. In addition to refereed publications, all datasets and source code will be made publicly available. The results will provide a foundation for improving forest management, Earth system models, and potential early warning systems to forecast future forest mortality.