Developing Spectral Metrics as Early Indicators of Water Stress Detection at the Canopy Level

Drought has become an increasing concern over the last few years in forest ecosystems. Understanding how forests respond to drought is critical to elucidate possible drought consequences for forest ecosystem structure and function. There is growing consensus that future climates will be characterized by extreme droughts and extreme precipitation events that will fall outside the historical range to which species and ecosystems are adapted. The limited information of how Northeastern U.S. forest tree species will respond to moderate to extreme drought events have promoted an increasing need to develop monitoring techniques which help us better understand the implications of future drought events on forest structure. Recent technological advances in remote sensing techniques have opened up new opportunities for forest health monitoring which are a great complement to the ecological and physiological data that have been collected within current monitoring programs. The aim of this project was to monitor the vegetation water content at the canopy level for 4 plots that are part of the DroughtNet project, a manipulative ∼50% throughfall removal experiment in the Northeastern U.S. forest. To obtain spectral information from the canopy level I used two different hyperspectral sensors: 1) the G-LiHT airborne hyperspectral sensor with a high spatial (1m) and spectral (4nm) resolution, and 2) the ASD FieldSpec Pro spectrometer that records reflected radiation within the 350–2500 nm spectral domain with a spectral resolution of 3 nm. Ten different water-sensitive hyperspectral indices were calculated to analyze differences between treatments looking for the most reliable way to detect water stress signals. The reflectance of the canopy was analyzed over time as well as the behavior of the two dominant species of the forest: White Pine and Red Oak. To estimate periods of time representing a range of water-stress conditions at the Thompson Farm throughfall experiment, soil moisture and soil water potential were also monitored continuously at multiple locations and depths within each plot. Our results suggest that for the weather conditions which occurred in 2017, the spectral comparison between treatments did not indicate that the spectral indices are more sensitive than the physiological measurements commonly used for water content estimations (i.e. gas exchange, sap flow data). Although none of the spectral indices showed early signals of water stress conditions, some of the indices performed well with correlation analyses for the leaf water content and specific leaf area, with the NDII and PRI standing out. This investigation reaffirms the importance of continuing with monitoring studies that can complement the DroughtNet experimental project and will be valuable for an overall evaluation of the experiment in the long term