Mapping Tree Water Deficit with UAV Thermal Imaging and Meteorological Data

Abstract

Abstract The mapping of forest stands and individual trees affected by drought stress is an important step in targeted forest management with the aim of creating resilient and diverse forests. UAV-based thermal sensing is a promising method to acquire high-resolution thermal data, yet the performance of typical UAV-adapted low-cost sensors are somewhat limited in deriving accurate temperature measurements. Uncertainty is evident in the effects of internal sensor dynamics as well as environmental variables such as solar radiation intensities, relative humidity, object emissivity and wind to name a few. Furthermore, the accurate assessment of drought stress in trees is challenging to quantify, and typical research station methods can be laborious and cost-intensive and particularly challenging when carried out in the field. In this study, we explored the possibility to acquire reliable tree canopy temperature using the thermal band of the Micasense Altum multispectral sensor while examining the prospect of quantifying drought stress by implementing point dendrometers and UAV-derived tree canopy temperature to model Tree Water Deficit (TWD). In an indoor environment we showed that the usage of a limited number of pixels (< 3) can result in temperature errors of over 1°C whereas increasing the spot size can reduce the mean difference to 0.02°C when using leaf temperature sensors as validation. Interestingly, leaves which were subjected to drought treatment (unwatered) resulted in a higher root mean squared error ((RMSE) (RMSE = 0.66°C and 0.73°C) than watered leaves (RMSE = 0.55°C and 0.53°C) due to most probably a lower emissivity of the dryer leaves. In a comparison of field acquisition methods, measuring the tree crown temperature of a selected tree from various incidence angles derived from typical gridded flights resulted in a mean standard deviation (SD) of 0.25°C and a maximum SD of 0.59°C (n = 12), where as a close-range hovering method resulted in a mean SD of 0.09°C and a maximum SD of 0.1°C (n = 8). Modelling the TWD from meteorological and point dendrometer data from the 2021 growth season (n = 2928) resulted with a R2 = 0.667 using a Generalised Additive Model (GAM) with the Vapor Pressure Deficit (VPD), wind speed and solar radiation as input features and a point dendrometer lag of one hour. When predicting individual tree TWD with UAV-derived tree canopy temperature, relative humidity and air temperature as input features, a RMSE of 4.92 (µm) and R2 of 0.87 was achieved with a GAM. The GAM with the Leaf-to-Air Pressure Deficit (LVPD) as an input feature resulted in a RMSE of 6.87 (µm) and a R2 of 0.71. This study presents a promising method to acquire thermal data for the purpose of mapping TWD of beech on an individual tree basis. Further testing and development is an imperative and more drought period point dendrometer data as well as higher resolution meteorological data is required.