Derive Accurate Water Thermal Maps from UAV-Mounted Thermal Camera for Ecological Application

Water temperature is a crucial parameter for freshwater ecosystems, considering its modification caused by climate change but also from more punctual anthropic activity (i.e., water abstraction, riverbank vegetation, etc.). The alteration of water absolute temperature (Tk) regime and its spatial distribution has a direct effect on freshwater ectotherm communities at all trophic levels. Thus, there is an increasing need for freshwater research and management purposes of methods to accurately monitor water temperature at high resolution. UAV- integrated radiometric thermal sensors represent an increasing popular method to map water Tk to a fine scale. However, the production accurate of Tk maps still suffer of three main limitations: (I) low water Tk estimation accuracy from camera-recorded thermal emissivity values, (II) difficulty of extracting solely watered areas in complex river morphology, (III) estimating impacts of shading on water Tk, when shaded areas are non-visible. Different methods have been proposed in literature to reduce Tk estimation error (I), but error up to 8°C are still reported using these approaches. Techniques are also available to extract the watered area (II), usually based on surfaces radiometric properties, but these are only suitable for relatively simple river channel morphology. Point (III) has rarely been addressed, despite the potential of shaded areas to determine temperature variation downstream, by changing the amount of radiation reaching the water. The aim of the study is to offer new solutions for each of these three points, integrating established methods with new techniques. Particularly for the point (I) we introduce a new vignetting-effect correction method to reduce the camera bias on the thermal image that coupled with a precise estimation of environmental parameters, allowing water Tk estimation with Average Mean Absolute Error <0.5 °C. To address the needs to have a map of the solely watered areas (II), we present here a method to filter out all non-aquatic pixels from the thermal map, based on scene’s differential reflectance behaviour in the Red and Near Infrared wavelength; in the resulting Tk map, all non-watered areas (including centimeter-sized features, i.e. emerging stones) are successfully removed. Further, we address point (III) by integrating in the map the Tk estimates for non-visible pixels, derived from the temperature of the nearest visible ones eventually corrected for the loss of heat due to consistent coverage (shading). These combined approaches greatly improve accuracy and reduce bias in the magnitude, spatial variation, and account for (non- visible) processes impacting of freshwater Tk estimates.