Optimizing Spectral and Spatial Resolutions of Unmanned Aerial System Imaging Sensors for Monitoring Antarctic Vegetation

• Main Authors: Turner, Darren, Malenovky, Zbynek, Lucieer, Arko, Turnbull, Johanna, Robinson, Sharon A
• Format: Article in Journal/Newspaper
• Language: unknown
• Published: Research Online 2019
• Subjects: antarctic; spectral; optimizing; monitoring; sensors; imaging; system; aerial; unmanned; resolutions; vegetation; spatial; Antarctic; Antarc*; Antarctica
• Online Access: https://ro.uow.edu.au/smhpapers1/926; https://ro.uow.edu.au/cgi/viewcontent.cgi?article=1940&context=smhpapers1

Antarctic moss communities, found in the spatially fragmented and fragile moss beds, can serve as indicators of the regional impacts of climate change. Unmanned aerial systems (UAS) carrying visible and near infrared (VNIR) sensors are a suitable nonintrusive mapping platform. UAS deployments in Antarctica are, due to weather and logistical restrictions, infrequent and short, thus it is essential that field time is optimized. This article identified the optimal spectral and spatial resolution of the UAS-based sensors to facilitate efficient data acquisition without jeopardizing the accuracy of remotely sensed moss health indicators. A hyperspectral line scanner was used to collect imagery of two moss study sites near the Casey Australian Antarctic base. The spectral and spatial data degradation simulated two lightweight sensors that could be used for more efficient spectral image acquisition in the future. These simulations revealed that the spectral quality deteriorated more definitively at the spatial resolution where moss spectra started to mix with spectra of surrounding rocks. Subsequently, random forest models (RFMs) were trained with lab measurements for predicting chlorophyll content and effective leaf density. The RFMs were applied to the UAS imagery of the reduced spectral and spatial resolutions to quantify decline in accuracy of both indicators. We identified the optimal UAS sensor capable of mapping a relatively large moss bed (∼5 ha) with the prediction accuracy similar to the hyperspectral system. This sensor would be a frame camera acquiring 25 VNIR spectral bands at a spatial resolution of 8 cm. This developed methodology has the potential to be adopted for other similar vegetation biophysical/chemical plant traits.