Single tree detection in agro-silvo-pastoral systems from high resolution digital surface models obtained from UAV- and gyrocopter-based RGB-imaging

Cork oak decline in Mediterranean silvo-pastoral systems is of clear evidence. The reasons for this decline range from climate change, unfavorable site-related factors, pathogens, pests, and mismanagement. However, cork oaks are economically and ecologically important. Therefore, monitoring the temporal vitality of single trees is of key importance to investigate and differentiate the different stresses. In a spatial or regional context, remote sensing is a powerful monitoring tool. It is well investigated that multi- and hyperspectral image analysis can be used to analyze tree vitality. To conduct remote sensing image analysis of cork oaks, a spatial resolution of smaller than 0.3 m is demanded to identify pure tree crown pixels. The latter are a precondition for multi-temporal monitoring of tree vitality development. While satellite- and airplane-borne remote sensing approaches hardly provide the desired spatial (< 0.3 m) and temporal (event- and phenological-driven) resolution, remote sensing sensors mounted on umanned aerial vehicles (UAVs) can provide very high spatial resolutions (< 0.05 m) and can be operated daily or weekly. A clear disadvantage of UAV-based monitoring approaches are limitations of regional coverage due to aviation regulations. In general, UAVs cover up to 100 ha per day. Consequently, there is a niche for low-weight, low-cost, and low-altitude operating manned aircrafts, such as gyrocopters, providing similar very high spatial and temporal resolutions like UAVs but can cover significant larger areas of up to 10.000 ha per day. A key advantage of both carrige systems, UAVs and gyrocopters, are below cloud operating altitudes which enable the acquisition of optical images despite cloud cover. The objective of this contribution is the improvement of single tree classification enabling an improved vitality monitoring with remote sensing techniques. We present a method to derive single trees from the analysis of very high resolution Digital Surface Models (DEMs) which are obtained from very high resolution RGB images and Structure from Motion (SfM) techniques. The study area is the Dehesa San Francisco in Andalucia were we conducted a UAV- and gyropcopter-campaign from March 3rd to 5th, 2016. We used a multirotor UAV for image acquisition in 2 cm resolution and a gyrocopter for 3 cm resolution. Besides, RTK-GPS field measurements were carried out to provide precise ground control points for georectification and data evaluation.