Silvilaser 2019 - Poster Presentations »
Choosing filtering parameters in LiDAR clouds collected over Amazon rainforest
With the advent of Airborne Laser Scanning (ALS), the possibility of obtaining a large number of geo-referenced points from the Earth's surface coverage has arisen. This technology has opened up new horizons in the field of forest measurement, especially in the quantification studies of forest biomass and structure. However, during the path traveled by the laser beam, some returns that must be considered as noise can be recorded by the sensor. Such noise may be caused by equipment interference, error in the interpretation of the return signal and physical obstacles between the forest and the aerial platform, such as birds and water vapor. These noises can be easily visualized, but manual removal becomes impractical, as the cloud of points can reach the magnitude of terabytes of data. In this way, the need to use automated algorithms to filter out the noises is perceived. The objective of this work was to optimize the cloud filtration parameters obtained by means of an airborne laser survey of the Amazon rainforest. Overflight was conducted between 2015 and 2016, with a minimum density of four returns per square meter (average of six returns per square meter) and a footprint of approximately 0.3 m. The study area is located in the northern channel of the Amazon River, in the state of Pará, in the Jari river basin. The processing was performed using the free software FUSION (v.3.8) and R (v.3.5.1), using the package lidR (v.2.0.1). The returns from the terrain were classified by means of the cloth simulation filter (CSF) algorithm and then the digital terrain model (MDT) was generated, with a resolution of 1 m. The cloud was normalized and then subjected to a local noise filter, testing standard deviation ranging from 1 to 7, with a 1-deviation interval. The standard deviation was computed locally considering a 10-meter window. The number of points removed by the filter decreased as the standard deviation increased. Up to 3 deviations, all the noises were removed, but points referring to the tree canopies were also filtered. Using 4 detours, it was possible to remove all visible outliers without compromising the forest canopy structure view. From 5 deviations up it was noticed that the filter was not efficient to remove the noises.