LAND COVER CHANGE DETECTION NEAR SMALL WATER BODIES BASED ON RGB UAV DATA: CASE STUDY OF THE POND BAROCH, CZECH REPUBLIC
Keywords: UAV, RGB Data, Maximum Likelihood, Support Vector Machine, Deep Learning, Supervised classification
Abstract. Monitoring changes of land cover near water bodies and water bodies themselves represents a part of environment protection and management. The management can be done at the global or local level. The local level requires more detailed data, which can be collected i.e. by means of aircraft or UAV. The paper describes a case study focused on the utilization of UAV-based RGB data to monitor land cover near the pond Baroch, which is located in the Czech Republic, near the city of Pardubice. The area is specific – it is a small pond accompanied by several smaller pools and connecting canals and surrounded by meadows (often watered), reeds, bushes and some trees Used data were collected by authors by in advance planned flights in August, September, October, November, and December 2021. Support Vector Machine, Maximum Likelihood, Random Trees, and Deep Learning are used as methods to process data and detect land cover changes. Manually interpreted data are used as reference data. Because of the nature of the data (only R, G, and B bands), classification into bare land, the water, vegetation, dry vegetation, and wet vegetation classes only was used. Very high heterogeneity of the observed area, availability of RGB bands only, and very high spatial resolution (1,9 cm per pixel) led to isolated cells.