Observations and analysis of Beimen sandbank surface change using multi-temporal UAV datasets

Understanding topographic change of sandbank is important in order to monitor ecological function and sustainability of human life along the seashore area. This paper presents the study on topographic change analysis at Beimen sandbank by using multi-temporal UAV (unmanned aerial vehicle) imagery. Due to the Beimen's coastline morphology over large spatial scales (4 km alongshore, i.e., approximately 4 km² of sand-dune system), we use a VTOL (Vertical Take Off and Landing) UAV to acquire high spatial resolution images. In order to reveal the sandbank topographical changes, we used five time series data acquired between February and July of year 2019 and propose a novel strategy to co-register multi-temporal datasets. After dense image matching, we found corrupted pointclouds with significant amount of noise and outlier in the water surface due to image matching failed. To overcome it, we apply Random Forest-Machine Learning (RFML) for removing these outliers in unordered pointclouds. This approach is efficient and robust to remove amounts of noise and outliers, especially in the water surface with sun reflection and wave, while being able to handle large and dense sampled pointclouds. According to its result, we generate DSM from the refined pointclouds and analyze the erosion and accumulation effects at Beimen's sandbank area. Finally, we calculate the erosion and accumulation volume and identify the area affected by erosion and accumulation. We noticed that the highest accumulation area occurs between May 19^th and July 4^th, with an estimated accumulation area of 1,111,282.20 m² and accumulation volume of 224,549.60 m³ due to the heaviest rainfall occurred at July 2^nd and 3^rd. Furthermore, we obtained a total of erosion area with 934,497.60 m² between July 4^th and July 17^th,.