Application of 3D point clouds derived from repeated UAV LiDAR surveys for erosion mapping

One of the major environmental problems of our time is soil erosion, which is a natural process reducing the fertility of soils and leading to land degradation. Soil erosion mapping helps to develop different control methods and is essential for sustainable agricultural practices. Among the modern remote sensing technologies, UAV laser scanning (ULS) is gaining importance as an accurate and efficient survey solution and can be applied in vegetated areas. It generates point clouds (PC) mapped in 3D coordinates during the survey, allowing the creation of detailed digital terrain models (DTM) which allow identification and mapping of soil runoff traces, rills and gullies.

In our research, we were conducting a monitoring study on a 10 ha vineyard in the Balaton Highlands, where three different mulching and row-cropping techniques (disking, natural cover and sown grass) were applied between the rows of vines, so that we could also investigate their impact on erosion.

The area was surveyed five times between October 2023 and December 2024 using UAV L1 LiDAR on board a DJI M350 drone to collect 3D PCs. A temporary GNSS base station was used in the northern part of the vineyard, placed in the same location for each survey. It acted as a local base station, providing RTK correction for the position calculation of the M350 drone and L1 sensor with centimetre-level accuracy. The raw PCs were later pre-processed using the R package lidR to filter for ground points using the Progressive Morphological Filter with window sizes of 1 and 3 m and distance thresholds of 0.1 and 0.3 m. The PCs representing the ground were compared for differences using the M3C2 module. The plugin allows users to select core points, adjust normal and projection scales, and calculate distances using precision maps. For the latter parameter, a constant value of 5cm was set for all PCs, representing the average measurement error of L1 LiDAR for 3D coordinates.

The results showed that rill erosion is present in all areas due to the direction of the rows planted parallel to the general slope of the vineyard, but is most dominant in areas cultivated by disking without cover crops. Surprisingly, the most severe rill erosion (~10 cm) was found along a small ditch created in a row using grass as a natural cover after the winter and spring period in 2024. We also detected a newly formed, early-stage gully in the south-western corner of the bare soil section, which channels the moving soil and sediment into a depot in this corner. Analysis of the robustness of the calculated distances showed that even low vegetation, such as grass, had a negative effect on the LiDAR measurements and PC quality, which (together with the average 3D coordinate accuracy of 5 cm) rendered the distances insignificant.

Despite its current drawbacks, the method is useful for detecting changes and formation of erosion features, and will later be tested against empirical soil erosion models to investigate similar patterns of erosion and also volumetric changes.