Identifying Soil Erosion Processes in Alpine Grasslands on Aerial Imagery with a U-Net Convolutional Neural Network

Abstract

Erosion in alpine grasslands is a major threat to ecosystem services of alpine soils. Natural causes for the occurrence of soil erosion are steep topography and prevailing climate conditions in combination with soil fragility. To increase our understanding of ongoing erosion processes and support sustainable land-use management, there is a need to acquire detailed information on spatial occurrence and temporal trends. Existing approaches to identify these trends are typically laborious, have lack of transferability to other regions, and are consequently only applicable to smaller regions. In order to overcome these limitations and create a sophisticated erosion monitoring tool capable of large-scale analysis, we developed a model based on U-Net, a fully convolutional neural network, to map different erosion processes on high-resolution aerial images (RGB, 0.25–0.5 m). U-Net was trained on a high-quality data set consisting of labeled erosion sites mapped with object-based image analysis (OBIA) for the Urseren Valley (Central Swiss Alps) for five aerial images (16 year period). We used the U-Net model to map the same study area and conduct quality assessments based on a held-out test region and a temporal transferability test on new images. Erosion classes are assigned according to their type (shallow landslide and sites with reduced vegetation affected by sheet erosion) or land-use impacts (livestock trails and larger management affected areas). We show that results obtained by OBIA and U-Net follow similar linear trends for the 16 year study period, exhibiting increases in total degraded area of 167% and 201%, respectively. Segmentations of eroded sites are generally in good agreement, but also display method-specific differences, which lead to an overall precision of 73%, a recall of 84%, and a F1-score of 78%. Our results show that U-Net is transferable to spatially (within our study area) and temporally unseen data (data from new years) and is therefore a method suitable to efficiently and successfully capture the temporal trends and spatial heterogeneity of degradation in alpine grasslands. Additionally, U-Net is a powerful and robust tool to map erosion sites in a predictive manner utilising large amounts of new aerial imagery.

Investigating Causal Factors of Shallow Landslides in Grassland Regions of Switzerland

“…This fully convolutional neural network approach for semantic segmentation in images allows for objective and efficient mapping. The U-Net model was trained to identify and map erosion sites on aerial images (Swisstopo, 2010) with the aid of digital terrain model information (Swisstopo, 2014), as described in Samarin et al (2020). The U-Net model was trained on a small area of 9 km 2 and tested on an area of 17 km 2 in the Urseren Valley (Samarin et al, 2020).…”

“…The U-Net model was trained to identify and map erosion sites on aerial images (Swisstopo, 2010) with the aid of digital terrain model information (Swisstopo, 2014), as described in Samarin et al (2020). The U-Net model was trained on a small area of 9 km 2 and tested on an area of 17 km 2 in the Urseren Valley (Samarin et al, 2020). For this study we use the same U-Net model without further training to map the new study sites and focus only on the erosion 80 class shallow landslides, as defined in the introduction.…”

“…For our shallow landslide inventory we map the eroded sites on aerial images (0.25 m resolution) using a U-Net deep learning approach (Ronneberger et al, 2015). The U-Net tool was trained by Samarin et al (2020) to identify and map the extent of soil erosion features on grassland. While this mapping tool is able to distinguish between different erosion processes/appearances (i.e., shallow landslides, livestock trails, sheet erosion and management effects Samarin et al (2020)), here, we focus on shallow landslides, as we aim to understand their causal factors and spatial patterns better.…”

“…The U-Net tool was trained by Samarin et al (2020) to identify and map the extent of soil erosion features on grassland. While this mapping tool is able to distinguish between different erosion processes/appearances (i.e., shallow landslides, livestock trails, sheet erosion and management effects Samarin et al (2020)), here, we focus on shallow landslides, as we aim to understand their causal factors and spatial patterns better. With the U-Net mapping tool, we can identify locations of shallow landslides in a very efficient and precise manner, increasing the possibilities for mapping but also future model validation of soil erosion studies (Samarin et al, 2020).…”

“…While this mapping tool is able to distinguish between different erosion processes/appearances (i.e., shallow landslides, livestock trails, sheet erosion and management effects Samarin et al (2020)), here, we focus on shallow landslides, as we aim to understand their causal factors and spatial patterns better. With the U-Net mapping tool, we can identify locations of shallow landslides in a very efficient and precise manner, increasing the possibilities for mapping but also future model validation of soil erosion studies (Samarin et al, 2020). The mapped shallow landslide sites are subsequently evaluated with a statistical model to identify the most important explanatory variables and gain a better understanding of causal factors as well as regional differences.…”