Soil erosion trends can be obscured in remote sensing data if inappropriate spatial resolutions are used: evidence from a high-latitude tundra landscape undergoing shrub expansion

The Arctic tundra biome is large and difficult to access. Satellite monitoring is essential for observing changes in these areas, but in most cases the resolution of imagery is larger than typical vegetation patch sizes. NDVI greening trends have been observed in most parts of the Arctic since the 1980s. These trends can be explained by the expansion of vegetated areas, higher biological productivity and changing vegetation composition. Low to tall stature shrubs are changing the spectral reflectance characteristics of the overall tundra vegetation. In particular, the NIR wavelengths are affected, resulting in higher NDVI values.

At the same time, soil erosion threatens parts of the tundra biome such as in Iceland. Soil erosion occurs due to biogeomorphological processes. Uncertainties remain in how these processes will adjust to a rapidly changing climate and to anthropogenic pressures such as grazing. This makes the future trajectory of these landscapes difficult to predict. Critical thresholds may exist in these landscapes, which if crossed, can lead to irreversible desertification. For these reasons, it’s important to be able to accurately assess the environmental state of these landscapes, especially the extent of soil erosion.

Satellite observation of NDVI greening in areas undergoing shrub expansion could mask other trends, such as increasing levels of soil erosion. This is because typical satellite datasets used for tundra monitoring have resolutions in the order of 10s of m, which means that each pixel tends to include multiple different land cover types. At the level of a sensed pixel, higher NDVI values of shrubs could lead to a net positive NDVI trend, despite the eroded area increasing.

To address this issue, we need to establish which spatial resolutions are appropriate for monitoring. We used a multi-scale study in a degraded tundra landscape in northern Iceland. Different satellite products from (PlanetScope, Sentinel-2, Landsat-8) were compared with 1.1 km² multispectral UAV imagery collected in 2021. The very high-resolution UAV imagery (0.05 m resolution) is used to classify land cover and allows us to explore how the composition of different land cover classes affects the overall NDVI value of a satellite pixel (3 – 30 m resolution) at the same location.

We find that for the same NDVI values in a satellite pixel, the UAV data reveals large variations in the degree of soil erosion. This can mainly be attributed to variability in the ratio of shrub cover to other vegetation cover, which alters the spectral signature of a pixel. This makes the interpretation of NDVI trends more difficult and stresses the importance of using an appropriate spatial resolution. Landsat-8 (30 m) revealed low accuracy in resolving tundra heterogeneity, while Sentinel-2 (10 m) and PlanetScope (3 m) showed significant improvements.

This study highlights the importance of using the right spatial resolution when monitoring highly fragmented environments, and the need to consider that an increase in NDVI may not reflect an improvement in environmental state.