Challenges in reconstructing seasonally driven landslide motion from optical satellite data: insights from the Del Medio catchment, NW Argentina

Abstract. Optical satellite images are a valuable resource for studying slow-moving landslides from space. However, monitoring displacement through pairwise image correlation and time-series inversion presents several challenges, including the impact of seasonality on measurement accuracy. Seasonal biases arise from systematic measurement errors related to variable illumination conditions and shadows. These errors manifest in the form of an oscillation pattern in the satellite-derived time series. This complicates the identification of a true seasonal component linked to feedback mechanisms between landslide displacement and a seasonally variable climate.

Here, we provide a comprehensive evaluation of different strategies to reduce the magnitude of seasonal biases. These include modeling the seasonal error component, restricting correlation to pairs with similar illumination, weighting, and upsampling optical images. We find that all methods can reduce the impact of systematic seasonal bias at different trade-offs: longer processing times, creation of sparse or disconnected networks, dependence on topographic data, or potential alteration of the true displacement signal.  

We evaluated the removal of seasonal bias from displacement time series derived from optical satellite data (Landsat-8, Sentinel-2, PlanetScope) over a large slow-moving landslide in the Río Del Medio catchment in northwest Argentina. The transition zone between the Altiplano-Puna plateau and the Andean foreland is characterized by a highly seasonal climate, with intense rainfall during the South American summer monsoon and semi-arid conditions throughout the rest of the year. Over the 10-year observation period, the landslide accumulated approximately 35 m of displacement at spatially and temporally variable rates. After the removal of the seasonal bias component, three acceleration phases remain. These date to early 2017, early 2018, and late 2021, and all fall within the rainy season of the respective year. The timing of the acceleration suggests precipitation as a major driver of a landslide that is already preconditioned by infiltration and sliding through inherited fault structures, weakened lithologies, and freeze-thaw processes at high altitudes of up to 4500 m above sea level.
 
Based on this example, our study provides a basis for selecting the appropriate correction methods to address seasonal biases. Considering all effects of seasonality is essential to improve the accuracy of satellite-derived displacement measurements and better constrain the feedback mechanisms between landslide velocity and a seasonal climate.