Remote sensing-informed zonation for capturing snow-plant-soil interactions and dynamics in the sub-alpine Rocky Mountains

Climate models predict that earlier snowmelt will intensify the annual early summer drought in the Colorado Rocky Mountains, a region where ecosystems are limited by water availability. Plant dynamics in turn influence the water budget through evapotranspiration. Long-term plot studies have shown that early snowmelt leads to a decrease in ecosystem productivity due to low soil moisture. Scaling the understanding from such plot-scale experiments to the watershed management scale is difficult, since snow, soil moisture and plant dynamics are heterogeneous in mountainous terrain, and particularly influenced by microtopography. Recent advances in satellite remote sensing have created an opportunity for monitoring snow and plant dynamics at high spatiotemporal resolutions. However, synthesizing temporally dynamic information into coherent understanding presents a challenge, particularly in conjunction with topography or other key factors.

In this study, we investigate the relationships between topography, snowmelt, soil moisture, and plant dynamics in the East River, CO watershed based on a time series of 3-meter resolution Planet Lab images. We hypothesize that we can identify regions with similar snowmelt and plant dynamics, and that these zones are associated with key topographic features and soil moisture. We use unsupervised clustering methods to identify such regions, which is mathematically equivalent to a dimensionality reduction of time series images into spatial patterns. Comparing different clustering methods (e.g. k-means, hierarchical clustering) and loss functions (e.g. Ward method, Complete method), we find that the identified regions are consistent between methods after removing correlations among the time series by principal component analysis. Results show that the derived zones are associated with particular microtopographic features in terms of flow accumulation and topographic position index. In addition, point source field data have confirmed (with Tukey's pairwise comparison test) that these zones have different soil moisture distributions. This cluster-based analysis is powerful to tractably analyze high resolution time series images for understanding snow-soil-plant interactions as well as to identify optimal locations for soil moisture sensor placement.