Supplementary Files for: "Structure Identification for High-Dimensional Data in the Vicinity of Bear Lake"

This report focuses on seven water quality measurements taken at 43 different depths on the Bear Lake for the months of June - November in the years 2018 - 2023. These measurements create a high-dimensional dataset on which we apply state-of-the-art machine learning (ML) techniques to look for low-dimensional structure in the data. A similar effort was made for weather measurements taken near the lake. Our analysis revealed that water quality measurements tend to cluster (i.e., group together) by year, while weather measurements tend to cluster by time of the year. This suggests that the structure observed in the water quality measurements cannot be fully explained by seasonal changes, since the weather data structure is fundamentally different than the water quality data structure.

This in mind, we explored potential drivers of this strong year to year clustering in the water quality data. This included an exploration of land use change (see Appendix A) as well as an exploration of water inflows/outflows (see Appendix B). The land use/land cover analysis revealed that land use near the Bear Lake has remained remarkably stable over the past two decades, which means that land use change cannot explain the stark differences we see in lake measurements in the platform data.

In contrast, we find that a combination of max inflows from the Causeway, and max outflows from the Lifton Pumps, can explain about 50% of the variability in the position of each year within the low dimensional representations of the platform data. With only six years to compare, it is difficult to know whether or not this phenomenon is due to chance, but the discovery motivates further exploration of the lasting impact of the maximum inflow and outflows from the Lifton Pumps and the Causeway on the water quality measurements for the following year.