LB 36-304 - Uniting remotely sensed and in-situ data records to understand seasonal dynamics of lake metabolism: a GLEON-NASA collaboration

The application of remote sensing data to studies of lake microbial community composition and metabolism in inland waters has been limited by the relatively low spectral resolutions of existing remotely sensed data compared to the relevant scales of variation in freshwater systems. The upcoming NASA Surface Biology and Geology (SBG) hyperspectral mission will provide routine high spectral resolution data globally. However, few studies have been conducted to explore the possible applications of these data to inland waters, partially due to the difficulty of collecting coincident in-situ datasets. In this study we take advantage of existing long-term limnological datasets and team science capabilities, to address this gap through a partnership between the Global Lake Ecological Observatory Network (GLEON) and NASA’s SBG Mission. The GLEON Lake Expedition 2022 team is a group of graduate students and principal investigators representing a diversity of disciplines in aquatic science and remote sensing, with the ultimate goal of connecting water quality in inland waters to remotely sensed hyperspectral data. Lake Mendota, a eutrophic lake in Madison, WI, USA has been routinely monitored by the North Temperate Lakes Long Term Ecological Research (NTL-LTER) program over the past 30 years. We used 27 scenes collected from airborne and space-based hyperspectral sensors, in conjunction with 16S rRNA microbial data from 2000-2019 and modeled metabolic data from 2010-2020, to describe the relationships between remotely sensed and in-situ datasets. Using a random forest model, we found that the seasonal dynamics of top microbial OTUs in Lake Mendota predicted lake metabolism variables, with members of the Actinobacteriota and Cyanobacteria phyla as particularly important drivers. Through a time series analysis of Sentinel-2 data, we found a notable seasonal relationship between the red band ratio and modeled net primary productivity (NPP). We used principal component analysis (PCA) to reduce the dimensionality of hyperspectral images. The first 2-4 PCA loadings, which capture approximately 30-50% of variance in the raw data, show temporal and spatial variability across the hyperspectral scenes. We will use the PCA results in redundancy analysis to examine how the primary loadings of the hyperspectral data relate to metabolic processes such as NPP as well as key microbial groups such as cyanobacteria. Given the growing need for water quality monitoring in inland water bodies and the increasing availability of hyperspectral data, our study serves as a relevant connection between high resolution remote sensing data and in-situ metrics of ecological function and water quality.