Because of rapid technological advances in sensor development, computational capacity, and data storage density, the volume, velocity, complexity, and resolution of these data are rapidly increasing. Machine learning, data mining, and other artificial intelligence approaches offer the promise for improved prediction and mechanistic understanding, and methods for fusing data from multiple sources into data-driven and hybrid models comprised of both process-based and deep learning components. At the watershed scale, in situ measurements must be combined with near-surface, airborne, and satellite remote sensing data to understand the structure and function of ecosystems in heterogeneous landscapes; their interactions with nutrients, water, and energy; and ecohydrological responses to environmental change. However, sampling in remote, dangerous, or topographically complex watersheds is often prohibitive, necessitating use of sensor optimization and scaling techniques for characterization of landscape properties, vegetation distributions, and responses to climate and weather events.
Results/Conclusions
We applied data mining and machine learning approaches to delineate ecoregions and characterize vegetation distributions in Arctic ecosystems from in situ, airborne, and satellite remote sensing data. First, we developed a quantitative methodology for stratifying sampling domains, informing site selection, and determining the representativeness of measurement sites and networks. We applied multivariate spatio-temporal clustering (MSTC) to gridded data for the State of Alaska at 2 km x 2 km resolution to define multiple sets of bioclimatic ecoregions across two decadal time periods. Representative sampling locations were identified on present and future ecoregion maps. A representativeness metric was developed and used to characterize the environmental similarity of each sampling site. Second, we combined high resolution multi-spectral remote sensing from the WorldView 2 satellite with LIDAR-derived digital elevation models to characterize polygonal ground in the the tundra landscape on the North Slope of Alaska. Vegetation distributions sampled in situ within 1 m x 1 m plots were used to derive relationships with the remote sensing data to upscale vegetation types for the larger Barrow Environmental Observatory (BEO) region. Third, we used airborne hyperspectral remote sensing data from NASA’s AVIRIS-NG and satellite remote sensing platforms to develop high resolution maps of vegetation community distributions. We built deep learning models, trained with field-based vegetation community survey observations, to classify vegetation across three watersheds on the Seward Peninsula of Alaska. The resulting maps are being used to design field sampling campaigns and to inform development of fractional plant functional type (PFT) distributions for models used to understand ecosystem responses to climate change.