Earth Lab EDS Seminar: Shashi Konduri, NEON - Scaling from Plot to Peninsula
Date: Tuesday, June 13th, 2023
Time: 11-11:50 AM Mountain Time
Location: ZOOM + 372A & B (VizStudio), SEEC, University of Colorado Boulder
Title: Scaling from Plot to Peninsula: High-resolution Vegetation Type Mapping in Alaska Low Arctic Tundra
Abstract: High-resolution maps of vegetation distribution are crucial for studying the impacts of climate change on Arctic tundra ecosystems and their associated feedbacks. Bridging scales and integrating field-scale observations, airborne remote sensing and environmental datasets, this study examines the current distribution and potential future redistribution of vegetation types across the low Arctic region of northwestern Alaska, USA. In a two-step upscaling approach, we first developed a deep neural network-based classifier, trained on field vegetation surveys and airborne hyperspectral imagery, to create high-resolution (5 m) landscape-scale maps of vegetation types with an accuracy exceeding 93%. In the second upscaling step, we used a Random Forest environmental niche model trained on the landscape-scale maps to determine the habitat suitability of vegetation types across the southern Seward Peninsula region, based on topography and climate variables. Using future climate projections from five earth system models (RCP 8.5, CMIP5) as inputs to the niche model, we predict that shrub cover is expected to increase from 32.2-34.5% (mean 33.4%, std dev 0.85%) in 2010-19 to 51.6-75.4% (mean 69.4%, std dev 9.1%) by 2050-59 at the regional scale, particularly along hillslopes and high-resource environments like floodplains and stream corridors, at the expense of other vegetation types. Generation of high-resolution maps of vegetation types for the data-scarce low Arctic tundra region is an important step towards improved parameterization of Arctic land surface models and simulation of Arctic vegetation in Earth system models.
Short speaker bio: Shashi Konduri is a remote sensing scientist at the National Ecological Observatory Network (NEON)/Battelle, Boulder, CO. He is interested in studying the structure, composition, and environmental drivers of vegetation distribution over large spatiotemporal scales using remote sensing data. He has experience working with earth observing data from different sensors (multispectral/hyperspectral/lidar), mounted on various platforms (spaceborne/airborne) and collected across different spatiotemporal scales to study diverse vegetation types (crops/Arctic tundra shrubs/tropical savannas). In collaboration with scientists at reputed government agencies such as NASA, Oak Ridge National Laboratory (ORNL) and the US Forest Service, he has developed and deployed spatiotemporal data mining methods for improved understanding of plant structure and composition and enhanced predictive modeling of vegetation productivity.