Remote Sensing of Vegetation Parameters for Modeling Coastal Marsh Response to Sea Level Rise

Monday, 15 December 2014
Kristin B Byrd1, Lisamarie Windham-Myers1, Bernhard Warzecha2, Rebecca Crowe3, Michael C. Vasey2 and Matthew Ferner2, (1)U.S. Geological Survey, Menlo Park, CA, United States, (2)San Francisco Bay National Estuarine Research Reserve, Tiburon, CA, United States, (3)San Francisco State University, Biology Department, San Francisco, CA, United States
Coastal planners are seeking ways to prepare for the potential impacts of future climate change, particularly sea level rise though management of future risks is complicated by uncertainty in the timing, distribution and extent of these impacts. Sea level rise impacts will be most evident at the regional level where decisions related to climate change adaptation including those related to land use planning and habitat management typically occur. To aid coastal managers with decision-making we are integrating remote sensing data with the marsh equilibrium model (MEM3) to project coastal marsh habitat response to future sea level rise. MEM3 is a 1-dimentional, calibrated Excel-based model that incorporates both physical and biological feedbacks to changing relative elevations. Modeled future elevations are then distributed at the regional scale with LiDAR DEMs to project changes to coastal habitats and dependent wildlife. Because plant biomass and structure influence both organic and inorganic accretion, MEM3 includes multiple vegetation input variables. Deriving these variables, including maximum and minimum elevations of marsh vegetation, peak aboveground biomass, and elevation at peak biomass from remote sensing will enable the model to have spatially variable inputs across sites. We are evaluating 30m Landsat 8 and 2m World View-2 (WV2) satellite data for mapping peak biomass at Rush Ranch, a highly diverse brackish marsh in the San Francisco Bay National Estuarine Research Reserve. The high spatial resolution of WV2 produces greater variability in plant reflectance at the pixel scale than Landsat 8. Initial results show the need for plant community-specific biomass models with WV2 to account for differences in plant structure and canopy architecture. When removing plots dominated by Salicornia pacifica and Lepidium latifolium, peak biomass is best estimated with an NDVI-type vegetation index based on WV2 near infrared bands 7 and 8 (R2 = 0.21, RMSE = 318 g/m2). Biomass values in the top 95 percentile (greater than 1385 g/m2) are estimated with an average absolute error of 705 g/m2. While error rates may decrease with the addition of more seasonal biomass data, these peak biomass values are within a suitable range of model performance based on a sensitivity analysis of the peak biomass variable in MEM3.