HR: 17:00h
AN: B34A-04    [Abstracts]
TI: Quantitative Analysis of Land Loss in Coastal Louisiana Using Remote Sensing
AU: Wales, P M
EM: pmwales@olemiss.edu
AF: University of Mississippi, 118 Carrier Hall, University, MS 38677 United States
AU: * Kuszmaul, J
EM: kuszmaul@olemiss.edu
AF: University of Mississippi, 118 Carrier Hall, University, MS 38677 United States
AU: Roberts, C
EM: croberts@pixsell.com
AF: PixSell Inc., 1014 Priory Place, McLean, VA 22101 United States
AB: For the past thirty-five years the land loss along the Louisiana Coast has been recognized as a growing problem. One of the clearest indicators of this land loss is that in 2000 smooth cord grass (spartina alterniflora) was turning brown well before its normal hibernation period. Over 100,000 acres of marsh were affected by the 2000 browning. In 2001 data were collected using low altitude helicopter based transects of the coast, with 7,400 data points being collected by researchers at the USGS, National Wetlands Research Center, and Louisiana Department of Natural Resources. The surveys contained data describing the characteristics of the marsh, including latitude, longitude, marsh condition, marsh color, percent vegetated, and marsh die-back. Creating a model that combines remote sensing images, field data, and statistical analysis to develop a methodology for estimating the margin of error in measurements of coastal land loss (erosion) is the ultimate goal of the study. A model was successfully created using a series of band combinations (used as predictive variables). The most successful band combinations or predictive variables were the braud value [(Sum Visible TM Bands - Sum Infrared TM Bands)/(Sum Visible TM Bands + Sum Infrared TM Bands)], TM band 7/ TM band 2, brightness, NDVI, wetness, vegetation index, and a 7x7 autocovariate nearest neighbor floating window. The model values were used to generate the logistic regression model. A new image was created based on the logistic regression probability equation where each pixel represents the probability of finding water or non-water at that location in each image. Pixels within each image that have a high probability of representing water have a value close to 1 and pixels with a low probability of representing water have a value close to 0. A logistic regression model is proposed that uses seven independent variables. This model yields an accurate classification in 86.5% of the locations considered in the 1997 and 2001 survey locations. When the logistic regression was modeled to the satellite imagery of the entire Louisiana Coast study area a statewide loss was estimated to be 358 mi² to 368 mi², from 1997 to 2001, using two different methods for estimating land loss.
DE: 0497 Wetlands (1890)
DE: 1640 Remote sensing (1855)
DE: 1855 Remote sensing (1640)
DE: 1890 Wetlands (0497)
SC: Biogeosciences [B]
MN: Fall Meeting 2005