Assessing Resolution Tradeoffs of Remotely Sensed Data for Invasive Species Detection
Lori
Mann
Bruce, Mississippi State University, bruce@ece.msstate.edu
(Presenting)
Wesley
Johnson, U.S. Army Corps Engineers, Wesley.Johnson@erdc.usace.army.mil
Abhinav
Mathur, Optech International Inc., m_abhinav@hotmail.com
In order to aid federal agencies and private companies in mission planning and data analysis for invasive species detection applications, the authors have developed software for generating invasives detection accuracy-resolution-cubes (ARCs) for use in determining sensor resolution requirements. The three-dimensional ARC is the result of an invasives detection model, where spectral, spatial, and temporal resolutions are varied to determine acceptable system specifications. The software has three layers: (1) data ingest and resolution modification, (2) invasives detection model, (3) accuracy cube construction and assessment. The software is flexible, such that various data sources and detection models can be utilized. Two case studies are presented: (i) the detection of Cogongrass (Imperata cylindrical) in habitats containing other grasses, such as Johnsongrass (Sorghum halepense), and (ii) the detection of Water Hyacinth (Eichhornia crassipes) in habitats containing other floating vegetation, such as American Lotus (Nelumbo lutea). The ARCs resulting from the two case studies reveal the trade-offs of spectral, spatial, and temporal resolutions on various models to accurately predict and/or detect the invasives. For example, in the aquatic vegetation case study, overall detection accuracies of 90% or higher can be obtained during late summer, e.g. August, for spectral sensors with 80 - 1000nm FWHM when target abundances are 70 - 100% per pixel. In this fashion, ARCs can be readily used in remote sensing mission planning, sensor design/selection, and data analysis for invasive species applications.