LS2SRC: Landsat Scene to State and Regional Landscape Ecological Classification and Mapping

Federal and state agencies have had significant achievements in regional and national land cover projects, such as GAP, Landfire, and NLCD. These data are based on 1999-2001 Landsat ETM+ data, which is outdated for current landscape conservation management. We propose a periodical updating approach for ecological system database generation. It is LS2SRC: Landsat scene to state and regional classification that is based on our imagery chain standardization (ICS) method.

The LS2SRC includes six steps: (1) state-wide Landsat imagery mosaic using ICS. (2) The generation of a primary training dataset by using the Ecological Systems of 2001 GAP dataset for each state and referencing to NAIP imagery. (3) Ecological system classification of a state by using SEE5/C5 with input of three seasonal images and environmental and geomorphologic variables. (4) The generation of secondary training data for neighboring states. For example, Mississippi (MS) and Alabama (AL), the classified land cover data of MS will provide the most training datasets for AL. However, there are still some ecological system classes that are not included in the overlap imagery areas of the two states. We will use the Step (2) procedure to generate additional samples that complete the whole training dataset for AL. We call this sampling and training process redundancy reduced sampling (RRS). (5) Input the combined sampled data from step 4 into SEE5/C5 classifier to generate the updated ecological system classes for neighboring states (such as Alabama to 2011). (6) Redo step 1 to 5 from a state to its neighboring states.