Final Project

The final project in Remote Sensing was an opportunity to take some of what we learned in class and put it into practice on our own. We were to come up with a question or two that could be answered using remote sensing data, whether data we collected ourselves or data that was provided.

As I was in the middle of a move during finals, I opted to use provided data for the Lake Tahoe Basin. Using 2010 Landsat 5 TM satellite imagery (30 m resolution), I created a land cover map of a subset study area around Lake Tahoe which was then compared with a subset of the same area from the 1992 National Land Cover Dataset (NLCD) (30 m resolution Landsat imagery used to create it). I was interested in answering two questions: what is the land cover type and what areas do they cover in my study area, and where has the land cover changed occurred geographically between 1992 and 2010.

My steps in the project were first in ERDAS Imagine, including preprocessing the 2010 Landsat imagery by adjusting the breakpoints within the histogram to widen the range of values in the image and enhance contrast and feature identification and clipping the 2010 data to match the pre-clipped 1992 study area. After preprocessing, I used automated unsupervised classification to create groupings or clusters of pixels with like values. I recoded those groups to a smaller number of classes using the NLCD modified Anderson classification scheme, noting there were about 5 "confused" classes where pixels from different classes were mixed (e.g. agriculture and herbaceous plants). Recommendations would be to further refine those classes through supervised classification and then mosaic the results back with the unsupervised classification.

After creating the land cover map I moved over to ArcMap and projected the 1992 NLCD to the UTM Zone 10N WGS84 coordinate system to match the 2010 data. I determined the area and percentage for each land cover type then created a difference raster using the Raster Calculator to find where there were areas of positive, negative, or no change between the 1992 map and 2010 map.

Limitations in this project included slight disagreement in area due to raster clipping/extent issues, lack of radiometric processing such as changing DN values to account for atmospheric interference, coarse resolution made land cover identification difficult, confused classes affecting land cover type area values, and finally errors in the 1992 classification map and changes in classification legend causing increased inaccuracies in area comparison and change calculations.

Further project steps should include adding ancillary data such as IKONOS 1 m resolution imagery or other high resolution imagery to assist with land cover identification, ground truth accuracy assessment, and further refinement of confused clusters/groups from unsupervised classification.

Land cover map comparing the 1992 NLCD land cover to a newly created 2010 land cover map. 

Map of overall geographic change in land cover type from 1992 to 2010 land cover maps. Blue areas are decreased change, red are increased, and gray is no change.