In order to allocate affordable housing, COAH needed the amount of vacant and developed land for each municipality. CRSSA decided to use satellite data classfication to compute vacant and developed land. Three LANDSAT TM scenes were stitched together, and a Level I land use/land cover map was generated (see above image).
One problem with the satellite-derived information was that residential land was greatly underestimated. This is due in part to the mixed spectral reflectance of residential areas. Residential areas, which are a land use, contain a mix of land cover types, such as impervious surfaces, trees, and grass. The satellite data classfication algorithm often divides residential land into these landcover types.
To alleviate this problem, CRSSA combined housing unit density information derived from the US Census Bureau's TIGER line files and PL94-171 redistricting data with the satellite data to improve the detection of residential land. An example of this is shown below.
The above images are for Essex Fells, a residential community in northern New Jersey. It is a small municipality, and consists mainly of residential homes with significant tree cover. Each image shows a land use/land cover map, where red = developed, dark green = forest, light green = nonforest vegetaion and blue = water.
The images show the increase in the amount of developed land when a decreasing housing unit density threshold is combined with the satellite data. Each image is as follows:
Amounts of vacant and developed land were then computed for each municipality, and tables were then given to COAH to compute affordable housing amounts.
There is a report to the state of NJ somewhere, and Jimmy G did his Master's thesis on this project. Contact Jimmy G or Teuvo Airola for more info.