Monitoring urban land cover change: An expert system approach to land cover classification of semiarid to arid urban centers

The spatial and temporal distribution of land cover is a fundamental dataset for urban ecological research. An expert (or hypothesis testing) system has been used with Landsat Thematic Mapper (TM) data to derive a land cover classification for the semiarid Phoenix metropolitan portion of the Central Arizona-Phoenix Long Term Ecological Research (CAP LTER) site. Expert systems allow for the integration of remotely sensed data with other sources of georeferenced information such as land use data, spatial texture, and digital elevation models (DEMs) to obtain greater classification accuracy. Logical decision roles are used with the various datasets to assign class values to each pixel. TM reflectance data acquired in 1998 [visible to shortwave infrared (VSWIR) bands plus a vegetation index] were initially classified for land cover using a maximum likelihood decision role. In addition, spatial texture of the TM data was calculated. An expert system was constructed to perform postclassification sorting of the initial land cover classification using additional spatial datasets such as texture, land use, water rights, city boundaries, and Native American reservation boundaries. Pixels were reclassified using logical decision roles into 12 classes. The overall accuracy of this technique was 85%. Individual class user's accuracy ranged from 73% to 99%, with the exception of the commercial/industrial materials class. This class performed poorly (user's accuracy of 49%) due to the similarity of subpixel components with other classes. The results presented here indicate that the expert system approach will be useful both for ongoing CAP LTER research, as well as the planned global Urban Environmental Monitoring (UEM) program of the Advanced Spacebome Thermal Emission and Reflection Radiometer (ASTER) instrument.