3. Results and discussion
3.1. Relationship between urban heat island and land use/
cover changes at the regional level
3.1.1. Land use/cover patterns in the PRD region
The threshold approach in combination with the Boolean
operators based on band DNs and indices worked very well in
identifying land use/cover types. In comparison with traditional
image classification methods, such as automated supervised and
unsupervised classifications, our method achieved relatively
high accuracy.Table 2represents the error matrix and accuracy
assessment for the classification of image on Nov. 1, 2000. The
sampling points were randomly taken across the study area, also
relatively uniformly distributed among these land use/cover
types. The overall accuracy of classification is approximately
92%. Among selected 7 land use/cover types, the classification
performed best for bare land, water, and fish pond, but less so
for forest and built-up area.