From the regionality viewpoint

this paper discussed the methods of built-up landcover change monitoring by remote sensing in large-scale regions. The methods of extracting built-up land-cover change information were generally discussed. The benefits and accuracy of 3 classification methods of MLC

frequency-based contextual classification with gray-level vector reduction

and ANN were compared in the test classification. The tests proved that the highest accuracy of classification method was the ANN

the kappa index was 0.6625. However

it need a long time to train the neural net

and the classification results didn’t have a significant improve than the other two methods. Frequency-based contextual classification considered the neighborhood pixels information

but the classification result depended on the window size greatly

it’s not easy to select a good window size for every kinds of land-cover. According to the test results

we selected the MLC to classify the TM imageries of the Pearl River Delta. With the expert knowledge

the computer classification results were improved based on GIS. The final results showed that the built-up land-cover of urban area changed greatly during 1988-1998 The buili-up land-cover area was only 556.08 km 2 in 1988

and increased to 1497.61 km 2 in 1998. Both core-oriented development and road-oriented development of the built-up land-cover were significant in the Pearl River Delta. During the study period

66.87% of the new built-up land-cover was in the 1 kilo-meter buffer of the main roads. But the general spatial structure kept relatively stable

83.46% of the built-up land-cover in 1988 was in the inner ring

and it was 84.18% in 1998. Finally

the results show that the built-up land-cover change destroyed the high quality soils greatly.