This paper aims to identify land cover changes of the western  part of the Horqin steppe by means of Landsat MSS satellite  data. The Horqin steppe is located in the temperate semi-arid  zone of North China, Inner Mongolia Autonomous Region. The  steppe is considered to be one of the most drastic examples of  land degradation in China's semi-arid region.  

Two Landsat MSS scenes, one for the year 1975, the other for  1989, with an overlapping area of around 20000 km2 were used  for the land cover change detection. After performing geometric  and radiometric corrections, an unsupervised cluster  classification revealed six major land cover classes including  water. Then a supervised Maximum Likelihood classification  was applied to both scenes using the results from the  unsupervised classification as well as the Horqin  desertification/land use map (Lanzhou Institute of Desert  Research, 1991) and filed data, which have been collected in  September 1996. 5-10 training areas for each class where  identified for the five land cover classes: (1) Moving sand with  less than 10% vegetation cover, (2) farmland, (3) grassland with  a vegetation cover of 10-60%, (4) grassland with a vegetation  cover of 60-100% and (5) scrub forest. Further a post  classification filtering process was applied to produce more  homogenous areas. Finally by means of visual interpretation and  a digital elevation model obvious misclassifications, such as  overrepresentation of cropland at the expense of scrub forest or  grassland with vegetation cover >60%, were changed.  

A comparison of the 1989 image classification with the  collected field data gave poor results. Most likely this is due to  the time difference between image acquisition and field data but  also the result of low precision when georeferencing the image.  Another source of error could be the actual sampling method,  which has occurred close to river systems, where land use and  land cover may undergo rapid changes.  

Comparing the 1975 and 1989 final classifications three main  types of changes may be observed. First, larger areas of  farmland are found in the 1989 image compared to 1975, mainly  distributed around the Xiliao River system. Second a significant  area of the class `grassland with 10-60% vegetation cover' in  1975 has become converted to the class `grassland 60-100%  vegetation cover' by 1989. Third, sandy surfaces with the lowest  vegetation cover, <10% have also decreased. Therefore this  study does not confirm any overall degradation of the area as far  as concerns vegetation cover. On the contrary, the study implies  a possible improvement of grassland, especially in the loess area  south of the Xilao River. Precipitation data seem not to explain  these changes. However, additional data including monthly  precipitation data should be analyzed.