A Monitoring Method of Land Cover/Land Use Change in Naiman,
Inner Mongolia Autonomous Region, China Using Landsat Data
Extraction methods by land use classification
When we extract desertified areas using Landsat data, the most precise method is to define the desertified areas and land use types based on existing maps and field surveys. Therefore, we examined the land use classification in a test area in Naiman using land use units of the desertification map of Naiman Qi. These units contained active dunes, grassland, cropland, forest, wetland, villages and water bodies.
First, in the test area we set up about 5 typical points of each unit where changes had not been recorded since the publication of the map. Second, we checked the reflectance of every point using Landsat data observed at 5 different times from May to September 1992. Fig 3 shows an example of the results. Some characteristics are summarized as follows. Active dune showed the highest reflectance compared with other units through bands and time. In contrast, water bodies showed the lowest reflectance on both 4th and 5th bands, which belong to the near infrared region, through bands and time. Therefore, active dune and water bodies can be easily distinguished from other units using only one datum for the above reasons.
Fig. 3 Seasonal changes in spectral reflectance of each land cover/land use type
We could separate the other units using 2 data of late May and late July, since the vegetation of cropland and grassland was poor and the forest area had already been covered with green leaves in late May.
Moreover, the density of vegetation of cropland was higher than that of grassland in late July. These phenomena indicate the difficulty in classification into cropland, forest and grassland by using only one image. However this problem could be solved by using 2 different images observed in spring and summer, because in late May cropland and grassland are almost bare before seeding and trees are covered with fresh green, while in late July and August cropland and grassland are covered with green and cropland has a higher density then grassland. These changes reflect the value of the red band which is 3rd and in TM and displays a high reflectance against the photosynthetic pigment chlorophyll.
In this study, we used fine band data of Landsat TM for classifying land use as shown below; namely the 3rd band on May 20 and 1st, 3rd, 4th and 5th bands on July 23. We selected the locations where we checked the reflection of each land use type as training areas and classified these types with supervised classification (Fig. 4). As a result there was a high probability of accuracy in the classification (Table 1). Each unit was discriminated from the other units with more than 80% of probability. Especially, cropland and grassland were discriminated with 95% of probability. As mentioned above, we confirmed that this method was useful for land use classification. If each year we classify the land use in the same area and compare the results, we will be able to obtain desertification trends along with data related to reclamation, amount of cropland, clearing and plantation of trees.
Fig. 4 Land cover/land use map classified by Landsat TM
