GISdevelopment.net ---> AARS ---> ACRS 1990 ---> Land Cover/Land Use

Landcover change detection using digital analysis of remotely sensed satellite data: A methodological study

Charat Mongkolsawat, Phanee thirangoon
Khin Kaon University, khon Kaen 4002 Thailand


Abstract
The objectives of this study were to develop a land cover change detection methodology through digital analysis of remote sensing satellite data. Two geocoded TM subscences of yasothon province northwest Thailand acquired during the wet season and the dry season were selected for analysis a supervised classification based on land cover and terrain was performed on each scène which yielded a September image with 11 classes and an April image with 14 classes. The two classified images were mathematically combined resulting in a unique linear combination of all 154 possible grey level image through a process of regrouping these grey levels in to classes the resultant image province meaningful land cover dynamics with respect to terrain type.

Introduction
Satellite remote sensing is widely accepted as a technique to study land use land cover change Dynamics .The use of satellite data for compiling land use change area is becoming substitute for data derived from time consuming aerial photo interpretation .In Thailand land use change is rapidly increased encroachment on forest reserve for agriculture is at alarming rates with significant impacts on soil erosion soil Stalinization rural communities and climate patterns. Better assessment of land use land cover change using digital analysis of remotely sensed satellite data can help decision maker to develop effective plans for the management of land.

Two main approaches to digital change detection have been reported. Both involve autitemporal images and can be categorized as single data set or separate data set analysis, separate data set analysis involves classification of each-date imagery separately into landcover classes. The results were subsequently compared 9gordon, 1980; willighton et al, 1986). Single data set analysis involves coregistering and resembling multitemporal images into the same dataset matamatical transformation, aminly image differencing and /or rationing, is then applied to the raw coregistered images to produce a residual image indicating the relative change in reflctance between the two dates. This technique is reported to give slightly more accurate result . (see , for example, Nelson, 1982; jenson and toll, 1982 : woodwell et al., 1983; Singh, 1986; Quarmby et al., 1987 ) Image differencing and image rationing have been used to detect abrupt changes in canopy reflectance from forest harvest and road construction (woodwell0 et al., 1983; tucker et al., 1984: Pelletier and Sader, 1985, ) as well as urban encroachment of agricultural land (Quarmby et al., 1987).

West of the digital change detection reported are two dates investigation. Sader (1988) ufllized Landsat imagery of three dates, supplemented by aerial photography, to study forest change dynamics in a troprical area The technique analyses the three date imageries as a simple data set and nvolves computation of normalized difference vegetation index of each data as data compression procedure. Modified parallelepiped classifier was employed to generate a multitemporal greenness image. Thus this technique is suitable for detection of forest change.

The objective of this study was to develop a landcover/landuse change detection methodology through digital analysis of satellite data.

Study Area
To study change detection methodology, substances of yasothon, Northeast Thailand, corresponding to central part of landsat tM path/row 127 /49 was selected.

Geologically, the area is underlain by a thick sequence of Mesozoic rock of the Maha sarakam formation gently undulating terrain with sparse trees and isolated patches of forest remants is commonly found. The loamy paleaquults and loamy paleustults are found extensively in lower terraces and upper terrace respectively, clayey Tropaquepts are present narrow strips in the flood plain. The soils are mainly derived from alluvium of sandstone origin. Aeolian deposits are found sporadically.

Methodology
Two Landsat Thematic Mapper scenes of diffrent dates, Septamber 1988 and April 1989, were acquired for the study site representing we and dry season respctively. Both were cloud-free for the substances selected. The two sets of Landsets TH data were coregistered to a 15-meter UTH projection resulting in a single date set.

A supervised classification was performed on the meridian system for the subscence of each date. Field survey was undertaken to study vegetation cover, terrain and other related information. This procedure allowed identifying the training samples to be selected. The training samples selected were based on landcover and terrain. Signature set components of bands 92,3,4,5 and 7) obtained from training sample were statistically tested to measure discrimination between classes. When satisfactory discrimination was obtained, classification was performed to assign the pixels into the classes having nearest mean vector. piele falling outside data boundary as defined by standard deviation were classified as unknown or full. Classification results give 11 classes for the September 1988 image and 14 classes for the April 1989 image. Thus, the digital number of pixels for the September 1988 classified image ranges from 1 to 11 and 1 to 14 for those of the April 1989 classified image.

The two classified images were digitally combined into one image. this is accomplished using a formula I1 x (ncls2) + I2 , where I1 and I2 represent the digital numbers of the September and April images respectively and nc1s2 is the number of classes in the April image. The matrix in figure 1 reveals how the formula produces a different digital number for each possible combination of classes. The result is a classified image created by a process that provides a unique method of referring back to the original images.

April Image classes (1-14)
September
Image Classe
(1-11)		   1 2 3 4 5 6 7 8 9 10 11 12 13 14
1 15 16 17 18 19 20 21 22 23 24 25 26 27 28
2 29 30 31 32 33 34 35 36 37 38 39 40 41 42
3 43 44 45 46 47 48 49 50 51 52 53 54 55 56
4 57 58 59 60 61 62 63 64 65 66 67 68 69 70
5 71 72 73 74 75 76 77 78 79 80 81 82 83 84
6 85 86 87 88 89 90 91 92 93 94 95 96 97 98
7 99 100 101 102 103 104 105 106 107 108 109 110 111 112
8 113 114 115 116 117 118 119 120 121 122 123 124 125 126
9 127 128 129 130 131 132 133 134 135 136 137 138 139 140
10 141 142 143 144 145 146 147 148 149 150 151 152 153 154
11 155 156 157 158 159 160 161 162 163 164 165 166 167 168

Figure.1. matrix of the possible range of numbers resulting from the dignitally combined September 1988 and April 1989 images

The large range of digtal numbrs in this image are than regrouped into a smamm number of more meaningful classifications

Results and Discussion
The change detection procedure used has involved a combination of classified images derived for each date. This approach as described above, was performed on an initian supervised classification of Land sat TM data from 2 dates and subsequently mathematically combined the two classified images. Hence the output image was greatly dependent upon the accuracy of the classified images field survey together with the knowledge of relationships among landform, soil and land use etailed the regrouping of the classes with a meaningful legend. The change image obtained from this analysis offered the landuse /landcover types for 2 season with respect to terrain. it should be noted that difficulties in distribution of riparian from other types of vegetation and floodplain from lower terrace found when taking into account only radiometric values. These problems can be solved by further analysis of the change image. figure 2 shows the false colour composite of the subscene for 2 date images contrasts in wetness and vegetation covers between 2 images could be discernible


Fig.2. False colour composite of the 2 substances.

Table.1 Area occupied by each Class of the change image (September 1988 and April 1989)
Glase landuse land cove Terrain Area (km)
1 -Null - 0.1056
2 Overlap - -
3 Water in April and September - 127043
4. Water in September, no water in April - 16,2781
5. water in April, no water in September - 4,4347
6 Rice cultvation in September
and rice stubble in April		 Lower terrace 405,3868
7 Moderate to dense ground cover
of vegetation in April and September		 Lower terrace 8,7387
8 Slight ground cover of vegetation
in April and September		 Lower terrace 0.045
9 Slight ground cover of vegetation
in April and September		 upper terrace 51.56
10 Moderate to dense ground cover of
vegetation in April and September		 Upper terrace 248.5287
11 Moderate to dense ground cover of
vegetation (tree +annual field crops) in April
and slight vegetation in September		 Upper terrace 248.5287
12 Moderate to dense ground cover of
vegetation September and slight
vegetation in April		 Upper terrace 125,695
Total     873,63127

The most extensive area of the September substances was covered by rice at the vegetative stage together with statement water. The field crops, forest remnants, shrub and woody weeds occupied the remaining area (upper terraces) was recognized. Table 1 summarizes the area occupied by each class of the change image (April and September). It was not the intention of this study to monitor the chgange of specific landuse type. The change of landcover/landuse of the entire study area was the main interest and concern . Figure 3 fillustrates the change image in the study area.


Fig. 3 Yasothon Landcover change

To assess the accuracy of the change it was compared with the existing maps in combination with field surveys. A randow check of the classes in the change image was done for both seasons and satisfactory results were obtained. In this regard, it should be observed that in term of landuse/ lancove change monitoring, the change image offered an advantage in identifying the vegetation density. Indentification of upper and lower tesraces was inferred from the soil soisture condition of the September image. The April suscene image had a relatively narrow range of reflectance in the identifcation of terrain: the upper terrace could not be discriminated from the lower terrace by using the dry season image.

Consideration of the radiometric value could only allow for the recoginition of a very limted type of terrain. The parameters used in the classified image should inlude, not only landcover and soil moisture conditions, out other sources of information as well . However this approach was based on the use of purely digital radiometric values.

Conclusion
There are a number of methods for change detection. However, many of these can only identify change and do not distguish in landuse and terrain type, or they specify only the change clause of interest. This method covers all kinds of landus/land cover change in the study area.

This method can be applied with more than 2 images. This will be very useful for monitoring the landuse dynamics different seasons.

Detailed study in the identification of terrain of the same reflectance is needed in order to improve map significance. Another source of information is needed

Improvement to this method should be made possible by mathematical condinations or by correcting the radiometric values of raw data rather than using the classified images. So far due to differences in atmosoheric conditions and sun angles for different seasons, the use of raw data for generating the change image was relatively difficult.

References
  1. Gordon, s., 1980 Utilizing Landseat imagery to monitor land use change : A case study in ohio. Remote sensing of Environment , 9, 189-196

  2. Jenson, J.R. and Toll, D.R., 1982. Detecting residential Land use development at the Urban fringe, Photogrammetric Engineering and Remote Sensing, 48, 629-643

  3. Millington, A.C., Jones, A.R., Quarmby , N.A. and Townsend , J.R.G. , 1986 monitoring geomorphological process in desetmarginal environments using multieporal sattelite imagery proceedings of the International Symposium on Remote sensing for Resource Developemt and Environmental management , A.A.Balkema.

  4. Nelson , R.E 1982 Detecting forest conopy change using Landsat. AgRISTARS Report TH 83918 , NASA /Gooddard Space Flight centr Greenbelt , MD 80 pp

  5. Pelletier , R.E. sader S.A. 1985 . The relationship between soils data and forest clearing and forest regrowth frends in costa Rica Proceedings of pecora 10, August 20-22 , 1985 colorado state University , Colorado pp276-285

  6. Quarmby, A.A. Townshend, J.R.G. and Cushine j.L. 1987 Monitoring urban land cover changes at the urban fringe from SPOT HRV imagery in South-east England SPOTI

    image utilization Assessments results, Centre National D" etudes Spatituble

  7. Sader S.A. 1988 satelite digital classfication of forest change using three landsat data sets . Remote sensing for resources Inventory , planning and Honitoring procedings of the Second Forest service remote sensing application conference-slidell louisiana and NSTL Mississippi April 1 15, 1988

  8. Singh A., 1986 change detection in the tropical environment of northestern India using Landsat remote sensing and tropical Land Management, edited by Eden H.J. Parry J.T. John wiley and sons, 36,365 pp

  9. Tucker C.J.Holben B.N. Goff T.E. 1984 Intensive forest clearing in Rondonia , Brazil as defected by satelite Remote ssensing , Remote sensing of Environment 15:255-261

  10. Woodwell G.H. Hobbie, J.E. Houghton, R.A. , Mellio J.H. peterson .B.J. Shaver G.R. stone , T.A., Moore B., Park A.B. 1983 deforestation measured by landsat : Steps toward a method DOE/EV 10468 NTTS , spring field Va 62 pp