Jihn-Fa JAN (Taiwan)
Associate Professor, Department of Land Economics National Chengchi University
64, Sec. 2, ZhiNan Road, Taipei 116, Taiwan
Telephone: +886-2-29393091 ext. 51647 Facsimile: +886-2-29390251 E-mail: jfjan@nccu.edu.tw
Yu-Ching HSU (Taiwan)
Master Student, Department of Land Economics National Chengchi University
64, Sec. 2, ZhiNan Road, Taipei 116, Taiwan
Telephone: +886-2-29393091 ext. 50621 Facsimile: +886-2-29390251 E-mail: 97257022@nccu.edu.tw
KEY WORDS: remote sensing, coastal zone monitoring, land use changes, landscape ecology
ABSTRACT: Coastal sand dunes are dynamic and fragile terrain often regarded as environmentally sensitive areas. The coastal zone environment is easily affected by many natural and man-made factors, especially improper land use that result in rapid disappearance of coastal sand dunes. In this paper, by using three Formosat-2 images acquired between 2005 and 2008 and supervised classification methods, the change detection analysis of sand dunes and land use show temporal and spatial variability in the costal zone of I-Lan Plain. Based on the result of image classification, the land use/land cover changes and spatial variability were analyzed by using Fragstats, a public domain software for spatial pattern analysis. The
objectives of this research include: (1) analyzing spatial variability of land-use and sand dunes area by landscape metrics; (2) testing differences of study area between two periods by Shannon t-test ; (3) building a Markov model to predict land-use/land-cover area for the next 15 years.
The result of this study showed that water and built-up areas increases between 2005 and 2008.
According to the comparisons of landscape indices and Shannon t-test results, the landscape structure of the study area during 2005 and 2008 had significant changes. Based on the
simulation of the land-cover changes between 2005 and 2020, water body and built-up areas will increase, and the other land-use types (sandy area/cropland/forest) areas will decrease.
1. INTRODUCTIONS
Land use/ land cover is regarded as one of important factors that affect global environment changes. The land-use types result in changes to the land cover, and the difference of land-use types are due to human activities and environmental effects, moreover, under the different temporal and spatial circumstances (Turner., 1995; Wu, 2006). Therefore, it is necessary to realize land-use situations in a regional area before researching what factors influence the regional environments. The direct way to describe the changes of land-use types is calculating total area and comparing the area of each kind of land-use areas. Several landscape models have been developed to describe, explain, or predict landscape dynamics. For landscape studies, indices are often used to describe or assess the structural condition of a landscape, by measuring spatial configurations of patches on a landscape such as their density, size, shape, edge, diversity, interspersion, and juxtaposition (McGarigal and Marks, 1995). Indices can be used to compare
approach applied to different categories of study, especially in land use/ land cover change research.
For example, Muller and Middleton (1994) compared 3 types of land-use patterns for 5 periods, including crop land, forest and urban land, and analyzed changes of each land-use forms in different time series by a Markov model. The change in land-use patterns are summarized by a series of transition probabilities from 1 state to another over a specified period. These probabilities can be subsequently used to project the landscape properties at alternate future time points (Burham, 1973).
The objective of this study was to integrate statistical methods of landscape indices and Markov
model to assess the change of land-use forms in landscape both temporally and spatially.
2. MATERIALS AND METHODS
2.1 Study site
The study area selected for the empirical analysis was part of I-Lan plain and costal zone located in north-eastern Taiwan (Fig.1). The study area is about 7571ha.
Affected by man-made factors, climate and topography conditions, the main land-use patterns of I-Lan plain are crop lands and aqualculture, and there are coastal forest and sand dunes in coastal area.
2.2 Data
Data used for the empirical analysis included Formosat-2 satellite images of the study area collected on 2005.8.26 and 2008.8.16 at about 10AM. The images are level 4 orthophotos with TWD97 coordinate system, and atmospheric and spectral correction were done by the provider. In addition, vector data of land use, streams, roads, and administrative boundaries were used in the study.
2.3 Methods
2.3.1 Land-use/Land-cover classification using supervised classification method
Surpervised classification approach with maximum likelihood algorithm was used to classify the study area into five land use categories, which included water body, built-up area, cropland, sandy area, and forest land. The training areas and validation areas were selected by comparing the images with land use vector maps.
2.3.2 Quantification of the Landscape indices
Based on the classification results, several indices of landscape structure were computed by using Fragstats. Four landscape indices were used to assess the temporal and spatial changes of patterns in the study area. Number of patches (NP) equals the number of patches in the landscape, as the area is held constant, it can convey patches density and mean patch size of the area.
Landscape Shape Index (LSI) equals the total length of edge in the landscape divided by the minimum total length of edge possible, and it provides a standardized measure of total edge or edge density that adjusts for the size of the landscape. LSI can be interpreted as a measure of patch aggregation or disaggregation, as LSI increases, the patches become increasingly disaggregated. To measure the shape effects, the shape index (SHAPE) was calculated to monitor the increasing of shape irregularity and shape complexity. The shape index was denoted as:
a
SHAPE 0.25p (1) Fig.1. Location of study area
where p is the perimeter of a patch, and a is the size of that patch. The shape index equals 1 if a patch is circular or square, and increases as the patch shape becomes more irregular. Larger patches play a dominant role in the landscape function relative to the phenomenon, so the area-weighted shape index may be more appropriate than unweighted mean shape index. The area-weighted shape index (SHAPE_AM) is computed by weighted patches according to their sizes. Shannon’s diversity index (SHDI) is a popular measure of diversity in community ecology, which was measured by: where Pi is proportion of the landscape occupied by patch type i. SHDI equals 0 when the landscape contains only 1 patch and SHDI increases as the number of different patch types increases and/or the proportional distribution of area among patch types becomes more equitable (McGarigal et al., 2002).
2.3.3 Evaluation of the change on landscape patterns by Shannon t-test
Magurran (1998) proposed a Shannon’s diversity index t-test method to test whether the land-cover patterns change significantly. The following algorithms are used for t-test:
land-cover categories in period i, and n is number of patches in period i. i2.3.4 Projection of the land-cover changes
It was assumed that the land cover changes of the study area could be depicted as a Markov process. A transition matrix, in which the element Tij represents the amount of land cover change from cover type i to cover type j during 2005 and 2008, were derived from the land cover maps. The transition probability P , which represents the fractions of land cover ij changes on each land cover type, was estimated by:
where m is the number of land cover types, which is equal to 5 in this case.To determine whether it is appropriate to apply the Markov model to the observed land cover changes, Goodman’s Chi-squared statistic (Goodman 1968) was used to test the null hypothesis that the land cover conditions in 2005 and 2008 were independent of each other:
where the definition of T and ij P are the same as in equation (7), and ij A j denotes the fraction of land cover in each of the 5 land cover types in 2008.
Assuming that the transition probabilities will remain constant in the future, the Markov model was then used to project the land cover at the next stage:
2
1 t
t P n
n (8) where P is the transition probability matrix, and nt1 and nt2are column vectors denoting the fractions of land cover types a t1and t2.
2.3.5 Analysis of land-cover area with spatial and temporal series
The land-cover maps were derived from 2005 and 2008 Formosat-2 images by supervised classification methods, using two statistical approaches respectively. One is quantifying the landscape indices and comparing indices to observe the land-cover type changes in landscape between two different dates; the other is to establish a Markov model based on statistics of each land-cover area between 2005 and 2008, 3 years for a period, to predict 5 states for the next 15 years of each land-cover types area, from 2008, 2011, 2013, 2015, and 2020.
3. RESULTS
3.1 Land-cover maps
The images were classified to 5 land-cover types including water body, built-up area, cropland, sandy area and forest land. Water body consists of river, aqualculture; built-up areas represent road, buildings, grave land and barren land; and cropland includes lands either harvested or not. Despite the overall accuracy of over 83%, it was difficult to distinguish between the spectral feature of barren land in built-up area and the harvest cropland. The results of land-cover maps are regarded as acceptable and as shown in Fig.2.
Table 1 shows the percentage of various land-cover types in the images of two different years.
In 2005, based on the area size of each land-cover type, the largest land-cover type is cropland (29.31%), followed by built-up area (26.68%), forest land (21.85%), water body (12.19%) and sandy area (10.05%). In 2008, the largest land-cover type is built-up area (32.38%), then cropland (28.2%), forest land (19.47%), water body (15.13%) and sandy area (4.92%).
Compareing the land-cover areas of two time periods, sandy area and forest land area had decreased, especially sandy area decreased about 5%. Many researches had proved that I-Lan coastal sand dunes area had a trend of reduction (Hsu and Chang, 2002), and it may be the reason why water area increased in 2008. Built-up area increased about 6% in 2008, and it is believed that the construction of freeway connecting Taipei and I-lan is the major factor.
Table.1. The percentage of each Land-cover type in 2005 and 2008
Land-cover Types
Year
Water body Built-up area Sandy area Cropland Forest land Total 2005 12.19% 26.68% 10.05% 29.23% 21.85% 100.00%
Fig.2. Land-cover maps in 2005 and 2008
2008 15.13% 32.28% 4.92% 28.20% 19.47% 100.00%
Difference 2.94% 5.59% -5.13% -1.03% -2.38%
3.2 Evaluation of the changes by landscape indices and Shannon t-test
Comparing the indices of two different years, all indices showed minimal changes and the patches tend to be less fragmented. As for the shape of patches, more irregularity and variation of shapes were observed. The SHDI value in 2005 was larger than that of 2008, which indicated that the proportional of area among patch types becomes more even. In addition, the t-test on SHDI of two periods were significant. Under 5% significant level, the computed t-value is 12.73, which is larger than the theoretical t-value 1.96. It indicates that the land-cover patterns of landscape had very significant changes. Based on the results of indices and Shannon t-test, the land-cover structure of the study area had changes but not obvious between 2005 and 2008.
Table.2. The landscape indices in 2005 and 2008 Landscape Indices
Year
NP LSI SHAPE_AM SHAPE_CV SHDI 2005 82568 123.3355 10.256 71.5982 1.321 2008 65901 106.7401 11.3608 74.7876 1.2935 3.3 Prediction of land-cover area in the future
Using the transition probability metrics shown in table.3 to compute chi-squared value to determine land-cover area is appropriate for Markov model. With degree of freedom equals 16 and 1% level of significance, the α2-value equals 220870049086 which is larger than
theoretical value (α2=32.00). It indicates that the change of land-use types conforms to Markov model between 2005 and 2008. To assess the accuracy of the prediction of Markov model, by using the percentage of land-cover types derived from the 2008 image, the difference between predicted values and observed values for all five land-cover types are all within 5%. Therefore, it is possible to predict land-cover area by using Markov model. The results of accuracy
assessment for model and area predictions are all shown in table.4.
Table.3. The transition probability of each land-cover types between 2005 and 2008
Transition Probability 2008
2005 Water body Built-up area Sandy area Cropland Forest land Total Water land 0.737 0.136 0.066 0.053 0.008 1 Built-up land 0.082 0.547 0.039 0.29 0.043 1 Sandy area 0.188 0.385 0.241 0.174 0.012 1 Crop land 0.069 0.348 0.021 0.486 0.076 1 Forest land 0.007 0.089 0.002 0.176 0.727 1 Predicting 5 states area of land-cover type with a period in 3 years and the initial data used for 2005. The results of land-cover area in the future are shown in table.4.
Table.4. Predictions of land-cover by Markov model
Results of image classification Results of Markov model predict Land-cover Types
Initial Year Year
2005 2008 2008 2011 2014 2017 2020 Water land 12.05% 15.13% 15.09% 16.76% 17.89% 18.71% 19.32%
Built-up land 26.72% 32.28% 32.26% 33.14% 33.36% 33.46% 33.53%
Sandy area 10.03% 4.92% 4.89% 4.05% 3.98% 4.03% 4.09%
Crop land 29.29% 28.20% 28.25% 28.16% 28.03% 27.88% 27.75%
Forest land 21.90% 19.47% 19.51% 17.89% 16.74% 15.91% 15.31%
Total 100% 100% 100% 100% 100% 100% 100%
4. CONCLUSIONS
The results of this study show that land cover types between 2005 and 2008 changed spatially.
Comparing the areas of different land-cover types between 2005 and 2008, water body and built-up areas raised in 2008; cropland, forest land and sandy area decreased, especially sandy area had most changes. The indices of landscape level show that the land-cover patterns in 2005 are more fragmented than in 2008. As for the shape index, the area-weighted shape index and coefficient of variation in shape index show that the shape of patches in 2008 are more irregular and more variable than in 2005. The number of land-cover types of two periods were the same, and the SHDI values in 2008 were more than in 2005; it shows that the proportional distribution of area among patch types becomes more even. According to the results of the comparison on the values of indices and the Shannon’s diversity t-test, the land-cover spatial structure in I-lan plain had changed between 2005 and 2008 but the change is not significant. Though the main
land-cover type was crop land in 2005, the prediction of Markov modelshows that built-up land area will increase and the other land-cover types will decrease.
5. LITERATURE CITED
Burnham, B. O. 1973. Markov intertemporal land use simulation model. Southern J. of Agri. Econ. 5:
253-258.
Hsu MY and Chang CL, 2002. A study of coastal retreat in I-lan plain. Bullet of the Geographical Society of China 30: pp.57~76.
Hulshoff, R.M. 1995. Landscape indices describing a Dutch landscape, Landscape Ecology 2:45-61.
Magurran AE. 1988. Ecological diversity and its measurement. Princeton (NJ): Princeton Univ Press.179 p.
McGarigal, K. & Marks, B. J., 1994. FRAGSTATS: Spatial Pattern Analysis Program for Quantifying Landscape Structure (Version 2.0). Forest Science Deartment, Oregon State University, Corvaills.
McGarigal, K., Cushman, S. A., Neel, M. C. & Ene, E., 2002. FRAGSTATS: Spatial Pattern Analysis Program for Categorical Maps. Computer software program produced by the authors at University of Massachusetts, Amherst. Available at he following web site:
http://www.umass.edu/landeco/research/fragstats/fragstats.html
Muller, M. R., and Middleton, J. 1994. A Markov of Land-use Change Dymanics in Niagara Region, Ontario, Canada. Landscape Ecology 9:151-157.
Wu CF . 2006. Model constructs of land use change and landscape ecology assessment. The Graduate Institute of Urban Planning, National Taipei University, Taiwan.