|
Results and Discussion
Land use analysis: Land use analysis using Gaussian Maximum Likelihood Classifier was done for multi-resolution data (MODIS, Landsat, IRS P6 and Ikonos) and the results are presented in Table 2 and figure 4. Overall accuracy of the classification was 88% using Landsat data, 91% accuracy using IRS-P6 data and 74% using Modis data respectively.
Table 3.a. Temporal Land use dynamics in Hectares
| Class |
Urban |
Vegetation |
Water |
Others |
| Year |
Ha |
Ha |
Ha |
Ha |
| 1973 |
5448 |
46639 |
2324 |
13903 |
| 1992 |
18650 |
31579 |
1790 |
16303 |
| 1999 |
24163 |
31272 |
1542 |
11346 |
| 2002 |
25782 |
26453 |
1263 |
14825 |
| 2006 |
29535 |
19696 |
1073 |
18017 |
| 2010 |
37266 |
16031 |
617 |
14565 |
Figure 2: Land use statistics a). Ikonos 4m, b).IRS-P6 5 m, c). Landsat-30m, d). Modis-500m.
Table 3.b. Temporal Land use dynamics in %
| Class |
Urban
% |
Vegetation
% |
Water
% |
Others
% |
| Year |
| 1973 |
7.97 |
68.27 |
3.40 |
20.35 |
| 1992 |
27.30 |
46.22 |
2.60 |
23.86 |
| 1999 |
35.37 |
45.77 |
2.26 |
16.61 |
| 2002 |
37.75 |
38.72 |
1.84 |
21.69 |
| 2006 |
43.23 |
28.83 |
1.57 |
26.37 |
| 2010 |
54.42 |
23.41 |
0.90 |
21.27 |
Landscape Metrics: Landscape metrics were computed for varied resolution of data for sample space in Greater Bangalore. The data was classified into 4 land use categories in a heterogeneous landscape, Urban category was considered for further analysis as the landscape is rapidly urbanizing and constitute a dominant class. Table 3 lists the quantified values of each metrics across resolutions of multi-resolution data.
Percentage of Landscape (PLAND) and Number of Patches (NP) as tabulated in table 3, Indicates the level of fragmentation. The results highlight the dependence on spatial resolutions evident from the refinement of values with finer spatial resolutions. PLAND indicates that the urban patches in this region is becoming a single patch. Figure 3a highlight the correlation of PLAND with the resolutions (r = 0.97). Number of Patches (NP) indicates that smaller patches aggregating to form a cluster of the urban surface. Figure 3b indicates that better spatial resolution reveals large number of smaller patches and as the resolution becomes finer the number of patch metrics becomes precise.
Largest patch index (LPI) indicate that the landscape is in the process of aggregation to a single patch indicating homogenisation of landscape. This metrics is not dependent on resolutions as in quantifies the largest patch and almost accurately in all resolutions as illustrated in figure 3c.
The Patch density (PD) indicates the densification of a particular patch. Figure 3d indicates of improved performance with finer resolution. This was verified with the ground truth data and validation of the classified land use data with spatial metrics along with the resolutions of the data.
Fractal dimension index (FRAC) indicates complexity of the shape, while FRAC_MN and FRAC_AM which indicates complexity of shape around the mean and with respect to area weighted mean (AM) which has very high values indicating complex geometry. Moderate and high resolution images were able to quantify these accurately (Figure 3e).
Clumpiness index (Clumpy), Aggregation index (AI), Interspersion and Juxtaposition Index (IJI) highlights the occurrence of same patch in the neighborhood. Clumpiness and aggregation indexes mainly highlight the nature of development of a particular class in the neighborhood. Clumpiness value of 1 indicates that the particular class is highly clumped in that region. Aggregation value close to 100 indicates the same. If the value of IJI is not obtained it means to say that the patch types distinctly pound is less than three. All resolution output for all these metrics indicates that higher or better resolution is necessary to obtain appropriate result. Figure 3f, 3g and 3h corresponding to these spatial metrics indicate of improved results with the improvements in the spatial resolution.
Figure 3: Correlation of spatial Matrices with resolutions of the data
Cohesion and Connect metrics measures the physical connectedness of the patches. It increases with increase in aggregation of among the patch type. Cohesion value of 100 indicates the clumpiness or connectedness of the patch values close to 0 indicates highly unconnected fragmented landscape. Earlier as indicated the aggregation level in the considered image is quite high hence the cohesion values should be on its higher side. Connect value of 0 indicates that the considered area is becoming a single patch and the values close to 100 indicates that every patch is highly connected and there are small fragmented patches. Figure 3i and 3j indicate that these metrics are independent of resolutions used and gives almost similar results.
Percentage of Like Adjacencies (PLADJ) calculated for the adjacency matrix indicates the frequency of different pairs of patch types occurring, measuring the degree of aggregation of the focal patch type. The values close to 0 indicates maximally dispersed pattern and values close to 100 indicates maximally contagious. Figure 3k highlight of dependence on spatial resolutions as lower resolution images fails to give an appropriate result.
Normalized Landscape Shape Index (NLSI) indicates the shape of the landscape. Values close to 0 indicates that the landscape under study has simple shape means to say it is further aggregating to become a single patch. Values close to 1 indicates that the landscape has a complex shape. Figure 3l highlight that the regions are becoming a single patch of the simple size and independent of resolutions.
|
