| Based on the thermal band of Landsat 7 ETM imagery (acquired on October 7,2002), land surface temperature (LST) is retrieved for the central region of Chengdu,China. Combining spatial lattice sampling with landscape pattern indices, the landscapepattern of the whole expressway rounding area in Chengdu is also analyzed. With thesetwo results, relationships between the landscape pattern and the LST are estimatedthrough the correlation analysis at multi-scales. Results could be arranged mainly as fol-lows:(1) The central region of Chengdu has significant heat island effect, urban residenceand industrial land are the mostly hot land-use types at the urban center, while the naturalland-use types show relative low temperature at the rural area. For the mean temperatureof various types at pixel level, water (23.03℃) and industrial land (28.44℃) are the low-est and highest type respectively. Forest (25.41℃) don't exhibit notable different tem-perature comparing with the agricultural types of paddy farmland (25.68℃) and dryfarmland (25.70℃), this indicates that the forest don't have more advantage when slow-ing the temperature raise in this study. The temperature of green space (26.17℃) ishigher than other vegetation types, but lower than the man-made types like urban resi-dence (27.57℃) and industrial land (28.44℃). With close temperature of green space,rural residence exhibites lower temperature than urban residence.(2) NDVI and LST exhibit clearly linear relationship at the patch level, and withhigher NDVI of 0.1, lower LST of 1.23℃would be observed. This indicates the highervegetation cover would introduce lower LST. While the one order coefficients of the re-gressions between NDVI and LST at different types imply that paddy farmland(-14.7287), forest (-12.1667) and dry farmland (-11.1844) have considerablely greatercooling effects than green space (-7.6548).(3) With the percentage of 70％, residence and farmland are the two main composi-tions of the central Chengdu landscape in 2002, and urban residence is the largest type with 31％of the whole landscape. Industrial land has the largest mean patch size (MPS),followed by urban residence and paddy farmland, and other types have relatively lowervalues of MPS. Except road with thelargest patch density (PD), rural residence showsdistinct highter PD than urban residence, and dry farmland, forest, green space and waterall have relatively higher PD than the man-made constructional types like urban resi-dencc and industrial land. It is indicated that this piece-form urban land-use induce thelow level fragmentation at the urban center.(4) In the spatial map of the sampled indices, the circling distributed index valuesindicate the development of Chengdu city follows the concentric zone theory, The spatialpatterns of PD, ED (edge density), MPS (mean patch size), and CONTAG (contagion)are similar with the patterns of urban man-made landscape types such as urban residenceand industrial land, and at these area indices can exhibit distinctive differences than sur-roundings, especially at the position of main road. This could be recognized that the ur-ban expansion should start more often around roads.(5) The fragmentation of the urban center represented by PD, ED and MPS is lowerthan the suburb, and MSI (mean patch shape index) and MFD (mean patch fractal index)which are supposed to represent the shape complexity show irregular trends along theurban-rural gradient, but both for lowest value at the urban center. With relatively lowervalue at the urban-rural ecotone along the gradient, LPI (largest patch index) andDIVISION (division) could measure the special peculiarity with lower fragmentation.SHDI (Shannon's diversity index) exhibits higher diversity at the ecotone, witch is gen-erally located at place of 4～10km distant from the urban center.(6) Within the 1200m sampling scale, LST represent a linear downtrend at the ur-ban-rural gradient. Adding 1 km from the urban center, the LST wilt reduce 0.2℃and theheat island intensity (difference between the min-max values) would be 3.2℃at thisscale. While the NDVI represent a more linear uptrend, with 1km added from the urbancenter, NDVI will raise 0.02. With the up-scaling, heat intensity exhibits a linear down-trend, and the correlation between LST and NDVI augment generally in a power-lawmanner.(7) Except MSI and MFD, all the indices can represent constant trends along theurban-rural gradient, no matter what the scale is. With the different peak positions of non-linear trends, MSI and MFD have different representations along the gradient at dif-ferent scales. Similarly, except MSI and MFD, all the indices have mainly constantcurves of semi-variances at multi-scales, and only the variation rate of the semi-variancecurves will be enlarged with the increase of scale. That is within a same separation dis-tance, the sell-correlation of pattern index will be decreased with up-scaling, and repre-sent more variation between separations. This result implies that with the increase ofscale, variations at small scale will be "fainted" and differences between plots will bemagnified. This would give prominence to the main characteristics of patterns withoutchanging the pattern forms essentially. The semi-variance curves of MSI and MFDaround the scale of 2400m (or within a more broad scale region) indicate that therewould likely be a measurable structural scale exists for these two indices.(8) Correlations between the linearly variated indices and LST (and NDVI) aregreater than the correlations related to the non-linearly variated indices. Various indicesintroduce different correlations, while the inherent area property of indices can have no-table influence to this pattern-process relationship. Among the scales, three differenttrends of correlations between indices and LST (and NDVI) can be identified with theincrease of scale:①PD, ED, AWMFD (area weighted mean patch fractal dimension),MPS, CONTAG, and AI (aggegation index) related correlations increase in a log func-tion;②SHDI related correlations increase more linearly;③AWMSI (area weightedmean patch shape index) related correlations decrease with no clear regulations;④Otherindices related correlations variated with fully irregulations.
|
PDF Full Text Request