緑地が都市内熱環境に及ぼす影響
(2) リモートセンシングによる緑地の抽出と表面温度の解析
1987 年 43 巻 1 号 p. 31-36
詳細
1987 年 43 巻 1 号 p. 31-36
Remote sensing is perhaps the most economical method for assessing the thermal effects of urban greenspaces. Since 1984 the thermal infrared band of Landsat TM sensor provides high resolution imagery data showing surface temperature distributions. In previous studies we showed Landsat TM data clearly exhibit characteristics of lower temperature in the urban greenspaces and investigated the intensity of temperature decrease in urban greenspaces by comparing the distribution of temperature with the distribution of greenspaces classified by using remote sensing data.
However as the scale of the urban components is usually smaller than the resolution of Landsat TM sensor (30m×30m), many pixels contain a mixture of cover classes. A pixel including multiple classes is called a mixel and it is very hard to classify mixels with conventional methods such as the maximum likelihood method or the cluster analysis. Therefore, the precision of the classification is restricted by the resolution of a pixel.
In the present study we introduce fuzzy clustering to examine the distribution of urban greenspaces more precisely. Fuzzy clustering is based on the theory of fuzzy sets which allows fuzzy boundary of sets. Using this method it is possible to calculate the ratio of greenspaces in a mixel (hereafter we call the ratio ‘greenness’). The distribution of greenspaces as the result of fuzzy clustering is compared with the temperature distribution.
As a test area, a region of 32×32 pixels in Minato-ward, Tokyo is selected from Landsat TM data of Kanto district (pass 107-row 35, acquired on Aug. 3rd, 1985). The ground truth of the greenness is calculated from the vegetation map of Minato-ward digitized by an image scanner.
To evaluate the precision of fuzzy clustering we compared the ground truth of greenspaces shown in Fig. 5(a) with the result of fuzzy clustering shown in Fig. 5(b). These patterns visibly show good agreement but the greenness of a corresponding pixel is not exactly the same because of various types of errors. After the process of flattening to ease the errors, the relation between the ground truth and the result of fuzzy clustering is shown in Fig. 6. The agreement is satisfactory considering the various errors included in the original data and difficulty of the classification.
To show the relation between the ratio of greenspaces and surface temperature, the distribution of surface temperature obtained from band 6 is also shown in Fig. 5(c). It seems that low temperature area generally coincides with the high greenness area.
