decreased significantly for the scale 250 m  250 m. Consequently, we set the area
250 m  250 m as the smallest unit for classification. Certainly, there were cases
where different types coexisted even in a 250 m  250 m area. In such cases, since
the major type is selected as the representative one, the classification result does not
accurately reflect an actual condition. Setting a smaller unit would have alleviated
this problem; however, this would have significantly increased the number of
meshes to be visually judged and caused the method to become impractical.
Considering the feasibility of visual detection and acceptable accuracy, we adopted
the area 250 m  250 m as a reasonable unit in this study.
The classification steps adopted by this study are as follows:
1. The entire urban area is divided into 250 m  250 m mesh units.
2. Meshes that can be considered nonresidential according to existing data, such as
land-use data, are removed.
3. The remaining meshes are visually classified into several types based on the
following four indicators: (1) the ratio of the area of green space and farmland to
unit area, (2) block shape and building layout, (3) building density, and (4) average building height in unit area. Areas that could visually be identified as being
nonresidential were categorized as “nonresidential area.”
4. The model used to predict visual detection results is constructed by applying the
multinominal logit model using existing data as explanatory variables.
5. If the hit rate to model prediction results to visual detection results is considered
valid, the visual detection results are adopted.
4.3 Application of the Residential Classification to Jakarta
4.3.1 Four Types of Classification and Spatial Distribution
In this section, the classification method is applied to Jakarta, the capital city of
Indonesia. Jakarta is administratively referred to as DKI Jakarta; however, the actual
urban area extends beyond the city’s administrative boundaries and is recognized as
a broad metropolitan area called Jabodetabek; it includes the surrounding municipalities, as well (Fig. 4.2). Its size is approximately 80 km from east to west and
70 km from north to south; the population of DKI Jakarta was approximately 9.6
million and that of Jabodetabek was approximately 28 million in 2010.
2 This study
classifies the residential areas in DKI Jakarta, Tangerang City, and South Tangerang
City in the west; Bekasi City in the east; and Depok City and Bogor City in the south.
The residential areas of these regions are divided into 250 m  250 m meshes and
classified according to the satellite image in Google Earth and four indicators:
2 Calculated from Sensus Penduk 2010 (Population Census 2010) published by BPS (Badan Pusat
Statistik, Central Agency on Statistics)
50
K. Hayashi et al.
250 m  250 m as the smallest unit for classification. Certainly, there were cases
where different types coexisted even in a 250 m  250 m area. In such cases, since
the major type is selected as the representative one, the classification result does not
accurately reflect an actual condition. Setting a smaller unit would have alleviated
this problem; however, this would have significantly increased the number of
meshes to be visually judged and caused the method to become impractical.
Considering the feasibility of visual detection and acceptable accuracy, we adopted
the area 250 m  250 m as a reasonable unit in this study.
The classification steps adopted by this study are as follows:
1. The entire urban area is divided into 250 m  250 m mesh units.
2. Meshes that can be considered nonresidential according to existing data, such as
land-use data, are removed.
3. The remaining meshes are visually classified into several types based on the
following four indicators: (1) the ratio of the area of green space and farmland to
unit area, (2) block shape and building layout, (3) building density, and (4) average building height in unit area. Areas that could visually be identified as being
nonresidential were categorized as “nonresidential area.”
4. The model used to predict visual detection results is constructed by applying the
multinominal logit model using existing data as explanatory variables.
5. If the hit rate to model prediction results to visual detection results is considered
valid, the visual detection results are adopted.
4.3 Application of the Residential Classification to Jakarta
4.3.1 Four Types of Classification and Spatial Distribution
In this section, the classification method is applied to Jakarta, the capital city of
Indonesia. Jakarta is administratively referred to as DKI Jakarta; however, the actual
urban area extends beyond the city’s administrative boundaries and is recognized as
a broad metropolitan area called Jabodetabek; it includes the surrounding municipalities, as well (Fig. 4.2). Its size is approximately 80 km from east to west and
70 km from north to south; the population of DKI Jakarta was approximately 9.6
million and that of Jabodetabek was approximately 28 million in 2010.
2 This study
classifies the residential areas in DKI Jakarta, Tangerang City, and South Tangerang
City in the west; Bekasi City in the east; and Depok City and Bogor City in the south.
The residential areas of these regions are divided into 250 m  250 m meshes and
classified according to the satellite image in Google Earth and four indicators:
2 Calculated from Sensus Penduk 2010 (Population Census 2010) published by BPS (Badan Pusat
Statistik, Central Agency on Statistics)
50
K. Hayashi et al.