Discussion on results of modeling the urbanisation and its relation to public policy

Prediction accuracy for each direction is used as a basis for rule calibration. Over/under estimation principle was implemented. If a set of rules for a particular direction produced underestimated results, this mean the growth rate is small and hence the rules are modified to increase the urban growth. For overestimation, the rules were modified to reduce the urban growth amount. The transition rules for a direction were repeatedly calibrated till the convergence criterion is met. The classified image provides the reference for calibration process. In table 2, the Fitness %, Total Error (∆E %) and f values for the year 1992 indicates a poor match of the simulated image with the real image (classified image), which is an indication of underestimation of urban pixels in various directions. However, the results for the 2006 simulated image (table 2), indicates very good spatial prediction accuracy. The spatial variability between the various directions as compared to the real image is small. This indicates the effect of spatial calibration in matching each direction with its realistic urban growth pattern through calibrating its rules. It also helped in capturing finer details in the modeling process while calibrating the model over smaller spatial units to reduce modeling uncertainty. Visually, calibration on a directional basis succeeds in preserving the urban pattern over space and over time. Rule values’ results at the end of the calibration process indicate some similarity between growth in various directions such as the east, west and the northwest. These wards have almost the same growth rate and pattern because of similar infrastructure, facilities, and more open area for outer growth and urban sprawl. Most of these similar wards have ring roads or highways passing through them that allow linear urban growth happening along.

The average fitness value for the 1992 image was ~ 60% and the total error was 29.09 with an approximate match of 71%. It is to be noted that for a highly accurate prediction, the total modeled urban count and ground truth urban count will be equal and therefore the fitness value (F) will be 1 or 100%. The total errors ∆E is the error of omission and commission. More the value of ∆E, more is the percentage of error count. There seems some mismatch between the urban pixels in 1992 that is without any visible pattern and therefore could not be assessed and captured by the change in population density contours and curve fits in various wards and different directions. Simulation and prediction urban modeling results, as shown in table 2 for the year 2006, show that the fitness results for prediction was close in terms of urban count (values close to 100%) between the modeled and real data with average fitness of 101.60 (little overestimate) and the average total error of 31.68% was achieved. This indicates an approximate match level of 69% on a pixel by pixel basis between modeling and reality. Therefore, higher the value of f in equation 8, higher is the modeling error. For the 2006 simulated image, the average f is 33.33 showing a more realistic result as compared to the actual urban growth pattern. This is a high accuracy level compared to the results shown in literature for the urban land spatial fit area that was only 28.15 to 44.6% (Yang and Lo, 2003). The close urban pattern match is also clear in figure 6 where the simulated images have urban distribution similar to those shown in their corresponding real images.   

The simulation results of urban growth should be accurate and should represent the actual local site specific patterns close to reality since urbanisation process is directly linked to society, infrastructure, level of services, etc. At this point of time, it would be appropriate to link people-and-pixels in remotely sensed images. One rationale for doing so is that, it might result in better social science research in several ways – such as measuring the context of social phenomena and their effects while providing additional measures, making connections across levels of analysis, providing time-series data on socially relevant phenomena. On the other hand, social science has also to play a major role for remote sensing. Social science makes several kinds of scientific contributions to remote sensing such as validation and interpretation of remote observations, data confidentially and public use, etc. Together remote sensing mapping technology and social science can improve understanding of human-environment interactions to a great extent. They help in interpreting, modeling, predicting the dynamics of natural resources, and in understanding the human consequences of climate flux, etc.

The change in land use such as agricultural fields, buildings, roads are often considered human artifacts and gets less importance and are therefore less interesting than the abstract variables that explain their appearance and transformations. Changing land use are regarded as manifestations of more important variables, such as government policies, land-tenure rules, distribution of wealth and power, market mechanisms, and social customs, none of which are directly reflected in the bands of the electromagnetic spectrum. The social utility argument posits that the interpretation of classified images obtained from remote sensing imageries becomes even more valuable to the extent that social scientist find useful, and that efforts should be made to identify and overcome the existing barriers to making this happen. From the perspective of social science, one important reason for using remotely sensed data is to gather information on the context that shapes social phenomena. The role of context has been central to the theories and empirical work of numerous statisticians, sociologists, economists, and anthropologists. In this context, remote sensing technology offers an additional source of contextual data for multilevel analyses. Another consideration involves the growing interdisciplinary community ranging from sustainable development, pollution prevention, global environmental change, to related issues of human-environment interaction who need to compare data on social and environmental phenomena at the same spatial and temporal scales (Liverman et al., 1998). Therefore, the consideration of spatial and temporal resolutions is very important.

Another critical issue in linking people with pixels and image is the decision on where to gereference individuals or other social units. The approach adopted in this work aggregates social data to larger geographical units; assigning individuals to larger areas in which their environmental effects are more likely to be confined. It is necessary at this point of time, to socialise the pixel and pixelise the social in land use and land cover change. Mining the pixel involves seeking social meaning in imaginary – information and indicators relevant to such concerns as economic well-being or criticality, perhaps signaling the underlying processes that give rise to land use and land cover change. This meaning is often hidden deep within the analysis of the imagery and this depth may impede such investigation. A paucity of spatially explicit data has constrained spatial modeling of human behavior and social structures, especially beyond the field of geography and has fostered modeling approaches that abstract the essential spatial nature of the problem. As a result, either aggregate relationships are specified, or the spatial components in a model are reduced to unidimensional variables, such as the distance between economic activities in location model, the wage differential in a migration model, or the cost of access in a transportation model. The increased availability of spatially explicit data, both remotely sensed and other data, and GIS (Geographic Information System) has begun to change the situation. Advances are being made to link on-the-ground human actions and consequences to imagery (pixels) through models, or modeling to the pixel, as in modeling the determinants of the decision of individual land managers on the basis of utility maximisation, satisficing, or other theories of human behavior. The use of pixels may extend to explain the dynamics of many indicators such as energy demand and conservation, environmental area assessment, disaster energy response, forecasting urban expansion that can be visualized as concentric rings, sectors or multiple nuclei.

The future interaction between societal studies and remote sensing depends on what kind of features can be detected and how often data can be obtained. Remote sensing technology may be used not only for monitoring change, but also for conducting surveillance. For example to count houses, to count the number of stories in each house, and detect changes in building structure. This may provide ability to check on building regulations and thus develop some new surrogates for social economic conditions. The key question is whether this type of information can be used to create more efficient urban environments and provide a more equitable distribution of resources and services?

Conclusion

This work explores the potential of implementing the cellular automata to model the historical urban growth over Bangalore city. The main goal is to design the model as a function of local neighbourhood structure to minimise the input data to the model. Satellite imagery represents the medium over which the model works. One special issue the model takes into account is the calibration process. Two modules were used namely, spatial and temporal calibration. Spatial calibration fits the model on a directional basis to its site specific feature while the temporal calibration adapts it to the urban growth dynamic change over time. This is a noticeable effect on producing a good spatial match between the real and simulated image data. On the other hand, GA is introduced to enhance the CA calibration process. GA makes the calibration process more efficient through manipulating a set of feasible solutions in the search space to find an optimal solution. This will reduce the search space for the optimal rules’ values on a directional basis. The above techniques are robust in predicting urban growth and visualizing them through pixels in images. Relating pixels in remote sensing data and people in society is important for studies on sustainable development, pollution prevention, global environmental change, and issues of human-environment interaction at different spatial and temporal scales.