Assessing the utility of DMSP-OLS night-time images to propose surrogate census

Countries, such as India, conduct a census collection every ten years. Currently census in India is carried out manually with enumerators visiting every household in the country. Being such a vast country (in terms of area) and with a population of more than 1 billion, manual data collection is a laborious and an expensive process. This thesis proposes a surrogate method for collecting key census metrics using satellite images which help overcome some of the problems such as inconsistency issues, the Modifiable Areal Unit Problem (MAUP) and large temporal acquisition timeframes. Satellite images from the Operational Linescan System (OLS) onboard the Defense Meteorological Satellite Program (DMSP) group of satellites were used for the study. The data processing section of the thesis describes the pre-processing of the census and satellite datasets used in this research and the various data quality issues that were encountered. The next part of the research examines models to propose census metrics from non – composited fixed gain radiance calibrated images. The next section of the study used the global composite stable light images and brightness images for the year 2001 to propose surrogate census for areas at different spatial scales. Linear regression and multivariate analyses were subsequently performed and models proposed for each of the selected census metrics with results ranging from r² of 0.8 to 0.9 at the 95% confidence interval. At Taluks the adjusted r² values range from 0.2 to 0.8 at the 95% confidence interval, with the majority of the metrics being moderately correlated (with r² between 0.4 and 0.7). Generally it was found that the observed lights and brightness of big rural settlements from DMSP-OLS images have the potential for predicting certain census metrics. Census metrics unavailable at spatial scales lower than districts were also predicted using the proposed models and maps were derived showing the predicted measures. The thesis concludes with a comparative assessment of the models and the utility of the DMSP-OLS night-time images in proposing census. The method proposed in this research will enable prediction of census metrics more frequently and determine the trends of change over the inter-censal periods.<br>