Remodeling census population with spatial information from LandSat TM imagery
Y. Yuan, R.M. Smith and W.F. Limp
    The goal of this study was, by using the dasymetric mapping principle, to apply multivariable regression to examine the correlation between population counts from census and land cover types. The first step is to reconstruct the census geographic entities and retrieve population data which is done by linking each BG in the map from TIGER to the appropriate census socioeconomic variables by combination codes from TIGER and key columns in STFs. The second step is to reclassify the land cover map to seven categories: residential; commercial-industrial; recreational; agricultural; other agricultural; forested; and uninhabited. The third step is to overlay the census population map and reclassified land cover map, breaking down the census population map units into smaller polygons. The fourth step is regression model testing and refinement of the population densities. The fifth and final step is to locally fit the estimates from regression to each county and each land cover type with scaling techniques. Further research may be to improve the resolution of result; remodel socioeconomic variables other than population and also improving the method to assess the reliability of the results.

Radiance Calibration of DMSP-OLS Low-Light Imaging Data of Human Settlements
D. Elvidge, K. Baugh, J. Dietz, T. Bland, P. Sutton, and H. Kroehl
   The data processing of the nighttime light data involves separating sub-orbits of usable light data from full orbit files. Six processes must be followed. First, a reference grid is established. Lights, clouds, and coverage areas are identified and geo-located. A digital number is established which will provide the radiance scale for the final product. Two overlapping gain ranges both high and low, are determined for cloud free compositing. A threshold is established which will eliminate isolated detections; and lastly, the calibrated images from the two gain images can be combined by averaging based on the number of detections (81). This article explores the implications of satellite observation of nocturnal lighting as a method of locating human activities and modeling population density. Between the years 1994 and 1995 stable lights have been produced for North and South America, Europe, Asia, and northern Africa. Stable lights simply refers to “the percent frequency with which lights were detected within the set of cloud free observation with no indication of the brightness of the light” (79).

   The data processing of the nighttime light data involves separating sub-orbits of usable light data from full orbit files. Six processes must be followed. First, a reference grid is established. Lights, clouds, and coverage areas are identified and geo-located. A digital number is established which will provide the radiance scale for the final product. Two overlapping gain ranges both high and low, are determined for cloud free compositing. A threshold is established which will eliminate isolated detections; and lastly, the calibrated images from the two gain images can be combined by averaging based on the number of detections (81).

    There are of course locations such as airports or industrial areas which have low population densities but high levels of nocturnal lighting. Although at this point this method is more useful for examining the extent of human activity versus the population density of an area, it is still rather useful in determining the distribution of population.

Refining population surface models: experiments with Northern Ireland census data D. Martin, N. Tate and M. Langford
D. Martin, N. Tate and M. Langford
    This paper examines the method of redistribution from enumeration district and postal code centroids with that of dasymetric mapping using remote sensed imagery and compares and contrasts the two. Based on the Martin and Bracken methodology in calculating population densities from Enumeration District (ED) centroids, a continuous surface of population density is created for Northern Ireland.

   The second method in mapping population density is by using remote sensed imagery. The first step is to geo-rectify and simultaneously resample the imagery to a 25m pixel resolution. Next, a standard unsupervised classification algorithm was applied to create a classified land cover map. Spectral clusters closely associated with urban areas were then cross referenced to the original image and to paper map sources. These were merged to form a class labeled as built while the rest was labeled unbuilt. A process of manual editing is then undertaken to remove obvious image classification errors. An analysis is performed to find the census population total, and a tally of built and unbuilt pixels contained within it. The regression equation of “Population = 1.20076x #built + 0.008695x #unbuilt” is used to redistribute the population of each ED amongst its constituent built and unbuilt pixels in such a way that the relative densities between the two categories are maintained.

Annotated Bibliography: Continuous Surface Model . Remotely Sensed Imagery . Population Distribution
...........................................References