Spatial Autoregressive Model for Population Estimation at the Census Block Level Using LIDAR-Derived Building Volume Information
Qiu, Fang, Sridharan, Harini, Chun, Yongwan, Cartography and Geographic Information Science
The distribution of population is one of the fundamental inputs for many research tasks (such as urban planning, facilities allocation, and quality of life assessment) because the dispersion of resources and energy among various geographical areas is strongly dependant on the population size (Hoque 2008). Traditional methods for the collection of population data involve a census, which entails extensive planning, surveying, and data processing, which is time consuming, laborious, and expensive (Lo 2006). Due to the huge financial investments involved, census data, such as those provided by the U.S. Census Bureau, are usually collected only at a fixed time period (e.g., once every ten years) and at a set of pre-defined geographic units (e.g., the census tract, block group, and block). The fact that census data are provided at a limited temporal and spatial scales restricts census data's applications to only those that do not critically rely on the most current information of the population distribution at detailed levels.
In addition to conducting a census every ten years, the U.S. Census Bureau also provides intercensal population count estimations at state, county, and city levels based on projection techniques (Smith 1998), but the estimates are not available at any finer geographical scale such as the tract or block level. The U.S. Census Bureau has recently introduced the American Community Survey (ACS) which aims at providing population information every year instead of every ten years. The ACS selects a sample ofhouseholds for surveying and provides yearly estimates based on this sample. Currently, only three-year estimates from 2006-2008 are available and five-year estimates are expected to be available by the end of 2010. These estimates will be available only at the tract level or above but will remain unavailable to the more detailed (block group and the block) levels. To support activities that need suitable population information at a fine-scale level for a given year between two consecutive decennial censuses (e.g., emergency response planning), population estimation is still needed. At this scale, reliance must be placed on third party, often expensive, complex, commercial demographic models. These models may provide reasonable estimation but often involve significant manpower for demographic analysis. Due to the requirement to collect multiple inputs, the success of the models relies heavily on the quality of these inputs and the performance of the models in earlier time periods, which are not always reliable (Qiu et al. 2003).
In order to estimate population at various levels of detail, remote sensing imagery and its derived datasets have been employed as viable alternatives in population prediction. Remote sensing provides a synoptic view of a large area which can be acquired at a small and regular time interval or within a short period of time when it is needed (Lo 2006). With the right techniques, remote sensing based population estimation can serve as a cheaper and less laborious replacement for commercial demographic models. Over the past 40 years, a variety of remote sensing products with different spatial resolutions have been employed to estimate population at different scale levels. For example, low to medium resolution satellite images, such as Landsat TM imagery, have been used to conduct city level population estimation (Lisaka and Hegedus 1982; Qiu et al. 2003; Wu and Murray 2007), while large-scale aerial photographs have been employed to support the modeling of population counts at a community level, such as the census tract (Porter 1956; Dueker and Horton 1971; Lo 1989). The advent of IKONOS, Quickbird and other very high spatial resolution satellite imagery enables us to drill down to even more detailed levels, such as the block and block group (Haverkamp 2004; Lu et al. 2006; Liu et al. 2006; Chen 2002). The increasing refinement in spatial resolutions for aerial and satellite images and their growing availability to the public have motivated a wider adoption of remote sensing technology for population estimation, aimed at achieving better accuracy at a finer scale level than was previously possible. …