Abstract
Landscan USA is a high resolution dasymetric model incorporating multiple ancillary variables to distribute populations. LandScan USA is a valuable tool in determining the population at risk during emergency response situations. However, a critical evaluation is necessary to produce user confidence regarding model accuracy through the verification and validation of model outputs. Unfortunately, dynamic models, such as population distribution, are often not validated due to the difficulty of having multiple input datasets and lack of validated data. A validated dataset allows analysis of model accuracy, as well as quantifying the benefits and costs of improving input datasets compared to find a balance for producing the best model.
This paper examines inaccuracies present within the input variables of two national school datasets incorporated in the model. Schools were chosen since a validated school dataset exists for Philadelphia County, Pennsylvania. Quality control efforts utilized throughout the LandScan USA process are quantified to determine the degree of quality control necessary to have a statistically significant effect on model output.
Typical LandScan USA quality control resulted in 43% of school enrollment values changed, compared to 89% for the validated dataset. Normal quality control methods resulted in 36% of schools being spatially relocated compared to 87% for the validated dataset. However, the costs of increasing quality control from normal to the validated dataset equated to a 600% increase in manual labor time for statistically insignificant improvements in LandScan USA daytime. This study enabled validation verification of not only the quality control process for LandScan USA, but also provides confidence in model output and use for policy issues, planning and emergency situations.
