Limits...
Methods for determining the uncertainty of population estimates derived from satellite imagery and limited survey data: a case study of Bo city, Sierra Leone.

Hillson R, Alejandre JD, Jacobsen KH, Ansumana R, Bockarie AS, Bangura U, Lamin JM, Malanoski AP, Stenger DA - PLoS ONE (2014)

Bottom Line: For five of those twenty sections, we quantized the rooftop areas of structures extracted from satellite images.Evaluations based either on rooftop area per person or on the mean number of occupants per residence both converged on the true population size.We demonstrate with this simulation that demographic surveys of a relatively small proportion of residences can provide a foundation for accurately estimating the total population in conjunction with aerial photographs.

View Article: PubMed Central - PubMed

Affiliation: Information Technology Division, Naval Research Laboratory, Washington, District of Columbia, United States of America.

ABSTRACT
This study demonstrates the use of bootstrap methods to estimate the total population of urban and periurban areas using satellite imagery and limited survey data. We conducted complete household surveys in 20 neighborhoods in the city of Bo, Sierra Leone, which collectively were home to 25,954 persons living in 1,979 residential structures. For five of those twenty sections, we quantized the rooftop areas of structures extracted from satellite images. We used bootstrap statistical methods to estimate the total population of the pooled sections, including the associated uncertainty intervals, as a function of sample size. Evaluations based either on rooftop area per person or on the mean number of occupants per residence both converged on the true population size. We demonstrate with this simulation that demographic surveys of a relatively small proportion of residences can provide a foundation for accurately estimating the total population in conjunction with aerial photographs.

Show MeSH
Occupancy-based bootstrap population estimations parameterized by sample size and random number seed.Results for (A) random number seed  = 12345 and (B) random number seed  = 67890. In both panels, the brown line is the total population count 25,954, the number of individuals counted in the 1,979 residences surveyed for DS01. The 0.50 CIs is shown in blue and the 0.95 CI in red. (C) A quartile boxplot shows the variation in the population bootstrap estimates  for 25 replicate samples initiated with different random number seeds.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC4232376&req=5

pone-0112241-g004: Occupancy-based bootstrap population estimations parameterized by sample size and random number seed.Results for (A) random number seed  = 12345 and (B) random number seed  = 67890. In both panels, the brown line is the total population count 25,954, the number of individuals counted in the 1,979 residences surveyed for DS01. The 0.50 CIs is shown in blue and the 0.95 CI in red. (C) A quartile boxplot shows the variation in the population bootstrap estimates for 25 replicate samples initiated with different random number seeds.

Mentions: SIM01. Figure 4AB shows the required convergence [19] of the occupancy-based population estimator as a function of the  = 1,979 residential structures in DS01. The variance (uncertainty) decreases uniformly as the sample size increases. The confidence intervals (CIs) are illustrated by the blue (0.50 CI) and red (0.95 CI) bars. The otherwise identical simulations were initiated with two different random seeds. Changing the random number seed will result in a different sequence of bootstrap replicate samples, and the estimated population as a function of sample size will vary. Using a different seed has little impact on the variance (CIs) but does result in modest variation in the estimated population as a function of sample size, as shown in the quartile boxplots in Figure 4C. Note that the expanded -axis exaggerates the absolute variation relative to the measured value of the population.


Methods for determining the uncertainty of population estimates derived from satellite imagery and limited survey data: a case study of Bo city, Sierra Leone.

Hillson R, Alejandre JD, Jacobsen KH, Ansumana R, Bockarie AS, Bangura U, Lamin JM, Malanoski AP, Stenger DA - PLoS ONE (2014)

Occupancy-based bootstrap population estimations parameterized by sample size and random number seed.Results for (A) random number seed  = 12345 and (B) random number seed  = 67890. In both panels, the brown line is the total population count 25,954, the number of individuals counted in the 1,979 residences surveyed for DS01. The 0.50 CIs is shown in blue and the 0.95 CI in red. (C) A quartile boxplot shows the variation in the population bootstrap estimates  for 25 replicate samples initiated with different random number seeds.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC4232376&req=5

pone-0112241-g004: Occupancy-based bootstrap population estimations parameterized by sample size and random number seed.Results for (A) random number seed  = 12345 and (B) random number seed  = 67890. In both panels, the brown line is the total population count 25,954, the number of individuals counted in the 1,979 residences surveyed for DS01. The 0.50 CIs is shown in blue and the 0.95 CI in red. (C) A quartile boxplot shows the variation in the population bootstrap estimates for 25 replicate samples initiated with different random number seeds.
Mentions: SIM01. Figure 4AB shows the required convergence [19] of the occupancy-based population estimator as a function of the  = 1,979 residential structures in DS01. The variance (uncertainty) decreases uniformly as the sample size increases. The confidence intervals (CIs) are illustrated by the blue (0.50 CI) and red (0.95 CI) bars. The otherwise identical simulations were initiated with two different random seeds. Changing the random number seed will result in a different sequence of bootstrap replicate samples, and the estimated population as a function of sample size will vary. Using a different seed has little impact on the variance (CIs) but does result in modest variation in the estimated population as a function of sample size, as shown in the quartile boxplots in Figure 4C. Note that the expanded -axis exaggerates the absolute variation relative to the measured value of the population.

Bottom Line: For five of those twenty sections, we quantized the rooftop areas of structures extracted from satellite images.Evaluations based either on rooftop area per person or on the mean number of occupants per residence both converged on the true population size.We demonstrate with this simulation that demographic surveys of a relatively small proportion of residences can provide a foundation for accurately estimating the total population in conjunction with aerial photographs.

View Article: PubMed Central - PubMed

Affiliation: Information Technology Division, Naval Research Laboratory, Washington, District of Columbia, United States of America.

ABSTRACT
This study demonstrates the use of bootstrap methods to estimate the total population of urban and periurban areas using satellite imagery and limited survey data. We conducted complete household surveys in 20 neighborhoods in the city of Bo, Sierra Leone, which collectively were home to 25,954 persons living in 1,979 residential structures. For five of those twenty sections, we quantized the rooftop areas of structures extracted from satellite images. We used bootstrap statistical methods to estimate the total population of the pooled sections, including the associated uncertainty intervals, as a function of sample size. Evaluations based either on rooftop area per person or on the mean number of occupants per residence both converged on the true population size. We demonstrate with this simulation that demographic surveys of a relatively small proportion of residences can provide a foundation for accurately estimating the total population in conjunction with aerial photographs.

Show MeSH