Open-source satellite enumeration to map households: planning and targeting indoor residual spraying for malaria
© Kamanga et al. 2015
Received: 20 May 2015
Accepted: 3 August 2015
Published: 17 September 2015
Defining the number and location of sprayable structures (houses) is foundational to plan and monitor indoor residual spray (IRS) implementation, a primary intervention used to control the transmission of malaria. Only by mapping the location and type of all sprayable structures can IRS operations be planned, estimates of spray coverage determined, and targeted delivery of IRS to specific locations be achieved. Previously, field-based enumeration has been used to guide IRS campaigns, however, this approach is costly, time-consuming and difficult to scale. As a result, field-based enumeration typically fails to map all structures in a given area, making estimations less reliable and reducing the enumerated coverage.
Using open source satellite imagery and Geographic Information System software, satellite enumeration was conducted to guide IRS operations in 15 districts (91,302 km2) in northern Zambia during the 2014 spray season. Cost of satellite enumeration was compared to standard enumeration. Enumerated households were sampled to estimate sprayable surface area and wall type from the satellite enumeration using linear and logistic regression, respectively.
In comparison to the traditional field-based enumeration procedure, satellite-based enumeration was 22 times faster, and 10 times less costly. An estimated 98 % of the satellite enumerated buildings correctly classified roof type. Predicted surface area of each household correlated at a value of 0.91 with measured surface area of each household.
For IRS campaigns, high quality and high coverage enumeration data aid in planning, through informed insecticide procurement. Through the identification of geographical areas and populations to target, enumeration data guide operations and assist monitoring and evaluation of IRS through the unbiased estimation of coverage achieved. Satellite enumeration represents a quick, cheap and accurate system to provide these data, and has potential applications beyond IRS for delivery of other targeted or non-targeted interventions (e.g. net distributions, mass drug administration, immunization campaigns, or even sampling frames for field studies).
KeywordsMalaria Indoor residual spraying Google Earth Enumerate Structure mSpray Ground-truth
Malaria is a disease caused by a parasite transmitted through mosquitoes of the Anopheles mosquitoes and is a leading cause of child mortality; in 2013, approximately 584,000 deaths were attributed to malaria, most of which were children . An estimated 1.5 billion individuals throughout the globe are at risk of Plasmodium falciparum malaria transmission, the species which causes the most mortality, with the greatest number at risk on the African continent . A primary intervention is the use of indoor residual spraying (IRS), which entails spraying an insecticide on the inside walls of households. Female Anopheles typically rest on the inside walls of houses after taking a blood meal, and with effective IRS those female Anopheles are killed. IRS thus reduces both the amount of mosquitoes circulating in a household and decreases the age of the population of mosquitoes, effectively reducing malaria transmission when coverage is sufficiently high .
Millions of dollars are spent across the southern African region every year on IRS to control malaria. In Zambia alone, the total IRS budget crested US 16 million to cover six high-risk provinces in 2015. The absence of household address systems and population counts in this region leads to uncertainty in estimating numbers of households needing to be sprayed as well as the amount of insecticide required during IRS planning sessions. Following implementation of IRS, uncertainty also exists in estimates of houses that are actually sprayed.
To date, IRS in Zambia is largely done in three phases. First, census records are consulted to estimate the number of households in each district or region targeted for IRS. Second, field-based enumeration teams travel through targeted districts to enumerate and locate households. Records are typically kept on paper, though some programmes have begun to transition to electronic-based enumerations. Third, IRS teams are mobilized and move through the targeted areas spraying households as they find them. The total process requires significant financial inputs in the first and second phases even before any insecticide is sprayed. Inaccuracies in the estimate of the number of houses and thus amount of insecticide to be sprayed leads to inefficient use of labor and funds, which can potentially lead to inadequate coverage. Furthermore, there has been no validation check that the ground-based teams effectively enumerate the entire targeted areas, an important consideration given the rural nature of the region.
Advances in the availability and resolution of satellite imagery facilitate the use of remotely sensed data to enumerate areas. This paper relates the development of an effective, efficient, low-cost and scalable methodology to enumerate houses in 15 districts of Luapula and Central Province, Zambia in preparation for the 2014 spray season using freely available remotely sensed satellite imagery and freely available software.
A team of nine recent University graduates with little or no experience in using Geographic Information System (GIS) were given a 2-day GIS training which focused on Google Earth and Quantum GIS (QGIS) version 1.8.0, a freely available and open source GIS software package (http://www.qgis.org). The training focused on image interpretation, digitization and creation of vector data. Google Hybrid and Bing imagery plugins were loaded to complement each other during the screen digitization of the study area. Google Hybrid imagery was consulted first and where Google Hybrid imagery had missing or low resolution, Bing imagery was applied. Most images, by both providers, were taken between 2011 and 2013. For this reason the images correlated well in terms of visible structures. However, the location of the structures often varied for up to five metres between the Bing and Google images. Considering that most GPS devices have similar or higher discrepancies, this was not considered to be a hindering factor. Interestingly, once the enumeration neared completion, Google updated many of their images to 2014/2015 data. It can be assumed that Google updates images based on use/load of images.
Estimating sprayable surface area and wall type from satellite enumeration
A two-stage cluster sample of 125 randomly selected structures from four districts within the study area was conducted to develop a model predicting sprayable surface area from the satellite enumeration. The survey collected the following measures: type of roof of each structure, the amount of sprayable surface area in each house, and whether the house had smooth or porous walls. The captured data was used to test the ability of the satellite-based enumeration process to predict the sprayable surface area at the household level as well as whether or not the household had smooth wall surfaces. These data were necessary to ensure that IRS could be planned for based on variables easily identified through satellite imagery.
The total sample size of 125 houses was divided into 75 model fitting households and 50 model testing households. Three predictive regression models were conducted among the 75 model fitting households. First linear regression predicted the number of rooms in each house from the roof are of the enumerated structure (footprint) digitized from satellite imagery and the composition of the roof being either thatch or non-thatched as determined from satellite imagery. Second linear regression predicted the measured sprayable surface area as the outcome variable with the structure’s footprint, roof type and predicted number of rooms. Third the wall type (smooth or roof) was predicted from the structure footprint and the roof type. For the third model, three structures in the model fitting households had both rough and smooth walls and so were excluded from the predictive regression. In order to facilitate a linear relationship, the footprint variable and the outcome of sprayable surface area were log transformed for all three models. Accuracy of predicting sprayable surface area was assessed by applying coefficients from model fitting regression to the model testing data and estimating the correlation between measured sprayable surface area and predicted sprayable surface area. A receiver operating characteristic (ROC) curve defined sensitivity and specificity of correctly identifying the wall type in the model testing data using the predicted values. QGIS was used for map making ; R version 3.1.0 was used for all analyses; the pROC package was used to generate the ROC curves [6, 7].
Cost comparison of field versus satellite enumeration methods
A comparative cost analysis was conducted reviewing costs from traditional field-based enumeration with satellite-enumeration practices. Budgets from previous field-based enumeration exercises were consulted and compared with the budget for satellite-based enumeration. The metric of houses enumerated per day was used to compare the two methods and associated costs (field costs, travel, person-days) were also considered in the comparison.
The nine GIS technicians enumerated 270,000 structures across a total area of 91,302 km2 within a period of 22 days. On average, each technician was able to enumerate 1,364 structures per day (range 1,150–1,500).
Model outcomes to estimate sprayable surface area and wall type from satellite enumeration
All structures sampled on the satellite imagery were successfully found in the field. The measured sprayable surface area among sampled structures ranged from 18.5 to 580.5 m2 with an interquartile range of 61.6–224.8 m2. Mean and median surface areas for non-thatched houses (typically steel-roofs) measured at 203.1 and 187.5 m2, respectively. Mean and median surface areas for thatched houses were measured at 68.3 and 52.2 m2, respectively.
Predicting number of rooms per household
Multivariate linear regression results predicting number of rooms per household, surface area of households and smooth wall surface within households
Predicting number of rooms per householda
Log-transformed area of roof
Thatched roof (non-thatched is reference)
Predicting surface area of householdsb
Log-transformed area of roof
Predicted number of rooms
Thatched roof (non-thatched is reference)
Predicting smooth wall surface within householdsc
Area of roof
Thatched roof (non-thatched is reference)
Predicting surface area of households
The log-transformed footprint (area) of the roof, predicted number of rooms, and roof composition (thatch versus non-thatch) were all associated with the surface area of the walls in bivariate analyses. Together they accounted for 75.8 % of the variation in surface area of the walls in the model fitting households (Table 1). Residual differences between predicted surface area of walls and measured surface area of walls among the model testing houses were typically small; correlation between measured and predicted surface area of walls among the 50 model testing households was 0.91. There was greater variation in the residuals of non-thatched roof households than in the residuals of thatched-roof households.
Predicted wall type
Smooth wall surfaces were found in 40 (32 %) of the households sampled and rough wall surfaces were found in 80 (64 %) of the households sampled. Five households (4 %; three in the model fitting data and two in the model testing data) had both rough and smooth walls in different part of the house and were excluded from further analysis. Households with non-thatched roofs (steel) were 12 times more likely than households with thatched roofs to have smooth walled surfaces [prevalence ratio = 12.23, 95 % confidence interval (CI) = 2.76–54.22]. Houses with larger roofs were also more likely to have smooth walls compared to rough wall surfaces; mean roof area was 58.5 m2 for houses with rough wall surfaces and 113 m2 for houses with smooth wall surfaces (t test = −5.75, P < 0.001).
In comparison to field-based enumeration, the satellite enumeration process was ten times less costly
GPS cost: 10 % contributionb
Training venue hire
TOTAL per district
In this study, use of satellite imagery for enumeration of structures was investigated for the planning and monitoring of IRS campaigns for malaria prevention and control. Previous studies have used satellite imagery and digitization for estimating population, creating sampling frames and building point feature class databases, however these studies did not use open-source imagery. Other studies that have used open-source imagery, did not use open-source GIS to process the imagery [8–13]. This study appears to be the first to use both freely available imagery and open-source GIS to conduct satellite enumeration. The expansion of freely available satellite imagery coupled with advancements in open-source GIS makes the methodology outlined herein applicable to various health sectors including estimating denominators for mass drug administration campaigns, net distributions, estimating coverage for IRS campaigns, estimating population size and creating sampling frames. The satellite enumeration methodology was found to be extremely cost-effective and efficient. Each technician was able to enumerate far more structures per day than an entire team of ground-based enumerators without the need for transport and field costs.
Aspects of households which are observable from satellite imagery (roof type and digitized surface area) were significantly associated with the type of wall surface, number of rooms and total sprayable wall surface area, characteristics important to consider for the procurement of insecticide commodities for IRS campaigns. This finding further identified the satellite enumeration methodology as an extremely useful and efficient tool for planning IRS campaigns.
Despite the advantage of being efficient, low-cost and scalable, various challenges were encountered during the satellite enumeration. In this study we ignored structures which had a roof surface area less than 9 m2 as these were most likely to be latrines, chicken coops or granaries which are not considered sprayable by the Government of the Republic of Zambia (GRZ). However, some animal shelters larger than 9 m2 could not be easily distinguished from structures where people slept. Also, it was difficult to isolate buildings which had contiguous roofing to determine if these were separate sleeping spaces, a problem which was also observed in a similar study . Unfortunately, the enumeration accuracy was not assessed in terms of enumerating the correct (sprayable) structures, however, a study in Malawi found that fewer than 5 % of enumerated structures were not households but rather separate kitchens, stores or animal pens . In a separate study in Zambia, only 6 (1 %) of 750 randomly selected sampled structures enumerated by satellite were not found . A similar level of accuracy is likely for the methodology presented herein in its ability to rely on satellite imagery to build maps of sprayable (enumerated) structures and for those enumerated structures to contain sleeping spaces.
From observations during IRS field operations, two factors are suspected to affect the accuracy of the satellite enumeration. First, publicly available satellite imagery is between 1 and 3 years old and hence satellite enumerations do not account for population growth. Second, in some cases, canopy cover may obstruct the view of rooftops and making the satellite enumeration process more difficult. Cloud cover was not a significant impediment during enumeration as we were able to switch between Google Hybrid and Bing satellite imagery, utilizing the most cloud free image available. Further studies are planned to determine the extent these two factors affect the accuracy of the satellite enumeration.
The satellite enumeration methodology described here represents a highly cost effective, scalable and efficient system for enumerating structures and describing their characteristics, e.g., footprint, roof type, wall area. Satellite enumeration data provides the foundation for determining an unbiased estimated of spray coverage during campaigns ensuring IRS impact is maximal. This system is superior to traditional ground enumeration, and should be considered for other applications requiring spatial targeting or planning e.g. mass drug administration, net distributions and for general sampling purposes.
indoor residual spraying
Government of the Republic of Zambia
Ministry of Health
Ministry of Community Development Mother and Child Health
AK and SR oversaw the satellite enumeration process, finalized the methodology and trained the enumeration team. AK and JP conducted the fieldwork. JP, DP, DB and BC contributed to methodology development and field consideration discussions. DL led model development and contributed to the manuscript. AW conceived the satellite enumeration application and contributed to the methodology and manuscript development. All authors read and approved the final manuscript.
We wish to acknowledge the hardworking team of enumerators for their tireless efforts in completing this process. Support for the satellite enumeration process was provided by the US Agency for International Development (USAID) and the Presidents Malaria Initiative (PMI). The opinions expressed herein are those of the author(s) and do not necessarily reflect the views of USAID or PMI.
Compliance with ethical guidelines
Competing interests The authors declare that they have no competing interests.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- World Health Organization (2015) Malaria Fact Sheet. http://www.who.int/mediacentre/factsheets/fs094/en/. Accessed 29 July 2015
- Gething PW, Patil AP, Smith DL, Guerra CA, Elyazar IR, Johnston GL, Tatem AJ, Hay SI (2011) A new world malaria map: Plasmodium falciparum endemicity in 2010. Malar J 10:378. doi:10.1186/1475-2875-10-378 PubMed CentralView ArticlePubMedGoogle Scholar
- Pluess B, Tanser FC, Lengeler C, Sharp BL (2010) Indoor residual spraying for preventing malaria. Cochrane Database Syst Rev (4):CD006657. doi:10.1002/14651858.CD006657.pub2
- Republic of Zambia Central Statistical Office: 2010 Census of Population and Housing (2010) http://unstats.un.org/unsd/demographic/sources/census/2010_phc/Zambia/PreliminaryReport.pdf. Accessed 31 July 2015
- QGIS: QGIS Geographical Information System (2015) http://qgis.osgeo.org. Accessed 29 July 2015
- R Project for Statistical Computing (2015) R: a language and environment for statistical computing. http://wwwR-project.org. Accessed 29 July 2015
- Robin X, Turck N, Hainard A, Tiberti N (2011) pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinform 12:77View ArticleGoogle Scholar
- Mohammad A, Shahid R, Jin-Kyung P, Shamoon S, Rion O, Qamaruddin N et al (2004) Use of satellite imagery in constructing a household GIS database for health studies in Karachi, Pakistan. Int. J Health Geogr 3:20View ArticleGoogle Scholar
- Veljanovski T, Kanjir U, Pehani P, Ostir K, Kovacic P (2012) Object-based image analysis of VHR satellite imagery for population estimation in informal settlement Kibera-Nairobi, Kenya. In: Escalante B (ed) Remote sensing–applications. InTech, EuropeGoogle Scholar
- Schmitt U, Sulzer W, Schardt M (1998) Analysis of settlement structure by means of high-resolution satellite imagery. In: Fritsch D, Englich M, Sester M (eds) Between visions and applications. ISPRS Commission IV Symposium on GIS, Stuttgart, pp 557–562Google Scholar
- Escamilla V, Emch M, Dandalo L, Miller WC, Martinson F, Hoffman I (2014) Sampling at community level by using satellite imagery and geographical analysis. Bull World Health Organ 92:690–694PubMed CentralView ArticlePubMedGoogle Scholar
- Lowther SA, Curriero FC, Shields T, Ahmed S, Monze M, Moss WJ (2009) Feasibility of satellite image-based sampling for a health survey among urban townships of Lusaka, Zambia. Trop Med Int Health 14:70–78View ArticlePubMedGoogle Scholar
- Checchi F, Stewart B, Palmer J, Grundy C (2013) Validity and feasibility of a satellite imagery-based method for rapid estimation of displaced populations. Int J Health Geogr 12:4PubMed CentralView ArticlePubMedGoogle Scholar