Test and develop high resolution mapping and modelling methods to support inter-censal population estimates

塞拉利昂最近完成了地理参照中期人口普查，收集了关于其居民的最新、准确和完整的人口信息。这为比较和验证不同类型的小区域人口估计模型提供了一个独特的机会，从而为今后编制两次普查之间的估计数提供了信息。通过利用完整的地理参考中期人口普查结果，拟议的研究将：㈠使用中期人口普查的子样本测试各种人口估计方法：测试“自底向上”模型应用中不同样本设计策略的性能（例如，分层、加权等）B.在使用可靠收集的调查数据作为投入时，对照中期人口普查，量化这种“自下而上”的估计数的业绩c。开发和测试广泛的地理空间数据（即，协变量），并确定最适合的人口估计。这将包括探索机器学习选项，以使用高分辨率卫星图像，建筑物足迹数据和标签以及各种调查数据来模拟建筑物使用情况（住宅/非住宅和其他特征）。确定以高分辨率估计年龄/性别结构的最佳方法，㈡测试上次人口普查中不同类型的国家以下各级预测方法，以审查哪些工作最准确，哪些辅助数据集最有价值，㈢比较各种自上而下的人口预测分类与地理参照普查数据。

Sierra Leone has recently concluded its georeferenced mid-term census that collected up-to-date, accurate andcomplete demographic information on its residents. This provides a unique opportunity to compare and validatedifferent types of small area population estimation models and therefore inform approaches to producingintercensal estimates going forward. By utilizing the full georeferenced mid-term census results, the proposedresearch will:(i) Test various population estimation methods using sub-samples from the mid-term census:a. test the performance of different sample design strategies in 'bottom-up' model applications (e.g., stratification,weighted, etc.)b. quantify the performance of such 'bottom-up' estimates against the mid-term census when using routinelycollected surveys as inputsc. develop and test a wide range of geospatial data (i.e., covariates) and identify the best suited for populationestimation. This will include the exploration of machine learning options to model building usage (residential/nonresidential,and other characteristics) using high-resolution satellite imagery, building footprint data and labels anddata from various surveys.d. identify the best method to estimate age/sex structures at high resolution,(ii) test different types of sub-national projection methods from the last census to examine which work mostaccurately and what ancillary datasets are most valuable, and(iii) compare various top-down disaggregation of population projections with the geo-referenced census data.