Professor Andrew J. Tatem, gave a seminar on his work on population modelling. He collaborates with NGOs and other organisations to better the lives of disadvantaged communities around the world so that the World Development Goals can be achieved. His work focuses on developing computational modells and data acquisition methods to monitor population distributions and movements.
The seminar started by listing the World Development Goals achieved so far and those to be achieved by 2030. The most notable goal achieved is the high reduction of extreme poverty and future goals will include the reduction of high inequality in developing nations. To achieve WDGs it is imperative to know who to target, where they are and how the populations are structured.
Population data provided by governments such as census and employment data is highly unreliable in regions that are most in need of intervention due to either the poor quality of procedures or outdated data. This data also does not provide information on population mobility and structure, which is important in the prevention of epidemics or deploying vaccines.
The government or survey data can be correlated across regions with other indicators such as satellite imagery of night lights and settlement sizes, mobile phone connections to cell towers, geolocated tweets or photographs posted online. This provides useful and reliable data for humanitarian or development organisations which they can use to manage resources better and rapidly intervene during crises. All this data is gathered into the WorldPop and Flowminder online open-source databases.
Two methods of counting population distribution are used: Top-Down - first the census data is projected on a grid and this is then correlated with satellite pictures of buildings to estimate the number of people living in a grid area. The Bottom-Up approach uses night satellite images (indicator of economic activity/wealth) or satellite/airplane images of agricultural developments, forrests, roads and rivers with machine learning algorithms and perform analysis on computer clusters.
The population structure can then be determined by assuming either by assuming that a certain percentage of the population has charcteristics noticed elsewhere, e.g. 20% is under 5, in a simple model, or more complex bayesian modelling can be used with statistical covariates from satellites. For instance images of large settlements with agricultural lands but very little night lights correlates with higher poverty and in consequence a larger proportion of young people.
Mobile phone data has started to increasingly provide more reliable data about the population structure (how many people) are near a cell tower) where they go (log-ins into a series of cell towers), how rich they are (top-ups, are friends rich?). This is useful in identifying how the disease will spread or how many people are in a place during a disaster. Data volumes are very large and confidentiality concerns mean that more servers have to be installed in phone operator towers to perform local analysis and send 'distilled' results for further processing. An example of the current capabilities of the WorldPoP and Flowminder capabilities was the quick response during the Nepal earthquake, when international agencies could quickly estimate how many people are in affected regions and send appropriate help.
|||(1, 2, 3) Image source WorldPop|