Serbia is facing severe depopulation where many rural areas of the country are devastated while at the same time larger urban areas are under growing pressure of in-migrations and work commuters. Internal migrations are very challenging to estimate using the traditional approach of tracking the number of people who changed their permanent residence. Namely, many people live and work outside their place of permanent residence, but due to the demanding administrative procedure to have their personal documents updated, they simply continue to use the old ones. To better understand the complex phenomena of depopulation and in migrations in Serbia, we are exploring the data from diverse sources such as telecom data, remote sensing data, vector based Land Change / Land Use data (such as Copernicus Land Monitoring Service data), national statistics, administrative data. Our aim is to make population distribution more informative by bringing other data sources that can measure pulse of the population and human activity, extract relevant indicators that can provide insights into depopulation trends and predictions.

The knowledge of human population distribution and dynamics is of crucial importance for regional and local development and policy making. The exact numbers are hard to estimate based on traditional techniques such as census and surveys. Researchers across the globe are exploring innovative approaches for estimating population based on diverse data sources such as remote sensing, GPS and telecom data.

The usage of telecom data as a global proxy for human activity and dynamics proved to be very efficient due to the high penetration rate of mobile phones even in developing countries. The state-of-the-art research results show that using call detail record (CDR) data to estimate migration flow is a promising technique that can complement traditional national statistics.

Human migrations and activity are highly dynamic and a single snapshot of data such as census results or one satellite image can be insufficient to evaluate daily based or weekly migrations which in many cases correspond to work commuters which can affect regional and local economy. Therefore, an intensive research effort is currently invested in exploration of different weekday/weekend patterns of human connectivity based on telecom data, fused with other sources, e.g. satellite data, and learning correlations between human dynamics and land use.

We focus on in-country distributions of population and migrations, based on analysis of activity, connectivity and mobility of people in Serbia reflected through telecom data together with data from different sources. From telecom data we have extracted numerous indicators regarding activity profiles in telecommunication traffic, graph properties from connectivity across the country, and mobility patterns. Based on satellite data we detect land use patterns and spatial changes in the land use and explore them jointly with population indicators derived from telecom data. Such data fusion enables better understanding of population trends and development of predictive models. Furthemore, sources providing information on points of interest POI are also included since they provide information of function and purpose of observed space.

Special attention is given to quantifying population pulse across the municipalities and interactions between them and finally selecting indicators for depopulation assessment. Extracted knowledge can serve for predictive modeling of population trends and developing decision support system for policy making.