What happens to a country when its population has been concentrating around a single point for decades?

Research question

How has the population of Belarus changed from the earliest modern censuses to current estimates—and what becomes visible when the country, regions, districts, and towns are examined separately? BY Maps was built to answer that question, not merely to display an attractive map.

The scope connects historical material from 1897, census records, official estimates, and explicitly separated scenarios through 2075. This scale makes it possible to study depopulation, concentration, and changes in the settlement hierarchy without reducing the process to one headline number.

Data problem

Long demographic series combine changing administrative boundaries, definitions, formats, and reliability levels. A compelling visual can hide missing comparability, interpolation, or confusion between an observed value and a projection.

Every number therefore needs visible provenance: source document, territorial unit, year, value type, and the transformation applied by the pipeline.

My role

• research framing and programme design;
• reproducible ETL and data normalisation;
• scenario design and backtesting;
• architecture of the interactive publication and research packages.

Architecture and method

Raw and curated data are stored separately. A versioned Python pipeline performs transformations, range checks, and consistency tests, while a Next.js and MapLibre application publishes the result without hiding the underlying data layer.

source → raw data → curated data → tests → analysis → map → reproducibility package

Historical observations and scenarios use different labels. Each research package includes code, explicit assumptions, reproduction instructions, and checksums, so an update cannot silently overwrite the previous result.

Reading spatial change

rural area → small town → district centre → regional capital → Minsk → abroad

This sequence is a research interpretation of spatial movement, not the route taken by every person. The map, source tables, and territorial definitions are required to test the claim.

History and scenarios through 2075

The three future trajectories are published as conditional scenarios, not forecasts. The historical series is visually and semantically separated from modelled values: a scenario changes with its assumptions and is not presented as an observed fact.

What the data cannot establish

• correlation does not prove the cause of migration or depopulation;
• changes in administrative boundaries constrain direct comparison;
• the 2026 estimate is not a census;
• a scenario through 2075 is not a promise about the future.

Eleven research tracks

The map opens a broader programme covering city hierarchy, ageing, wages and population, service accessibility, migration, mono-towns, Chernobyl, night lights, twentieth-century shocks, model challengers, and the material overhang of infrastructure. Each track has its own question, scope, and limitations.

How quality was evaluated

The pipeline checks ranges, types, duplicates, aggregate consistency, and differences between releases. Scenarios are backtested and compared with a simple baseline. Test results remain part of the artifact rather than an internal note available only to the author.

What the project demonstrates

The result is an independent research service and a public repository. A reader can move from map to record, source, code, and limitations without access to the author's private environment. The study does not ask readers to trust the researcher; it lets them inspect the work.

Why multiple scenarios

The project publishes multiple explicit scenarios rather than one context-free forecast. The interactive map is paired with methodology and versioned packages because the visualisation is an interface to the research, not a substitute for it.

Limitations

Comparability depends on boundary changes, population definitions, and historical-source quality. Correlation does not establish causality. A scenario is a conditional model result, not a promise about the future.