Prof. John Gibson (p), Geua Boe-Gibson

Concerns about declining towns and uneven regional development are central to regional science, yet remain difficult to study empirically over long horizons. Population change is often gradual and spatially uneven, unfolding within towns as well as across larger administrative units. Conventional data, such as censuses, are infrequent and tied to fixed boundaries, limiting their usefulness for detecting long-run contraction or expansion in the spatial footprint of local economic activity. Satellite-detected luminosity offers a spatially continuous alternative, but use of these data has involved a fundamental trade-off between temporal coverage and spatial resolution.

Early work relied on data from the Defense Meteorological Satellite Program (DMSP), which allows analysis back to the early 1990s but suffers from coarse effective sensor resolution, spatial blurring, and other well-documented measurement issues. Newer systems, such as VIIRS, provide far more accurate observations at finer resolution, but only from around 2012 onwards. As a result, researchers face a choice between a long time series with substantial spatial distortion, or a short but more reliable series.

This paper advances how declining towns can be measured by applying a deep learning reconstruction of night-time lights that bridges this gap. Using information learned from modern high resolution VIIRS data, we generate a consistent annual series of luminosity from 1992 to 2021 on a fine spatial grid (cells of 0.7 km2). Unlike most existing harmonisation approaches, which rely on cross-sectional overlap between sensors, the model explicitly incorporates temporal change during training, allowing it to recover long-run dynamics while substantially reducing the spatial blurring characteristic of DMSP data. This sharpens inference about contraction and expansion at the extensive margin of towns.

We apply the reconstructed data to New Zealand, where there is ongoing policy concern about the uneven growth experiences across the urban hierarchy, with claims of ‘zombie towns’ facing long-run contraction. With the reconstructed night-time lights data we examine whether some towns exhibit persistent declines in their extensive margins of luminosity over three decades, how widespread such declines are, and whether sharp rises in energy costs have exacerbated any such decline, especially for ‘mill towns’. While the application focuses on New Zealand, the contribution is methodologically portable. By combining deep learning with satellite data to recover long-run spatial detail, the paper advances the measurement of urban decline in settings where traditional data are limited, with direct relevance for regional science research on shrinking towns and long-run uneven development.

Dr. Francois Hermet (p)

Dr. Réka Horeczki (p)

Our analysis focused on the municipal-level dynamics of spatial inequalities in Hungary. It was based on local GDP estimations compared to the EU average. This approach can be explained by the importance of GDP as the main indicator of European cohesion policy and our aim to evaluate the Hungarian spatial development processes in a European context. As main contribution, a more refined picture compared to the NUTS-2 or NUTS-3 level analyses should be emphasized. In European context all the 19 NUTS-3 level regional units outside the capital city can be regarded as underdeveloped or peripheral: their GDP per capita values are under the EU average (mostly under 75% of it) during the examined period. But our analysis demonstrated that there are some higher developed islands beyond the administrative borders of Budapest as well. It also revealed a colourful patchwork of GDP convergence with the predominance of municipalities getting really closer to the EU average. However, particularly the convergence pattern of higher developed larger cities shows that the success of the chosen economic development model is largely depending on the global turbulences (many cities showed convergence only in the second decade, after the economic crisis). On the other hand, the case of Győr (having convergence only in the first period) regarded as one of the most successful cities of Hungary might demonstrate the limitations of this model often connected with the so-called middle-income trap of the factory economies.
Current analysis discovered a definite convergence on the municipal level inequalities of economic power (decomposed and estimated GDP). This convergence process is indicated according to the European averages. However, the trend of convergence has tended to slow down within the country and inequalities have begun to slightly increase again. Besides of this, the spatial development structure of Hungary is expressively fixed with segregated central and peripheral zones.
The conceptualisation might serve as a basis for identifying new growth paths in peripheral regions (both in growing urban and rural areas) in the post-pandemic world used for challenge-oriented local development.