3 seminaria ośrodka badawczego Spatial Warsaw z udziałem młodych badaczy WNE
Polecamy uwadze 3 seminaria ośrodka Spatial Warsaw.
12 stycznia br. badanie pt. „Beyond the mean: exploration of nighttime light intensity based features in regional economic prosperity prediction” zaprezentuje doktorant Szkoły Doktorskiej Nauk Społecznych UW - Michał Künstler. Spotkanie rozpocznie się o godz. 17:00, a odbędzie się w sali B111 (WNE). Do seminarium będzie można dołączyć również za pośrednictwem platformy Zoom.
Z kolei 15 stycznia br. o godz. 15:00, w formule zdalnej, o prowadzonym badaniu opowie doktorant Muhammad Usman. Prezentacja dotyczyć będzie tematu „Disentangling the heterogeneous effects of climate shocks and conflict exposure on child malnutrition”.
Podobnie 15 stycznia przewidziane jest kolejne spotkanie seminaryjne – tym razem z udziałem studenta WNE, Norberta Jaworskiego. Odbędzie się o godz. 16:45. Prelegent przedstawi wyniki badania pt. „Spatial Herfindahl–Hirschman Index (sHHI): A New Density-Aware Market Competition Index Separating Inter-Type Rivalry from Intra-Type Cannibalisation”. Spotkanie zaplanowane jest w formule hybrydowej. Zapraszamy do udziału zarówno stacjonarnie - w sali B102, jak i online.
Osoby zainteresowane udziałem zdalnym prosimy o wiadomość mailową i kontakt z dr Kateryną Zabariną pod adresem: }QCv0bRiFlk7OujBs&S~9Yt]#[iL{YlU8X~&c$1~`6|i4tFJa.
Z abstraktami wystąpień można zapoznać się poniżej.
---
Beyond the mean: exploration of nighttime light intensity based features in regional economic prosperity prediction
This study investigates best practices for modeling regional economic prosperity using nighttime light (NTL) satellite data, with a focus on subnational GDP and GDP per capita prediction across European regions at the NUTS2 and NUTS3 levels. Despite growing interest in remote sensing for economic applications, many methodological uncertainties remain, particularly concerning model architecture, feature selection, and urban-rural heterogeneity. To address these, we construct nearly 200 machine learning models spanning multiple time periods and urbanization types, and evaluate them using $R^2$ and normalized dropout loss (NDL) to assess model performance and variable importance. Our findings confirm that tree-based ensemble models—particularly XGBoost and random forests—consistently outperform linear approaches, highlighting their suitability for high-dimensional, nonlinear prediction tasks. Evidence on whether to model urban and non-urban areas separately is mixed, suggesting that the optimal approach may be context-dependent and warrants empirical testing. Most notably, features that capture the shape of the NTL distribution, including inter-quantile ranges, skewness, and the commonly used sum of lights, emerge as the most informative predictors. These results demonstrate the value of moving beyond aggregate measures and emphasize the importance of distributional characteristics in understanding spatial economic disparities. The study offers both methodological contributions for researchers and practical implications for policy, particularly in identifying economically lagging areas through NTL-derived proxies.
Disentangling the heterogeneous effects of climate shocks and conflict exposure on child malnutrition
Child malnutrition remains a persistent public health challenge in low-and middle-income countries, driven by complex and spatially uneven environmental and socio-political stressors. This study examines the spatially heterogeneous effects of climate variability and conflict exposure on childhood malnutrition in Pakistan focusing primarily on stunting using high-resolution gridded data from 2001 to 2017. By integrating satellite-derived climate indicators, geolocated conflict events, and socioeconomic covariates, we employ a suite of spatially explicit modelling approaches including Geographically Weighted Regression (GWR), Multiscale GWR (MGWR), and Mixed GWR. Results indicate that the impacts of climatic shocks especially drought intensity and extreme precipitation are highly location-dependent and operate at different spatial scales. Conflict exposure further intensifies these adverse effects in vulnerable and socioeconomically deprived regions, with MGWR revealing stronger fine-scale variations compared to traditional GWR. Mixed GWR identifies a combination of global drivers (e.g., long-term socioeconomic deprivation) and locally varying predictors (e.g., seasonal drought anomalies and conflict proximity). Collectively, the findings highlight substantial spatial heterogeneity in the determinants of child malnutrition and demonstrate that ignoring local variations can mask critical pockets of vulnerability. This study provides a nuanced understanding of how climate and conflict interact across space to influence child health outcomes. The results underscore the need for spatially tailored and multisectoral policy interventions that address both environmental and political fragility to effectively reduce child malnutrition.
Spatial Herfindahl–Hirschman Index (sHHI): A New Density-Aware Market Competition Index Separating Inter-Type Rivalry from Intra-Type Cannibalisation
We introduce a new spatial competition index that blends the classical Herfindahl–Hirschman Index (HHI) with ideas from ecological competition. The metric is density-aware: it uses Voronoi catchments to assign population to the nearest unit (e.g., store or school) and summarises concentration at both the unit and type levels. We also extend the index to separate inter- and intra-type competition. As a proof of concept, we use simulations to test the behaviour of the index for different numbers of units and population-density schemes. The simulations show intuitive patterns: concentration rises when units are sited unevenly or when people are clustered; doing both amplifies the effect. Case studies from Warsaw—Żabka as a monopoly-like network, the Biedronka–Lidl duopoly, and public primary schools—are based on real data and serve as an empirical proof of concept. We also frame location choice as a practical optimisation problem for planners (minimise concentration) and managers (strengthen a focal type), showing how a density-aware approach changes the diagnosis and guides siting decisions. Results are robust across data and modelling choices and are reproducible with open data.