Noktasal veri dağılımının mekânsal birleştirme performansına etkisi: GeoPandas ve DuckDB Spatial karşılaştırması

Öz

Bu çalışmada GeoPandas ve DuckDB Spatial, 16 GB bellekli bir kişisel bilgisayarda nokta-poligon (point-in-polygon) mekânsal birleştirme performansı bakımından karşılaştırılmıştır. 10 000 eşit ve çakışmayan poligondan oluşan bir ağ üzerinde, 100 000 ile 5 000 000 nokta içeren düzgün ve kümelenmiş nokta kümeleri üç bağımsız örnek olarak üretilmiştir. Her iki araç aynı GeoParquet girdilerini okumuş, noktanın poligon içinde bulunması (within) koşulunu uygulamış ve her veri kümesinde doğrulanmış biçimde aynı eşleşme çiftlerini üretmiştir. Ana protokolde tek performans çekirdeğine sabitlenmiş ayrı süreçlerde yürütülen 240 tek iş parçacıklı koşuda okuma, birleştirme ve toplam süreler ile işletim sisteminin bildirdiği en yüksek bellek kullanımı ölçülmüştür. GeoPandas’ın birleştirme medyanı DuckDB medyanının 0.46 ile 0.63 katı olurken DuckDB dosyaları GeoPandas süresinin 0.14 ile 0.20’sinde okumuştur. GeoPandas’ın en yüksek bellek kullanımı DuckDB değerinin 2.2 ile 4.9 katına ulaşmıştır (beş milyon noktada GeoPandas için 2.0 GB, DuckDB için 0.41 GB). Sonuç sayısının sabit olduğu bu ağ iş yükünde nokta dağılımı birleştirme sürelerini en fazla yaklaşık %5 değiştirmiştir. Ayrı bir duyarlılık deneyinde DuckDB’nin varsayılan 16 iş parçacıklı yapılandırması 500 000 ve daha fazla noktada toplam süre sıralamasını tersine çevirmiştir. Buna karşılık test edilen Natural Earth idari poligon katmanında tek iş parçacıklı DuckDB birleştirmesi GeoPandas süresinin 21 ile 37 katı sürmüştür.

Anahtar Kelimeler

• Mekânsal birleştirme
• Nokta poligon sorgusu
• GeoPandas
• DuckDB Spatial
• Performans karşılaştırması

Effect of point data distribution on spatial join performance: A comparison of GeoPandas and DuckDB Spatial

Öz

This study compares GeoPandas and DuckDB Spatial for point-in-polygon spatial joins on a 16 GB personal computer. Uniform and clustered point sets of 100 000 to 5 000 000 points were generated in three independent realizations over a grid of 10 000 equal, non-overlapping polygons. Both tools read identical GeoParquet inputs, evaluated the same within predicate and returned identical match-pair sets, verified for every dataset. In the main protocol, 240 single-thread runs in isolated processes pinned to one performance core measured load, join and total times and the peak working set reported by the operating system. The GeoPandas join median was 0.46 to 0.63 times the DuckDB median, whereas DuckDB loaded the files in 0.14 to 0.20 of the GeoPandas time. GeoPandas peak memory was 2.2 to 4.9 times the DuckDB value (2.0 GB against 0.41 GB at five million points). In this fixed-cardinality grid workload, the point distribution changed join times by at most about 5%. In a separate sensitivity set, the default 16-thread DuckDB configuration reversed the total-time ranking from 500 000 points upward. On the tested Natural Earth administrative polygon layer, however, the single-thread DuckDB join took 21 to 37 times the GeoPandas time.

Anahtar Kelimeler

• Spatial join
• Point-in-polygon
• GeoPandas
• DuckDB Spatial
• Performance benchmarking

Kaynakça

1. E. H. Jacox and H. Samet, “Spatial join techniques,” ACM Transactions on Database Systems, 32 (1), Article 7, 2007. https://doi.org/10.1145/1206049.1206056
2. A. Eldawy and M. F. Mokbel, “The era of big spatial data: A survey,” Foundations and Trends in Databases, 6 (3-4), 163-273, 2016. https://doi.org/10.1561/1900000054
3. J. Van den Bossche, K. Jordahl, M. Fleischmann, M. Richards, J. McBride, J. Wasserman et al., geopandas/geopandas (software, version 1.1.4), Zenodo, 2026. https://doi.org/10.5281/zenodo.20938197
4. W. McKinney, “Data structures for statistical computing in Python,” Proceedings of the 9th Python in Science Conference, pp. 56-61, Austin, USA, 2010. https://doi.org/10.25080/Majora-92bf1922-00a
5. C. R. Harris, K. J. Millman, S. J. van der Walt, R. Gommers, P. Virtanen, D. Cournapeau et al., “Array programming with NumPy,” Nature, 585 (7825), 357-362, 2020. https://doi.org/10.1038/s41586-020-2649-2
6. S. Gillies, C. van der Wel, J. Van den Bossche, M. W. Taves, J. Arnott, B. C. Ward et al., Shapely (software, version 2.1.2), Zenodo, 2025. https://doi.org/10.5281/zenodo.17193310
7. A. Guttman, “R-trees: A dynamic index structure for spatial searching,” ACM SIGMOD Record, 14 (2), 47-57, 1984. https://doi.org/10.1145/971697.602266
8. S. T. Leutenegger, M. A. Lopez and J. Edgington, “STR: A simple and efficient algorithm for R-tree packing,” Proceedings of the 13th International Conference on Data Engineering, pp. 497-506, Birmingham, UK, 1997. https://doi.org/10.1109/ICDE.1997.582015

9. T. Brinkhoff, H.-P. Kriegel and B. Seeger, “Efficient processing of spatial joins using R-trees,” Proceedings of the 1993 ACM SIGMOD International Conference on Management of Data, pp. 237-246, Washington, USA, 1993. https://doi.org/10.1145/170035.170075
10. M. Raasveldt and H. Mühleisen, “DuckDB: An embeddable analytical database,” Proceedings of the 2019 ACM SIGMOD International Conference on Management of Data, pp. 1981-1984, Amsterdam, Netherlands, 2019. https://doi.org/10.1145/3299869.3320212
11. M. Raasveldt and H. Mühleisen, “Data management for data science: Towards embedded analytics,” Proceedings of the 10th Conference on Innovative Data Systems Research (CIDR), Amsterdam, Netherlands, 2020.
12. DuckDB contributors, Spatial extension, DuckDB documentation. [Online]. Available: https://duckdb.org/docs/current/core_extensions/spatial/overview.html, [Accessed Sep. 27, 2026].
13. M. Gabrielsson, Spatial joins in DuckDB, DuckDB official blog, 2025. [Online]. Available: https://duckdb.org/2025/08/08/spatial-joins, [Accessed Aug. 7, 2026].
14. Open Geospatial Consortium GeoParquet Standards Working Group, GeoParquet specification, version 1.0.0, 2023. [Online]. Available: https://geoparquet.org/releases/v1.0.0/, [Accessed Oct. 1, 2026].
15. M. Saeedan and A. Eldawy, “Spatial Parquet: A column file format for geospatial data lakes,” Proceedings of the 30th International Conference on Advances in Geographic Information Systems (SIGSPATIAL 2022), pp. 1-4, Seattle, USA, 2022. https://doi.org/10.1145/3557915.3561038
16. V. Pandey, A. Kipf, T. Neumann and A. Kemper, “How good are modern spatial analytics systems?,” Proceedings of the VLDB Endowment, 11 (11), 1661-1673, 2018. https://doi.org/10.14778/3236187.3236213
17. J. Yu, Z. Zhang and M. Sarwat, “Spatial data management in Apache Spark: The GeoSpark perspective and beyond,” GeoInformatica, 23 (1), 37-78, 2019. https://doi.org/10.1007/s10707-018-0330-9
18. T. Vu, A. Eldawy, V. Hristidis and V. Tsotras, “Incremental partitioning for efficient spatial data analytics,” Proceedings of the VLDB Endowment, 15 (3), 713-726, 2021. https://doi.org/10.14778/3494124.3494150
19. S. Acharya, V. Poosala and S. Ramaswamy, “Selectivity estimation in spatial databases,” Proceedings of the 1999 ACM SIGMOD International Conference on Management of Data, pp. 13-24, Philadelphia, USA, 1999. https://doi.org/10.1145/304182.304184
20. A. Georges, D. Buytaert and L. Eeckhout, “Statistically rigorous Java performance evaluation,” Proceedings of the 22nd ACM SIGPLAN Conference on Object-Oriented Programming Systems, Languages and Applications (OOPSLA 2007), pp. 57-76, Montreal, Canada, 2007. https://doi.org/10.1145/1297027.1297033
21. T. Hoefler and R. Belli, “Scientific benchmarking of parallel computing systems: Twelve ways to tell the masses when reporting performance results,” Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC15), Article 73, 1-12, Austin, USA, 2015. https://doi.org/10.1145/2807591.2807644
22. GeoPandas developers, geopandas.sjoin, GeoPandas 1.1.4 documentation. [Online]. Available: https://geopandas.org/en/v1.1.4/docs/reference/api/geopandas.sjoin.html, [Accessed Oct. 1, 2026].
23. J. R. Herring (ed.), OpenGIS Implementation Standard for Geographic Information - Simple Feature Access - Part 1: Common Architecture, OGC 06-103r4, Open Geospatial Consortium, 2011. [Online]. Available: https://www.ogc.org/standards/sfa/, [Accessed Aug. 7, 2026].
24. N. V. Kelso and T. Patterson, Natural Earth, 1:10m cultural vectors: Admin 1 - States and Provinces (version 5.1.1) and Populated Places (version 5.1.2). [Online]. Available: https://www.naturalearthdata.com/, [Accessed Aug. 7, 2026].
25. B. Šavrič, T. Patterson and B. Jenny, “The Equal Earth map projection,” International Journal of Geographical Information Science, 33 (3), 454-465, 2019. https://doi.org/10.1080/13658816.2018.1504949
26. PostGIS developers, ST_Subdivide, PostGIS documentation. [Online]. Available: https://postgis.net/docs/ST_Subdivide.html, [Accessed Sep. 15, 2026].
27. GeoPandas developers, geopandas.sindex.SpatialIndex.query, GeoPandas 1.1.4 documentation. [Online]. Available: https://geopandas.org/en/v1.1.4/docs/reference/api/geopandas.sindex.SpatialIndex.query.html, [Accessed Oct. 1, 2026].