Open Access

Generative Adversarial Approach to Urban Areas’ NDVI Estimation: A Case Study of Łódź, Poland

Maciej Adamiak
Maciej Adamiak
, 
Krzysztof Będkowski
Krzysztof Będkowski
 and 
Adam Bielecki
Adam Bielecki
   | Jan 29, 2023
Quaestiones Geographicae's Cover Image
About this article
Previous Article
Next Article
Cite
Published Online: Jan 29, 2023
Page range: 87 - 106
Received: Sep 27, 2022
DOI: https://doi.org/10.14746/quageo-2023-0007
Keywords
© 2023 Maciej Adamiak et al., published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.

Fig. 1

Research area, based on the data from the Head Office of Geodesy and Cartography (Land and Building Records).
Research area, based on the data from the Head Office of Geodesy and Cartography (Land and Building Records).

Fig. 2

The main steps of the investigation.
The main steps of the investigation.

Fig. 3

Visualization Visualisation of a dataset item, based on Geoportal (2021). From left to right: RGB composition, R, G and B bands that form the input tensor and normalised difference vegetation index, which serves as a target tensor.
Visualization Visualisation of a dataset item, based on Geoportal (2021). From left to right: RGB composition, R, G and B bands that form the input tensor and normalised difference vegetation index, which serves as a target tensor.

Fig. 4

Modified Pix2Pix discriminator model visualisation.
Modified Pix2Pix discriminator model visualisation.

Fig. 5

Modified Pix2Pix generator sub-model visualisation.
Modified Pix2Pix generator sub-model visualisation.

Fig. 6

Colour palette used during visual inspection of Figs 7–15.
Colour palette used during visual inspection of Figs 7–15.

Fig. 7

Inference result – abandoned land; based on Geoportal (2021). From left to right: PAN, NDVItrue, NDVIartificial and NDVIdiff.
Inference result – abandoned land; based on Geoportal (2021). From left to right: PAN, NDVItrue, NDVIartificial and NDVIdiff.

Fig. 8

Inference result – wooded area: PAN, NDVItrue, NDVIartificial and NDVIdiff; based on Geoportal (2021).
Inference result – wooded area: PAN, NDVItrue, NDVIartificial and NDVIdiff; based on Geoportal (2021).

Fig. 9

Inference result – small parking lot: PAN, NDVItrue, NDVIartificial and NDVIdiff; based on Geoportal (2021).
Inference result – small parking lot: PAN, NDVItrue, NDVIartificial and NDVIdiff; based on Geoportal (2021).

Fig. 10

Inference result – buildings: PAN, NDVItrue, NDVIartificial and NDVIdiff; based on Geoportal (2021).
Inference result – buildings: PAN, NDVItrue, NDVIartificial and NDVIdiff; based on Geoportal (2021).

Fig. 11

Inference result – residential area: PAN, NDVItrue, NDVIartificial and NDVIdiff; based on Geoportal (2021).
Inference result – residential area: PAN, NDVItrue, NDVIartificial and NDVIdiff; based on Geoportal (2021).

Fig. 12

Inference result – commercial facility: PAN, NDVItrue, NDVIartificial and NDVIdiff; based on Geoportal (2021).
Inference result – commercial facility: PAN, NDVItrue, NDVIartificial and NDVIdiff; based on Geoportal (2021).

Fig. 13

Inference result – commercial facility parking lot: PAN, NDVItrue, NDVIartificial and NDVIdiff; based on Geoportal (2021).
Inference result – commercial facility parking lot: PAN, NDVItrue, NDVIartificial and NDVIdiff; based on Geoportal (2021).

Fig. 14

Inference result – farmland: PAN, NDVItrue, NDVIartificial and NDVIdiff; based on Geoportal (2021).
Inference result – farmland: PAN, NDVItrue, NDVIartificial and NDVIdiff; based on Geoportal (2021).

Fig. 15

Sliding window inference (NDVIartificial) of an orthophoto used to compute the test dataset); based on Geoportal (2021). The scene (51.76174 E, 19.42149 N) presents Łódź, Poland. Red values indicate high NDVI values (closer to 1). Blue ones represent small values (closer to −1).
Sliding window inference (NDVIartificial) of an orthophoto used to compute the test dataset); based on Geoportal (2021). The scene (51.76174 E, 19.42149 N) presents Łódź, Poland. Red values indicate high NDVI values (closer to 1). Blue ones represent small values (closer to −1).

Fig. 16

Sliding window inference (NDVIartificial) of an archival 1966 greyscale aerial image; based on GUGiK (Head Office of Geodesy and Cartography b.d.). The scene presents Łódź, Poland. Red overlay indicates values where 0.5 < NDVI < 1.
Sliding window inference (NDVIartificial) of an archival 1966 greyscale aerial image; based on GUGiK (Head Office of Geodesy and Cartography b.d.). The scene presents Łódź, Poland. Red overlay indicates values where 0.5 < NDVI < 1.

Normalised difference vegetation index threshold values used in urban studies in Poland.

NDVI Threshold for vegetation Image data used Research area References
0.3 Landsat TM, GSD 30 m, 3 Jul. 2006 Warsaw Tomaszewska et al. (2011)
0.1 MODIS, GSD 250 m, 3 Jul. 2006 Warsaw Tomaszewska et al. (2011)
0.1 Digital orthophoto, GSD 0.1 m, May 2014 Wroclaw Kubalska and Preuss (2014)
0.2 IKONOS-2, GSD 1(4) m, 18 Aug. 2005 Lublin Krukowski et al. (2016)
0.2 Landsat 8, GSD 30 m, 3 Jul. 2015 Łódź Będkowski and Bielecki (2017)
0.1 Pléiades 1A, GSD 0.5 m, May 2012 Warsaw Pyra and Adamczyk (2018)
0.1 CIR-orthophoto, GSD 0.25 m, 2015 Łódź Pluto-Kossakowska et al. (2018)
0.2 IKONOS-2, GSD 1(4) m, 18 Aug. 2011 Lublin Krukowski (2018)
0.1 CIR aerial orthophoto, GSD 0.25 m, 2015 Łódź Worm et al. (2019)
0.6 Sentinel 2, GSD 10 m, summer 2018, 2019 Poland Łachowski and Łęczek (2020)
0.2 IKONOS-2, June 2005, GSD = 0.8 m PAN (3.2 m MS)QuickBird-2, September 2006, GSD = 0.6 m PAN (2.4 m MS)WorldView-2, October 2014, GSD = 0.5 m PAN (2.0 m MS)Aerial orthophotomap (CIR), May 2017, GSD = 0.25 m Poland Zięba-Kulawik and Wężyk (2022)

Test set evaluation metrics.

SSIM PSNR RSME
AVG 0.7569 26.6459 0.0504
STD 0.1083 3.6577 0.0193
MIN 0.3589 16.3343 0.0026
MAX 0.9987 51.7674 0.1525

Structure of land use in Łódź [km2].

Total Agricultural land Forest, woody, and bushy land Residential areas Industrial areas Transport areas Groundwater Other
293.25 113.76 24.67 47.13 13.91 42.37 1.33 1.15
eISSN:
2081-6383
Language:
English
Publication timeframe:
4 times per year
Journal Subjects:
Geosciences, Geography
Journal RSS Feed