LCZ-Generator Training Areas

This dataset contains all training areas (TA) submitted to the LCZ Generator (Demuzere et al. 2021) since 2021-04-14. The LCZ Generator follows a crowdsourcing approach, making fast and easy LCZ-mapping available to the public, while collecting LCZ maps and TAs in a centralized, easy to access, location. The crowdsoucing approach overcomes the limitations of previous approaches where a manual review was mandatory before publication. While this improved the quality of individual LCZ-maps, the number of cities mapped during this period remained low. The LCZ Generator removed the manual review process, allowing for faster collection of LCZ maps and TAs, however, sacrificing some quality since any person can submit to the LCZ Generator without prior training or review.

This dataset is based on crowdsourcing, hence LCZs may be mislabelled, polygon shapes may not be perfect etc. also city names may not be correct. Some contributors chose to name their city e.g. "..", "....amsa" etc. also some author names may not be correct. The automated quality control (see table below and section 2.3 in Demuzere et al. 2021) may help filter out some of the incorrect TAs. We intentionally included all available TAs to allow for (the development of) custom filtering.

The data was extracted from the LCZ-Generator database taking into account:

Whether or not the submitting author agreed to show their name (if not, it is also left blank in this dataset)

The license the TA was submitted under (a license change happened with version 2.0.0 of the LCZ Generator)

Duplicate geometries were dropped, since multiple (re-)submission may have the same geometries. Only the first submitted version is kept and attributed to the submission_id of the first submission.

The data, up to December 2021, was used during creation of the global LCZ Map (Demuzere et al. 2022).

Additional TAs were extracted from the WUDAPT Portal and processed using the LCZ-Generator.

Note: The data is updated periodically, but not on a fixed schedule.

Data Description

The data is provided as GeoPackage (.gkpg) which can be used with most GIS.

Column Name
Description

geometry
The polygon geometry of the training area (TA) in EPSG:4326

submission_id
The ID of the corresponding submission in the LCZ Generator

submission_date
The date and time in UTC the TA was submitted to the LCZ Generator

city
The city the TAs are for. Note: This is sometimes incorrect due to users entering incorrect information and the LCZ Generator following a crowdsourcing approach.

reference
The submitting author may have provided (additional) references via this field. This can be a scientific paper or a citation of the original creator of this TA

remarks
General information: e.g. co-authors, information about the study/framework the TAs were generated for

representative_date
The date the TAs are representative for (i.e. the date the aerial image was taken)

firstname
First name of the submitting author (if the author did not agree to publish their name, this is left blank and the submission is treated anonymously)

lastname
Last name of the submitting author (if the author did not agree to publish their name, this is left blank and the submission is treated anonymously)

license
The license this specific polygon is licensed under. With version 2.0.0 of the LCZ Generator the license was changed from CC BY-SA to CC BY-NC-SA 4.0

cite_as
A suggestion how to cite the TA (-set) based on the name, year, and city information. If the author submitted anonymously, this is left blank

version
The version of the LCZ Generator the polygon was submitted to. Detailed information can be found in the Changelog

class
The LCZ Class the TA-polygon has been labelled (1 - 17)

area
The area of the TA-polygon in km2

perimeter
The perimeter of the TA-polygon km

shape
The shape of the TA-polygon calculated as: (perimeter2) / (4 π · area)

vertices
The number of vertices of the TA-polygon

qc_step1
Whether the TA-polygon passed the automated quality control (QC) step 1: Surface area below 0.04 km2 (too small) or a shape ratio 3 (too complex shape) are flagged. More information about the QC can be found in the FAQ and the corresponding paper Demuzere et al. 2021

qc_step2
Whether the TA-polygon passed the automated quality control (QC) step 2: Average spectral value of a polygon of LCZ class is considered as an outlier compared to the average spectral values of all other polygons of that class. Note that this is done on a per-submission basis More information about the QC can be found in the FAQ and the corresponding paper Demuzere et al. 2021

qc_step3
Whether the TA-polygon passed the automated quality control (QC) step 3: Considers all individual pixel values of all polygons in each LCZ class compared to the polygon average approach from QC Step 2. More information about the QC can be found in the FAQ and the corresponding paper Demuzere et al. 2021

oa
The overall accuracy of the submission based on Demuzere et al. 2020

oau
The overall accuracy for the urban LCZ classes only of the submission based on Demuzere et al. 2020

oabu
The overall accuracy of the built versus natural LCZ classes only of the submission based on Demuzere et al. 2020

oaw
A weighted accuracy taking the similarities of LCZs into account based on Demuzere et al. 2020

f1_1
Class-wise metric F1 for LCZ class 1

f1_2
Class-wise metric F1 for LCZ class 2

f1_3
Class-wise metric F1 for LCZ class 3

f1_4
Class-wise metric F1 for LCZ class 4

f1_5
Class-wise metric F1 for LCZ class 5

f1_6
Class-wise metric F1 for LCZ class 6

f1_7
Class-wise metric F1 for LCZ class 7

f1_8
Class-wise metric F1 for LCZ class 8

f1_9
Class-wise metric F1 for LCZ class 9

f1_10
Class-wise metric F1 for LCZ class 10

f1_11
Class-wise metric F1 for LCZ class 11

f1_12
Class-wise metric F1 for LCZ class 12

f1_13
Class-wise metric F1 for LCZ class 13

f1_14
Class-wise metric F1 for LCZ class 14

f1_15
Class-wise metric F1 for LCZ class 15

f1_16
Class-wise metric F1 for LCZ class 16

f1_17
Class-wise metric F1 for LCZ class 17

Acknowledgements

We acknowledge all WUDAPT contributors and community members for providing the training areas via the LCZ Generator.