Using 3D GIS to estimate population and refine census maps

The remote estimation of a region’s population has for decades been a key application of geographic information science in demography. Most studies have used 2D data (maps, satellite imagery) to estimate population avoiding field surveys and questionnaires. As the availability of semantic 3D city models is constantly increasing, in our new paper we investigate to what extent they can be used for the same purpose:

Biljecki, F., Arroyo Ohori, K., Ledoux, H., Peters, R., & Stoter, J. (2016). Population Estimation Using a 3D City Model: A Multi-Scale Country-Wide Study in the Netherlands. PLOS ONE, 11(6), e0156808. doi:10.1371/journal.pone.0156808

Based on the assumption that housing space is a proxy for the number of its residents, we use two methods to estimate the population with 3D city models in two directions: (1) disaggregation (areal interpolation) to estimate the population of small administrative entities (e.g. neighbourhoods) from that of larger ones (e.g. municipalities); and (2) a statistical modelling approach to estimate the population of large entities from a sample composed of their smaller ones (e.g. one acquired by a government register). Starting from a complete Dutch census dataset at the neighbourhood level and a 3D model of all 9.9 million buildings in the Netherlands, we compare the population estimates obtained by both methods with the actual population as reported in the census, and use it to evaluate the quality that can be achieved by estimations at different administrative levels. We also analyse how the volume-based estimation enabled by 3D city models fares in comparison to 2D methods using building footprints and floor areas, as well as how it is affected by different levels of semantic detail in a 3D city model. We conclude that 3D city models are useful for estimations of large areas (e.g. for a country), and that the 3D approach has clear advantages over the 2D approach.

The paper is Open Access.


j j j

The variants of an LOD of a 3D building model and their influence on spatial analyses

The level of detail (LOD) concept conveys the grade of 3D city models, however, it still allows flexibility for different modelling choices. For instance, consider the following four (valid) variants of LOD1:



These variants, which we term geometric references, are a topic of our new paper which has been published in the ISPRS Journal of Photogrammetry and Remote Sensing:

Biljecki, F., Ledoux, H., Stoter, J., & Vosselman, G. (2016). The variants of an LOD of a 3D building model and their influence on spatial analyses. ISPRS Journal of Photogrammetry and Remote Sensing, 116, 42–54. doi:10.1016/j.isprsjprs.2016.03.003

The freely available author’s version PDF is available here. Please use the publisher’s version if available to you.


Abstract: The level of detail (LOD) of a 3D city model indicates the model’s grade and usability. However, there exist multiple valid variants of each LOD. As a consequence, the LOD concept is inconclusive as an instruction for the acquisition of 3D city models. For instance, the top surface of an LOD1 block model may be modelled at the eaves of a building or at its ridge height. Such variants, which we term geometric references, are often overlooked and are usually not documented in the metadata. Furthermore, the influence of a particular geometric reference on the performance of a spatial analysis is not known.

In response to this research gap, we investigate a variety of LOD1 and LOD2 geometric references that are commonly employed, and perform numerical experiments to investigate their relative difference when used as input for different spatial analyses. We consider three use cases (estimation of the area of the building envelope, building volume, and shadows cast by buildings), and compute the deviations in a Monte Carlo simulation.

The experiments, carried out with procedurally generated models, indicate that two 3D models representing the same building at the same LOD, but modelled according to different geometric references, may yield substantially different results when used in a spatial analysis. The outcome of our experiments also suggests that the geometric reference may have a bigger influence than the LOD, since an LOD1 with a specific geometric reference may yield a more accurate result than when using LOD2 models.

j j j

Speed of publication of 18 GIS journals (publication delay)

TL; DR: I have analysed the submitted/accepted/online dates for 1023 papers published in 18 GIScience journals to determine the median lag from submission to publication for each journal. Several ranked lists of journals are presented as plots. On average it takes 6.9 months to get a paper reviewed and published online since submission, and about 2.8 more months to have it included in an issue.

Edit on 18 Sep 2015: this blog post attracted substantial interest (thanks everyone for spreading it around). In the meantime two new journals have been added to the analysis.

Edit on 18 Jan 2016: I have published a related (scientometric) study in IJGIS which investigates many more different aspects related to GIScience publishing.


Scholarly publishing can be an annoying process due to an often slow peer-review stage, editorial decision, and inert typesetting procedure. This processing time is known also as publication delay: the chronological distance between the moment a paper is submitted and its publication (Amat, 2009). It can take several months (or years?) to see one’s paper published in an issue of the journal. Naturally, publishing great papers takes time, but long publication delays are frequently caused by the inefficiency of journals and publishers (not to mention reviewers), and are frowned upon by authors. They have been cited to considerably influence one’s decision where to submit a paper (Strevens 2003, Dong et al 2006, Carroll 2001, Swan and Brown 2003, Rousseau and Rousseau 2012, Solomon and Björk 2012). Therefore it comes as no surprise that publishers started to boast about their processing times.

Publication delay is a principal topic in scientometrics, with many analyses and papers written on this topic. Long story short, while a long publication delay is a source of frustration for authors, actually it can benefit journals to boost their impact factor (Tort et al, 2012), hence the interest from the scientometric community. The goal of this analysis is to expose the publication delay of journals in GIScience (Geographical Information Science), as done in similar analyses, e.g. analysis for PLOS journals, and various disciplines.


There is not a firm consensus on the measures about the publication delay. Some researchers consider the publication delay the time from submission to publication, and some distinguish the publication online and publication in an issue. Some people consider only the delay from the paper available online to the date when it becomes paginated and in print (allocated to a volume, assigned with page numbers, and distributed to readers).

In the digital age the publication in issue is becoming less and less important, and to me (as author) what matters most is the moment the paper becomes available online. However, when the paper is published in print is still the authoritative epoch for determining the impact factor, hence it is of relevance to scientometrics.

Let’s consider all the aforementioned measures:

  • A: submission to acceptance (includes peer review, editorial decision, and revision; often in multiple rounds).
  • B: acceptance to published online; i.e. online posting (includes proof reading and typesetting the paper by the publisher, among other things).
  • C: published online to published in an issue (the period the paper is available online but not yet part of an issue with page numbers and not yet distributed to readers).
  • PD: the publication delay (from submitted to the journal to published online; A+B).
  • TPD: the total publication delay (from submitted to published in an issue; A+B+C).

The following figure illustrates the typical process of publishing, with the considered lags.


Data has been acquired from publishers’ records. 18 GIScience journals have been considered for this analysis.  These are the journals I usually consider when preparing a paper, with some additions for added diversity. The list is a subset of the extensive list compiled at my department. The selected journals are listed below (alphabetically):

  1. AAG: Annals of the Association of American Geographers (Taylor and Francis)
  2. CaGIS: Cartography and Geographic Information Science (Taylor and Francis)
  3. CEUS: Computers, Environment, and Urban Systems (Elsevier)
  4. C&G: Computers and Geosciences (Elsevier)
  5. EPB: Environment and Planning B: Planning and Design (Sage) – date of acceptance is missing
  6. GEAN: Geographical Analysis (Wiley)
  7. GEIN: Geoinformatica (Springer)
  8. G&RS: GIScience and Remote Sensing (Taylor and Francis)
  9. JAG: International Journal of Applied Earth Observation and Geoinformation (Elsevier)
  10. IJDE: International Journal of Digital Earth (Taylor and Francis)
  11. IJGI: ISPRS International Journal of Geo-Information (MDPI)
  12. IJGIS: International Journal of Geographical Information Science (Taylor and Francis)
  13. P&RS: ISPRS Journal of Photogrammetry and Remote Sensing (Elsevier)
  14. JGS: Journal of Geographical Systems (Springer)
  15. JSS: Journal of Spatial Science (Taylor and Francis)  – not fully transparent (does not state the submission and acceptance date)
  16. PE&RS: Photogrammetric Engineering & Remote Sensing (ASPRS) – date of online posting is missing
  17. SCC: Spatial Cognition and Computation (Taylor and Francis)
  18. TGIS: Transactions in GIS (Wiley) – not fully transparent (does not state the submission and acceptance date)

All articles published in 2014 have been selected, except those published in special issues. This results in 1023 papers included in the analysis, enough to draw a solid conclusion about the publication delay.

Luckily most publishers are fully transparent about the chronological record of each paper (although for some papers the data was missing). However, the analysis for is limited for some journals because their publishers do not state the dates of the submission and acceptance of each paper:


Not enough information to derive the publication delay; only C can be calculated. Example of Wiley / Transactions in GIS


Because of the multiple metrics, multiple ranking lists of journals can be derived, and there are multiple aspects to look at.

The results are shown graphically, with a table with all the data in the end. The plots show the distribution of individual papers in each journal. The small white point denotes the median, and the thick stroke the interquartile range.

Let’s start with the A: the median time from submission of the paper to its acceptance per journal, i.e. peer review:

Exhibit A: Time from the submission of the paper to its acceptance. Entries are sorted by median. (Violin plot made possible by Matplotlib and Seaborn, and inspired by Daniel Himmelstein.)

The ISPRS International Journal of Geo-Information is a clear winner here with a median of 79 days: less than 3 months from a paper submission to acceptance. No wonder this fairly new journal is rapidly gaining popularity. The plot also shows that there is a substantial difference between journals.

After a paper is accepted, it takes some time to get it published online (B):

Exhibit B: Delay of online posting after acceptance

The ranking differs from the previous one, and Computers & Geosciences emerges as the quickest one (median of 11 days). This can be explained by the practice, that upon acceptance, the journal posts online the accepted manuscript while the final publisher-formatted version is being prepared. This is not the final version of the paper, but having the accepted version available in the meantime is a favourable measure to alleviate publication delay. Kudos to those journals that make their papers accessible as soon as possible.

AAG and GEAN do not fit the plot due to their pitiful performance: their medians are 7.3 and 14.7 months, respectively. Yes, it took Geographical Analysis more than a year to publish an already accepted paper. Shameful. However, it seems that this practice was not continued in 2015.

The lag from submission to publication online of a paper (A+B) is what most people care about (PD: publication delay). Another ranking summing the two above metrics is exposed:

Exhibit C: Publication delay (from submission to online publication) of the considered GIS journals

IJGI is first with a median of 92 days, JAG comes second with 161 days, and C&G is third with 186 days. Again, a substantial discrepancy between journals is exposed, ending with Geographical Analysis with a performance of almost two years.

The presented three plots also reveal something else of interest: it seems that some journals have a limit on the duration of the process, e.g. for IJGI and JAG almost no paper took more than approx. 6 and 12 months, respectively, to get published. Furthermore, their narrow distributions reveal a consistent process. For other journals, some papers seem to be stuck in peer review much much longer than others.

After a paper is accepted and published online most authors don’t care anymore. But we are not done yet: the following plot shows the lag from publication online to publication in an issue and print (C). A long online-to-print lag can artificially boost a journal’s impact factor, so technically, editors may hold a paper online for a very long time to manipulate their impact factor.

Exhibit D: Time from online posting to publication in print

Again, another ranking is exposed: with GIScience & Remote Sensing, Cartography and Geographic Information Science, and Annals of the Association of American Geographers being the quickest three when it comes to the time between online posting and publication in print. IJDE does not fit the plot owing to its slow reaction, with the median of 18.6 months. The story with GEAN is a bit different. Apparently the papers are published online only when they are published in the issue, so technically C is always zero. Of course this poor practice is reflected in the very long B. Note that this plot debuts JSS and TGIS, the latter not being particularly quick. However, the impact factor of TGIS has recently increased by nearly 40%. Coincidence?

The C (online-to-print) delay, while overlooked by most authors, appears to be exploited by journals to boost their impact factor. For instance, JAG, which has recently increased its impact factor by 36.7%, at the time of this blog post (September 2015) has already its February 2016 issue almost ready with a dozen papers:

2015-09-06 at 18.05

I don’t see a reason why to hold unpaginated papers online for months, except for impact factor related reasons. Not to mention the lack of other explanation behind publishing a 2016 issue when we are still chronologically far from it.

Finally, the total publication delay (TPD) is given below (from submitting the paper to get it published in issue):

Exhibit E: Total publication delay (time from paper submission to publication in print/issue)

As illustrated in the plot, a paper submitted to a GIS journal can take anywhere from a few months to a few years to see it published in an issue.

The results for some journals are stimulating and for some are appalling, prompting me to reconsider my list of GIScience journals. The key takeaways:

  • IJGI is the quickest journal from submission to online posting, followed closely by JAG and C&G (or G&RS and CaGIS if you consider the complete cycle). MDPI (publisher of IJGI) is definitely disrupting the GIS publishing scene, for the better.
  • Assuming that the records are correct, Geographical Analysis takes 15 months to publish a paper after acceptance. Ridiculous.
  • Within a journal there may be considerable differences (e.g. IJGIS exhibits a significant variation), and one cannot always accept the medians as a guarantee when submitting a paper, rather as a guideline.
  • Some editors appear to artificially boost impact factors by deliberately holding papers online for a long time (in press without pagination), e.g. International Journal of Digital Earth takes 18.6 months to assign a paper to an issue, the slowest of all analysed journals. Coincidence or not, IJDE has seen the highest increase of the impact factor in GIS.
  • Most papers in GIS take around 7 months to get published, but again, there are differences:

Distribution of the publication delay of GIS journal papers. Joint data for all journals.

Finally, here are the medians in a table. First the three metrics A, B, C, and the publication delays PD and TPD.

Journal A B C PD TPD
AAG 8.0 7.3 1.4 16.9 17.9
CaGIS 4.9 1.3 1.3 6.3 8.3
CEUS 8.5 0.9 2.3 9.4 11.8
CG 5.6 0.4 2.9 6.1 9.0
EPB 3.7 20.5 24.0
GEAN 11.5 14.7 0.0 23.3 23.3
GEIN 8.6 1.2 8.4 9.6 17.9
GRS 4.6 1.5 1.1 6.6 7.3
IJDE 7.0 0.9 18.6 7.6 27.7
IJGI 2.6 0.5 1.8 3.0 4.1
IJGIS 6.5 1.4 3.8 8.5 12.3
JAG 4.0 1.0 4.9 5.3 10.6
JGS 9.9 0.8 5.9 10.7 15.8
JSS 1.3
PERS 6.1 11.4
PRS 5.8 1.0 1.6 7.0 8.4
SCC 5.6 0.5 2.2 5.9 9.1
TGIS 10.5
All papers 5.6 1.0 2.8 6.9 10.7


And here is a stacked bar plot with the medians of the three components (A, B, C):

Decomposing the metrics A, B, and C (medians) for the selected journals. The sum of these metrics corresponds to the total publication delay (TPD), however, the sum of medians may slightly deviate from the median of TPD.

Decomposing the metrics A, B, and C (medians) for the selected journals. The sum of these metrics corresponds to the total publication delay (TPD), however, the sum of medians may slightly deviate from the median of TPD.

The last line of the table above exposes the median of all analysed GIS papers. How does that compare to other disciplines? Björk and Solomon (2013) conducted an extensive cross-disciplinary experiment. I incorporated my findings in their data:

Comparing the publication delay in GIS with other disciplines: could be better. Data source: doi:10.1016/j.joi.2013.09.001


Comparing the results reveals that the total publication delay of GIS is exactly at the average of the considered disciplines (slightly longer than one year). However, the metric A is longer than the cross-disciplinary average. Furthermore, the delay found in GIS is considerably longer than the average of engineering, but not far from earth sciences, to which GIS could be partly assigned. The longer delay may also be explained by the fact that some GIS journals are closer to social sciences, which rank among the disciplines with the longest publication delay.

Now let’s examine individual papers. What are the fastest and slowest papers? Here are some extremes:


I didn’t know it was possible to get a paper accepted the same day (ironic considering that GEAN is the slowest journal in the sample; note that it took it one and half year to publish it after acceptance). An explanation is that it could be a paper resubmitted to the same journal after a prolonged revision. Our champion is followed by a few papers accepted within a month:







On the other side of the axis, we have a Geoinformatica paper that took more than 4 years from its submission to be published in an issue. Mind you, it takes less time to complete a PhD.



Finally, an interesting thing to explore is the relation between the following two measures:

  • A, which includes a possible delay by authors, peer-reviewers, and editors; and
  • B, which includes a possible delay by authors, editors, and publication office.

They are not so much related (weak correlation of 0.20).

(Cor)relation between A and B


Publishing in GIS can be slow. Choose your journal wisely.

References and further reading

Amat, C. B. (2009). Editorial and publication delay of papers submitted to 14 selected Food Research journals. Influence of online posting. Scientometrics, 74(3), 379–389.

Björk, B.-C., & Solomon, D. (2013). The publishing delay in scholarly peer-reviewed journals. Journal of Informetrics, 7(4), 914–923.

Carroll, R. J. (2001). Review times in statistical journals: Tilting at windmills? Biometrics, 57(1), 1–6.

Dong, P., Loh, M., & Mondry, A. (2006). Publication lag in biomedical journals varies due to the periodical’s publishing model. Scientometrics, 69(2), 271–286.

Rousseau, S., & Rousseau, R. (2012). Interactions between journal attributes and authors’ willingness to wait for editorial decisions. Journal of the American Society for Information Science and Technology, 63(6), 1213–1225.

Strevens, M. (2003). The role of the priority rule in science. The Journal of Philosophy, 100(2), 55–79.

Solomon, D. J., & Björk, B.-C. (2012). Publication fees in open access publishing: Sources of funding and factors influencing choice of journal. Journal of the American Society for Information Science and Technology, 63(1), 98–107.

Swan, A., & Brown, S. (2003). Authors and electronic publishing: what authors want from the new technology. Learned Publishing, 16(1), 28–33.

Tort, A. B. L., Targino, Z. H., & Amaral, O. B. (2012). Rising Publication Delays Inflate Journal Impact Factors. PLOS One, 7(12), e53374.

Trivedi, P. K. (1993). An analysis of publication lags in econometrics. Journal of Applied Econometrics, 8(1), 93–100.

Woolston, C. (2015): Long wait for publication plagues many journals. Nature, 523(7559), 131.

Disclaimer and further information:

  1. Data has been prepared with care, but I cannot guarantee that it’s 100% correct as I noticed many errors in the publisher records, e.g. date of acceptance is 2 years before the date of submission, which I filtered out, but other errors could have passed unnoticed.
  2. Many predatory and low-quality journals are very quick in publishing papers. With this analysis I don’t want to imply that there is a correlation between publication delay of a journal and its quality. All of the analysed journals are ISI journals so a reasonable degree of quality is ensured.
  3. The presented data is an aggregation of hundreds of records, so it goes without saying that your mileage may vary: this is no guarantee that your submission will fall into these averages.
  4. The date of the publication in issue is somewhat ambiguous in scientometrics. For instance, an issue may bear the date of March, but that can deviate from the actual of distribution: it could have been distributed in February or April. Moreover, some journals prepare issues well in advance, and keep them filling until the cover month. Therefore, for the authoritative date I have taken the first day of the month of the time on the cover  (i.e. 1 March). An exception are JSS and IJGI, as it is obvious it releases the issues at the end of the month or later; those have been adjusted to the actual release date as I have received them.
  5. Many papers are delayed because of authors (e.g. long time to revise the paper), but scientometric researchers assume this is uniform for all journals, so it’s not specifically elaborated.
j j j

Analysis of the new (2015) Impact Factors of GIS journals

[Update: for the impact factors released in June 2016 read the new blog post]

Thomson Reuters has recently released the 2015 edition of the Journal Citation Reports, a publication that contains the impact factors of journals they index.

I have compared the new impact factors to the previous ones (from the 2014 edition of JCR) for several GIS journals indexed in Web of Science. The result is shown below. Both the relative and absolute IF change are included.

Old vs. new impact factors for several GIS journals

Old vs. new impact factors for several GIS journals

It’s interesting to see that almost all GIS journals have seen the increase of their impact factor, some of them considerably: Cartography and Geographic Information Science has almost doubled it!

ISPRS International Journal of Geo-Information is the newest addition to Web of Science, but it is too recent to have an impact factor. According to the editors of the journal, the JCR that will be published in July 2016 will contain the IF for IJGI.

Please note that this is my personal list of indexed GIS journals, and it may differ from your preferences. For a more comprehensive list please see the internal list of my group.

Update: Martin Tomko has a related analysis about the temporal trends in Google Metrics of GIScience periodicals.

j j j

A paper in IJGI about the 3D+LOD (4D) integration

I am relaying the news from my group’s blog:

We are happy to announce that the paper Modeling a 3D City Model and Its Levels of Detail as a True 4D Model, authored by Ken Arroyo Ohori, Hugo Ledoux, Filip Biljecki and Jantien Stoter, has been published in the ISPRS International Journal of Geo-Information.

The topic of the paper is investigating and developing a three-step approach to model the level of detail (LOD) as an extra geometric dimension perpendicular to the three spatial ones. The result is a true 4D model in which a single 4D object (a polychoron) represents a 3D polyhedral object (e.g., a building) at all of its LODs and a multiple-LOD 3D city model is modeled as a 4D cell complex.

The paper (open access) can be accessed here.

ijgis-4d-1 ijgis-4d-2

j j j

A new GIS journal: Open Geospatial Data, Software and Standards

I am relaying the news about Open Geospatial Data, Software and Standards, a recently established open-access journal from Springer.

From its website:

Open Geospatial Data, Software and Standards provides an advanced forum for the science and technology of open data, crowdsourced information, and sensor web through the publication of reviews and regular research papers. The journal publishes articles that address issues related, but not limited to, the analysis and processing of open geo-data, standardization and interoperability of open geo-data and services, as well as applications based on open geo-data. The journal is also meant to be a space for theories, methods and applications related to crowdsourcing, volunteered geographic information, as well as Sensor Web and related topics.

For an extensive list of GIS journals, please see our inventory.

j j j

Release of Solar3Dcity, a utility to estimate the yearly solar irradiation of buildings stored in CityGML

In the last a few months I had been busy with examining the theory of solar radiation and the estimation of the solar irradiation of roofs with 3D city models. Estimating the solar irradiation of buildings with 3D city models is one of the prominent use-cases of 3D GIS, and it is used to assess the feasibility of installing a photovoltaic panel on a roof (i.e. how much sun energy would a panel get over a year if installed).

I have been doing this in order to build my own software. Now I am happy to announce the release of Solar3Dcity, an open-source tool for estimating the yearly solar irradiation of roofs from 3D city models stored in CityGML.

Screenshot of the results of the Solar3Dcity estimations on 36 buildings in CityGML

Screenshot of the results of the Solar3Dcity estimations on 36 buildings in CityGML

I have decided to create my own software since software packages that are used nowadays are not free, and usually do not support CityGML, which is one of the primary formats of 3D city models.

Long story short, the software extracts the roof surfaces from CityGML buildings. It computes the tilt, orientation, and area of each surface. The tilt and orientation (azimuth) of a surface have a big influence on the radiation, as it’s obvious from the following plot (which was also generated with the Solar3Dcity package):

The tilt-orientation factors computed by Solar3Dcity for Delft in the Netherlands. This plot shows how drastic the influence of the tilt and orientation of the roof can have onto the received solar energy.

The tilt-orientation factors computed by Solar3Dcity for Delft in the Netherlands. This plot shows how drastic the influence of the tilt and orientation of the roof can have onto the received solar energy. This plot can be computed for any location on Earth since Solar3Dcity utilises a global weather database.

The above location-dependent plot is a key to do the estimations (the software just samples the value from it), but it takes some time and effort to obtain it. In short, the solar radiation is a three-component function over time (due to the different position of the sun every day; and reflections and refractions of the sun rays), and its values have to be integrated over the whole year. Further, the cloud cover has to be taken into account to adjust the estimations. The next plot shows the solar radiation for Delft during two days (1 Mar and 21 Jun) for three differently oriented and tilted surfaces (A, B, and H). Big difference…


This experimental research software is in the development phase, but I had a chance to compare its results with a commercial software and it seems accurate. So far I have used it for investigating the propagation of acquisition errors, about which I am currently submitting a paper.

Please head to the Github page for more information about this tool.

Edit: In the meantime I have created an animation:


Edit2: a paper has been published:

Biljecki, F., Heuvelink, G. B. M., Ledoux, H., & Stoter, J. (2015). Propagation of positional error in 3D GIS: estimation of the solar irradiation of building roofs. International Journal of Geographical Information Science. doi:10.1080/13658816.2015.1073292

j j j

Random3Dcity – the first CityGML procedural modelling engine and multi-LOD building generator


I am happy to announce the release of Random3Dcity, an open-source experimental CityGML procedural modelling engine that I have developed within my PhD research at the 3D geoinformation group at the Delft University of Technology. It is an experimental software prototype that was created for research purposes, but potential applications are not limited to it.

I have built Random3Dcity entirely from scratch with a custom grammar, and implemented it in Python. The source is available on Github. A prepared collection of sample datasets is available on the datasets page, with extensive technical details, so I invite you to visit it if you are interested in further details and/or interested in the data without the need to run the software.

This is the first procedural modelling engine that generates buildings and other features in CityGML, and one that is designed to do so in multiple levels of detail. The engine generates buildings according to a novel series of 16 refined levels of detail (“Delft LODs”) that I have developed during my research on this topic:

This specification will be detailed soon in a research paper that is currently under submission. The program supports five types of roofs:


The number of unique buildings is virtually unlimited, and the datasets are suited for a number of application domains, from error propagation analysis to the testing of validation and repair software.

Random3Dcity supports interior (see the image in the header), and also vegetation and roads:


Further, the engine generates different geometric references within each LOD (e.g. LOD2.0 with the walls at their actual position and another [photogrammetric] LOD2.0 with the walls as projections from the roof edges), and different geometry (solid vs. b-rep). This results in almost 400 representations of a building. I believe that this is the most thorough CityGML dataset available to date. Solids are assembled by using the surfaces that define the usable volume of the building:


The composite rendering below shows an example dataset of 100 buildings in four LODs.


A research paper is under submission to the journal Computers, Environment, and Urban Systems, describing the engine and the refined LOD specification. I will update this post when the paper becomes available. If you are interested in using the engine, please contact me to give you the reference to cite.

For more information about this project please head to my personal page. Please let me know if you encounter a bug and/or have a suggestion. Note that this is an experimental software under continuous development.

As a bonus, check a video of a sample dataset of 10000 buildings:

Happy CityGML-ing!

j j j

CityGML2OBJs – a robust semantic-aware tool for converting CityGML to OBJ


I have released the code of CityGML2OBJs, a tool for converting CityGML files to OBJ. The source and manual of the code can be found on Github. The methodology is published in a research paper:

Biljecki, F., & Arroyo Ohori, K. (2015). Automatic Semantic-preserving Conversion Between OBJ and CityGML. Eurographics Workshop on Urban Data Modelling and Visualisation 2015, Delft, Netherlands, pp. 25-30. [PDF] [DOI]

OBJ is probably the most supported 3D format, and converting your CityGML files to OBJ opens a door to a large number of software packages and use(r)s.

Besides the conversion, the tool features some additional options reflected through the suffix “s” in its name:

  • semantics–decoupling of thematically structured surfaces in CityGML and converting them into separate OBJs (that’s where the “OBJs” in the name come from). For instance, below you can see the rendering of an OBJ that contains only WallSurface polygons.sem-tri
  • structured objects–separation and storage of buildings into multiple objects in OBJ by structuring faces that belong to a building into the same group.
  • “see” the attributes from a CityGML file–the utility converts quantitative attributes into colours to support their visualisation. OBJ does not really support the concept of attributes, hence if the CityGML file contains an attribute, this is generally lost in the conversion. However, this converter is capable of converting a quantitative attribute to OBJ as a texture (colour) of the feature. For instance, if the attribute about the yearly solar irradiation is available for each polygon in the CityGML file, it is converted to a graphical information and attached to each polygon as a surface, so now you can easily visualise your attributes in CityGML.
  • sturdy–it checks polygons for validity, considers different forms of geometry storage in GML (e.g. gml:pos and gml:posList), detects for lack of boundary surfaces, etc.

Other features:

  • Supports polygon holes by triangulating all surfaces. Besides the holes, this is done by default because some software handles OBJs only if the faces are triangulated, especially when it comes to the texture, so not only holey polygons are triangulated.
  • It re-uses repeating vertices, resulting in a reduced file size and redundancy
  • Batch processing of multiple files

Visit the Github page, and please let me know if you encounter a bug and/or have a suggestion.

j j j