Generating WUDAPT Level 0 data – Current status of production and evaluation

TLDR: The protocol by which LCZ maps generated by different members of the community are produced and evaluated is outlined, which supports the assumption that the current level 0 products are already of sufficient quality for certain applications.

Abstract: The World Urban Database and Access Portal Tools (WUDAPT) project has grown out of the need for better information on the form and function of cities globally. Cities are described using Local Climate Zones (LCZ), which are associated with a range of key urban climate model parameters and thus can serve as inputs to high resolution urban climate models. We refer to this as level 0 data for each city. The LCZ level 0 product is produced using freely available Landsat imagery, crowdsourced training areas from the community, and the open source SAGA software. This paper outlines the protocol by which LCZ maps generated by different members of the community are produced and evaluated. In particular, the quality assessment comprises cross-validation, review, and cross-comparison with other data sets. To date, the results from the different quality assessments show that the LCZ maps are generally of moderate quality, i.e. 50–60% overall accuracy (OA), but this is much higher when considering all built-up classes together or using weights that take the morphological and climatic similarity of certain classes into account. The training data contributed by researchers from around the world also vary in quality and in the interpretation of the landscape, which affects the final quality of the LCZ maps. The acceptable level of quality needed will depend heavily on the application of the data. However, initial modelling studies that use the level 0 products as inputs showed improved performance in simulating the urban climate when replacing the default surface descriptions with the WUDAPT level 0 data. This is also promising for the application of level 0 data in regional and global climate and weather models and supports the assumption that the current level 0 products are already of sufficient quality for certain applications. Moreover, there are various ongoing developments to improve the methods used to produce LCZ maps and their accuracy.

Frequently asked questions

Q1. What are the contributions in "Generating wudapt level 0 data – current status of production and evaluation" ?

The World Urban Database and Access Portal Tools ( WUDAPT ) project has grown out of the need for better information on the form and function of cities globally. This paper outlines the protocol by which LCZ maps generated by different members of the community are produced and evaluated. This is also promising for the application of level 0 data in regional and global climate and weather models and supports the assumption that the current level 0 products are already of sufficient quality for certain applications. 

Q2. How many sites could be clearly assigned to specific LCZ classes?

Approximately 50 sites could be clearly  assigned  to  specific  LCZ  classes  that  are present  in  the Augsburg  region  (LCZs  2,  5,  6,  8, A, B, D). 

Q3. What is the LCZ methodology used in the case of Tandil city, Argentina?

Preparing urban climate maps using the LCZ methodology for  improving  communication with  urban  planners:  the  case  of  Tandil  city, Argentina. 

Q4. What is the reference for Europe?

A good reference  for Europe  is  the EEA  (European  Environment  Agency)  soil  sealing  data  set, which  is  freely  available  at  a  20 m  and  100 m  resolution. 

Q5. What were the quality measures used to evaluate the LCZ maps?

The  consistency  and  accuracy  measures  used  to  evaluate  the  quality  indicated  that  both  the  quality  of  the  TA  data  sets  and  the  resulting LCZ maps were poor to moderate. 

Q6. What was the main approach used to train the classifiers?

Finally,  a  combination  of  the  co‐training  approach with other classifiers, such as a support vector machine (SVM), was used to further enhance the  final classification accuracy. 

Q7. What is the impact of the operators on the classification accuracy?

The  impact  of  the  operators  on  the  classification  accuracy  was  found  to  be  considerable  based  on  experiences  from  the HUMINEX, which  stresses  the necessity of  strict quality  assessment  and  review. 

Q8. What is the process of releasing a LCZ map?

Before  the  LCZ  maps  are  released  to  the  wider  community via the WUDAPT portal, they must undergo quality assessment and are then published with  metadata related  to  the quality achieved. 

Q9. How many TA sets should be used for a city?

From these results, the HUMINEX concludes that at  least  ten  individual TA sets  from untrained operators should be used  for one city  to produce an LCZ map of  good quality, although this aspect needs further investigation (Bechtel et al., 2017b). 

Q10. What is the advantage of using online processing platforms?

Such  online  processing  platforms  can  also  facilitate  implementation  of  the  online  cross‐ validation bootstrapping procedure (see section 3.1), and provide the ability to scale‐up from single cities  over countries to continents. 

Q11. Where are the LCZ maps and metadata made available?

The  LCZ maps  and  metadata  are  made  accessible  via  the  the  WUDAPT  portal  (https://wudapt.cs.purdue.edu/),  where  several additional  tools are available. 

Q12. What is the way to evaluate LCZ maps?

For example, the LCZ maps  might be resampled to a coarser resolution for input into global atmospheric models and be less sensitive  to  accuracy  at  the  100 m  resolution. 

Q13. What are the main components of the LCZ quality assessment procedure?

In  particular  the  bootstrapping  cross‐validation  and  the  review were  described,  which  are  currently  part  of  the  standard  LCZ  quality  assessment  procedure.