CrowdVal

The collection of in-situ data for training and validation of land cover products is expensive, e.g. the LUCAS exercise for 2018 will cost more than 12 million euros in total for the collection of land cover and land use at around 300K locations across the EU. Increasingly, training and validation data are collected through visual interpretation of satellite imagery but Landsat and Sentinel-2 are still too coarse for identifying many important land cover features. At the same time, very high resolution imagery from commercial providers is too expensive and it is not viable to purchase the imagery for large areas, e.g. for Africa. Crowdsourcing is now starting to be used to validate land cover and to develop reference data sets using tools such as Geo-Wiki and LACO-Wiki (Fritz et al. 2012; See et al., 2015, 2017). This approach has the potential to collect large amounts of data, both in situ and using online visual interpretation, which can be used to validate ESA/Copernicus products and provide training data for the development of new products. The use of crowdsourcing has yet to be fully exploited by ESA so this project represents one investigation in this direction. 

The overall aim of the CrowdVal project is to enhance existing crowdsourcing tools for in situ data collection, i.e., LACO-Wiki Mobile, and for online visual interpretation of satellite imagery, i.e., LACO-Wiki (https://laco-wiki.net) with the explicit purpose of collecting validation data for land cover/land use products and training data for the development of new products. The additional functionality includes: optimization of sampling strategies, taking road networks  into account; opportunistic data collection, i.e. seeing the land cover map on the phone as you are in the field or driving along a road and correcting the land cover if it is incorrect; and the addition of auxiliary data sets such as geo-tagged photographs from the internet (e.g., Flickr) and NDVI (Normalized Difference Vegetation Index) time series from different sensors. The enhanced crowdsourcing tools have been demonstrated through a number of planned data collection exercises, where the data hve been used to validate the first 20 m land cover product for Africa and collect training and validation data.

LACO-Wiki Mobile has now been made open source (in github) so that it can be used by the European Space Agency (ESA) as well as other companies or organizations for in situ data collection.

Objectives of the Project

• To develop new innovative sampling schemes that allow a stratification and bias  removal via road networks and that take other constraints into account for in-situ data collection
• To enhance LACO-Wiki and LACO-Wiki Mobile with the new sampling strategies, functionality for opportunistic map evaluation on the ground and additional of auxiliary data sets including Flickr geo-tagged pictures and time series of NDVI
• To create an open source version of LACO-Wiki Mobile
• To demonstrate the enhanced tools through crowdsourcing data collection campaigns (online and in-situ) to validate the first land cover map of Africa at a 20m spatial resolution
• To investigate the possibility of developing a business model around an open source version of LACO-Wiki Mobile with a payment model around access to enhanced features, e.g. additional data sources, commercial satellite imagery, increased sample size, etc. 

  Achievements to Date

  Three workshops were held: (1) one at RCMRD in Kenya at the end of September 2018; (2) one in Gabon early December 2018; and (3) one in Abidjan, Ivory Coast. All of the workshops involved training volunteers in the collection of data using LACO-Wiki and LACO-Wiki Mobile. The results have been used to undertake an accuracy assessment of the ESA CCI African 20m land cover product including maps of spatial accuracy.

  We have also made presentations at three conferences: the ESA Phi-Week in 2018; the EGU 2019 meeting in Vienna, Austria; and the ESA Living Planet Symposium in May 2019 in Milan, Italy. 
  

Last edited: 30 August 2019