In 2011, the number of smartphone users in the US was closing in on 100 million, making up about 40% of the mobile phone market. Most late-model smartphones come embedded with megapixel-resolution cameras and GPS technology. Additionally, there are a plethora of mobile applications (apps) running location-based services (LBS) such as Twitter, Geoloqi and Foursquare. In fact, several websites, including Facebook and Google, have enabled users to perform online “check-ins” on mobile versions of their websites. To illustrate the potential of crowdsourcing, one such company (Foursquare), mapped user checkins between Halloween and Christmas of 2010. When I saw that map, the first thing that sprung to my mind was “validation”, and by that I don’t mean validating a person’s whereabouts at a particular moment in time (although this has been used by tech-savvy burglars). I am referring to validating the accuracy of land use and land cover (LULC) maps. Think about it: millions of people using LBS apps, checking in, commenting (e.g. “Just landed in JFK” or “There used to be a park here last year“), and taking photos (many with captions, e.g. “Park Avenue, facing west“). Furthermore, the recent advent of GPS-enabled point-and-shoot cameras such as Nikon’s AW100 and the Fujifilm FinePix XP150 have further expanded this opportunity. At any one time there are hundreds, if not thousands, of individuals using either their cameras or smartphones to take pictures of their surroundings. Most of the time, these photographs are posted on websites such as Flickr (which has been transformed into a social media site and contains 3.5 billion photos), Twitter’s native photo platform, Twitpic, or yfrog, allowing for spatial data-mining opportunities using an application programming interface (API). This means there is a wealth of geolocated and time-stamped, photographs and “checkins” that can be used to validate local, regional, and even global LULC maps.
Click! This geolocated & time-stamped photo can be used to validate a land use and land cover map of the area.
Having said that, this opportunity also has important limitations. For one, the positional accuracy of these consumer products often leaves a lot to be desired. In most cases users have to wait some time until the unit acquires enough satellites (or in the case of smartphones, cell towers) to provide acceptable location information. Zandbergen (2009) discusses this in detail, and concluded that A-GPS locations from the iPhone 3G had a root mean square error (RMSE) of 8.3 meters with a maximum error value of 18.5m. This means that positional accuracy of these datasets needs to adequately scrutinized prior to using them for validation purposes. Additionally, the choice of sampling unit for use in the accuracy assessment process is another factor that decides whether to use social media for validation. For instance, if a single ETM+ pixel (30m) is used as a sample unit, and considering that geometric accuracy of ETM+ is between 50 and 100m, then using the above mentioned iPhone 3G might be feasible. However, if the pixel is from a higher resolution satellite, say IKONOS (4m), with an accuracy of 4m, then data with an 8m RMSE are inadequate for validation purposes because it will adversely influence the thematic accuracy of maps.
At present, I know of just one paper (Fritz et al 2009) that specifically addresses crowdsourcing as a method to improve LULC maps. The platform presented in that paper, Geo-Wiki, is a brilliant concept that needs to be expanded, improved on and simplified further.
Abdulhakim Abdi
References
- Zandbergen, P. A. (2009). Accuracy of iPhone locations: A comparison of assisted GPS, WiFi and Cellular Positioning, Transactions in GIS, 13(s1):5-26.
- Fritz S., McCallum I., Schill C., Perger C., Grillmayer R., Achard F., Kraxner F., Obersteiner M. (2009). Geo-Wiki.Org: The Use of Crowdsourcing to Improve Global Land Cover. Remote Sensing, 1(3):345-354.
