Alternative methods to validate the Terra-i system data: evaluating imagery sources from Google Earth and Landsat Viewer

Figure 1. Anthropogenic factors such as the colonization model, illegal mining, logging of trees and the expansion of the agricultural frontier, as a consequence of illicit crop production, are some of the causes of deforestation in the Caquetá study area (Colombia). SOURCE: Natura

Land cover change and land use maps generated by the Terra-i tool provide valuable information to users. These maps are vital in understanding and predicting the effects of anthropogenic interactions on the environment which may extend across local, regional and global scales [1] and are thus crucial for decision-making processes and policy design. For these reasons the generation of spatial information must be consistent and accurate to correctly analyze phenomena such as deforestation.

Therefore, it is important to assess the accuracy of different data sources [2]. There are different methods available to validate satellite data including field corrections which are accurate but time expensive, and less time expensive methods such as the use of imagery with high and/or medium resolution e.g. QuickBird (2.4m) and Landsat (30m).

Methodology
An approach was designed based on the visual interpretation of satellite images (high and/or medium resolution) such as QuickBird and Landsat provide by Google Earth and Landsat Viewer (http://landsatlook.usgs.gov/) respectively in the study area of Caqueta, Colombia (Figure 2). Essentially, the methodology, based on the reference [3], was to visually identify "hotspots" areas of land cover change by exploring high/medium resolution images which correspond to detections identified by the Terra-i system. Further information about the workflow is shown in Figure 3.

Figure 2. Study area with Terra-i detections from January 2004 to April 2013. Red boxes 1 and 2 correspond to the validation area with Google Earth and Landsat Viewer images, respectively.

Figure 3. Workflow methodology

Results
Google Earth Images
It was concluded that Google Earth was the most successful and accurate method due to the spatial resolution of the Digital Globe QuickBird images which allows accurate identification of changes in vegetation cover, and these changes correlated strongly with Terra-i’s detections.

Figure 4. Hotspot area detection with Google Earth images to assess land cover loss detected by the Terra-i system in 2005.

Using Google Earth as a validation tool has further advantages as it is easily and freely accessible and it allows you to visualize large datasets of high resolution images. However, Google Earth has limited provision of historical data which prevents multi-temporal analysis in impacted regions detected by the Terra-i system. Also, the time series of the Digital Globe images only is only available from 2002 to 2010, making it impossible to validate the most recent Terra-i detections (2011, 2012 and 2013). Finally, the high resolution imagery provided by Google Earth could be expensive if a user is planning to analyze it on a remote server or using GIS software.

Landsat Images
The Landsat Viewer tool allows users to select Landsat images to identify the land cover changes. This tool has images with a 30 m spatial resolution and they are displayed in the viewer using the band composition 5, 4, 3 which is exposed as a "natural" image of the surface [4].

Using the data from coarse resolution sensors (10-60m) such as Landsat is a well-known data source used by researchers to map land cover changes. Its spatial resolution level is high enough to identify land use changes and its 35 year data record allows users to perform multi-temporal analysis, thus providing an alternative method to validate Terra-i system detections.

However, the image quality is lower than Quickbird due to cloud presence as well as the banding in Landsat 7 ETM+ imagery since May 2003. Furthermore, the spatial resolution (30 meters) doesn’t allow users to directly identify the cause of the change, instead only algorithms are used to classify it.

Figure 5. Hotspot area detection with Landsat imagery shows how Terra-i detections follows similar patterns of disturbance events.

In conclusion, the validation method with visual interpretation of high/middle resolution images should use data sources with high temporal and spatial resolution, also the image quality must be cataloged well, which allows an accurate hotspots area detection of land cover and land use change.

Bibliography
[1] Clark , et al, A scalable approach to mapping annual land cover at 250 m using MODIS time series data: A case study in the Dry Chaco ecoregion of South America, Remote Sensing of Environment 114 (2010) 2816–2832, 2010 , [Online http://www.desertificacion.gob.ar/wp-content/uploads/2013/01/Clarketal2010_RSE.pdf]

[2] Olofsson , et al., A global landcover validation data set, part I: fundamental design principles, International Journal of Remote Sensing, 33:18, 5768-5788, 2012, [Online http://www.tandfonline.com/doi/abs/10.1080/01431161.2012.674230#.UjcZPz9F1Fg]

[3] Bontemps, et al, Operational Service Demonstration For Global Land- Cover Mapping: The GlobCover And GlobCorine Experiences For 2005-2009. En Giri y Weng y, Remote Sensing Of Land Use And Land Cover: Principles and Aplication (pp. 243-264). CRC Press.

[4] USGS, Landsat Missions, [Online http://landsat.usgs.gov/LandsatLookImages.php]

[5] Knorn, J., et al., Land cover mapping of large areas using chain classification of neighboring Landsat satellite images, Remote Sensing of Environment (2009).