Improvements in algorithm design and computational power have steadily reduced the analytic obstacles for leveraging this imagery, yet the cost of commercial imagery remains prohibitive for many science applications. This cost landscape changed in 2005, when Google began hosting high-resolution commercial imagery at reduced spectral and spatial resolution on its cost-free Google Earth and Google Maps applications.GE high-resolution imagery does not contain an infrared band and sometimes has a slightly coarser spatial resolution than the native images provided directly from the sensor operators, yet a user of the GE environment is often able to readily discern land cover type, disturbance events, and other relevant attributes based solely on the imagery.

Users also have a number of additional resources to rely upon, including: detailed digital elevation models which allow three-dimensional viewing of the imagery, more than five million geo-referenced photos from services such as Panoramio [12], and a rapidly expanding set of vector and image-based overlays from a wide range of commercial geospatial services companies, scientific and government organizations, and millions of individual members of the GE community.To make Google’s high-resolution imagery as useful as possible, it is necessary to more fully characterize the temporal, spectral, and spatial properties of the archive. Up to this point, Google has been unwilling to release detailed information regarding any of these aspects of their holdings.

Of all the desired attributes, georegistration is the most readily tested.

Errors in image Dacomitinib alignment are apparent in the form of disjoint linear features such as roads and coastlines (Figure 1). In the face of these errors, the question arises: how trustworthy is the horizontal positional information in this archive? Large errors in georegistration would limit the scientific utility of the GE archive. In this analysis, we address this important question of horizontal positional accuracy.Figure 1.Apparent georegistration problems (highlighted with dashed yellow circles) between adjacent high-resolution images from Google Earth in: (a) Sao Paolo, Brazil, (b) San Salvador, El Salvador, (c) Chonan, South Korea, and (d) Anqing, China.

The Anqing example …2.?Data and MethodsThe dataset under review in this analysis is the GE high-resolution imagery archive��a Cilengitide global collection of images at roughly 2.5-meter resolution. The bulk of the high-resolution images in GE are from DigitalGlobe’s QuickBird satellite, a polar orbiting sensor that produces sub-meter resolution imagery with a horizontal accuracy of 23 meters (90% confidence interval; [13]).