High-temporal frequency monitoring of urban land cover

Alexandre Boucher, Karen Seto

Earth-observing satellites have collected remote sensing data for more than 30 years, yet most urban mapping studies do not take full advantage of the historical record and the temporal frequency of the observations available. That information is ever more important as remote sensing images are increasingly being used with other types of data such as demographics, economics, and policy to understand the link between human activity and impacts on the landscape. Linking social processes with spatial patterns observed in remote sensing has been the subject of numerous studies . Yet, it is almost without exception that the spatial patterns in these studies are observed in only two or three periods. The underlying assumption is that the relationship between landscape dynamics and social processes can be understood with several observations in time. Although this may hold true for relatively slow land use and land cover changes, the assumption is not valid for rapidly urbanizing landscapes.

Despite decades of research in geography, economics, and urban studies, urban land-use hange patterns are still not well understood nor characterized. Although the consensus is that compact urban form is critical for sustainable development, very little is known about how urban form evolves.  Due to the complex nature of urban growth patterns, using only a few observations in time will not provide accurate or meaningful information about the ways in which urban land-use change volves. Rather, urban growth and associated land cover changes are complex nonlinear rocesses that require frequent observations through time to properly monitor and describe their morphology.

Mapping is an exercise undertaken to answer specific questions. Where do the land covers of interest exist? How are they distributed geographically? What is the spatial configuration of the landscape? The choice of a classification algorithm, the land cover types, as well as the validation metrics must be selected according to how these questions are to be answered. The classifi ed map must provide information about land cover at temporal and spatial resolutions sufficient for the objectives.When the mapping exercise is not an end in itself, but only a starting point to linkobserved landscape patterns to their drivers or their consequences, the mapping output should:

  • Provide high temporal resolution land cover transitions;
  • Accurately map land cover/use change patterns in space and time.

The resulting multitemporal classification should yield the area, the spatial distribution, and the rate of change as well as the trajectories of the land covers. In addition, the accuracy of the change map, including the area estimates andspatial patterns of change, must be assessed.


  • 2006   Boucher A., Seto K.C. , Journel A.G., 2006 A novel method for mapping land cover changes ; Integrating space and time with geostatistics. IEEE transactions on geosciences and remote sensing, Vol. 44 no 11, 3427-3435.