Fabrice DUBERTRET, CNRS, France
The recent availability of cloud computing platforms, such as Google Earth Engine (GEE), has started a new era in remote sensing. Their ability to process enormous datasets in very little time has lifted most existing calculation barriers in traditional desktop-based land use/land cover (LULC) classification and change detection algorithms.
This presentation proposes a method to mobilize the computing capacities of GEE, as well as emerging literature exploring its potential, to create annual LULC maps since 1986 over the Tucson metropolitan area, USA. We chose to work with Landsat data because they have provided spatially and spectrally consistent imagery of the Earth’s surface at a medium/high spatial resolution of 30 meters for over 40 years, and on to the future with the recent launch of Landsat 9. We used a combination of corrections to Landsat scenes later merged into multi-date composites, derived transformation and indices, texture analysis and other ancillary data fed to a Random Forest classifier. Hosting our entire workflow in GEE allowed us to process a large amount of information (124 raster bands) for each year and to achieve high overall classification accuracy, ranging from 86.7% (1986) to 96.3% (2020). We also used conservative post-processing techniques to mitigate persistent confusions characteristic to our region of interest and to smooth year-specific LULC changes in order to better identify general trends.
This semi-automated tool is key for monitoring the important landscape transformations of the Southeastern Arizona Sonoran Desert fragile arid ecosystem and scarce resources, increasingly under pressure by both a massive population growth and rapid climate change. As we will see, policies to lessen urban sprawl in the area had little effects so far…
While our GEE workflow was developed to fit the needs of our study, it can easily be adapted to, or combined with, other sensors and regions of interest.
Mots clés : land use land cover classification|Landsat|Random Forest (RF)|Google Earth Engine (GEE)|cloud computing
A105466FD