Vegetation seasonality, particularly in the tropical dry regions, can cause conventional land cover classification that rely
on satellite data to “misclassify” land cover types, because a single satellite image might reflect only a particular stage
within a natural phenological cycle. To address this, we developed a phenology-based synthesis (PBS) classification algorithm that
maps land cover by analysing full time series, rather than temporal composites, of satellite images using the Google Earth Engine cloud computing platform.
The PBS classifier operates through occurrence rules applied to a series of single date image classifications of a study area to assign the most appropriate
land cover class. Since the launch of Landsat 8 in 2013, every point on Earth is imaged every 16 days with exceptional radiometric quality.
By feeding Landsat 8 data to a PBS classifier, we mapped the land cover of four protected areas and their 20-km buffer zones in different
ecoregions of Sub-Saharan Africa.
The Google Earth Engine script of the PBS classifier can be found at
https://code.earthengine.google.com/fcc04d2a79aef5f123008d6c178a18c0 (for GEE trusted user only)
Validation of the maps through a visual interpretation of coincident very high resolution images and a web-GIS showed that the combined overall
accuracy exceeded 90 %. The Sentinel 2 satellites will drastically increase the frequency of global image acquisitions, which, along with the
Landsat 8 programme and open data policies, will enable near real-time monitoring of Earth’s surface at a 10-30 m resolution. Using the first
series of Sentinel 2a images over one of the test areas, we demonstrate that Landsat 8 and Sentinel 2 data streams can be jointly exploited by
PBS classification to provide exceptional spatial and temporal detail for the mapping and monitoring of global land cover.
... more info ...
The JRC TREES-3 project aims at estimating forest cover changes at continental and regional levels for the
tropical belt for the periods 1990-2000 and 2000-(2005)-2010 based on a systematic sample of forest cover
change maps. An operational system has been developed for the processing and change assessment of a large
data set of multi-temporal medium resolution imagery (sample units of 20 km x 20 km size analysed from with
Landsat imagery). The main task is to assess as accurately as possible for each sample unit the forest cover
and forest cover change between two dates.
The analysis includes a crucial final step of visual verification and final assignment of land cover labels
which is carried out by forestry national officers or remote sensing experts from tropical countries. The
visual interpretation is conducted interdependently on two-date imagery to verify and to adjust the labels
pre-assigned to each segment for the different dates. A dedicated stand-alone application has been developed
for this purpose. The application is a graphical user interface, called the JRC Land Cover Change Validation
Tool. The aim of this tool is to provide a user-friendly interface, with an optimised set of commands to
navigate through and assess a given dataset of satellite imagery and land cover maps, and to correct easily
the land-cover labels as appropriate. FAO is collaborating with JRC in this work under the Global Forest
Resource Assessment (FRA) Remote Sensing Survey. JRC has added functionality to the tool to enable labelling
of land-use classes that are part of the FRA classification.
The technical document, entitled “User Manual for the JRC Land Cover/Use Change Validation Tool”
describes the steps for the installation of the tool on a personal computer, as well as the detailed features
of this dedicated graphical user interface. The authors welcome feedback from potential users of the tool,
in particular reporting of any potential software issue or providing suggestions for improvements of future
versions of the tool.