MapInfo Pro

 View Only

Global Land Cover Facility, University of Maryland – Tree Canopy Cover Version 4

  • 1.  Global Land Cover Facility, University of Maryland – Tree Canopy Cover Version 4

    Posted 08-24-2018 02:12

    Global Land Cover Facility, University of Maryland – Tree Canopy Cover Version 4

    The GLCF have released a new version of their Tree Canopy Cover raster dataset. The dataset is global and free to download. It provides “estimates of the percentage of horizontal ground in each 30-m pixel covered by woody vegetation greater than 5 meters in height”. Raster datasets are provided for the years 2000, 2005, 2010 and 2015.

    Unlike discrete classification, this dataset provides a continuous estimate, from 0 – 100, of the percentage of tree coverage. As far as I know, it is the highest quality and highest resolution public domain dataset for analysis of changes in forestation over time.

    The data is supplied in TIFF format – one raster per Landsat scene – and in each epoch there are 8,488 scenes. Acquiring and preparing this data for analysis and visualisation is a challenge and there are a number of issues that make it difficult to work with in MapInfo Pro Advanced.

    1.      For each year there are 34,000 files to download spread over 8,500 directories.

    2.      The TIFF files are individually compressed and must be extracted prior to use.

    3.      The TIFF files use no internal compression and so, once decompressed, balloon to several Terabytes on disk.

    4.      The TIFF raster contains imagery utilising a color palette. Both land use classification and tree cover percentage are encoded in the color index. Consequently, it is difficult to consume this data for analysis. In MapInfo Pro Advanced there is little you can do with a raster with field type of “ImagePalette”, apart from display.

    5.      The rasters use either a UTM projection or a polar stereographic projection and unfortunately MapInfo Pro Advanced fails to recognise the UTM projection and fails to support the polar stereographic projection. [Update: These two coordinate system issues have been fixed and will be shipped in the 17.02 Pro patch later in 2018.]

    6.      MapInfo Pro Advanced does not support opening more than a thousand rasters simultaneously. There are two workarounds – firstly to merge the rasters into one (or a manageable number) of rasters or secondly to create a GDAL virtual raster (.vrt).

    With the benefit of having access to the raster engine code, I have managed to prepare this data for analysis. In my workflow, the original TIFF is converted into three separate rasters in MRR format – a continuous field raster containing the tree cover percentage, a classified field raster containing the land use classification and an image field raster containing the supplied color imagery. As MRR uses compression internally, this collection of rasters is of manageable size on disk and I do not need to use the original TIFF’s thereafter.

    I have prepared data for the Snowy-Monaro council area in New South Wales, Australia. In this region there has been a mysterious large scale die-back of Ribbon Gum trees. The trees are being eaten by an evil native weevil and dying as a result, but the root cause of the dieback is thought to be related to climate change and changes to rainfall patterns and fire regimes which are stressing the Ribbon gums and making them vulnerable to weevil attack. If you drive through this area the scale of the dieback is very apparent – thousands of square kilometres of gum tree skeletons litter the farming landscape. I am hoping to be able to quantify the extent using the GLCF data.

    The raster calculator can be used to compute the difference in tree canopy coverage between 2015 and 2000, where an increase of coverage is positive, a decrease of coverage is negative and no change in coverage (or no coverage) is an empty cell. For reference, I executed two calculator operations and the expressions I used were -



    A rendering of the canopy change raster is attached where green shows gains and brown shows canopy loss. Summary statistics from the rasters can be used to make the following assertions -

    ·        Mean canopy coverage in 2000 is 31% and effective total coverage is 4636.3 square kilometers.

    ·        Mean canopy coverage in 2015 is 33.2% and effective total coverage is 4972.1 square kilometers.

    ·        Mean change of canopy coverage is 2.5% and change in effective total coverage is +335.8 square kilometers.

    ·        The standard deviation from 2000 to 2015 has increased, seeming to indicate that the canopy increases have been in dense coverage – perhaps new plantation forest.

    Features of note are a visible loss of canopy in regions affected by the 2003 alpine bushfires, a gain in canopy above the snow line in the mountains and significant changes in coverage associated with logging and plantation activity in the south east of the region.

    Looking more specifically at the region of ‘severe dieback’ indicated in the diagram, I observe that mean change of canopy coverage is 0.1 % and change in effective total coverage is +0.5 square kilometers, so even in this region the dieback appears to have been offset by other canopy gains. Visually, the regions of canopy loss are pretty evenly interspersed with regions of canopy gain and it is hard to see any pattern to the changes.

    I have attached a TIFF file of the rendered canopy change as well as preview below.

    The dieback extent diagram is sourced from the ABC -

    See AttachmentSee Attachment


    I have included histograms of the canopy coverage for the years 2000 and 2015 and also a histogram of the change in coverage.

    See Attachment

    See Attachment

    See Attachment

    Apart from a major gain in coverage from 4 - 6% which I don't know how to explain, the histogram shows there has been a consistent loss of canopy cover below about 50% and a consistent gain in canopy cover above 50%. Maybe this could be interpreted as a loss of trees that are isolated and unprotected and consequently under greater climate change stress.

    It is possible to dig further into these numbers and build histograms for each percentage bin which would show what cells of a particular kind of coverage in 2000 changed into in 2015. This might provide a way to build (supervised) classifications for land use change events such as Bushfire, Deforestation, Reforestation, Dieback, Forest thickening, Forest thinning etc.

    See Attachment

    The image above illustrates this. It is a series of stacked histograms (logarithmically scaled and colored). On the vertical axis is the canopy coverage in the year 2000 and read across to see where cells of a particular coverage value ended up in 2015. If there was little change then we would expect the data to be clustered about diagonal line from (0,0) to (100,100). If there was general thickening of cover we would see data skewed to the right of this line. The diagram shows several distinct populations, as well as significant deviations from the 'expected' trends.







    CanopyChange2000to2015.tif   16.00 MB 1 version