A screenshot from Global Forest Watch showing a global dataset on tree cover loss from the University of Maryland. The data goes down to 30 meter resolution, fine enough to see change in even extremely small plots of forest.
GUEST POST: Response to FORCLIME article on UMD high-resolution global map of 21st-century forest cover change
Dear colleagues, Please find below a response to an article distributed a few days ago—The high-resolution global map of 21st-century forest cover change from the University of Maryland (‘Hansen Map’) is hugely overestimating deforestation in Indonesia. As lead author of the original global forest change study and lead for Global Forest Watch (GFW) respectively, we welcome FORCLIME’s discussions and critiques of the data. We appreciate this and other efforts to report on local, in-depth analyses as they help us understand global datasets in their proper context and identify new research priorities. We also welcome and encourage scrutiny of all of the data we present on GFW, as this type of response sparks productive discussion and clarifies the value of each data layer. The concerns raised by the FORCLIME article are important, and have been addressed in the original paper, as well as in technical comments on the paper (Tropek at al.; Hansen et al.) http://www.sciencemag.org/content/344/6187/981.4.full http://www.sciencemag.org/content/344/6187/981.5.full The equating of all change in the map as “deforestation”, as performed in the FORCLIME study, is an erroneous interpretation and application of the dynamic mapped in the Hansen et al study. The global data set maps the biophysical presence of tree cover, defined as canopy cover of 5m or taller trees per Landsat pixel. There is no reference to natural forest as the targeted theme and there is similarly no consideration of forest land use. In the original paper, forest is defined by tree cover and includes plantations, agroforestry systems, and regrowth in addition to long-lived natural forests. As we state in our response to Tropek et al., “If … users interpret the results based on the forest cover definition specified, the results cannot be misleading.” Unfortunately, the misinterpretation of forest cover and loss as depicted in the Hansen et al. paper by FORCLIME has led to confusion concerning the forest change dynamic in Indonesia. Loss in tree cover vs. deforestation The University of Maryland data provide spatially and temporally explicit information on where trees are being lost (referred to as “forest loss” in the Hansen et al. paper and “tree cover loss” on Global Forest Watch.) Any stand-replacement disturbance, whether the clearing of a 5-year old acacia plantation or a clearing of a long-lived natural forest, is mapped as loss. Accordingly, contextual data sets – such as maps of primary forests, protected area boundaries, or burned area maps – are critical for drawing deeper, additional insights about what is happening in forests from local to global scales. We have stated this repeatedly, including in the original paper and on the Global Forest Watch website. The issue of natural forest loss in Indonesia was specifically addressed in a new paper and dataset Primary forest cover loss in Indonesia over 2000–2012 authored by Belinda Margono and others from the University of Maryland and World Resources institute. This dataset adds context to the Indonesian forest cover change dynamic by quantifying the portion of UMD loss that occurred within primary forests. This dataset is also available for visualization, download, and analysis (on the GFW and UMD sites), and shows the extent and subsequent loss and degradation of primary forests in Indonesia from 2000-2012. The result of Margono et al. is the one FORCLIME should use as a basis of comparison, and Figure 2 of the FORCLIME paper documents similar results when masking the gross forest loss data by a natural forest reference layer. Margono et al. applied this idea at the national scale to illustrate the spatio-temporal variation of Indonesia primary forest loss from 2000 to 2012. We see Figure 2 of the FORCLIME paper as a positive validation of the global map as a gross indicator of forest disturbance, where 23k ha and 25k ha of natural forest loss in West Kalimantan are quantified using the data from the global and FORCLIME studies, respectively. Submitting the FORCLIME study to a peer-review science journal likely would have pointed out this fact and resulted in a more proper interpretation of the FORCLIME study results. Data applications and next steps While no remotely sensed dataset will ever be truly definitive, the University of Maryland data and Global Forest Watch can play an integral role in improving our understanding about the world’s forests when applied in an informed way. A core strategy of Global Forest Watch is to put as much credible spatial data about forests as possible into the public domain, or generate these datasets if they don’t exist yet. Recognizing the clear need for additional contextual map layers, the Global Forest Watch team has already commissioned several research products and enhancements in functionality to address data gaps and concerns, and these will be added to the GFW platform as they become available. These include maps that will differentiate between forests and plantations and the ability to adjust the definition of “forest” by tree cover density. We appreciate the feedback on the data and look forward to working with the FORCLIME team and others to improve how we monitor and manage forests worldwide. Sincerely, Dr. Matthew Hansen, Professor, Department of Geographical Sciences, University of Maryland Dr. Nigel Sizer, Global Director, Forests Program, World Resources Institute