Technical Comments

Comment on “High-resolution global maps of 21st-century forest cover change”

See allHide authors and affiliations

Science  30 May 2014:
Vol. 344, Issue 6187, pp. 981
DOI: 10.1126/science.1248753

Abstract

Hansen et al. (Reports, 15 November 2013, p. 850) published a high-resolution global forest map with detailed information on local forest loss and gain. We show that their product does not distinguish tropical forests from plantations and even herbaceous crops, which leads to a substantial underestimate of forest loss and compromises its value for local policy decisions.

The high-resolution global map of forest cover loss and gain in Hansen et al. (1) is a fascinating and much-needed tool for both research and conservation planning. The authors claim that “[t]he information content of the presented data sets...provides a transparent, sound, and consistent basis on which to quantify critical environmental issues, including...(iv) the status of remaining natural forests of the world and threats to biodiversity…(v) the effectiveness of existing protected-area networks…(vi) the economic drivers of natural forest conversion to more intensive land uses.” After studying the supplementary data application (http://earthenginepartners.appspot.com/science-2013-global-forest) in detail, we express serious concerns about the appropriateness of the product for these purposes.

The main problem lies in Hansen et al.’s definition of forest as “all vegetation taller than 5m in height” [supplementary materials for (1)]. Such a structural definition includes types of plantations that have already replaced substantial parts of tropical (and also extratropical) forests. Monocultures of oil palm, rubber, or Eucalyptus are recognized as some of the biggest threats to tropical biodiversity (24), and their expansion into forest systems continues at an alarming rate [see (5) for details]. Although these plantations are technically “forests” in the definition above, they do not provide the benefits of forest vegetation as enumerated by the authors—i.e., “ecosystem services, including biodiversity richness, climate regulation, carbon storage, and water supplies” (69). Classifying plantations as forests confuses an endangered habitat with its greatest threats and thus underestimates real forest loss.

To evaluate Hansen et al.’s forest map, we compiled sites for which we had detailed information (e.g., from our previous fieldwork). We compared these validation sites to the forest map and identified three ways in which the product failed to accurately assess forest cover gain and loss (see Table 1 and Fig. 1 for specific cases):

Table 1 Examples of serious misclassifications by Hansen et al. (1).

Geographic coordinates allow easy checking of current vegetation in the supplementary online map application (http://earthenginepartners.appspot.com/science-2013-global-forest). Hundreds of additional examples of similar errors can easily be found in the application map by simply following plantation maintenance roads. Often, even details such as individual oil palms or soybean rows are clearly visible.

View this table:
Fig. 1 Selected examples of Hansen et al.’s (1) failures in classifying of tree plantations (A to E) and herbal crops (F to H) as forest.

All the maps are screenshots from Hansen et al.’s supplementary online map application (http://earthenginepartners.appspot.com/science-2013-global-forest) taken in November 2013 and modified to highlight details by adding the yellow squares. The colors in the right halves of each panel indicate stable forest (green), forest loss (red), forest gain (blue), and forest loss and gain (magenta). See Table 1 for more details, including coordinates, and for several additional examples.

(1) Areas deforested and converted into plantations before 2000 are classified as forests (cases 1 to 19, Table 1, and Fig. 1, A to E), which leads to an overestimation of total forested area. Furthermore, plantation management such as cutting old growth plantation and replanting with new crops is interpreted as forest gain/loss.

(2) Areas deforested around 2000 and replanted by tree plantations before 2012 are identified as “forest gain,” although their conservation value has been largely lost in such cases (case 20, Table 1).

(3) Contrary to the given definition of forest, vegetation lower than 5 m (e.g., pineapple, soybeans, or tea plantations) is often classified as forest. Including these types of vegetation as “forest” further biases estimates of forest cover gain and loss (cases 21 to 24, Table 1, and Fig. 1, F to H).

The ease with which we found classification errors in every examined tropical region suggests that these represent systematic misinterpretations with substantial consequences for inferences based on the product. Following our personal knowledge of several tropical regions and a survey of Hansen et al.’s map application, we tentatively estimate that the forest loss may be underestimated by tens of percents in the tropics. Similar issues may also occur outside the tropics, where species-poor  wood plantations are widespread (5, 10).

We warn that classification of high-resolution satellite data based on a single and simplistic algorithm can provide only limited insight into real forest dynamics at local scales. This forest map may provide preliminary identification of ongoing changes [e.g., (11)], but without locally specific calibration and evaluation and/or accompanying maps of pixel-specific classification uncertainty, it will mislead conservation policy-makers and managers, with potentially serious consequences for biodiversity and socioeconomic issues. The fact that the product comes with an easy-to-use online application further enhances the potential for uncritical use by nonspecialists and various interest groups.

Although the global loss of tree cover reported by Hansen et al. (1) represents a serious environmental issue, the replacement of natural forests by plantations (often with comparable tree cover) is a more important environmental and biodiversity problem at the local scale. Plantations are often characterized by considerably lower diversity than extensively used open countryside, including nonintensive pastures and small fields (2, 12). In this respect, the results of Hansen et al. are misleading and can potentially lead to abuse by local policy-makers who could consider an increase of tree cover a conservation success, even if this change is accompanied by decreases in biological diversity. The stated conservation relevance and utility of the approach of Hansen et al. is thus seriously compromised and calls for a critical reevaluation.

References and Notes

  1. Acknowledgments: We thank K. Mertes for English proofreading and valuable comments on the manuscript. This work was partly supported by the Czech Science Foundation (14-36098G).
View Abstract

Subjects

Navigate This Article