Technical Comments

Comment on “The extent of forest in dryland biomes”

See allHide authors and affiliations

Science  10 Nov 2017:
Vol. 358, Issue 6364, eaao0369
DOI: 10.1126/science.aao0369

Abstract

The study by Bastin et al. (Reports, 12 May 2017, p. 635) is based on an incomplete delimitation of dry forest distribution and on an old and incorrect definition of drylands. Its sampling design includes many plots located in humid ecosystems and ignores critical areas for the conservation of dry forests. Therefore, its results and conclusions may be unreliable.

The recent Global Drylands Assessment of forests (1) clearly responds to a worldwide concern for better management and conservation of these ecosystems. Uncertainties about the contribution of these vast ecosystems (covering more than 40% of Earth’s surface) to the global carbon balance, and about the value of their services, make these estimations necessary and timely. The presumed significant underestimates of forest cover presented by Bastin et al. (1) have implications for strategies of mitigation and adaptation to global warming. However, a careful analysis of their data and methodology reveals some drawbacks that limit the reach of these conclusions and their implications.

Bastin et al. claim that they use “the delineation adopted by the United Nations Environment Programme World Conservation Monitoring Centre [(UNEP-WCMC)]—lands having an aridity index (AI) lower than 0.65” (1, 2). This affirmation is wrong. The delineation of UNEP-WCMC was based on the United Nations Convention to Combat Desertification (UNCCD)—i.e., the AI criterion—and on some additional ecoregions (3) or “relevant areas” from the point of view of the Convention on Biological Diversity (CBD). As Bastin et al. show [map in fig. S1 of (1), areas in dark gray], they ignored both zones “presumed included” and zones “to review” in the UNEP-WCMC proposal. Therefore, they are using the UNCCD criterion (i.e., AI < 0.65) and not the UNEP-WCMC delineation.

This choice has had two consequences. First, the “global dryland extent” considered was based on the UNEP Global Resource Information Database (GRID) (4, 5) aridity zones data set. This is an old climatic database, with some errors, that induced Bastin et al. (1) to sample within non-dry areas. Even if they had desired to delineate drylands exclusively on the basis of AI, they should have used a corrected version (6) or, even better, they should have computed it by themselves. Computing AI on the basis of WorldClim, the standard global database used extensively by ecologists and biogeographers (79), shows that 10% of the plots evaluated by Bastin et al. (1), around 22,000 plots, were instead placed in areas with AI ≥ 0.65—non-dry areas according to UNCCD. These mistakes become apparent with a superficial inspection of the Global Drylands Assessment. For instance, most of the Ecuadorian easternmost rainforests (e.g., the Yasuní forest in the Napo region) and extensive parts of the southeastern Peruvian Amazonian forests were included as drylands. This error is also apparent when comparing the UNCCD aridity map (2) and its corrected version (6). This is a serious issue for the assessment of global dry forest cover because, for example, there are significant differences in tree cover between plots in non-dry (average 0.47) and genuine dry (average 0.11) locations (Kruskal-Wallis χ2 = 15,233, df = 1, P < 0.0001).

There is a second caveat, a consequence of the drylands delimitation used: In addition to excluding the relevant areas for CBD, where some dry biomes could be found [some Mediterranean, desert, grassland, and savanna types (2)], the study also ignored the area occupied by the seasonal dry forests [i.e., the “tropical and subtropical dry broadleaf forests” biome (3)], which seems critical for the assessment of forest in drylands. The results of this exclusion are evident if the map of ecoregions (3) is overlaid on the map of dry areas used by Bastin et al. (1). At least 52% of the area considered by the World Wildlife Fund (WWF) as dry forest is not contained in the UNCCD delineation of drylands (2) and therefore is not considered by Bastin et al. (1). As a consequence, some critical dry regions were omitted (10, 11). A simple comparison between them shows, for instance, the exclusion of extensive regions in Central America and Mexico (Fig. 1) and from most of the Asian southwest. On the contrary, the strict application of the erroneous AI map without further considerations resulted in some awkward facts, such as including 56% of world flooded grasslands, 10% of moist broadleaf forests, 23% of mangroves, and 51% of temperate forests as “dryland areas.” In fact, placing the studied plots over the ecoregions map (3) shows that almost 35,000 plots (i.e., around 17%) fall within “non-dryland” biomes, including 11,887 plots located in moist forests (5.6%), 13,313 in temperate forests (6.2%), 5192 in taiga forests (2.4%), and even 362 plots in mangroves (Table 1).

Fig. 1 Examples of omitted areas.

Distribution of tropical and subtropical dry broadleaf seasonal forests in North America, Central America, and part of South America. Cyan areas are included in the original study. Yellow areas are omitted.

Table 1 Distribution of plots identified as dry and non-dry biomes.

Location of the 213,795 sampling plots among WWF terrestrial biomes and over a global map of aridity index (AI) computed on modern climatic databases. Each cell represents numbers of plots within each combination (biome × AI category). Two plots were not considered because of incomplete coordinates in the database, and there were no aridity index data for another 142 plots.

View this table:

Because any map will include errors, it is critical to understand their nature because they can propagate to other data sets (12, 13). Bastin et al. (1) included an evaluation of the sampling and measurement errors attributable to estimation of forest area, but they did not make any attempt to validate the quality of the basal map used to define drylands (2). This could bring into question the validity of their results as a whole. A simple cross-tabulation of a sample of predicted classes against their corresponding observed classes would have been enough (14). Many measures can be derived from such a confusion matrix and used to test the validity of the work done. Bastin et al. (1) should have tested this before designing their sampling scheme and should have corrected it by excluding non-dry ecoregions and/or by recomputing the aridity index based on modern climatic databases. Other global assessments of the conservation status of dry forests (15) have taken these problems into consideration and have combined a corrected version of the aridity index (6) with the selection of ecoregions of interest (3).

Taking all these limitations into account, it is likely that the reported estimation of forest extent in dryland biomes, the 40 to 47% increase over previous estimates, and the potential 9% increase in the global area with >10% tree canopy cover suggested by the results of Bastin et al. (1) would be substantially different than claimed. This limits the usefulness of the implications of their results in developing and reconsidering strategies of mitigation and adaptation to global warming at a global scale.

References and Notes

  1. Acknowledgments: Supported by Comunidad de Madrid project REMEDINAL3 grant S2013/MAE-2719 (M.d.l.C., L.C., and A.E.), Spanish Ministry of Economy and Competitiveness project ROOTS grant CGL2015-66809-P (A.E.), and Program Prometeo (SENESCyT, Gobierno de Ecuador) (P.F.Q.-A.). All the numbers and statistics cited in the text have been computed using publicly available data sets [i.e., WorldClim (7, 8), ENVIREM (9), World Atlas of Desertification (6), WWF Terrestrial Ecoregions (3), and UNEP WCMC Drylands database (2)] and the open source programming language R. A script with the code to access the data sets and perform all the computations is available at https://doi.org/10.13140/RG.2.2.29636.32645.
View Abstract

Subjects

Navigate This Article