Dry-season greening of Amazon forests

Evidence from ecological studies1,2, eddy flux towers3–5, and satellites3,6 shows that many tropical forests ‘green up’ during higher sunlight annual dry seasons, suggesting they are more limited by light than water. Morton et al.7 reported that satellite-observed dry-season green up in Amazon forests is an artefact of seasonal variations in sunsensor geometry. However, here we argue that even after artefact correction, data from Morton et al. show statistically significant increases in canopy greenness during the dry season. Integrating corrected satellite with ground observations indicates that dry-season forest greening is prevalent in Amazonia, probably reflecting large-scale seasonal upregulation of photosynthesis by canopy leaf dynamics. There is a reply to this Brief Communication Arising by Morton, D. C. et al. Nature 531, http://dx.doi.org/10.1038/nature16458 (2016). Variations in sun-sensor geometry induce artefacts in remotely sensed vegetated surfaces8. Satellite studies thus typically use models to correct artefacts (for example, Moderate Resolution Imaging Spectroradiometer (MODIS) leaf area index9, and multiangle implementation of atmospheric correction (MAIAC) enhanced vegetation index10 (EVI)) or compositing algorithms designed to minimize artefacts (standard MODIS EVI11). Morton et al.7 used a modelling approach to correct MODIS satellite data, which they state removed seasonal changes in surface reflectance, and redefined debates over how climate controls forest productivity in the Amazon. Setting aside arguments that the remote sensing analysis by Morton et al. is faulty12, we take their correction7 at face value, and ask two questions. First, we ask whether the corrected results support their core conclusion that dry-season green up, previously observed by MODIS EVI, is eliminated. The hypothesis that Amazon forests green up in the dry season3 can be rigorously evaluated by formal statistical tests. Morton et al.7 showed that their correction reduces estimated dry season green up, Δ EVI (the EVI change during the dry season, Δ EVI = October EVI − June EVI; figure 3 in ref. 7 and Fig. 1). As the corrected mean Δ EVI was smaller than an a priori estimate of error for individual EVI observations, they concluded that the corrected mean Δ EVI was indistinguishable from zero. We find that this comparison, however, is not appropriate for assessing whether corrected EVI can resolve a basin-wide green up. The correct comparison, of mean Δ EVI to the error of the mean of the whole population of observations, is accomplished with standard statistical tests that lever the probability theory ‘law of large numbers’13. For example, the 95% confidence interval13 for basin-wide mean of corrected Δ EVI significantly excludes zero (Fig. 1). Alternatively, the corrected Δ EVI distribution7 can be compared to the binomial distribution generated by the null hypothesis that pixels are equally likely to exhibit positive or negative Δ EVI (Fig. 1), which is analogous to treating ‘green up’ or ‘brown down’ as the outcome of the flip of a fair coin. These standard tests show that corrected Δ EVI7, though substantially smaller in magnitude than uncorrected, nonetheless shows a highly significant increase in forest greenness. Second, we ask whether the smaller, but statistically significant, green up seen in the data from Morton et al. (Fig. 1) is biologically meaningful in terms of consistency with mechanisms and magnitude of seasonal changes in canopy-scale biophysics observed on the ground. We find that at an intensively measured site, significant dry-season increases in leaf area index are driven by coordinated flushing of new leaves, which have higher near-infrared reflectance (Fig. 2a) (mechanisms that Morton et al.7 hypothesized could drive true increases in satelliteobserved EVI). Leaf flushing is followed, after 1 to 2 months, by increases in photosynthetic capacity derived from CO2 fluxes measured at eddy flux towers (Fig. 2a). This correlation—1-month-lagged photosynthetic capacity with leaf area index, r = + 0.90, and with MAIAC EVI, r = + 0.89, where r is Pearson’s correlation coefficient, and the time lag is for new leaves to develop their photosynthetic capacity14—establishes a link between eddy flux measurements and biophysical properties observable from satellites. On the basis of this link, we find that increases in dry-season greenness seen by corrected EVI products (whether those of ref. 7 or the MAIAC EVI of Lyapustin et al.10; Fig. 2b) are real and consistently correlated with photosynthetic capacity increases seen at towers within the region analysed by Morton et al. (including adjustment for possible sun-angle effects on canopy illumination). This suggests that even the smaller corrected Δ EVI7 reflects mechanisms of canopy changes actually observed on the ground, and is therefore biologically meaningful. The analysis in Morton et al.7 is, notably, stimulating a productive re-examination of the methodology, meaning and magnitude of remote sensing indices, their artefacts, and their relation to field studies on the ground6,12. However, we believe that the primary substantive finding of Morton et al. of consistent canopy structure and greenness is incorrect. Both satellite remote sensing and ground-based observations show dry-season increases in greenness and biophysical properties associated with canopy photosynthesis across scales, from individual leaves to ecosystems to regions, in support of the conclusion that Amazon forests green up with sunlight in the dry season3,14.