Height-related changes in forest composition explain increasing tree mortality with height during an extreme drought

Recently, Stovall et al. 1 showed that during an extreme drought, remotely sensed mortality of tall trees was more than double that of short trees. They interpreted this to be a consequence of inherently greater hydraulic vulnerability of tall trees, and suggested that tall-tree vulnerability should generalize more broadly. Here we reassess their conclusions using contemporaneous, ground-based data from near their study sites. We find that 90% of trees belong to taxonomic groups showing declining, not increasing, mortality with height, and that the overall increase in mortality with height is instead a consequence of height-related changes in forest composition, not intrinsically greater vulnerability of tall trees. Similar mechanisms likely explain mortality patterns at Stovall et al.’s sites, and, regardless, we show that their conclusions should not be accepted in the absence of robust tests of alternative mechanisms. Because Stovall et al.’s remote-sensing approach did not distinguish among tree taxonomic groups, they could not test plausible alternative mechanisms. For example, consider the following two scenarios, each of a drought-stricken forest comprising two species. In the first scenario, mortality of both species declines with increasing tree height. However, at any given height, species B has substantially higher mortality than species A. In addition, the relative abundance of species B increases markedly with height. The net effect is that mortality in the forest as a whole increases with height (Supplementary Table 1). But because mortality declines with height for each species individually, we must reject explanations invoking intrinsically greater drought vulnerability of tall trees. In the second scenario, species C’s mortality declines gradually with height, but species D’s mortality increases sharply with height. Even without height-related changes in relative species abundances, mortality in the forest as a whole can increase with height, even if species D is the minority species (Supplementary Table 2). But in this scenario, we must seek mortality mechanisms that can explain opposite height-related drought responses of cooccurring tree species. To explore whether one or both of these scenarios could explain Stovall et al.’s results, we analyzed data from 89 randomly located forest plots distributed across a 1705-ha mixed-species, old-growth forest landscape, roughly 45–65 km southeast of Stovall et al.’s study areas in California’s Sierra Nevada2,3. During the last year of the drought (2016, also Stovall et al.’s last year of analysis), we recorded 5855 living and dead trees ≥5 m tall belonging to 15 species, which we assigned to three groups of species (hereafter: taxonomic groups) according to magnitude of mortality during the drought2,4–6 (Supplementary Table 3). Height classes of individual trees (5–15 m, 15–30 m, and >30 m, following Stovall et al.) were estimated from trunk diameter using species-specific allometric equations (Supplementary Table 3). Numbers of trees alive in 2013, and 2014–2016 mortality, were calculated as described in ref. 2 and as summarized in “Methods” section. When all trees were considered together, our results were similar to Stovall et al.’s: mortality of the tallest trees was ~2-fold greater than that of the shortest trees (Fig. 1a). But this simple analysis masked profound—and consequential—differences among taxonomic groups in both the magnitude of mortality and its relationship to tree height. For example, across the three height classes, mortality was low (<0.09) in angiosperms, intermediate (0.17–0.26) in non-Pinus conifers, and high (0.17–0.56) in Pinus (Fig. 1b). Within each taxonomic group, individual species had magnitudes and patterns of mortality that were largely similar to one another (Supplementary Fig. 1). Notably, variation in mortality was greater within height classes (among taxonomic groups) than among height classes (within taxonomic groups). In addition, only 10% of trees belonged to a taxonomic group (Pinus) in which mortality increased with tree height. The remaining 90% belonged to groups (angiosperms and non-Pinus conifers) in which mortality declined slightly with height. With increasing height, angiosperms, with their low mortality in all height classes, declined in relative abundance, whereas the intermediate-mortality non-Pinus conifers and high-mortality Pinus increased (Fig. 1c). To explore the effects of these changing relative abundances, we calculated hypothetical mortality for all https://doi.org/10.1038/s41467-020-17213-5 OPEN

A final key condition of our two scenarios is that, even if mortality of all trees considered together increases with height, within some or all individual taxonomic groups mortality remains constant or declines with height. This was clearly the case at our low-elevation study site, where mortality increased with height for Pinus but declined slightly with height for angiosperms and non-Pinus conifers (Fig. 1b). Although sometimes hampered by small samples, other studies in low-and mid-elevation forests during the drought similarly found that Pinus mortality typically increased with tree size, whereas mortality of angiosperms and non-Pinus conifers usually showed no consistent trend or declined with size [7][8][9][10]12 . Additionally, elsewhere we analyzed thousands of trees from 12 large plots mostly in mid-elevation forests above our 1524-1829 m study landscape 7 . Although we did not include these plots in our current analyses (because we wished only to include randomly located plots), mortality of angiosperms and non-Pinus conifers changed little or declined with increasing tree size, as in our low-elevation plots analysed here 7 .
Thus, available evidence suggests that the key elements driving our results and conclusions likely occurred broadly in low-and mid-elevation forests of the Sierra Nevada.
During the drought, tree mortality in high-elevation forests (ranging from ~2400 m to upper treeline at ~3400 m) was generally much less than in low-and mid-elevation forests, and this low mortality resulted in researchers collecting or analysing fewer ground-based data sets. Regardless, the composition of high-elevation forests allows us to frame a priori expectations for height-related mortality patterns there. Tree diversity declines at high elevations, often resulting in near monocultures of Abies magnifica or various Pinus species (P. albicaulis, P. balfouriana, P. contorta, or P. monticola) 5,6 . As noted in the main text, even in high-elevation Pinus monocultures we would expect mortality to increase with tree height during drought, but not because tree hydraulic vulnerability increases with height. Rather, the outbreaking Dendroctonus bark beetle species that kill Pinus in the Sierra Nevada preferentially mass-attack large trees, regardless of those trees' stress (reviewed in ref. 7).
In contrast, mortality of A. magnifica may have increased, decreased, or changed little with height, depending on the relative abundances of the three species of Scolytus bark beetles that typically attack A. magnifica: S. ventralis, which preferentially attacks large trees, and S. praeceps and S. subscaber, which preferentially attack small trees 7 . If relative abundances of the three Scolytus species were similar to those that attacked A. concolor during the drought 7 , the near monocultures of A. magnifica could represent one of the few Sierra Nevada forest types for which overall mortality during the drought did not increase with tree height. However, a systematic remote-sensing bias means that Stovall et al. 13 may have spuriously detected apparent increases in mortality with height even for forest types that had gradual declines in mortality with height (Supplementary Note 4).

Supplementary Note 4:
To define individual tree crowns, Stovall et al. applied standard algorithms to their LiDAR data 13 . For open-grown trees, this approach usually successfully identifies individual crowns of trees of all heights. In contrast, when trees are crowded, in our experience the crowns of shorter trees are often algorithmically merged with those of taller trees, or are missed entirely when the shorter trees' crowns are obscured by those of taller trees.
Indeed, Stovall et al.'s data suggest they substantially undersampled short trees. Numerically, Sierra Nevada forests are heavily dominated by small trees 11 ; for example, in our plots trees 5-15 m tall were the most abundant ( Supplementary Fig. 3), comprising 62% of all trees alive in 2013. In contrast, in Stovall et al.'s data set trees 5-15 m tall were the least abundant of the three height classes, comprising only 17% of trees they identified 13 .
Thus, Stovall et al.'s sample of short trees could be systematically biased toward the particular subset of short treesthose growing in the less-crowded conditions that allowed successful identification by LiDARthat had the lowest mortality. Specifically, Sierra Nevada trees growing in crowded conditions often suffer elevated mortality 14,15 , both from the direct and indirect effects of competition, and from greater tree-to-tree transmission of pathogens and bark beetles (the last of which were responsible for most tree mortality during the drought 7,8 ). Thus, systematic undersampling of short trees growing in crowded conditions (i.e., undersampling the high-mortality subpopulation of short trees) could introduce a bias toward finding increasing mortality with tree height, even in forest types in which mortality may have declined with height.  (4) and (5). Values of Mi were calculated using Equation (5), for x = 2 taxonomic groups.  (4) and (5). Values of Mi were calculated using Equation (5), for x = 2 taxonomic groups.  Table 4. Calculation of hypothetical height-specific mortality rates of all taxonomic groups combined, assuming constant abundances of taxonomic groups across the three height classes (Fig. 1d).  (4) and (5).

Supplementary
Values of Mi were calculated using Equation (4).
Supplementary Table 5. Calculation of hypothetical height-specific mortality rates of all taxonomic groups combined, assuming constant mortality of Pinus across the three height classes ( Supplementary Fig. 2).    . 1c) and actual height-and taxa-specific mortality values for angiosperms and non-Pinus conifers (Fig. 1b), but assuming constant Pinus mortality (that of the Pinus population as a whole) across the three Pinus height classes (Supplementary Table 5). Because these results are hypothetical, no credible intervals are shown. Supplementary Fig. 3: Numbers of trees in our 1705-ha study landscape, by taxonomic group, height class, and drought mortality status. Raw numbers are given in Supplementary Table 6. Gray shading at the tops of the bars represents numbers of trees that were estimated to be alive in 2013 but that died in 2014-2016 (with 95% credible intervals), calculated using Equations 6 and 7. As described in Methods, values of mi,t used in Equation 6 were derived from the posterior distributions of parameters estimated from 45,000 Markov Chain Monte Carlo iterations (three 15,000-iteration chains). Unshaded parts of the bars show actual numbers of living trees recorded in 2016. Thus, total bar heights represent numbers of trees estimated to have been alive in 2013 (numbers differ from those in Supplementary Table 3, which represent all living and dead trees in our sample regardless of year of death). Small trees dominate, as is typical for Sierra Nevada forests 11 , but in contrast to Stovall et al.'s data 13 (see Supplementary Note 4).