Herbivory, plant resistance, and climate in the tree ring record: Interactions distort climatic reconstructions

Abstract

To understand climate change, dendrochronologists have used tree ring analyses to reconstruct past climates, as well as ecological processes such as herbivore population dynamics. Such reconstructions, however, have been hindered by a lack of experiments that separate the influences of confounding impacts on tree rings, such as herbivores and the interactions of multiple factors. Our long-term experiments with scale insects on resistant and susceptible pines demonstrate three major points that are important to the application of this commonly used tool. (i) Herbivory reduced tree ring growth by 25–35%. (ii) The impact on ring growth distorted climate reconstruction, resulting in the overestimation of past moisture levels by more than 2-fold. Our data suggest that, if distortion because of herbivory has been a problem in previous reconstructions, estimates of the magnitude of recent climate changes are likely to be conservative. (iii) Our studies support a detectible plant resistance × herbivore × climate interaction in the tree ring record. Because resistance and susceptibility to herbivory are known to be genetically based in many systems, the potential exists to incorporate plant genetics into the field of dendrochronology, where it may be used to screen distortions from the tree ring record.

One of the greatest challenges to research in the physical and biological sciences is the relatively short life span of the researcher, in comparison to the processes studied. Biological processes such as outbreaks of phytophagous insects may occur over periods of centuries, and many physical processes such as climatic variation may require millennial length data sets to determine the extent and significance of variation. Consequently, there is a great need for proxy data sets that can be used to infer or reconstruct past conditions for which historical records are not available.

Tree rings provide what are probably the highest resolution, continuous, long-term proxy data sets available. Trees record in the density, width, structure, and chemical composition of their annual rings information about both biotic and abiotic processes in the surrounding environment. Since A. E. Douglas's 1920 publication (1) in Ecology, in which he documented the relationship between tree ring width and precipitation, the field of dendrochronology (i.e., the study of tree rings referenced in time) has expanded into a variety of subdisciplines including dendroecology, dendroclimatology, and dendropyrochronology, to name only a few (2–4). Yet, as each of these disciplines has greatly expanded our understanding of long-term processes, there remain several key issues that must be addressed as we seek to refine and expand this data-rich resource.

The first is the lack of experimental tree ring data sets. Reconstructions of biotic factors, specifically herbivory, have largely been dependent on using observational data to infer the herbivore signal (usually manifest as reduced ring width) within the tree ring record (but see ref. 5). Of the observational methods developed for reconstructing herbivore outbreaks, there are two strategies commonly used (3, 6, 7). The first is the correlation of observed periods of pest outbreak with concurrent ring growth, using the observed reduction in ring width as the herbivore signal. The second method requires the comparison of host and non-host tree species. By comparing growth between the two groups, variation because of shared factors such as climate can be removed, and periods of reduced growth observed only in the host species can be inferred as periods of pest outbreak. Although these methods have provided much information describing herbivore population dynamics, the lack of experimentally determined herbivore signals has remained a weakness in this field.

The second and perhaps more critical issue is the lack of information on how multiple factors such as herbivory and climate interact in the tree ring record, despite the recognition that multiple factors may simultaneously impact plant growth (8). The accurate reconstruction of climate is dependent on locating trees that are climatically sensitive (2, 3). Precipitation, for example, is best reconstructed by using trees growing in water-limited environments (such as slopes, ridges, and well drained soils). However, whereas selecting water-stressed trees does maximize the precipitation signal, previous work (9–12) has shown that there is a strong positive association between susceptibility to herbivory and water stress. Consequently the trees selected for climate reconstruction may include both an herbivore signal as well as a climate signal. Although some authors have suggested herbivory may make trees unsuitable for climate reconstruction (5, 7), the interactions of these factors, to our knowledge, have not been assessed, and the importance of this potential interaction is furthered by the fact that herbivory (or lack thereof) is not currently a part of the field criteria for selecting wood samples used for the reconstruction of climate.

Finally, there has been little investigation into the impact of genetics on tree ring-based reconstructions. In the context of herbivory and climate, trees that express varying levels of genetically based resistance to herbivory may have differential suitability for use in the reconstruction of climate.

Here, we address these issues by using a long-term herbivore experiment on the ash and cinder fields of Sunset Crater, east of Sunset Crater National Park (15 miles east of Flagstaff, Arizona). Several attributes of the ecosystem at Sunset Crater make it ideally suited to address these issues. First, Sunset Crater is a hot, dry environment (13, 14), making the region ideal for the selection of climatically sensitive trees. Yet, this water-stressed environment also supports several insect species, including the pinyon needle scale, (Margarodidae, Homoptera) Matsucoccus acalyptus, which reaches high population densities, and chronically attacks pinyon pine (Pinus edulis) (11–13). In addition to providing anecdotal evidence that the impact of two factors (climate and herbivory) can be concurrent in space, the pinyon pine at Sunset Crater provide an excellent system in which to assess this potential herbivore × climate interaction.

Second, the presence of naturally scale-resistant trees provides a control group whose growth should be influenced primarily by climate, whereas the growth of adjacent scale susceptible trees should be influenced by both climate and herbivory. Within this system, several lines of evidence suggest that resistant and susceptible phenotypes are genetically based. (i) Adjacent trees with their branches interdigitated exhibit marked differences in their scale populations; susceptible trees support millions of scales, whereas resistant trees support few or none. (ii) During more than 15 yr of continuous study, these resistance and susceptibility traits have remained stable. (iii) Transfer experiments have shown that scale survival on resistant trees is only 11%, compared with 73% on susceptible trees (11). (iv) Even when susceptible trees are protected and allowed to recover for 8 yr, subsequent scale transfer experiments demonstrate that these trees remain susceptible, arguing that scale susceptibility is a stable, innate characteristic (15). (v) Although the phenotypic mechanisms that result in resistance have not yet been identified in this system, these findings (11, 15, 16) are consistent with the results of other studies of homopterans (17), including other members of the same family (18) in which the genetic basis of resistance has been clearly established.

Finally, by using these two groups as controls, scales have been experimentally removed from a subset of the susceptible group since 1985. These experimental trees provide a long-term experiment that allows us to critically evaluate the relative roles and interactions of climate and herbivory in the tree ring record. Using this system, we addressed the following hypotheses. (i) Chronic herbivory results in the reduction of radial growth, and so produces an herbivore signal. Removal of the herbivore results in a release in radial growth. (ii) Herbivory alters the sensitivity of tree rings to climate, indicating a climate × herbivore interaction. (iii) The interaction of herbivory with climate in the tree ring record can result in a distorted reconstruction of climate.

Methods

Research Site.

The cinder fields on which our experimental trees grow are the result of a 200-yr-long series of volcanic eruptions at Sunset Crater that ended ≈800 years ago, and that covered 1,200 km² in lava, cinders, and ash (19). This cinder soil, which has been recolonized by the surrounding pinyon-juniper forest, has been demonstrated to be deficient in water availability (13, 14).

Our experimental site is located between the boundaries of Sunset Crater and Wupatki National Monuments (T23N R9E SEC 10), at an elevation of approximately 1795 m. The site has no history of active management, no history of fire (because fires cannot carry on bare cinder substrates), and no evidence of anthropogenic influences on stand composition or structure, (i.e., no cut stumps). The stand is composed almost entirely of pinyon and juniper in an approximately 75/25 pinyon/juniper mix, with a few rare ponderosa pines. The canopy is open, with only 25–40% of the ground covered by tree canopy. Pinyon within the stand range in ages from <5 yr to >270 yr, with the majority of the trees being between 40 and 130 yr old (R.T.T., unpublished data). The surrounding terrain is characterized by low rolling hills (<5 m in height) and lava ridges; however, the area within which the trees were selected is predominantly flat, with slopes ranging from about 0–5°.

Herbivore Life History and Removal Experiment.

Associated with this high water stress, the cinder fields of Sunset Crater support pinyon pines suffering high levels of herbivory by the pinyon needle scale. The pinyon needle scale is a mesophyll-sucking herbivore that passes through one complete life cycle per year. Adult female scales emerge in early April, are mated, and crawl to the base of the tree where they lay eggs in an aggregated mass (11, 20). Two to four weeks later, first instars (crawlers) emerge, climb the tree, and insert their mouth-parts through the stomata of needles. Once scales have settled and begun to feed, they remain stationary through the first two instars.

Feeding by the sessile life stages results in the chlorosis of needles, and ultimately in premature senescence. Although feeding by the scales results in extensive foliage loss, the trees are never completely defoliated. Scales emerge and settle early in the spring before bud break; consequently, the current year's needles are almost always scale free. This chronic loss of older needles while young needles escape herbivory results in a distinctive “French poodle” architecture, such that uninfested resistant trees maintain an average of 7 yr of needles, whereas infested susceptible trees support only 1 to 2 yr of needles (11).

This sessile life style, combined with the aggregate egg masses and flightless females, makes this insect ideal for experimental manipulation. By removing the egg masses from the base of the tree, and placing a band of Tanglefoot (Tanglefoot, Grand Rapids, MI) around the base of the trunk, it is possible to remove all of the scales from an individual tree within a single generation without the application of insecticides. The presence of naturally occurring resistant and susceptible tree phenotypes further increases the utility of this system. This type of scale-removal experiment was initiated in 1985 at Sunset Crater, at which time 108 susceptible trees, growing within a 600 × 300 m area were selected (based on the presence of scales) and divided randomly into two groups. One group was subjected to scale removal, while members of the other group were left untouched as controls. Fifty additional scale-resistant trees, growing randomly intermixed with the susceptible trees (determination of random association based on 99% C.I. envelopes by using K12(t) with 199 simulations; K12(t) program courtesy of R. Duncan, Lincoln University, Canterbury, New Zealand), were marked and monitored for presence/absence of scale herbivory.

Tree Ring Sampling.

In 1999 and 2000, we collected increment cores from 17 resistant, 15 susceptible, and 16 scale-removed trees within this experimental group. Because of our concern that these trees would be damaged by the removal of incremental cores, coring was limited to trees with basal trunk diameters in excess of 6 cm, and only one core was taken per tree. We used these cores to assess the impact of herbivory on growth rates, and to develop three models (based on resistant, susceptible, and scale-removed groups) calibrating growth with climate. Cores were prepared and cross-dated by using standard methods (21). Rings were then measured by using a Velmex sliding stage with an Acu-Rite encoder and Metronics digital readout (Velmex, Bloomfield, NY). Measurements were logged into a computer by using the program MEASURE J2X VERSION 2.2.6 (Voortech Consulting, Holderness, NH). Quality control of cross-dated series portions used for the models was checked by using the tree ring program COFECHA (courtesy of R. Holmes Univ. of Arizona, and the International Tree Ring Data Bank Program Library). Comparison of raw growth rates was conducted by using an ANOVA, followed by Tukey's honestly significantly different (HSD) test for pair-wise comparisons.

Modeling Tree Ring Response to Climate.

To assess climatic sensitivity and develop the resistant, susceptible, and scale-removed models, we calibrated ring growth with climate by regressing standardized rings (averaged by year) against the Palmer Drought Severity Index (PDSI, data from Arizona Districts 2, 3, and 4 of the National Oceanic and Atmospheric Administration, representing northern and northeastern Arizona, available from the National Climate Data Center at http://lwf.ncdc.noaa.gov/oa/ncdc.html), averaged over the May–October growing season (hereafter GSPDSI), during the experimental period. The PDSI is an index that measures relative water stress (22), and past work in this system has shown the GSPDSI to be a good predictor of growth (23). We used standardized rings, because standardization puts all ring measurements on the same scale and allows the comparison of the relative responses of trees with different growth rates (2, 24). Standardization values were based on tree-specific variation (i.e., each tree was standardized against itself), and values are centered on 0, such that above average growth is denoted by positive values, and below average growth is denoted by negative values. The slope of the resulting calibration model is indicative of the climatic sensitivity of the trees, such that a steep slope indicates high sensitivity, whereas a shallower slope indicates the opposite.

Statistical comparison of the resistant and susceptible model slopes was conducted by using all rings in all trees for all years within groups to verify population level differences in response to climate. Models of the climate/growth relationship were developed by averaging rings within years to reduce white noise and determine the average tree response to climate within groups. To compare the resistant and susceptible climate/growth calibration models produced by the regressions, we tested for differences in slopes by using a test analogous to the Student t test, such that t = (b₁ − b₂)/s_b1−b2 (25). The model generated by the scale-removed group was not statistically compared, because the trees did not recover a full compliment of needles for 7 yr, and we made no a priori predictions as to the climatic sensitivity of the trees during the recovery period.

If herbivory changes the calibration model relating ring growth to climate, the application of the wrong model to sampled trees for reconstruction could result in a distorted reconstruction. Because the sampling of dead remnant wood for climatic reconstruction would be conducted without knowledge of the herbivore loads experienced by that tree while alive, a mix of 5 resistant and 17 susceptible trees (i.e., the proportion naturally found in the field) growing intermixed within the same site, and on which herbivores have been monitored since 1990, were cored. These trees were combined to produce a single indexed tree ring series. This series was then applied to the previously developed resistant and susceptible models to reconstruct the same 14 years of GSPDSI values.

Addressing Alternative Hypotheses.

To verify that herbivory is the primary factor influencing tree ring growth, we collected data on two environmental variables, the slope of the surrounding terrain, and proximity to neighboring trees, because both can influence tree ring growth (2, 26). The average slope of the surrounding soil for each tree was determined by using a clinometer at three haphazardly selected points along the drip line. Averages were compared by using a one-way ANOVA. We estimated competition by measuring the distance from the drip line of the tree to the drip line of the three nearest woody plants >1 m tall. Comparisons of overall differences were made by using a multivariate ANOVA. We also compared age among the three groups based on core samples, and compared the three tree groups by using a one-way ANOVA.

Results and Conclusions

Herbivore Signal.

Two lines of evidence support our hypothesis that scale herbivory leaves a signal in the tree ring record. First, during the years before the experimental removal of scales from susceptible trees (1972–1985), resistant trees produced rings 25% wider than those produced by susceptible trees (mean resistant ring width = 0.551 ± 0.023 mm, susceptible = 0.440 ± 0.023, scale-removed = 0.423 ± 0.021; resistant vs. susceptible, and resistant vs. scale-removed groups P < 0.001, susceptible vs. scale-removed, P < 0.86; Fig. 1A). Because there is a significant difference in the growth rates of scale-resistant and susceptible trees, these findings demonstrate that resistance and susceptibility traits can be expressed in the tree ring record. Furthermore, because growth during the preexperimental period did not differ between susceptible trees that would serve as controls, and susceptible trees that would later have their scales removed for 14 yr, we validated that, at the initiation of the removal experiment, treatment and control groups had the same growth histories.

Figure 1.

(A and B) Before the removal of scales, the susceptible and pretreatment scale-removed groups exhibited reduced growth relative to the resistant group. After scale removal, growth among previously scale-infested trees increased to equal that of the resistant trees. Groups within time periods were compared by using a one-way ANOVA on all rings within the described time period. Differing letters indicate P < 0.001 using Tukey's honestly significantly different (HSD) test for pair-wise comparisons.

Second, after scale removal, previously infested trees demonstrated a release effect, producing rings equal in width to those produced by resistant trees (mean resistant ring width = 0.915 ± 0.032 mm, susceptible = 0.589 ± 0.030, scale-removed = 0.883 ± 0.034; resistant vs. susceptible, and susceptible vs. scale-removed groups P < 0.001, resistant vs. scale-removed, P < 0.75; Fig. 1B). As predicted, trees with scales continued to produce rings smaller than those produced by both resistant and scale-removed trees. These data demonstrate experimentally that herbivory is the causal mechanism of reduced radial growth, and that the removal of an herbivore can result in a 25–35% increase in growth. This evidence that herbivores act as a limiting factor agrees with, and provides experimental evidence for, the assignment of reduced growth not attributable to climate to pest outbreaks in other systems. Further, because we pooled all years for all trees across the three groups in a simple ANOVA, the difference between these groups argues that the variation because of herbivory (between group) is greater than the variation because of climate (within group). These data therefore suggest that variation introduced by herbivory may degrade the quality of climatic reconstructions by the introduction of extraneous variation.

Herbivore Mediated Sensitivity to Climate.

Although the above experiment clearly demonstrated an impact of herbivory on growth rates, it did not evaluate the impact of herbivory on climatic sensitivity. We made two predictions of how resistant and susceptible trees would respond to climate during the experimental period. First, because resistant trees have an average of 7 yr of needles, while susceptible trees support only 1–2 as a result of herbivory, we predicted that resistant trees would be better able to take advantage of favorable (i.e., wet) years. However, because increased foliage may result in increased water loss through evapotranspiration, we predicted that resistant trees would show a relatively greater negative response during dry periods.

Our second prediction is based on the hypothesis that scale herbivory acts as a limiting factor within susceptible trees, reducing the role of climate in determining radial growth. As a result of this disconnection from climate, susceptible trees should demonstrate a reduced goodness of fit between radial growth and climate, expressed as a smaller adjusted model coefficient of determination (r²).

The regression of standardized growth against the GSPDSI demonstrates that the magnitude of climatic sensitivity (indicated by the slope of the climate-ring regression) in resistant trees is nearly twice that of the susceptible trees (slope = 0.274, susceptible = 0.167, P < 0.05; Fig. 2, Table 1). These data argue that defoliation by herbivory reduces the tree's ability to take advantage of wet periods. In combination, the reduced response to favorable periods and reduced growth rates may place susceptible trees at an evolutionary disadvantage by limiting reproduction and growth.

Figure 2.

Resistant, scale-free trees express twice the climatic sensitivity of scale-infested as denoted by the difference in slope. Resistant and Susceptible model slopes differ significantly at P < 0.05. Model parameters are listed in Table 1.

Table 1.

Model parameters

Model	Slope	Adjusted r2	Model ANOVA
Resistant	0.274	0.515	P < 0.003
Susceptible	0.167	0.252	P < 0.04
Scale-removed	0.208	0.45	P < 0.008
Verification	Slope	Intercept	ANOVA
Resistant	0.882	−0.221	P < 0.003
Susceptible	1.445	0.649	P < 0.003
Perfect	1	0

The Model data indicate the slope (climatic sensitivity), adjusted r² (goodness of fit between growth and climate), and the significance of the regression between climate and standardized growth. Resistant and Susceptible slopes are significantly different at P < 0.05. The Verification data measure the relative quality of the climate reconstructions based on resistant and susceptible models. A perfect reconstruction would have a slope of 1 and an intercept of 0. 

Yet, we must acknowledge that the high proportion of susceptible trees within this environment (≈80%) suggests that the mechanisms of natural selection acting on herbivore susceptibility are complex. It is possible, for example, that the reduced relative response to drought is the result of reduced evapotranspiration potential, as we hypothesized above. If this is the case, the susceptible trees may be buffered to some extent against extreme droughts, the frequency of which is likely higher on this already water-stressed soil. Although these findings demonstrate the complexity of ecological interactions recorded by tree rings, further work within this system is needed to determine the specific mechanisms by which herbivory alters climatic sensitivity.

In addition to altering the magnitude of the response to climate, herbivory also acts to disconnect growth from climate, as shown by the adjusted r² values (Table 1). This disconnection indicates that, within resistant trees, climate is the primary predictor of growth, whereas susceptible trees are likely limited by climate and herbivore load, as well as other factors.

To separate cause and effect between herbivory and climatic sensitivity, we calibrated climate with tree rings in the scale-removed group. As predicted, this model indicates that, for the scale-removed group, both the slope and adjusted r² are higher than those of susceptible trees, but lower than those of resistant (Table 1). Thus, the experimental removal of scales and the resulting change in climatic sensitivity provide another line of evidence arguing that scales are the causal mechanism of reduced climatic sensitivity.

We recognized that there are three major alternative hypotheses that might account for differences in the growth rates and climatic sensitivity of resistant and susceptible trees, variation in proximity to neighboring plants (26), the slope of the surrounding ground, and age (2). Proximity to other plants can reduce available moisture, nutrients, and light by competitive uptake, resulting in reduced mean radial growth. Comparison of the mean distances to the first, second, third, and overall nearest neighbors, however, did not differ significantly (whole model MANOVA P < 0.22) arguing the three groups are experiencing similar competitive interactions (Fig. 3A).

Figure 3.

(A–C) Comparison of mean distance to 1st, 2nd, and 3rd nearest neighbors, soil slope, and age. Although each of these factors can impact the radial growth rates and climatic sensitivity of tree rings, none of these values differed among the three tree groups. Similar letters indicate lack of statistical significance (P > 0.05) using a MANOVA (A) or one-way ANOVA (B and C).

Fritts (2) argued that high soil slope results in an increased sensitivity to drought as a result of reduced incidence of vertical precipitation and increased runoff. Within our study, however, we found no differences in the mean slope of the surrounding substrate between the three groups (resistant = 3.8° ± 0.35°, susceptible = 3.8° ± 0.85°, scale-removed = 2.4° ± 0.32° P < 0.16; Fig. 3B), arguing that variation in localized moisture availability is not the cause of the observed differences in climatic sensitivity.

Age is also known to impact both climatic sensitivity and radial growth rates (2, 23). Specifically, growth rates and interannual variation in growth are commonly high when trees are young, but typically slow and stabilize as the tree matures. Comparison of the mean ages in our study trees, however, did not indicate any differences among the three groups (resistant = 61 ± 2.8, susceptible = 65 ± 2.1, scale-removed = 60 ± 3.2 yr, P < 0.41; Fig. 3C). Overall, none of these alternative hypotheses were supported.

Distortion of Climatic Reconstruction.

The difference in our calibration models argue that the inclusion of insect-susceptible trees can result in the distortion of the climate signal in the tree ring record. To test this possibility, we reconstructed climate (GSPDSI) by using a new set of 22 similarly aged trees (5 resistant and 17 susceptible) growing intermixed at the same site.

The application of these verification trees to the model based on susceptible trees results in the severe distortion of the reconstructed climate (Fig. 4). In Fig. 4, the solid line shows the actual climate (i.e., GSPDSI) values, whereas the dashed and dotted lines indicate values reconstructed by using the resistant and susceptible models, respectively. Note the lines placed at GSPDSI values 6 and −6, which represent the expected bounds of extreme wet and dry events. The year 1992, which was an unusually wet year, serves as an excellent example of the potential for herbivory to distort climate reconstructions. The actual GSPDSI value in 1992 was 4.23, and the value generated by the resistant model was 5.56, which, although high, is within the realm of reason. During this same period, however, the susceptible model generated a GSPDSI value of 10.12, which greatly exceeds expected extremes. Indeed, this reconstructed value would suggest that, in 1992, northern Arizona experienced a precipitation event of epic proportions.

Figure 4.

Reconstruction of climate by using models based on resistant and susceptible trees. Note the severe distortion of amplitude shown by the susceptible model. The trees applied to the models to generate the reconstruction were selected to mimic the proportion of resistant and susceptible trees observed in this system.

To quantify the magnitude of this distortion, we regressed the reconstructed GSPDSI against the actual values generated by the National Oceanic and Atmospheric Administration. A perfect set of reconstructed values would result in a slope of 1, an intercept of 0, and a coefficient of determination (r²) of 1. As Table 1 (Verification section) shows, the slope generated by the resistant model was 0.88, indicating a reasonable reconstruction of the magnitude of variation in climate. Use of the same series for verification of the susceptible model, however, results in a slope of 1.45 (Table 1), which is further from a perfect reconstruction slope of 1. Additionally, the intercept is further from 0, indicating an overestimation of all values. Because of the 60% difference in the amplitude of response to climate and a 50% reduction in precision (i.e., r² for the association of ring growth and climate in Table 1), these findings argue that climate will be best modeled and reconstructed by using insect-resistant trees.

Discussion

Although these data argue that reconstructions of climate will be most accurate when based on resistant or uninfested trees, the relative importance of herbivory within a specific reconstruction will be based on the sampling strategies used. Because distortion due to herbivory is limited to samples collected within the herbivore's range, pooling tree samples across stands, populations, and regions should reduce its impact. However, two problems remain. First, pooling samples regionally may not avoid distortion caused by herbivore outbreaks that are regional in dimension, and previous work has shown that outbreaks can be regional with large-scale impacts on the host population (27). Second, such regional pooling prevents the reconstruction of local climates, which are important if we are to model and understand both the spatial and temporal patterns of climate change. Unfortunately, local chronologies may be particularly sensitive to the role and impact of herbivore populations.

In addition to having an impact on spatially oriented sampling strategies, herbivory should also be considered when using dead remnant wood to extend chronologies through time. The use of ancient dead wood, anchored in time through cross-dating, can greatly lengthen chronologies, allowing them to be extended further into the past than is possible using only living trees. The impact of herbivory, however, may reduce the utility of some dead samples, and so samples collected from regions in which herbivory is likely should be screened for herbivory to ensure the highest quality reconstructions. Punctuated outbreaks may be most easily detected in remnant wood by the tell-tale herbivore signal of suddenly reduced and then released growth. Chronic herbivory, however, may be more difficult to detect, because it will likely result in reduced growth that is stable through time. Without some basis for comparison, such as a controlled experiment, the slow growth rates associated with chronic herbivory may be difficult to distinguish from alternative exogenous factors such as an extended drought or competition.

Genetics may provide the key needed to determine the value of wood samples, as well as developing sampling strategies used in dendrochronological analyses. Previous dendrochronological work has largely ignored the role of genetics in the reconstruction of past biotic and abiotic factors. Here, however, we suggest that tree ring analyses may be affected by genetically based variation in a population. Yet recent work has shown that genetic material can be extracted from some ancient wood fragments (28). Continued development of these techniques may yield genetic markers useful for separating and filtering wood samples for different uses: e.g., resistant trees best suited for reconstructing climate and susceptible trees best suited for reconstructing herbivory.

The need for high-resolution climate data is likely to increase as we move into a period of predicted anthropogenic climate change (for example, ref. 29). The most recent report by the Intergovernmental Panel on Climate Change indicates that increases in greenhouse gases and the associated changes in global climate will result in an increase in extreme climate events such as drought (30). Here, we have shown that herbivory by insects can result in the overestimation of the amplitude of variation. If overestimation has been a problem in past reconstructions, recent periods of extreme wet and dry (e.g., the wet spring of 1992, and the one hundred-year drought of 1996 in the southwestern U.S.) may be further outside of the bounds of normal variation than previously realized, making current estimates of the magnitude of climate change overly conservative. This potential for underestimating the amplitude of recent changes in the global climate suggests that further research on the role and prevalence of herbivory is warranted at sites from which dendroclimatic reconstructions have been or will be generated. Further, because of the interaction between herbivory, host sensitivity, and climatic variation, these data argue that future dendroclimatological models need to consider both biotic and abiotic factors. By developing an understanding of how these factors interact, we can increase the resolution of reconstructions, and, by incorporating long-term experiments and greater biocomplexity (31) in reconstructions of the past, we can expect to gain a better understanding of current and future climate changes, and the impact these changes will have on ecosystems.

Abbreviation

GSPDSI

Palmer Drought Severity Index averaged over the May–October growing season