δ¹³C and water-use efficiency in Australian grasstrees and South African conifers over the last century

Abstract

Annual or biannual time courses of plant δ¹³C (δ¹³C_p) over the last century (70–100 years) were recorded for leafbases of four grasstrees (Xanthorrhoea preissii) at four sites in mediterranean Australia and wood of four conifers (Widdringtonia cedarbergensis) at two sites in mediterranean South Africa. There was a strong downward trend of 2–5.5^‰ from 1935 to 1940 to the present in the eight plants. Trends were more variable from 1900 to 1940 with plants at two sites of each species showing an upward trend of 1–2.5‰. Accepting that δ¹³C of the air (δ¹³C_a) fell by almost 2‰ over the last century, the ratio of leaf intercellular CO₂ to atmospheric CO₂ (c _i/c _a) rose in five plants and remained unchanged in three over that period. Changes in c _i/c _a rather than δ¹³C_a were more closely correlated with changes in δ¹³C_p and accounted for 6.7–71.8% (22.6c _i/c _a) and 28.2–93.3% (δ¹³C_a) of the variation in δ¹³C_p. We doubt that possible changing patterns of rainfall, water availability, temperature, shade, air pollution or clearing for agriculture have contributed to the overall trend for c _i/c _a to rise over time. Instead, we provide evidence (concentrations of Fe and Mn in the grasstree leafbases) that decreasing photosynthetic capacity associated with falling nutrient availability due to the reduced occurrence of fire may have contributed to rising c _i/c _a. Intrinsic water-use efficiency (W _i) as a function of (c _a–c _i) usually increased linearly over the period, with the two exceptions explained by their marked increase in c _i/c _a. We conclude that grasstrees may provide equivalent δ¹³C_p and W _i data to long-lived conifers and that their interpretation requires a consideration of the causes of variation in both c _i/c _a and δ¹³C_a.

Introduction

Use of ¹²CO₂ is favoured over ¹³CO₂ in photosynthetic carbon fixation (Farquhar et al. 1982). Physical discrimination (fractionation) against ¹³CO₂ occurs when CO₂ diffuses from the atmosphere through the stomates and intercellular spaces to the carboxylation sites within the chloroplasts. Further chemical discrimination favours ¹²CO₂ at the carboxylation sites in the chloroplast matrix where reduction of CO₂ to phosphoglycerate occurs. Isotopic discrimination measurements can be expressed by the quantity delta, δ. For the plant, δ¹³C_p is given by:

$$ \delta ^{{\rm{13}}} {\rm{C}}_{\rm{p}} {\rm{ = [}}\delta ^{{\rm{13}}} {\rm{Ca + }}a{\rm{ - (}}a{\rm{ - }}b{\rm{)\cdot}}c_{\rm{i}} {\rm{/}}c_{\rm{a}} {\rm{]}} $$

(1)

(Farquhar et al. 1982), where δ¹³C_a is discrimination in air compared with a standard (see below), a is the coefficient of diffusion discrimination (4.4‰), b is the coefficient of carboxylation discrimination (27.0‰), and c _i/c _a is the ratio of intercellular CO₂ to atmospheric CO₂ concentration (which varies).

There has been a gradual increase in atmospheric CO₂ levels (c _a) since the start of the industrial revolution. Since fossil fuels are of plant origin, they are enriched in ¹²C (Flanagan and Ehleringer 1998). This means that δ¹³C_a has been declining (Friedli et al. 1986, Fig. 1a). Work on tree rings in the Northern Hemisphere over the last hundred years or more has shown an overall decline in δ¹³C_p (Freyer 1979; Stuiver et al. 1984; Feng 1999). Since the associated decline in ¹³C_a over this time is well established, it is this component of Eq. 1 that has usually been held responsible for the fall in δ¹³C_p. This assumes that any change in c _i/c _a has been negligible (Fig. 1b). We examined this assumption here.

Fig. 1. a

δ¹³C_a over time from Friedli et al. (1986) and extrapolated from 1977 based on data for 1956–1977, b δ¹³C_p vs δ¹³C_a for constant c _i/c _a, c δ¹³C_p vs c _i/c _a for constant δ¹³C_a, d W _i vs c _a at constant c _i/c _a, and e W _i vs c _i at three levels of c _a, with stars set at W _i = 55 and show how W _i may change as c _a and c _i increase. See Appendix for calculations

Where photosynthetic capacity is not limiting, c _i will increase at a lower rate (or remain constant) than the increase in c _a, with a reduction in c _i/c _a and a rise in δ¹³C_p (Fig. 1c). However, if mineral nutrients for example are in short supply, fixation rates may be limited, c _i/c _a will rise, and δ¹³C_p will fall. Thus, c _a has both predictable (on δ¹³C_a) and unpredictable (on c _i) effects on δ¹³C_p over time. Note that at constant c _i/c _a, δ¹³C_p increases linearly with increasing δ¹³C_a (slope =1, Fig. 1b). At constant δ¹³C_a, δ¹³C_p increases linearly with decreasing c _i/c _a (slope =a−b =22.6, Fig. 1c). These slope values indicate that δ¹³C_p is far more sensitive to variation in c _i/c _a than to variation in δ¹³C_a

A physiological attribute related to isotope discrimination is the ratio of net assimilation rate to transpiration rate, or water-use efficiency (W) (Farquhar and Richards 1984). A rise in W can be due to a fall in stomatal conductance, say from a reduction in water availability (Pita et al. 2001). Here, c _i is drawn down strongly relative to c _a, c _i/c _a declines and thus δ¹³C_p rises (Eq. 1). Where stomatal conductance remains constant but c _a increases, W should also increase. Since we are specifically interested in the latter case, there is merit in defining W as a function of (c _a –c _i) (Ehleringer et al 1993, see Appendix for estimating c _a, c _i and W). Since this expression assumes that transpiration is constant, it is more correctly termed 'intrinsic' water-use efficiency (W _i). This yields the expected increase in W for an increase in c _a, but constant c _i/c _a (Fig. 1d), and an increase in W for a declining c _i/c _a and fixed c _a (Fig. 1e). Recent tree ring work in the North Hemisphere has shown an increase in W _i over time associated with the increasing c _a (Bert et al. 1997; Duquesnay et al 1998; Feng 1999).

Grasstrees in Australia are arborescent monocotyledons in the genera Xanthorrhoea, Kingia and Dasypogon (Liliales) that survive for centuries, even in the presence of frequent fire (Lamont and Downes 1979; Staff and Waterhouse, 1981). They consist of a simple or branched caudex (stem) surrounded by dead leaves and a terminal spray of dark green leaves. The needle-like leaves, often exceeding 1 m in length, progressively die off in Xanthorrhoea as the grasstree grows and bend downwards to form a skirt that is burnt back to the leafbases during fire. Grinding off the charred surface of the persistent leafbases reveals seasonal banding—regular cream and brown bands 1–2 cm in the vertical plane (spring/summer and autumn/winter growth, respectively), with occasional black bands replacing the cream or brown bands due to the effect of fire (Colangelo et al. 2002). This technique has recently been used to determine the fire history of X. preissii in mediterranean Australia over the past 250 years (Ward et al. 2001; Lamont et al. 2003).

We thought that the phenomenon might also be able to provide data on δ¹³C_p levels through time, equivalent to tree rings, except that the rings of leafbases run up the stem rather than concentrically; they are much more convenient to sample as they are an order of magnitude larger and on the outside of the stem. We therefore removed leafbases in successive colour bands from 1900 onwards and analyzed their δ¹³C_p levels. Since fire rejuvenates the nutrient content of the leaves (Ward et al. 2001) we analyzed leafbases of unburnt and frequently burnt plants for Fe and Mn (required in photosynthesis but immobile and poorly leachable) at the same time. To increase the generality of our study, the grasstree data were compared with δ¹³C_p data for annual tree rings of Widdringtonia cedarbergensis (Cupressaceae), a conifer endemic to the Cedarberg Mountains near Cape Town, South Africa. This provided equivalent data using the new and traditional techniques, for two mediterranean regions on either side of the Indian Ocean, both in the Southern Hemisphere and subject to the same air circulation system. We interpreted the results in terms of expected (Feng 1999) or observed changes in δ¹³C_a, c _i/c _a, rainfall, and internal Fe and Mn (as they might affect c _i/c _a ) over the last century. We calculated W _i and compared the results with those for North American conifers (Feng 1999).

Materials and methods

Stable carbon isotope ratios (δ¹³C_p) of leaf bases were determined along a Xanthorrhoea preissii stem harvested from a site not burnt since 1931 (Amphion unburnt) near Dwellingup, southwestern Australia, and another in an adjacent block subjected to the usual fire management regime for eucalypt forest (Ward et al. 2001) (Amphion burnt 1931–1997). A third plant was obtained from a regularly burnt site at Gyngoorda between Bannister and Boddington (1900–1987). A fourth also subjected to frequent fires was collected from Boonerring also near Dwellingup (1900–1997). All plants were at least 100 years old by 1900. While X. preissii never stops growing, leaf production is least in autumn/winter, and leaves take a year to stop elongating (D. Korczynskyj, personal communication). Since leafbases are the last part of the leaf to mature, we assume that their C arises from both self and previous leaf growth.

The outermost, longitudinal 1-cm sections of leafbases comprising the autumn/winter (brown), and sometimes the spring/summer (cream) and fire (black) bands, were removed sequentially from each stem, after melting the surrounding resin with a heat gun. Each sample contained 3–6 leafbases per colour type. Resin was removed with laquer thinner from the leafbase material that was then dried to constant weight at 70–80°C. The leaf base material was milled with a ball mill (Retsch MM-2, Haan, Germany) and analyzed for ¹³C/¹²C using a Roboprep connected to a Tracermass isotope ratio spectrometer (Europa Scientific, Crewe, UK) at the Biogeochemistry Centre, University of Western Australia. Samples were compared with a secondary reference sample of Acacia pulchella leaves (50.10% C) standardised against primary analytical standards from the International Atomic Energy Agency in Vienna. Analytical variance in the reference never exceeded 0.2‰. Plants were aged relative to the last known fire corresponding to the uppermost black band and all cream-brown bands were treated as annual increments (Ward et al. 2001; Colangelo et al. 2002; Lamont et al. 2003).

In southwestern South Africa, Dunwiddie and La Marche (1980) reported on an annual ring width index chronology for Widdringtonia cedarbergensis in the Cedarberg Mountains. This chronology was derived from 52 radii of 32 trees from Die Bos. An average of two cores from each of six of these trees, all of which had previously been aged by Dunwiddie and La Marche (1980), were selected for stable carbon isotope analysis from 1890 to 1976 (February and Stock 1999). For ease of comparison, this is considered as one plant in the rest of the text. Three trees of W. cedarbergensis were harvested from Krakadouw (here called 38, 39 and 40) and treated in the same way for 1900–1994. All of the Die Bos trees were 150–250 years old by 1900, while the Krakadouw trees established in 1898. All but two of the conifers were between 150 and 250 years old by 1900. The individual rings were removed from the cores with a scalpel under a stereomicroscope. Samples from corresponding annual rings for the six trees at Die Bos were combined for analysis but the Krakadouw samples were kept separate. W. cedarbergensis grows in spring-summer and we assumed that wood C arose from current growth.

The sodium chlorite delignification technique of Leavitt and Danzer (1993) was used to extract holocellulose from wood. See February and Stock (1999) for further preparation details. Stable carbon isotope analysis of the conifer samples was carried out in the Archaeometry laboratory at the University of Cape Town on a Micromass 602E spectrometer. The standard was a laboratory reference gas related to the Chicago PDB marine carbonate standard calibrated against six NBS reference standards (Sealy 1986). The ¹³C/¹²C ratio was expressed relative to the standard using the formula, δ=[(R _sample/R _standard)–1]×1,000, where R signifies the isotopic ratio. For three grasstree stems, two ratios per year (autumn/winter and spring/summer) were determined for at least part of the total time sequence along the stem.

δ¹³C_a values for 1900–1977 were obtained from Friedli et al. (1986) collated from various parts of the world and the straight part of the relationship for 1955–1977 extrapolated to 1997 (Fig. 1a, Appendix). Rainfall for the wettest and driest 6 months per year was obtained for Dwellingup and Bannister-Boddington since 1930 from the Western Australian Bureau of Meteorology. Annual rainfall was obtained from the Weather Bureau, Department of the Environment, Pretoria for Algeria (1900–1997) and Wupperthal (1900–1977). These were plotted against time and any linear trends noted. Knowing δ¹³C_p, δ¹³C_a, a and b, the ratio c _i/c _a was calculated (Appendix). These were plotted against time and any trends noted. The linear correlation coefficients between δ¹³C_p and δ¹³C_a and δ¹³C_p and c _i/c _a were determined. The slopes (b) of the regression lines against time for δ¹³C_p and δ¹³C_a were compared: 'apparent' contribution of c _i/c _a to δ¹³C_p was given by the expression X=[(b _p–b _a)/b _p]/(a–b) and of δ¹³C_a by 1–X. 'Actual' contribution of c _i/c _a to δ¹³C_p was given by (b _p–b _a)/b _p and of δ¹³C_a by b _a/b _p.

In an attempt to explain variations in c _i/c _a at the level of c _i rather than c _a, concentrations of Fe and Mn [highly immobile once deposited (Marschner 1995) and strongly resistant to leaching from dead grasstree tissues (R. Wittkuhn, unpublished data)] were determined on the same material as for δ¹³C_p for the Amphion (unburnt) and Gyngoorda grasstrees. Material was digested with concentrated HNO₃/HClO₃ and analyzed by inductively-coupled plasmascan-atomic emission spectroscopy at the Chemistry Centre, Perth. Concentrations were plotted against time and c _i/c _a. W _i was calculated as 0.625(c _a–c _i) (Ehleringer et al. 1993), where c _a was obtained from a best-fit curve to data from Mauna Loa, Pacific Ocean (Friedli et al. 1986) and c _i from Eq. 1 (Appendix), plotted against time and any linear trends noted. Finally, all records for which δ¹³C_p, δ¹³C_a, rainfall, c _i/c _a and W _i were complete at any time (67 years) were combined for the four grasstrees, and separately for the four conifers. These were subjected to (geometric) partial correlation coefficient analysis as the starting point for principal components analysis (Podani 1995).

Results

δ¹³C for the eight plants trended downwards, especially since 1930–1940 (Fig. 2). The Gyngoorda grasstree trended downward from 1900, although the Amphion (burnt) and Boonerring plants and the Krakadouw 40 and Die Bos conifers trended upwards to 1940 before falling. There was no consistent seasonal (colour band) trend for the most comprehensive data sets, although autumn/ winter-initiated leafbases had higher δ¹³C_p than spring/summer bases in the Amphion (burnt) grasstree (Fig. 2c).

Fig. 2.

δ¹³C_p vs time for grasstrees at a Amphion (unburnt), b Amphion (burnt), c Gyngoorda (burnt), and d Boonerring (burnt) in southwestern Australia, and for conifers at e Krakadow (tree 38), f Krakadow (tree 39), g Krakadow (tree 40), and h Die Bos in South Africa. Open circles are samples from leaves initiated in autumn-winter, and filled circles in spring-summer

Wet season rainfall fell substantially over time at one grasstree station but was not significant at the station used for the other three (Table 1). Annual rainfall rose markedly for the station used for the Krakadow plants but there was no trend for the other. There was a rise in c _i/c _a for five of the eight plants while the other three showed no trend (Table 2). Fe and Mn concentrations fell over time in the Amphion (unburnt) grasstree but trends were insignificant (Fe) or negligible (Mn) in the Gyngoorda plant (Fig. 3a, b). The c _i/c _a fell as Fe and Mn concentrations increased for the unburnt grasstree but there were no relationships for the burnt tree (Fig. 3c, d).

Table 1. Best fit lines to annual rainfall (mm) for four closest stations to collection sites for 60–100 years in the period 1900–2000. Dwellingup (closest to Amphion and Boonerring) Bannister-Boddington (Gyngoorda), Algeria (Krakadouw) and Wupperthal (Die Bos)

Table 2. Best fit lines to calculate c _i/c _a from Eq. 1, Figs. 1a, 2 and 3. Trees as in Fig. 2

Fig. 3.

Log_e Fe and Mn concentration (mg kg⁻¹) vs time for a Amphion unburnt grasstree (Fe: Y=113.9–0.057X, r ²=53%, P<0.0001, Mn: Y=46.4 –0.023X, r ²=46%, P<0.0001) and b Gyngoorda, (Fe: P>0.05, Mn: Y=8.43–0.004X, r ²=8%, P=0.002), and c _i/c _a vs log_e Fe and Mn concentrations for c Amphion (Fe: Y=0.81–0.031X, r ²=62%, P<0.0001, Mn: Y=0.84–0.069X, r ²=59%, P<0.0001) and d Gyngoorda (Fe, Mn: P>0.05)

All but the die Bos δ¹³C_p values were significantly correlated with δ¹³C_a (with c _i/c _a as the other independent variable) and the level of correlation was ≥c _i/c _a for six of the eight trees (Table 3). Variation in δ¹³C_a accounted for 82.7–99.7% of variation in δ¹³C_p and c _i/c _a accounted for 0.5–17.3% of variation in δ¹³C_p. Since the contribution of c _i/c _a to δ¹³C_p is multiplicative [by(b–a)=22.6 in Eq. 1] while δ¹³C_a is additive, the contribution of c _i/c _a to δ¹³Cp increased to 7.7–71.8% while δ¹³C_a fell to 28.2–93.3% (Table 3). W _i rose for five plants, remained unchanged for one, and fell for two (those with the greatest increase in c _i/c _a) over time (Table 4).

Table 3. Best-fit lines to W _i (µmol mol^-1) calculated from Appendix and data for Table 2. Trees as in Fig. 2

Table 4. Contribution of δ¹³C_a and c_i/c_a to variation in δ¹³C_p over the twentieth century in four grasstrees and four conifers. Rank = ranking of correlation with δ¹³C_p; Apparent contribution = relative slope of lines fitted to δ¹³C_p, with c _i/c _a corrected for effect of (a−b) in Eq. 1; actual contribution = relative slope with (a−b) left in. NS Non-significant, all other correlations were significant at P<0.05. Trees numbered as in Fig. 2

The correlation matrix shows that c _i/c _a was more strongly correlated (negatively) with δ¹³C_p than was δ¹³C_a (positively) (Table 5). Annual winter rain was strongly correlated with that of rainfall summed with the 12-month rainfall on either side (i.e. three winters) and less strongly with year. Both were unrelated to δ¹³C_p or c _i/c _a although c _i/c _a was weakly correlated with year. W _i showed a strong positive relationship with δ¹³C_p and strong negative relationship with c _i/c _a. For the four conifers collectively, c _i/c _a was more strongly correlated (negatively) with δ¹³C_p than was δ¹³C_a (positively). Annual rainfall was correlated with year but was unrelated to δ¹³C_p or c _i/c _a, while c _i/c _a was correlated with year. W _i showed a strong positive relationship with δ¹³C_p and strong negative relationship with c _i/c _a.

Table 5. Correlation matrix for five attributes measured on the four combined grasstree (n=222) over 67 years (lower rectangle) and the four combined conifer (n=279) over 67 years (upper rectangle). P<0.001 for r=±0.221 and df =200 (2-tailed)

Discussion

From 1935 to 1940 to the present, there was a clear downward trend of 2–5.5‰ in δ¹³C_p for the eight plants. The trend was less clear for 1900 to 1940, partly because records for five plants only started from 1920–30. Nevertheless, four plants showed an upward trend of 1–2.5‰ from 1900–30 to 1940 and one a slight fall of 0.5‰. These trends agree with many of the conifers reported for Europe and the west coast of North America (Stuiver et al. 1984; Leavitt and Long 1992; Feng and Epstein 1995) although their rates were lower, 1–3‰ (Francey and Farquhar 1982; Feng 1999). However, while most showed a downward trend from 1940, few showed a clear increase for 1900 to 1940. Conifers in Tasmania, SE Australia, showed no overall change in δ¹³C_p for the period analyzed (Pearman et al. 1976).

All these trends might be explained by local phenomena (e.g. change in precipitation) or a combination of local and global phenomena (e.g. decreasing δ¹³C_a). For those with a dominant downward trend the most obvious likely cause is the δ¹³C_a component of δ¹³C_p (Fig. 1a). Only one site (Die Bos) did not show a strong positive correlation between δ¹³C_p and δ¹³C_a and this may be significant as it is the mean of six trees. The Friedli et al. (1986) data show a slight decline in δ¹³C_a up to 1955, followed by a stronger decline (1.5‰) thereafter, but provide no expectation of 1935–40 as a turning point in δ¹³C_p. It has been suggested that warmer summers should increase the level of ¹³C in the atmosphere, but there is no evidence of this for the Southern Hemisphere in the 1900–40 period (Pearman et al. 1976). Physiological and microenvironmental changes with plant stand age have been used to explain initial increases in δ¹³C_p (Duquesnay et al. 1998) followed by a decrease (Marshall and Monserud 1996). This cannot explain why the plants peak in our case, and simultaneously at that, as the grasstrees and Die Bos trees would have been already old (100–250 years) for the oldest date analyzed.

δ¹³C_p is inversely correlated with rainfall and with more direct measures of water availability in woody plants (Tans and Mook 1980; Pita et al. 2001; Lamont et al. 2002). Rainfall data were only available from 1930 for the grasstrees, i.e. covering the downward trend line for δ¹³C_p. Winter rainfall showed a slight but non-significant downward trend for the station nearest three of the grasstrees and a significant downward trend for the fourth (i.e., in the 'wrong' direction). Even when combined with the previous summer-winter (to account for a likely source of initial C) and/or the subsequent summer-winter (because the leaves take 12 months to mature, D. Korczynskyj, personal communication) the lack of correlation with δ¹³C_p was maintained (Table 5). These factors no doubt explain why there was no consistent seasonal effect on δ¹³C_p either (Table 1). While there was a full rainfall record for the conifers, the stations were 7 km away and other nearby stations with less complete records showed quite different patterns and annual totals, as expected in this rugged landscape. There was no overall correlation between rainfall and δ¹³C_p for the conifers either (Table 5).

Any rainfall effect on δ¹³C_p would have been expressed through c _i/c _a: an increase in water status should increase stomatal conductance and hence c _i. This did occur for the Krakadouw conifers but this may be a coincidence, for their actual rainfall may have been more similar to Wupperthal's with which there was no relationship. There was no overall correlation between rainfall and c _i/c _a for both sets of plants (Table 5). It is postulated that there has been a marked increase of water availability in eucalypt forest due to treefelling (Davison 1997) while grasstree water relations are sensitive to water availability (Crombie 1988). The Amphion (unburnt) block has not been logged since 1931 but showed the most marked increase in c _i/c _a with time for the entire analysis period, while the other grasstrees, with a small or no trend, grew in blocks subjected to selective logging of large eucalypts. The conifers grew in remote nature reserves not subject to logging with three showing a marked rise in c _i/c _a since 1920. Forest management as it might affect water availability therefore seems to have played no role in accounting for the trends in δ¹³C_p.

Increasing shade could have contributed to the rise in c _i/c _a of the unburnt grasstree (Francey and Farquhar 1982), as mid-level trees gradually overtopped shorter grasstrees. However, the same argument could not hold for the rise in c _i/c _a of the matching Amphion grasstree where fires would have opened up the overstorey every 5–10 years (Ward et al. 2001), more consistent with the other two grasstrees that showed no trend. It also could not apply to the conifers that were by far the largest plants in the landscape and their shoot architecture and scale leaves ensured that self-shading did not increase with time. All sites were at least 80 km from a city and deep within the water catchment so that air pollutants were unlikely to be responsible for any of the rises in c _i/c _a over time (Francey and Farquhar 1982). Decreasing temperatures also provide a mechanism for increasing c _i/c _a. The temperature data were no more comprehensive or reliable than those for rainfall, but it is reasonable to assume that it has been trending upwards as part of global warming (Hughes 2000).

Deserving of more serious consideration is the effect of mineral nutrient availability on photosynthetic capacity and thus c _i/c _a (Ehleringer et al. 1993). We report on Fe and Mn levels here as (1) they are directly involved in photosynthesis, (2) they are not readily absorbed by plants, and (3) they are highly immobile once deposited in the tissues and resist leaching out of the leafbases over time (Marschner 1995; R Wittkuhn, unpublished data). The unburnt grasstree showed a clear downward trend in Fe and Mn concentrations since the fire. This is consistent with a marked decline in Fe and Mn levels in dead leaves within individual grasstrees over 15 years since fire in the same vegetation type (R. Wittkuhn, unpublished data). Thus, the increase in c _i/c _a can be attributed to a declining photosynthetic capacity with time since fire as nutrients become less available for uptake. The same interpretation applies to the three Krakadouw conifers that have escaped fire for over 70 years (the species is killed by fire), but not the Die Bos tree which showed no trend in c _i/c _a. The slight decline for Mn and no trend for Fe over time in the burnt grasstree reflect bouts of increased nutrient availability with each fire (whose frequency has been declining, especially since 1940, Ward et al. 2001; Lamont et al. 2003) and are consistent with the slight increase or no trend in c _i/c _a for the three burnt grasstrees.

Recent research has emphasized the relationship between δ¹³C_p and δ¹³C_a (February and Stock 1999; Feng 1999). However, our results show that c _i/c _a may individually (Table 3) and collectively (Table 5) be better correlated with δ¹³C_p. In absolute terms, δ¹³C_a still contributes most to δ¹³C_p, but the fact that c _i/c _a is multiplied by 22.6, i.e. (a–b), in Eq. 1 means that its contribution exceeded 30% in five plants and reached at least 60% in two. Our calculations assume that the δ¹³C_a and c _a values we used, gathered from a Pacific Island location in particular (Friedli et al. 1986; Feng 1998), applied to the air around our plants. Both locations were in indigenous vegetation remote from industry so must have had similar values to these as well as to each other. Clearing for agriculture within 10 km of the grasstree stands may have released CO₂ at a faster rate than we assumed (Freyer 1979). Underestimating the rate of c _a rise will have overestimated the rate of rise of c _i/c _a—this trend was universal in the Cape plants post-1930–1940 where their remote, mountain location suggests such an underestimate is unlikely. The tendency for δ¹³C_p to rise up until then in several trees cannot be related to expected changes in c _a (Fig. 1a). Any cause of this peak must have operated on both sides of the Indian Ocean yet have been unevenly expressed in different trees.

Has the escalating increase in c _a led to a higher level of C fixation as expressed by a higher W _i? Overall, there has been no trend with time (Table 5). For five plants the increase in W _i has occurred at an overall rate of 7–17 μmol mol⁻¹ decade⁻¹. For the two plants with a marked increase in c _i/c _a there has been a decrease in W _i of 22–26 μmol mol⁻¹ decade⁻¹. Thus, a marked decline in δ¹³C_p can imply a decrease in W _i when c _i increases at a much faster rate than c _a (Fig. 1e, Tables 2, 4). W _i increases when c _i/c _a remains constant (Fig. 1d), increases slowly (e.g. Tables 2, 4), or decreases (not observed here). This is because W _i is a function of c _a(1–c _i/c _a). The increases in W _i for our five trees (3–18.5%) were somewhat less than those recorded by Feng (1999) over the same period (7.5–25.5%) for 21 of 23 trees. Underestimating the rate of c _a rise will have underestimated the already clear rises in W _i for six trees, and exaggerated the fall in the other two. For nine grasstrees in the same forest type (J. Eldridge, unpublished data) and these conifers (E. February, unpublished data) we were not able to detect an increase in productivity (band widths) that might be associated with higher W _i as suggested by Feng (1999). Despite marked trends in c _a, and sometimes rainfall and nutrient levels, overall growth rates have been conserved.

Our results show that the leafbases of grasstrees in southwestern Australia are suitable for time-series analysis of δ¹³C_p, producing similar patterns as for conifer wood in southern Africa over the last 100 years. While grasstrees are much more convenient to use than wood, the technique is limited by plant longevity to the last 250 years. The universal downward trend in δ¹³C_p since 1935–1940 is more closely correlated with a rise in c _i/c _a than a fall in δ¹³C_a. We attribute the unexpected rise in c_i/c_a to declining nutrient availability associated with reduced fire frequency, and the topic deserves further research.

Appendix

Calculation steps

• Step 1. Calculate c _a:

$$ c_{\rm{a}} {\rm{ = 277}}{\rm{.78 + 1}}{\rm{.350\cdot exp[0}}{\rm{.01572\cdot(year - 1740)]}} $$

(2)

(Feng 1998).

• Step 2. Calculate carbon isotope discrimination by the plant:

$$ \Delta {\rm{13 = (}}\delta ^{{\rm{13}}} {\rm{C}}_{\rm{a}} {\rm{ - }}\delta ^{{\rm{13}}} {\rm{C}}_{\rm{p}} {\rm{)/(1 + }}\delta ^{{\rm{13}}} {\rm{C}}_{\rm{p}} {\rm{/1,000)}} $$

(3)

(Feng 1999, citing Farquhar et al. 1989).

• Step 3. Calculate c _i using Δ13 and c _a above:

$$ {\rm{ }}c_{\rm{i}} {\rm{ = }}c_{\rm{a}} {\rm{\cdot[(}}\Delta {\rm{13 - }}a{\rm{)/(}}b{\rm{ - }}a{\rm{)]}}\;{\rm{with}}\;a{\rm{ = 4}}{\rm{.4}}{\rm{,}}\;b{\rm{ = 27}}{\rm{.0}} $$

(4)

(Feng 1998).

• Step 4. Finally, calculate W _i using c _a and c _i above:

$$ W_{\rm{i}} {\rm{ = (}}c_{\rm{a}} {\rm{ - }}c_{\rm{i}} {\rm{)/1}}{\rm{.6}} $$

(5)

(Feng 1999, citing Ehleringer et al. 1993).