The oxygen isotopic composition of phytolith assemblages from tropical rainforest soil tops (Queensland, Australia): validation of a new paleoenvironmental tool

Phytoliths are micrometric particles of amorphous silica that form inside or between the cells of higher plant tissues throughout the life of a plant. With plant decay, phytoliths are either incorporated into soils or exported to sediments via regional watersheds. Phytolith morphological assemblages are increasingly used as proxy of grassland diversity and tree cover density in inter-tropical areas. Here, we investigate whether, along altitudinal gradients in northeast Queensland (Australia), changes in the δ18O signature of soil top phytolith assemblages reflect changes in mean annual temperature (MAT) and in the oxygen isotopic composition of precipitation ( δOprecipitation), as predicted by equilibrium temperature coefficients previously published for silica. Oxygen isotopic analyses were performed on 16 phytolith samples, after controlled isotopic exchange (CIE), using the IR Laser-Heating Fluorination Technique. Long-term mean annual precipitation (MAP) and MAT values at the sampled sites were calculated by the ANUCLIM software. δOprecipitation estimates were calculated using the Bowen and Wilkinson (2002) model, slightly modified. An empirical temperature-dependant relationship was obtained: 1Owood phytolith-precipitation (‰ vs. VSMOW) =−0.4 (±0.2)t (C) + 46 (±3) (R2 = 0.4, p < 0.05;n = 12). Despite the various unknowns introduced when estimatingδOprecipitationvalues and the large uncertainties onδOwood phytolith values, the temperature coefficient (−0.4± 0.2 ‰◦C−1) is in the range of values previously obtained for natural quartz, fresh and sedimentary diatoms and harvested grass phytoliths (from −0.2 to −0.5 ‰◦C−1). The consistency supports the reliability of δOwood phytolithsignatures for recording relative changes in mean annual δOsoil water values (which are assumed to be equivalent to the weighted annual δOprecipitation values in rainforests environments) and MAT, provided these changes were several ‰ and/or several C in magnitude.


Introduction
Phytoliths are micrometric particles (<60-100 µm of diameter) of amorphous silica that form within a matter of hours to days (Perry et al., 1987) inside or between the cells of higher plant tissues throughout the life of a plant.With plant decay, phytoliths are either incorporated into soils or exported to sediments via regional watersheds.Phytolith morphological assemblages are increasingly used as proxy of grassland diversity and tree cover density in inter-tropical areas (e.g.Alexandre et al., 1998;Boyd et al., 2005;Bremond et al., 2005a,b, 2008a,b Piperno, 2006;Lentfer and Torrence, 2007;Neuman et al., 2009).In parallel, pioneering studies of the oxygen isotope composition (δ 18 O) of phytoliths have demonstrated that in non-transpiring grass stem tissues, the equilibrium fractionation between water and phytolith is temperature-dependent (Shahack-Gross et al., 1996;Webb and Longstaffe, 2000, 2002, 2003, 2006).The obtained temperature coefficient is −0.33 ‰ • C −1 (recalculated from A. Alexandre et al.: The oxygen isotopic composition of phytolith assemblages from tropical rainforest soil tops Shahack-Gross et al., 1996), in the −0.2 to −0.5 ‰ • C −1 range of temperature coefficients previously measured for biogenic silica (diatom frustules) and quartz (Clayton et al., 1972;Matsuhisa et al., 1979;Juillet-Leclerc and Labeyrie, 1987;Shemesh et al., 1992;Sharp and Kirschner, 1994;Brandriss et al., 1998;Moshen et al., 2005;Dodd and Sharp, 2010;Crespin et al., 2010).Those studies also evidenced that in transpiring tissues, 18 O enrichment of sap water increases with the inverse of relative humidity, which limits the use of the δ 18 O signature of phytoliths from grass leaves as a function of temperature and the isotopic composition of soil water (δ 18 O soil water ) (Webb and Longstaffe, 2000, 2002, 2003, 2006).These calibration studies, although extremely useful for our understanding of the temperature and soil-water δ 18 O signals carried by phytoliths, have not lead to paleoenvironmental reconstructions due to the fact that phytoliths from both transpiring and non-transpiring grass tissues are not morphologically distinguishable.
In this paper, we explore if the δ 18 O signature of rainforest phytolith assemblages (Fig. 1.), which consist of more than 80 % of a single phytolith type produced in non-transpiring wood (Alexandre et al., 1997;Runge, 1999;Bremond et al., 2005b), can be used as climate proxy.We investigate whether, along altitudinal gradients in northeast Queensland (Australia), changes in the δ 18 O signature of soil top phytolith assemblages reflect changes in mean annual temperature (MAT) and in the oxygen isotopic composition of precipitation (δ 18 O precipitation ), as predicted by equilibrium temperature coefficients previously published for silica.
With the overall objectives of paleoenvironmental reconstructions, we consider the soil top phytolith assemblages as modern reference assemblages for the following reasons: (1) in northeast Queensland, lake catchments are small (e.g.Haberle, 2005;Kershaw et al., 2007) and the majority of phytoliths found in lake sediments is expected to come from soil tops; (2) the time span recorded by soil top phytolith assemblages should fit with the time span recorded by fossil phytolith assemblages sampled from sedimentary cores.Indeed, the weak concentration of phytoliths in rainforest soil tops (a few ‰ in weight, Alexandre et al., 1997) and in the suspended load of tropical rivers (less than 5 ‰ in weight, Carry et al., 2005) suggests that the mean phytolith concentration in lake sediments is also on the order of a few ‰ in weight.Taking into account the amount of phytoliths required for δ 18 O analyses (several mg) and using the lake sediment accumulation rates commonly observed in Queensland rainforest environments (cm/100 yr; e.g.Haberle, 2005;Rieser and Wust, 2010), the fossil phytolith assemblages provided using sampling intervals of several centimeters in lake sediment cores should encompass 100s of years.Mean age of 100s of years was measured for bulk organic matter (OM) from one of the soil top samples investigated here ( 14 C mean age of sample #18 was cal.605 ± 36 yr BP; AMS-14 C UCI-AMS #75077, calibrated after Danzeglocke et al., 2011), and can be reasonably expected for phytoliths (Alexandre et al., 10µm 1999); and (3) soil top phytolith assemblages of hundreds of years may have been subject to early selective dissolution (Alexandre et al., 1997), alike fossil phytolith assemblages.
In the absence of measured δ 18 O precipitation and δ 18 O soil values, and in order to limit the uncertainty associated with δ 18 O precipitation estimates, we deliberately chose to sample 16 sites along 4 altitude gradients with close environmental characteristics: (1) podzolic soils are developed at the expense of granitic and metamorphic parent-rocks; (2) all sites (except site 21) are covered by a closed rainforest with a very low range of vapor pressure deficit for much of the time (0-0.2kPa; Hutley et al., 1997); (3) most of the precipitation comes from air masses brought by the dominant NW monsoonal winds; (4) altitude is the main control on temperature (DASETT, 1986) and presumably on δ 18 O precipitation and δ 18 O soil values variations.
Phytolith δ 18 O values, measured using the IR Laser-Heating Fluorination Technique, were correlated with mean annual temperature (MAT) values obtained using the ANUCLIM software (McMahon et al., 1995) and with δ 18 O precipitation estimates using the Bowen and Wilkinson (2002) model slightly modified.Although this approach requires to introduce various unknowns and to deal with large uncertainties, it shows that δ 18 O wood phytolith values from the leeward slopes record the modern combination of weighted annual δ 18 O precipitation values and MAT.

Geomorphic features
The Wet Tropical rainforests of Queensland extend from 16 • 27 to 17   Tracey (1982).δ 18 O precipitation values are estimated from Eqs. ( 7) and ( 8) slightly modified after Bowen and Wilkinson (2002).distribution of rainforests in northern Queensland straddles three geologic and geomorphic regions that are NW-SE or N-S oriented (Fig. 2a and b): the basaltic tablelands, with an average altitude of 760 m a.s.l. on the west; the alluvial, granitic, and metamorphic lower coastal belt (0-900 m a.s.l.) on the east that includes a coastal plain and a coastal range; and the intermediate granitic and metamorphic eastern highlands with isolated peaks up to 1545 m a.s.l.(Mt.Belleden Ker) and 1622 m a.s.l.(Mt.Bartle Frere).The eastern highlands and the tablelands belong to the Great Dividing Range.Red or yellow loams, red podzolic soils, xanthozems, and krasnozems were developed throughout the weathering of metamorphic rocks, granite, and basalts.

Climate features
The region is located at the southern limit of an area that is influenced by the Australian Summer Monsoon (ASM).
The climate pattern is controlled by the position of the Intertropical Convergence zone (ITCZ) and the monsoon circulation.Over 80 % of mean annual precipitation falls during the November-March interval, supplied by NW monsoonal winds during the earlier phase of the ASM, by S-E trade winds during the later phase of the ASM (Godfred-Spenning and Reason, 2002) and by occasional cyclones.In the NW-SE oriented eastern highlands, there is a very pronounced orographic influence with higher precipitation averages centred around the highest peaks and their eastern slopes, windward to the moist prevailing SE airstream.Mean annual precipitation reaches 4000 mm yr −1 along the coast, increasing with elevation to reach 11000 mm yr −1 at Mt. Bartle Frere, and falls dramatically inland (1200 mm yr −1 on the western Tableland) (Fig. 2c).At higher ranges (from 1000 m a.s.l.), fog and orographic cloud layers shroud the summits of the mountains and maintain moist conditions throughout the year.Fog deposition accounts for approximately 40 % of the water reaching the forest floor (Hutley et al., 1997).
The amount of water derived from the canopy interception, also called "cloud stripping" by these high-altitude "cloudforests", can account for 66 % of precipitation during the dry season (McJannet et al., 2007a,b).Coastal humidity averages 78 % in the summer but often reaches values higher than 90 %.Mean annual temperature (MAT) exceeds 24 • C along the coast and falls to below 21 • C on the tableland, and to below 17 • C in the highest ranges (DASETT, 1987;Moss and Kershaw, 2000;Godfred-Spenning and Reason, 2002;Robertson et al., 2005).

Vegetation features
Rainforests of the wet tropics of Queensland are classified into 27 structural categories that contain more than 3000 plant species from 210 families (Tracey, 1982;Webb and Tracey, 1994).Soils were sampled from areas that support seven rainforest categories (Table 1), numbered and described by Tracey (1982)

Northern Cairns Transect
Atherton a AUSTRALIA

Materials
Sixteen samples were collected from poorly drained podzolic soils developed at the expense of granitic and metamorphic parent-rocks, along four altitudinal transects from 70 to 1283 m a.s.l.(Table 1, Figs. 2 and 3).The sampling method consisted of collecting individual sub-samples of the upper 2 cm of the soil humic horizon (litter excluded), at random intervals, over an area of 5 × 5 m.Sub-samples were mixed together.
Three transects where samples were collected are located in the eastern highlands, east Atherton, while the fourth transect (Northern Cairns) is located on the coastal belt.The "Bartle Frere" transect is located on the North-Western slope of Bartle Frere South Peak (1615 m a.s.l.), above Russell River oriented NW-SE when its upstream section crosses the upper rainforest area.All of the sampled points are leeward relative to the dominant S-E trade winds.The "Palmerston Highway" transect is located in the eastern section of the eastern highlands, above the Johnston River valley (NW-SE).The sites sampled are windward to the dominant S-E trade winds.The "Mt.Edith" transect is located on the southern slope of Mount Edith (1149 m a.s.l.).The sampled sites are windward, but the S-E trades first pass over the Bellenden Ker range (maximum altitude of 1545 m a.s.l., mean altitude of 1100 m a.s.l.) before reaching the tablelands and flowing up to Mt. Edith.The "Northern Cairns" transect is located on  1).
the coastal belt.The sampled points are windward (Fig. 3) relative to the dominant S-E trade winds.
Long term climate means at the sampling sites (Table 1) were obtained from regional, digital maps of bioclimatic variables that were created using the ANUCLIM software (McMahon et al., 1995), which uses a Digital Elevation Model (DEM) and meteorological data from a large number of stations over variable time periods (several decades) to estimate the climate variables for each grid cell in the DEM (0.1 km 2 ).

Phytolith counting
The recovered fraction was mounted on microscope slides in Canada Balsam, for counting at 600X magnification.More than 200 identifiable phytoliths with a diameter greater than 5 µm and with a taxonomic significance were counted per sample.Repeated counting gave an error of ±3.5 % (SD).Phytoliths were classified following Twiss (1992), Mulholland (1989), Fredlund and Tieszen (1994), Kondo et al. (1994), Alexandre et al. (1997), Barboni et al. (1999), Runge (1999), andBremond et al. (2008); and named using the International Code for Phytolith Nomenclature 1.0 (Madella et al., 2005).Phytolith types were categorized as follows: (1) dicotyledon tree and shrub types mainly represented by the Globular granulate type produced by the wood (Scurfield et al., 1974;Kondo et al., 1994); (2) a palm Globular echinate type; (3) grass (poaceae) types comprising Acicular, Elongate echinate, bulliform cells, and short cells types; and (4) types without taxonomic significance (unclassified).Abundances of the classified phytolith categories were expressed as a percentage of classified phytoliths, while the abundance of unclassified types was expressed as a percentage of the sum of counted phytoliths (Table 2).Two to 60 µm size particles of quartz, iron oxide and charcoal, as well as larger thin remains of OM, were sometimes recovered and hence counted (Table 2).Samples with more than 15 % of OM remains were re-oxidized.

δ 18 O silica measurements
Phytoliths are hydrous silica particles that contain exchangeable oxygen mostly in hydroxyl groups (Labeyrie and Juillet, 1982;Perry and Keeling-Tucker, 2000).In order to evaluate the amount of exchangeable oxygen, to fix the isotopic composition, and to calculate the isotopic composition of non-exchangeable oxygen (δ 18 O silica ) a controlled isotopic exchange procedure (CIE) was carried out.Two aliquots of 1.6 mg of the same sample were exchanged with vapor from two waters of a known isotopic composition (Crespin et al., 2008).Oxygen extractions were then performed using the IR Laser-Heating Fluorination Technique as described in Alexandre et al. (2006) and Crespin et al. (2008).Oxygen gas samples were directly sent to and analyzed by a dual-inlet mass spectrometer (ThermoQuest Finnigan Delta Plus).The oxygen isotopic results are expressed in the standard δ-notation relative to V-SMOW.The measured δ 18 O values of each sample (δ 18 O measured 1 , δ 18 O measured 2 ) were corrected on a daily basis using a quartz lab standard (δ 18 O Boulangé 50−100 µm = 16.36 ± 0.09 ‰).Additionally, we checked that δ 18 O measured 1 and δ 18 O measured 2 values obtained for one to three aliquots of the phytolith lab standard MSG 40 were always in the standard deviation of the mean value measured during a long term calibration (Crespin et al., 2008).During the calibration period, replicate analyses of the international standard NBS 28 gave an average of 9.6 ± 0.17 ‰ (1 SD, n = 13).Replicate analyses of the soil top phytolith samples yielded to a reproducibility for δ 18 O measured better than ±0.5 ‰, except for one sample (#19) (Table 2).
The recent inter-laboratory comparison for oxygen isotopic composition of hydrous biogenic silica has evidenced that, when the CEREGE CIE is performed, a methodological bias occurs leading to abnormally high fractionation between the vapour and the exchanged oxygen ( 18 O vapour-O exchanged ) (Chapligin et al., 2012).Tests were conducted to examine any possibility of systematic error: before the start of the CIE, the working standards were heated at 350 • C in order to remove possible labile organic remains that may increase the surface area of the materials; the amount of silica subjected to exchange was increased by 6; watervapour exchange temperature (and associated fractionation factor) was changed; time of vapour-silica exchange was increased while the volume of exchange was reduced by 1/3; the rate of silica dehydration was reduced.None of these tests produced significant changes in the δ 18 O measured values.The difference between the two labelled waters was reduced by 8 ‰, which did not change the δ 18 O silica values either.Additionally, for two working standards (MSG60, BFC), vapour-silica exchanges were carried out at three temperatures (140 • C, 200 • C (usual temperature) and 246 • C).Changes in δ 18 O measured values were conform with those expected from the temperature-dependency of the equilibrium fractionation factors α water-vapour and α vapour-exchanged .The methodological bias assumed to occur during vapoursilica exchange remains unexplained but was reproducible and could be quantified.For this purpose, pooled values from Chapligin et al. (2012) were taken as true values.CEREGE 1000 ln α vapour-O exchanged values obtained for 140 and 200 • were compared with 1000 ln α vapour-O exchanged values previously obtained for diatoms exchanged at similar temperatures by Labeyrie and Juillet (1982).Difference was invariant with temperature and did not show any relationship with the measured percentage of exchangeable oxygen (R 2 = 0.08) but decreased with increasing δ 18 O silica values (R 2 = 0.4 and 0.9 after outlier removal).The relationship was used in a first step to correct the values of vapour-O exchanged@CEREGE ( vapour-O exchanged was assumed similar to 1000 ln α) as follows:   Corrected vapour-Oexchanged@CEREGE = vapour-Oexchanged, Labeyrie and Juillet (1982) − 1.3 × δ 18 O silica@CEREGE + 78. (1) In a second step, the corrected vapour-Oexchanged values were used to correct the δ 18 O silica values.The consistency of this correction was verified using an independent data set previously obtained at CEREGE for fresh water diatoms from Annecy Lake (Crespin et al., 2010).The obtained relationship between diatoms-lake water and temperature of the lake allowed defining a corrected relationship: Corrected diatoms-lake water (‰ vs. VSMOW) with a R 2 of 0.7 and a p-value of 0.002, instead of diatoms-lake water (‰ vs. VSMOW) In a diatomslake water vs. t diagram, the obtained corrected fractionation line was shifted towards lower values of diatoms-lake water , close to fractionation lines previously obtained for fresh water diatoms (e.g.Brandriss et al., 1998;Moschen et al., 2005;Dodd and Sharp, 2010).
Difference between corrected δ 18 O silica and the pooled δ 18 O silica values, however, still increased from −0.1 ‰ (MSG60) to +2.9 ‰ (G95), with the content in total organic carbon measured during the inter-laboratory comparison (Chapligin et al., 2012).This relationship suggested that organic remains increase the CEREGE abnormally high fractionation between the vapour and the exchanged oxygen ( 18 O vapour-Oexchanged ).
For each of the samples, mean and standard deviation (SD) of X (exchangeable oxygen), δ 18 O silica , and corrected δ 18 O silica were calculated using R and a Monte Carlo simulation: X, δ 18 O silica and corrected δ 18 O silica were computed 10 000 times, using 10 000 simulated values of the variables taken into account in the equations (Crespin et al., 2008 and this section).The simulated uncertainty (SD) on corrected δ 18 O silica ranged from ±0.5 to 1.2 ‰ for 15 over 16 samples and reached ±2.7 ‰ for one sample (#38; Table 2).

Corrections on δ 18 O silica for obtaining δ 18 O wood phytolith
Since some rainforest phytolith assemblages contained a small amount of 2-60 µm particles that were not wood phytoliths, corrections were made to calculate δ 18 O wood phytolith values from δ 18 O silica values.
-Correction for the presence of quartz particles: Weight correction was made for the amount of 2-60 µm size quartz particles given their abundance (less or equal to 3 % of the counted particles in 7 samples) and the respective densities of quartz (2.6) and phytoliths (2.3).A value of 8 ‰ was attributed to quartz particles as it is in the lower range of δ 18 O values measured worldwide for detrital quartz of metamorphic origin (e.g.Garlick and Epstein, 1967;Savin and Epstein, 1970;Clayton et al., 1972;Eslinger et al., 1973;Blatt, 1986;Graham et al., 1996;Alexandre et al., 2006).
-Correction for the presence of grass phytoliths: Phytoliths from the grass under-storey may originate from transpiring grass leaves and may have δ 18 O values slightly enriched relative to wood phytoliths.Webb and Longstaffe (2002) previously demonstrated for a grass species collected from across the North American prairies that the 18 O enrichment of leaf phytoliths relative to stem phytoliths increases with the inverse of relative humidity (h): 18 O leaf silica-stem silica = 12.5/ h − 13. (4) As underlined by Webb and Longstaffe (2002), this relationship is very similar to that reported by Yapp (1979) for the evaporation of body fluids in land snails.It is applied here, as a general equation, to phytoliths from the under-storey grasses.According to Eq. ( 4), for the 0.6-1 range of relative humidity calculated by the ANU-CLIM software for the sampled sites (Table 1), leaf phytolith 18 O enrichment should range from 5 to 0.5 ‰.In the absence of measured data on relative humidity in the under-storey, values from Table 1 were used in Eq. ( 4) to correct the obtained δ 18 O silica values for the presence of grass phytoliths.This correction is expected to be maximal.Indeed, in the lower canopy vapour pressure deficit is low (0-0.2kPa; Hutley et al., 1997) and 18 O enrichment of leaf water should be weak, as measured in the Amazonian rainforest (Ometto et al., 2005).Moreover, grass phytolith types (Table 1) gather phytoliths from both leaves and stems; the latest being not subject to transpiration (Webb and Longstaffe, 2002).

-Correction for the presence of palm phytoliths:
To our knowledge, δ 18 O water signatures in forest palm stems and leaves have never been measured.If 18 O enrichment is similar for palm and grass phytoliths, Eq. ( 4) should also be used to correct the obtained δ 18 O silica values for the presence of palm phytoliths.
-No correction for the presence of unclassified phytoliths and charcoal: Unclassified phytoliths were assumed to mainly originate from tree wood and were not corrected for.The occurrence of charcoal particles was not corrected either since charcoal is largely made of carbon and should not contribute to the oxygen yield.Less or equal to 5 % of thin organic matter remains, most of them with surface ranging from 100 to 200 µm, were counted in 5 samples (Table 2).Given the low density of organic remains (<1), their weight concentration is expected to be lower than 0.2 % weight.
-Related uncertainties: Given a counting error of ±3.5 % (SD), the uncertainty on δ 18 O wood phytolith due to quartz weight correction ranged from ±0.5 to 0.9 ‰ for the 7 samples where some quartz particles were counted.Uncertainties due to grass phytolith corrections ranged from ±0 to 0.2 ‰.Uncertainties due to palm correction were lower than 0.5 ‰, except for samples #7 and #11 with high content of palm phytolith types.

Estimation of MAP, MAT and δ 18 O precipitation values
Ideally, matching the time span recorded by the soil phytolith assemblages would require us to obtain, for each of the sampled sites, measurements for hundreds of years for MAP, MAT, δ 18 O soil water and/or δ 18 O precipitation values, which is unrealistic.Therefore, estimates were made, as justified below.
Regarding atmospheric temperature, measurements back to 1910 revealed an increase of MAT of +0.1 • C/10 yr, mostly during the second half of the 19th century in Queensland (Suppiah et al., 2001).To our knowledge there is no other continuous record of temperature from any tropical Australian site.A dendroclimatological study showed evidence of a slower increasing trend of +1.5 • C since the 16th century in New Zealand (Cook et al., 2000).From these records, the assumption was made that modern long term MAT values generated by the ANUCLIM software (Table 1) should only slightly overestimate (by less than 1-2 • C) the mean value for the last 100s of years; this in a similar way for all the sampled sites.Given the <0.5 • C standard errors in monthly maximum and minimum temperature values generated by the ANUCLIM software a standard error of around 0.2 • C in MAT would be a conservative estimate (M.F. Hutchinson, personal communication;Hutchinson, 1991Hutchinson, , 1995) ) (Table 1).
Regarding MAP, tree ring reconstruction from the Atherton tableland (Queensland) revealed no long trend in precipitation since 1861 (Heinrich et al., 2008).Making due allowance for varying conditions, station density and standardisation to 30 years, the predictive errors in MAP from the ANUCLIM surfaces are around 10-15 % across the continent (M.F. Hutchinson, personal communication;Hutchinson, 1991Hutchinson, , 1995) ) (Table 1).
In the absence of long term δ 18 O precipitation measurements close to the sampled sites, direct δ 18 O measurements could have been performed from non-evaporative surface water (e.g.river, spring water) (Lachniet and Patterson, 2009).However, given the confined sampled area, surface waters did not show sufficient variations in distance from the water source to record rapid changes in δ 18 O precipitation values with elevation.Weighted annual δ 18 O precipitation estimates were thus calculated using as a basis the Bowen and Wilkinson (2002) model established from the International Atomic Energy Agency-World Meteorological Organization Global Network for Isotopes in Precipitation (GNIP) database (IAEA/WMO, 1998).δ 18 O precipitation values are controlled by the latitude (LAT) and by the altitude (ALT).For stations located <200 m a.s.l.(Eq.5) and >200 m a.s.l.(Eq.6) the relationships are respectively expressed as: Equations ( 5) and ( 6) were slightly modified to take regional conditions into consideration.The magnitude of the altitude effect estimated as 0.002 ‰ m −1 a.s.l. by Bowen and Wilkinson (2002) was also measured worldwide with an uncertainty lower than ±0.0005 ‰ m −1 (Siegenthaler and Oeshger, 1980;Chamberlain and Poage, 2000;Gonfiantini et al., 2001;Lachniet andPatterson, 2002, 2009).However, in humid tropics, δ 18 O values of precipitation from air masses lifted over high mountains appear to be controlled by the cumulative rainout upwind of collecting stations (Lachniet and Patterson, 2009).At leeward stations, this rainout process leads to measured δ 18 O precipitation values lower than those predicted by the altitude effect alone (Rietti-Shati et al., 2000;Longinelli et al., 2006;Lachniet and Patterson, 2009).This process is also called "shadow effect".In order to take into account such a cumulative rainout, the cumulative change in altitude ( ALT) along a SE-NW transect (i.e.parallel with the trajectory of the dominant trade winds) was used to estimate δ 18 O precipitation values at both windward and leeward sites (Fig. 3; Table 1).This procedure resulted in an inversed isotopic vertical gradient for the leeward slope of Mt.Bartle Frere, neglecting the role of increasing temperature as elevation decreases.This may have led to an underestimate for δ 18 O precipitation values, especially at leeward low elevation sites.Additionally, in agreement with the few isotopic studies investigating the water cycle in rainforest areas, an inland 18 O gradient of −0.08 ‰/100 km reflecting the influence of recycled continental moisture (mainly from evapotranspiration) (Salati et al., 1979;Gat and Matsui, 1991;Martinelli et al., 1996;Njitchoua et al., 1999;Lachniet and Patterson, 2002) was added (Eqs.7 and 8).However, given the sites proximity to the coast, the inland gradient has limited impact on δ 18 O precipitation values.
Finally, for the <200 m a.s.l. and >200 m a.s.l.sites, the calculation of δ Uncertainty associated with the latitude parameter in Eqs. ( 7) and ( 8) is not known but should be similar for all the sites, as they are located at similar or close latitude (Table 1).At the global scale, the average difference between δ 18 O precipitation values estimated from Eqs. ( 5) and ( 6) and the measured values is 0.21 ‰ (σ = 2.49 ‰) (Bowen and Wilkinson, 2002).Added to the altitude effect uncertainty (±0.0005 ‰ m −1 ), it leads to a maximum propagated uncertainty on δ 18 O precipitation estimates from Eqs. ( 7) and ( 8) ranging from 0.2 to 0.8 ‰ ( • 25 E; 594 m a.s.l.) located in the Atherton tablelands (Fig. 2).Those were the only measured data available for the area.Since most of the precipitation occurs during the rainy season, weighted seasonal values were expected to be close to weighted annual estimates.They were indeed very close -respectively −8.68 ‰ and −8.50 ‰ at Malanda and −7.77 ‰ and −7.85 ‰ at Walkamine -which strengthened the reliability of the δ 18 O precipitation estimations.
There is no available data on long term δ 18 O soil water signature from the Australian rainforests.However, as those rainforests are characterized by low radiation levels due to frequent occurrence of fog and low clouds and by a low range of the vapor pressure deficit for much of the time, soil evaporation and understorey evaporation are expected to be low (Hutley et al., 1997).For comparison, δ 18 O soil water values measured in the Amazonian rainforest were shown to be close to δ 18 O precipitation values and to lie along the local isotopic meteoric water line (Girard et al., 2000).On steep slopes developed at the expense of granitic and metamorphic parent-rocks, shallow groundwaters may locally occur at the boundary between arenite and soils.However, if this is the case at the sample sites, local δ 18 O groundwater signatures should be close to the long term δ 18 O soil water values.Finally, given the conditions described above, the assumption was made that long term depth-weighted δ 18 O soil water signatures were close to weight annual δ 18 O precipitation values.
Quartz, OM, Fe oxides and charcoal particles were present in low proportions in few samples (Table 2).

δ 18 O silica , δ 18 O wood phytolith and environment
Mean, reproducibility and uncertainties of δ 18 O measured 1 , δ 18 O measured 2 , the percentage of exchangeable oxygen, calculated δ 18 O silica values corrected for the CIE methodological bias and for the presence of quartz are presented in Table 2. Corrections for the presence of grass and palm phytoliths are presented in Table 3 for comparison.
δ 18 O wood phytolith values, corrected for the presence of quartz only, range from 21.1 ‰ to 31.9 ‰ for the assemblages from the Bartle Frere transect, from 32.1 ‰ to 34.8 ‰ for assemblages from Palmerston Highway, from 28.5 ‰ to 31.1 ‰ for assemblages from the Mt.Edith transect, and from 30.8 ‰ to 33.9 ‰ for assemblages from the Northern Cairns transect.The estimated weighted annual 18 O precipitation values ranged from −8.97 ‰ to −3.67 ‰, for a mean annual temperature range of 17-24.1 • C and a precipitation range of 1500-4942 mm yr −1 (Table 1).When all points are taken into account, no direct correlation appear between δ 18 O wood phytolith values, weighted annual  (R 2 = 0.56) occurs, while there is still no trend either with MAT, MAP or elevation.Similar patterns are found when δ18 O wood phytolith values are additionally corrected for the presence of grass and palm phytoliths.

The relationship between δ 18 O wood phytolith values, δ 18 O precipitation values, and temperature
The δ 18 O value of a mineral grown in isotopic equilibrium is a function of the temperature and the oxygen isotopic composition of the water from which it forms (δ 18 O forming water ), as expressed by 18 O mineral-forming water ∼ 1000 ln Over a limited temperature range, the relationship is satisfactorily approximated by a straight line in the plot of ln α vs. t, expressed as 18 O mineral-forming water ∼ 1000 ln α = a t + b, given that 18 O mineral-forming water is equal to δ 18 O mineral − δ 18 O forming water , α is the fractionation factor, A and a are the fractionation coefficients, B and b are the constants, and T and t are, respectively, the temperature in • K and • C.
In the case of rainforest phytoliths, the δ 18 O wood phytolith value should be a function of the atmospheric temperature and, if there is no 18 O enrichment between precipitation, soil water, and xylem water, of the δ 18 O precipitation value:  1 and 2) and limits of the 95 % confidence interval.
If we focus on the entire temperature-dependant relationships (fractionation lines), Fig. 7 recalls that there is no uniform relationship for the different silica-water couples.Several factors were previously suggested to account for these discrepancies: diatom frustules that show higher dissolution rate than that of phytoliths (Fraysse et al., 2009) may have been subject to early diagenesis which could have impacted their isotopic composition after deposition (Schmidt et al., 2001;Dodd and Sharp, 2010); approximations of temperature and δ 18 O water value may have led to uncertainties on 18 O sedimentary diatom-water values (Moschen et al., 2005); recrystallization during experimental high temperature quartzwater exchange may have led to kinetic effects and lowered obtained 18 O quartz-water values (Sharp and Kirschner, 1994).The extent of these effects are still to be assessed.At the same time, Fig. 7 indicates that the line obtained from Eq. ( 12) is located in between the ones obtained for natural quartz (Sharp and Kirschner, 1994) and sedimentary diatoms (Juillet-Leclerc and Labeyrie, 1987;Shemesh et al., 1992) and the ones obtained for grass phytoliths (Shahack-Gross et al., 1996), fresh water diatoms (Brandriss et al., 1998;Moschen et al., 2005;Dodd and Sharp, 2010;Crespin et al., 2010) and values extrapolated from the high temperature quartz-water fractionation (Clayton et al., 1972;Matshuhisa et al., 1979).For the considered temperature range, our δ 18 O precipitation estimates (from Eq. 12) are lower by only 0.2 to 0.9 ‰ than estimates using the fractionation relationship from Juillet-Leclerc and Labeyrie (1987).They are lower by 1.5 to 2.2 ‰ than estimates using the fractionation relationship from Sharp and Kirschner (1994) and higher by 2.5 to 3.2 ‰ than estimates using the fractionation relationship from Shahack-Gross et al. (1996).The relative shift to harvested grass phytoliths may result from the uncertainties associated with both relationships.This shift can also be explained if calculated 18 O wood phytolith-precipitation values are higher than actual 18 O wood phytolith-forming water values, due to evaporative 18 O enrichment of the soil water absorbed by roots.This may occur if a significant part of the water comes from the first 10s of centimeters of soil during the dry and more evaporative season.However, a greening during the dry season, when it occurs, involves rainforest tree roots' capacity for absorbing non-18 O enriched deep soil water rather than shallow water (Huete et al., 2006).Additionally, although biased estimations of δ 18 O precipitation and/or discrepancies between δ 18 O precipitation and δ 18 O soil water may have occurred, they should be reproducible in order to explain systematic shifts of several ‰, which is rather unlikely.An underestimation of δ 18 O precipitation may also be involved.Although agreement between estimated δ 18 O precipitation and measured δ 18 O precipitation values at Malanda and Walkamine support the accuracy of our estimations, obtaining long term δ 18 O precipitation and/or δ 18 O soil water records (rather than direct measurements only instructive for short term hydrological conditions) for the studied area would help to further verify this accuracy.Finally, superficial dissolution of phytoliths in litter and soil (Alexandre et al., 1999) may also lead to slight 18 O enrichment as the lighter isotope, forming weaker bonds and having a higher diffusion velocity than the heavier isotope, goes preferentially to the liquid phase.Dissolution figures are difficult to detect on the granulated surface of the globular granulate phytolith type which prevents verifying the later hypothesis.

δ 18 O wood phytolith values obtained from the leeward Bartle Frere transect
Data obtained from Bartle Frere transect did not follow the above relationship (Eq.12), due to particularly low 18 O wood phytolith-precipitation values (Table 2) drastically decreasing with elevation in contrast to a low temperature gradient (Table 1).Several points are discussed below to account for this discrepancy: (1) a low temperature gradient may emphasize variations of δ 18 O wood phytolith values in relation to local environmental changes (e.g.soil evaporation, depth of water uptake, phytolith production), but would unlikely explain a shift as high as 8-9 ‰ • C −1 in δ 18 O wood phytolith and 18 O wood phytolith-precipitation values; (2) the four highest sites of the Bartle Frere transect are covered by a cloud forest characterized by cloud striping (Mc-Jannete et al., 2007a,b).As previously noted, cloud striping may increase the inland isotopic gradient, enhance the cumulative rainout of 18 O depletion in air masses and decrease δ 18 O precipitation values.However, cloud forest sites of the windward Mt.Edith transect do not show unexpectedly low 18 O wood phytolith-precipitation values; (3) the Bartle Frere transect diverges from the three other transects in that it is located leeward relative to the S-E trade winds.Potential δ 18 O precipitation underestimation at leeward sites would imply lower 18 O wood phytolith-precipitation values and cannot account for the discrepancy.N-W monsoonal winds and cyclones with westerly tracks may contribute to precipitation on the N-W slope of Bartle Frere South Peak to a greater extent than on slopes oriented windward.The associated high amount effect (Nott et al., 2007) would decrease the δ 18 O precipitation value (and increase associated 18 O wood phytolith-precipitation values).However, such an effect would likely impact the whole slope and cannot account for decreasing 18 O wood phytolith-precipitation values with decreasing altitude; (4) the Bartle Frere slope may be subject to higher rain shadow effect than the one taken into account in Eq. ( 2), which may lead to overestimate δ 18 O precipitation values all the more so altitude decreases.In the literature, the only δ 18 O precipitation measured value obtained from a leeward tropical site was about 2 ‰ lighter than the one expected for a windward site of similar altitude (Gonfiantini et al., 2001).
Finally, without further data, potential combined impacts of canopy interception, storms and rain shadow effects on δ 18 O precipitation values are difficult to assess but are rather interesting tracks to investigate in a near future.

Implications for paleoenvironmental reconstructions
In the absence of a uniform temperature-dependant relationship for different silica-water couples, the use of one or another fractionation equation for reconstructing precise values for δ 18 O forming water and temperature from fossil samples appears subject to caution.On the other hand, the small range of empirical temperature coefficients (from 0.2 to 0.5 ‰

Fig. 2 .
Fig. 2. (a) Location of the rainforests area (in green) in the Wet Tropics of Queensland.(b) Altitude and (c) mean annual precipitation (MAP) maps generated by the ANUCLIM software (McMahon et al., 1995).Sampled transects and stations of Malanda and Walkamine for which measured δ 18 O precipitation data is available are positioned.

Table 1 .
Location of the sampled sites and associated climate parameters provided by the ANUCLIM software (D. coast: distance from the coast; Alt: altitude; Max Alt: mean maximum altitude crossed by air masses; MAT: mean annual temperature; MAP: mean annual precipitation; H: relative humidity).Rainforest types from

Table 3 .
δ 18 O wood phytolith values obtained after weight correction for the presence of quartz particles, grass and palm phytoliths.

Clim. Past, 8, 307-324, 2012 www.clim-past.net/8/307/2012/ A. Alexandre et al.: The oxygen isotopic composition of phytolith assemblages from tropical rainforest soil tops 319 This study (eq. 12) Shahack-Gross et al., 1996
• C −1 ) supports their use for reconstructing, from continuous fossil sequences, relative changes in δ 18 O forming water and temperature.Consistence of the empirical temperature coefficient obtained in the present study (−0.4 (±0.2) ‰ • C −1 ) highlights the reliability of soil top phytolith assemblages from rainforests of northeast Queensland to reflect changes in MAT and δ 18 O precipitation .If no significant fractionation occurs when phytoliths are exported and buried, continuous fossil phytolith sequences from lakes in northeast Queensland can be used for reconstructing past relative changes in δ 18 O soil water and δ 18 O precipitation values (when both isotopic compositions are assumed to be close) and/or in MAT values.Given the large uncertainties obtained on 18 O wood phytolith-precipitation values, it is plausible to expect reproducible uncertainties of several ‰ and several • C on reconstructed changes in δ 18 O soil water and/or δ 18 O precipitation values, and MAT values.Moreover, given the low amount of phytoliths recovered from soils and sediments, we expect the time resolution from sedimendary phytolith records to be limited to hundreds of years.However, such large uncertainties and low time resolution are still sufficient for investigating significant terrestrial changes that occurred during the Quarternary glacial/interglacial transitions.If the range of temperature changes can be constrained by other proxies, such as pollen transfer functions, reconstruction of relative changes in δ 18 O soil water values using δ 18 O wood phytolith values should become straightforward.Additionally, the association of morphological and δ 18 O analyses on similar tropical forest phytolith assemblages should allow one to assess whether past forest dynamics were or were not synchronous with climate changes.At least a comparison of terrestrial δ 18 O records from tropical forest phytolith assemblages with deep-sea reference curves should help to further investigate the relationship between global oceanic transgression/regression phases (e.g.revealed by δ 18 O benthic foraminifera records) and local changes in the water cycle (revealed by δ 18 O phytolith records).The obtained empirical temperature-dependant relationship Eq. (12) suggests that top phytolith assemblages from rainforests of northeast Queensland reflect changes in MAT and δ 18 O precipitation , in the range predicted by equilibrium temperature coefficients previously published for quartz, diatoms and harvested grass phytoliths.This, despite the various unknowns introduced when estimating δ 18 O precipitation values and the large uncertainties on δ 18 O wood phytolith values.The consistency supports the reliability of δ 18 O wood phytolith signatures for recording changes in mean annual δ 18 O soil water values (which are assumed to be equivalent to the weighted annual δ 18 O precipitation values in rainforests environments) and MAT, provided these changes were several ‰ and/or several • C in magnitude.Morphological phytolith analysis of Quaternary continuous sedimentary sequences from the Western Australo-Pacific area should help to select rainforest phytolith assemblages suitable for δ 18 O analysis.The combination of both methods should provide simultaneous insights regarding rainforest dynamics and climate change.