Response of grassland ecosystem to monsoonal precipitation variability during the Mid-Late Holocene: Inferences based on molecular isotopic records from Banni grassland, western India

Banni, located in the arid western India, is one of the largest tropical grasslands of the Asian continent. The net primary production in this grassland ecosystem is currently mediated by precipitation during the Indian summer monsoon (ISM). However, timing of the grassland expansion and its link to the intensity of monsoonal precipitation remains enigmatic due to the paucity of datasets. The major objective of this study is to understand the changes in monsoonal precipitation and vegetation for the last 4600 cal yr BP using hydrogen and carbon isotopic composition of n-alkanes (δDn-alkane and δ13Cn-alkane) measured from two core sediments (Chachi and Luna) in Banni region. The δ13CC29 and δ13CC31 values for Chachi core sediments vary from −30.9 ‰ to −27.2 ‰ and −34.4 ‰ to −25 ‰ respectively. The δ13Cn-alkane values from the core sediments are converted into %C4 plants based on a binary mixing model using the end-member δ13Cn-alkane values derived from the dominant modern vegetation in the Banni region. The prominent feature of the paleovegetation curve is the marked increase in the δ13Cn-alkane values after 2500 cal yr BP, which suggests proliferation of C4 grasses in this region. Similar changes after 2500 cal yr BP have also been observed in the δDn-alkane values. The δDC29 values are used to calculate δD value of paleoprecipitation that varied from 10 ‰ to −60.2 ‰. A significant increase in the δD values of paleoprecipitation (ca. 25 ‰) indicates a weakened ISM precipitation after ca. 2500 cal yr BP. The regional aridification and frequent fire events may have helped the expansion of C4 plant dominated grassland ecosystem in Banni region. Correlation between paleoclimatic records suggests that the southward migration of intertropical convergence zone and more frequent warm phases of El-Nino Southern Oscillation have triggered the weakening of monsoonal precipitation in the tropical region.

δ 13 C n-alkane values from the core sediments are converted into %C 4 plants based on a binary mixing model using the end-member δ 13 C n-alkane values derived from the dominant modern vegetation in the Banni region. The prominent feature of the paleovegetation curve is the marked increase in the δ 13 C n-alkane values after 2500 cal yr BP, which suggests proliferation of C 4 grasses in this region. Similar changes after 2500 cal yr BP have also been observed in the δD n-alkane values. The δD C29 values are used to calculate δD value of paleoprecipitation that varied from 10 ‰ to −60.2 ‰. A significant increase in the δD values of paleoprecipitation (ca. 25 ‰) indicates a weakened ISM precipitation after ca. 2500 cal yr BP. The regional aridification and frequent fire events may have helped the expansion of C 4 plant dominated grassland ecosystem in Banni region. Correlation between paleoclimatic records suggests that the southward migration of intertropical convergence zone and more frequent warm phases of El-Nino Southern Oscillation have triggered the weakening of monsoonal precipitation in the tropical region. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111

Introduction
Grasslands are globally important as they comprise nearly one-fifth of the world's land surface and 80% of the total agriculturally productive land [1][2]. Geographical distribution, community composition and net primary production of grassland ecosystems not only play crucial roles in global carbon budget [3] but also influence regional climate by modulating the evapotranspiration flux [4]. The vegetation composition of grassland ecosystems reflects the competing influence of precipitation, fire and atmospheric CO 2 concentration. For instance, changes in CO 2 level and fire events in the grassland of West Africa during the Late Holocene shifted the vegetation composition towards woody plant cover, despite the increased regional aridity [5]. It is important to note that the interaction between above-mentioned forcing factors varies across the globe and thus produces a site-specific structure of the grassland ecosystem. Furthermore, the magnitude of the climatic factors and their effects on vegetation composition are also globally variable, depending upon duration of the rainy season, precipitation amount and number of precipitation events [6]. Prolonged and recurrent droughts predicted in a warming environment, combined with human-induced changes in the natural hydrological regime, will likely have an adverse effect on biodiversity of the grasslands [7,8]. The understanding of past climate-ecosystem coupling is thus vital to predict future ecosystem response to changing natural variables.
The available Mid-Late Holocene paleoclimate records from Indian summer monsoon (ISM) domain suggest significant modifications in seasonal distribution of monsoonal precipitation in response to shifts of the Intertropical Convergence zone (ITCZ) [9][10][11]. Changes in hydrological regime will likely to have a major impact on the biodiversity of the Banni grasslands in western India-one of the largest tropical grassland of Asian continent. The modern day precipitation of Banni grassland is characterized by extremely sparse rain events (n = 10-12), short length of monsoon season and very high temperature difference (ca. 40˚C) between summer and winter months. This extreme climate setting makes the Banni grassland an ideal site to investigate the past climate-grassland interactions.
A previous study from the Banni grassland used the carbon isotopic composition of bulk organic matter (δ 13 C org ) and oxygen isotopic composition of gastropod shells (δ 18 O shell ) to decipher paleoenvironmental changes [12]. The δ 13 C org values from the Banni grassland suggested Late Holocene expansion of a C 4 grass dominated ecosystem in response to changes in moisture availability and local fire-events [12,13]. However, δ 13 C org value based estimates of %C 4 plants have uncertainties due to thermal degradation of selective organic compounds [14], which can alter the pristine isotopic character [15]. Furthermore, δ 13 C org values integrate organic matter from numerous sources (each with distinct δ 13 C value) that can often led to bias in paleovegetation reconstruction [16]. Likewise, uncertainty has also been associated with δ 18 O shell based climate interpretation due to interplay of factors such as local precipitation variability, changes in the moisture pathway, lake water evaporation, residence time, temperature as well as biological effects controlling the δ 18 O values [17]. Therefore, more robust hydrological and vegetational proxies are required for an improved assessment of the paleoenvironmental changes from this region.
To achieve a better understanding of the past vegetation composition and its linkage with climatic factors, the present study measured carbon and hydrogen isotopic composition of nalkanes (δD n-alkane and δ 13 C n-alkane ) extracted from two seasonal wetland sediment cores (Chachi and Luna) (Fig 1). The relative phasing of vegetation and precipitation are assessed from the isotopic composition of same compound (i.e., n-alkane). Comparison of records from multiple seasonal wetlands provides a synoptic view of past environmental changes in this region. The objectives of the present study include (i) understanding the potential sources of organic matter (OM) in the Banni region; (ii) reconstruction of the climatic conditions and local vegetation history; (iii) recognizing the spatial variability of ISM; (iv) interpreting the role of teleconnections on the intensity of ISM precipitation and; (v) understanding the feedback mechanism loop between monsoonal precipitation, fire and CO 2 . To our best of knowledge, this is the first continuous biomarker-based record of climate change from western India during the Mid-Late Holocene.

Modern climate
The Banni terrain (total area: 3,847 km 2 , 23˚19' to 23˚52' N and 68˚56' to 70˚32' E) is a flat transitional land which lies between the mainland Kachchh and the Great Rann, one of the world's largest hypersaline marshy area. This region accounts for 8.4% of the total area of Kachchh Desert, western India [19]. The Banni region receives the dominant fraction (80-90%) of annual precipitation (avg. 310 mm) during ISM (Fig 1D). The average seasonal temperature in this region varies from 50˚C in summer (May-July) to 10˚C in winter (December-February) [18].

Geomorphology
Banni, a tectonically raised mudflat, is located on the southern edge of the Great Rann of Kachchh [20]. This unique geomorphic terrain is slightly elevated (ca. 4 to 20 m) from the Great Rann with sediments comprising of fluvial origin [20]. The sediments in the Banni region are derived from the Mesozoic sedimentary rocks of northern Kachchh Mainland brought down by the north flowing drainage system comprising of Khari, Chhari and Kaila rivers [21][22]. The fine textured Banni sediments are composed of stratified silt and clay with thickness more than 30 m [23]. Soil salinity is highly variable, and pH ranges between 6.5 and 8.5 [23].

Modern vegetation
The modern-day vegetation in Banni region is characterized by the mixed C 3 -C 4 plants [13] ( Fig 1A). Vegetation of Banni comprises of herbs (89 species, accounts for ca. 46% of all the plant species), grasses (37 species, ca. 19% of all the plant species,) shrubs (31 species, ca. 16% of all the plant species), trees (17 species, ca. 9% of all the plant species), and sedges and climbers (ca. 10% of plant species) [24]. The Chachi site comprises of Suaeda scrub and grassy vegetation with sparse cover of Prosopis juliflora and other herbaceous vegetation, whereas the catchment of Luna is composed of grassland with moderate density of Prosopis juliflora [13]. The region is covered with coarse and low perennial grasses like the Desmostachya bipinnata, Sporobolus marginatus, Dichanthium annulatum, Cenchrus ciliaris, Sporobolus fertilis and Chloris barbata [25]. Prominent tree species of this region are Acacia nilotica subsp. indica, Prosopis juliflora, Prosopis cineraria, Salvadora oleoides and Salvadora persica [25]. The region is currently degraded mostly due to frequent droughts events coupled with intense grazing and invasion of Prosopis juliflora [19].

Field sampling
Core sediments were collected from two closed shallow seasonal wetlands (Chachi and Luna) during a field expedition in summer 2012 when the region was completely dried up ( Fig 1B  and 1C). The cores (ca. 138 cm and 85 cm depth for Chachi and Luna, respectively) were retrieved using 6 cm diameter, long PVC pipes. No specific permissions were required for the sample collections from these locations. All the samples were stored at 4˚C before the analysis. The sediment cores collected from Chachi and Luna were clayey with occasional silt laminations [12]. The Chachi core was sub-sampled at 5-8 cm intervals until 70 cm depth followed by 4 cm intervals and Luna core was sub-sampled at 6 cm intervals. In addition to core sediment samples, leaf samples of dominant vegetation (grasses, herbs and trees) were also collected. The surface of the leaf samples were cleaned with 0.5 N HCl and distilled water in an ultrasonic bath to remove adhered contaminants. Leaf samples were dried at 40˚C for 48 hours, and core sediment samples were dried at 50˚C for 8 hours. Leaves and sediment samples were subsequently powdered in an agate mortar for further analysis. All the biomarkerbased analyses were conducted at the stable isotope laboratory of Indian Institute of Science Education and Research Kolkata.

Extraction and analysis of n-alkane
Total lipid extract (TLE) is obtained via accelerated solvent extractor (Dionex, ASE350) using dichloromethane/methanol (93:7) mixture at 100˚C and 1600 psi pressure for 15 min (2 cycles). For this purpose, 0.3-0.5 g leaf and 7-8 g sediment samples were used. Sample preparation and lipid extraction methods have been described in detail elsewhere [12, 26,27]. TLE fraction was concentrated in a Rotavapour (R-210, Buchi) by evaporating the dichloromethane and methanol mixture. The non-polar hydrocarbon fraction (n-alkane) was separated from the TLE using short-column silica gel column chromatography, using activated silica gel (100-200 mesh) in Pasteur pipette plugged with glass wool and hexane as eluent. Then n-alkane fraction has been concentrated up to 0.5 ml under a stream of dry N 2 .
The concentration of individual n-alkane was measured in a gas chromatograph (Agilent 7890A, GC system) equipped with a split/splitless injector, non-polar capillary column (HP5-MS, 30 m × 250 μm × 0.25 μm) and flame ionized detector (FID). Samples were injected in 1:1 split mode with an initial inlet temperature of 320˚C. The GC oven temperature was started at 60˚C (held 2 min) and then increased to 320˚C at 8˚C/min (held 12 min). The characteristic retention time obtained from the external n-alkane standard mixture (SUPELCO C 8 -C 40 alkanes) was used for the identification and quantification of individual n-alkanes.

The δD and δ 13 C analyses of n-alkane
For the Chachi core sediments, a total of 19 and 16 samples were measured for δ 13 C n-alkane and δD n-alkane values respectively. For Luna core sediments, δ 13 C n-alkane analysis has been carried out on 14 samples. All samples were not analyzed for δD n-alkane values compared to δ 13 C n-alkane due to limited sample material.
The δ 13 C n-alkane values were measured in Thermo Trace GC Ultra, connected via combustion interface (GC-Isolink) and Conflo IV interface to a MAT 253 isotope ratio mass spectrometer (GC-IRMS). The GC oven temperature was raised from 40˚C (held 1 min) to 320˚C (held 12 min) at 10˚C/min. 1 μl samples were injected in splitless mode with an inlet temperature at 280˚C and He as carrier gas at 1 ml/min. Individual compounds were combusted over Nickel and Copper wire in the presence of O 2 with He (1% v/v) at 950˚C to produce CO 2 . Calibration of δ 13 C n-alkane value was performed by injecting several pulses of CO 2 at the beginning and the end of each GC run. Isotopic ratios are expressed in ‰ relative to the Vienna Pee Dee Belemnite (VPDB) standard. A standard mixture of n-alkanes (C 16 -C 30 ) procured from Prof. Arndt Schimmelmann of Indiana University, with a known isotopic composition was used at different dilution (0.5, 1, 1.5, 7.5 and 15 ng/μl) to monitor the accuracy of the measurement.
The δD n-alkane values were measured using a Thermo Trace GC Ultra coupled via a GC Isolink pyrolysis interface operated at 1420˚C to a Thermo MAT 253 GC-IRMS. The δD n-alkane values were measured against calibrated H 2 reference gas. The δD n-alkane values are reported in ‰ notation against the Vienna Standard Mean Ocean Water (VSMOW). The Schimmelmann standard was used at different dilution (35,50,75 and 105 ng/μl) to assess the accuracy of the analysis. Long-term repeated analysis of the external standard mixture rendered a precision of ±3 ‰ (1σ) and an average accuracy of 1 ‰. The H 3 + factor varied around 11.1 ± 0.02 ppm nA -1 throughout the measurement period.

Chronology
The published 14 C activity measurements made on bulk organic matter are used to constrain an age model for Chachi and Luna core sediments [12]. However, using the radiocarbon activity of bulk organic carbon to date the deposition of lake sediments is often hampered by i) the "reservoir effect", carbon in the organic fraction derived from old groundwater or dissolution of limestone [28,29,30]; (ii) contamination by reworked organic materials from weathering and erosion in the upstream catchment can dilute the 14 C content of organic materials with respect to the atmosphere at the time of deposition [31,32]. The chronology based on the AMS dating of bulk organic sediments in Chachi and Luna core sediments are stratigraphically consistent [12]. The bulk organic matter in the Banni core sediments has dominant source derived from the terrestrial vascular plants [12] that incorporate 14 C in cellulose sourced from 14 CO 2 in the atmosphere [33]. Therefore, the bulk organic matter in the Banni core sediments is relatively free of any "reservoir effect" making it an ideal material for applying 14 C to date the sediments. The 14 C dates were calibrated using OxCal 4.1 software [34] with the IntCal 09 calibration curve [35]. The average sedimentation rates of the Chachi and Luna sediments are 0.03 and 0.06 cm yr -1 respectively. The age-depth relationship was modeled using linear interpolation between adjacent radiocarbon measurements.

Quantification of organic source using n-alkane indices
The n-alkane indices have been used to obtain information on OM sources using the sourcespecific carbon chain length [36,37]. The long-chain n-alkanes (C 27 -C 33 ) tend to dominate terrestrial leaf waxes, short-chain n-alkanes less than C 21 are more characteristic of algae, and mid-chain n-alkanes (C 21 , C 23 , and C 25 ) are mainly produced by aquatic macrophytes [38,39].
In the present study, P aq , Terrigenous versus aquatic ratio (TAR) and carbon preference index (CPI) have been calculated. The P aq and TAR indices have widely been used to estimate the relative proportion of terrestrial flux in the sediments [39,40].
Following equations have been used in this study:

Back-trajectory computation
The probable moisture source of the air masses has been deduced based on the backward trajectories computed from the database of the National Oceanic and Atmospheric Administration (NOAA, http://www.arl.noaa.gov/ready/hysplit4.html) using Hybrid Single-Particle Lagrangian Trajectories (HYSPLIT-4). Trajectories were computed for 48 hours at 500, 1500 and 2500 m.a.s.l. as precipitation in India is expected to originate from these altitudes [42]. In daily time-scale, trajectories did not show much variation with altitude, and hence 500 m.a.s.l. was considered to be cloud base height for precipitation over the Banni region.

The δD value of precipitation in Banni grassland
In the absence of meteorological station in Banni grassland, the Isotope Global Spectral Model version 2 (IsoGSM2) prediction of the stable isotopic composition of regional precipitation [43] has been used in the present study. This model provides monthly averaged δD and δ 18 O values of precipitation for the years 2011 to 2015. The spatial resolution of the model is 1.9l atitude 1.8˚longitude. In IsoGSM2, atmospheric processes and temperature fields are constrained by spectral nudging. The detailed description of the model has been provided elsewhere [43]. In this study, 5 years weighted averaged δD and δ 18 O values of precipitation have been used as a reference dataset to interpret paleoclimatic signals.

Statistical analyses
The Mann-Whitney U test has been used to assess statistically significant differences between δD n-alkane and δ 13 C n-alkane values of the Chachi core sediments before and after 2500 cal yr BP. This test is commonly used to identify the significantly differing groups, which are not normally distributed [44]. The null hypothesis of this non-parametric test is that the difference of location between the samples is equal to zero, while the alternative hypothesis is that the difference of location between the samples is different to zero. The analysis has been conducted at a significance level of 95%. In this study, this analysis has been conducted on both δD C29 and δD C31 values for two time-slices; 4600 to 2500 cal yr BP and 2500 cal yr BP to present. Mann-Kendall trend analysis, a nonparametric test, has been employed to identify the trends in timeseries of δD n-alkane and δ 13 C n-alkane values of the Chachi core sediments. The null hypothesis of this non-parametric test is that there is no trend in the time-series. The alternative hypothesis is that there is a trend in the time-series. The analysis has been conducted at a significance level of 95%. The correlation between dependent and independent variables has been calculated using linear regression analysis. All the statistical analyses have been conducted using PAST (Paleontological Statistics) and Sigma Plot software (version 12.5).

Molecular level characterization of the OM
The n-alkane distributions in leaves of modern plants showed the dominance (ca. 80% of total concentration) of C 29 and C 31 homologues (S1 Fig). On the contrary, the n-alkanes distribution in the core sediments displayed a distinct bimodal distribution with carbon numbers ranging from C 16 -C 35 (S2 Fig). In the Chachi core sediments, total n-alkane concentration ranged from 120 μg g -1 to 700 μg g -1 and reached the maximal value at the topmost section of the core. The short-chain and long-chain n-alkanes accounted for~40% and~25% of the total n-alkane concentration respectively. The CPI values for the Chachi and Luna core sediments varied from 2.9 to 15.7 (Fig 2A) and 3.2 to 4.2 respectively. The P aq values for the Chachi and Luna core sediments varied from 0.07 to 0.33 ( Fig 2B) and 0.10 to 0.25, respectively. The TAR  values of the Chachi sediments ranged from 1.1 to 6.5 (Fig 2C), whereas the Luna core sediments varied from 4.1 to 5.1.

The δD and δ 13 C values of long-chain n-alkane in modern plants
The δD n-alkane and δ 13 C n-alkane values of the modern vegetation from Banni region are listed in

The δ 13 C value of long-chain n-alkanes in core sediments
The δ 13 C C29 and δ 13 C C31 values of the Chachi core sediments varied from −30.9 ‰ to −27.2 ‰ and −34.4 ‰ to − 25.0 ‰ respectively ( Table 2). The δ 13 C n-alkane values of both the homologues showed more negative values during 4600 to 2500 cal yr BP (Fig 3A). Afterwards, δ 13 C C31 values displayed a progressively increasing trend with maximum values at 1500 cal yr BP. Similar trends are also observed for the δ 13 C C29 values. The δ 13 C C29 and δ 13 C C31 values of the Chachi core sediments exhibit strong and significant positive correlation (R 2 = 0.66, p < 0.05). The p-values of Mann-Kendall analyses for the both homologues are less than 0.05 and Kendall's tau value for δ 13 C C29 and δ 13 C C31 values are 0.45 and 0.61. The δ 13 C n-alkane values of Luna core sediments for the last ca. 1000 parallels the Chachi record (Fig 3). The δ 13 C C29 and δ 13 C C31 values of the Luna core sediments ranged from -25.3 to -27.7 ‰ and -23.9 to -28.0 ‰ respectively (Table 3). The δD values of long-chain n-alkanes in core sediments  (Table 2). In most of the samples, the δD C29 values were more positive compared to δD C31 value. The most prominent feature of the curve is the shift towards positive values after 2500 cal yr BP ( Fig 3B). Accordingly, we have compared the values for two time-slices; 4600-2500 cal yr BP (phase-1) and 2500 cal yr BP to present (phase-2). The average δD C29 and δD C31 values during the phase-1 were −113.0 ± 10.7 ‰ and −121.8 ± 8.8 ‰ (n = 10, 1σ) respectively. The average δD C29 and δD C31 values during phase-2 were −86.0 ± 13.5 ‰ and −90.7 ± 12.3 ‰ (n = 6, 1σ). The difference in the δD C29 (and δD C31 ) values for phase-1 and phase-2 has also been exhibited by Mann-Whitney test statistics. For both the homologues, we have found the p-values are less than significance level of 0.05 and thus reject the null hypothesis.

The stable isotopic composition of modern precipitation in Banni region
Monthly-weighted averaged δD values are positively correlated (R 2 = 0.9, p < 0.05) with the δ 18 O values of precipitation. The Local Meteoric Water Line for the Banni region is: ( Fig 4A) Due to the significant correlation (R 2 = 0.9), only the results of δD values of precipitation are further discussed. During 2011 to 2015, the δD values of precipitation varied from −4 ‰ to −77 ‰, and lower δD values coincide with the peak ISM precipitation (Fig 4B).

Sources of organic matter
The n-alkane assemblage in the sediments is an effective biomarker tool for evaluating sources of organic matter due to the distinct carbon chain-lengths of terrestrial higher plants and microorganisms [26,45]. In general, cuticular waxes of higher plants have a strong odd over even predominance in the long-chain n-alkanes and show CPI values >1, whereas the bacteria and algae are characterized by weak odd over even predominance and produce low CPI values (~1) [38]. Terrestrial higher plants are characterized by relatively higher abundances of odd-  numbered long-chain n-alkanes, whereas microorganisms contribute short-chain n-alkanes [16,36]. In Banni core sediments, the higher abundance of the short-chain n-alkanes (S2 Fig) indicates a significant contribution from microorganisms to the n-alkane pool in the sediments. On the other hand, peaks observed in the long-chain n-alkanes maximise at C 29 and C 31 . The CPI values of both the sediment cores are consistent with the modern higher plants growing in the Indian subcontinent [27]. High CPI values (Fig 2A) suggest that long-chain nalkanes in the sediments were derived from terrestrial higher plants. Further, the low P aq values of Chachi core sediment indicate higher contribution of terrestrial biomass to the organic matter compared to aquatic productivity, and this observation is concordant with the TAR values (Fig 2B and 2C). Therefore, the long-chain δ 13 C n-alkane values would be suitable for the paleovegetational reconstruction.

Estimates of shifts from C 3 to C 4 plant type
A previous study from Banni grassland using bulk δ 13 C org values showed that %C 4 plant varied from 16% to 43% [12] (S1 Table). However, bulk δ 13 C org values reflect contribution from multiple organic matter sources [16,27], which are characterized by distinct isotopic composition. Furthermore, C 4 plant derived components degrades faster compared to its C 3 counterpart [46]. Therefore, paleovegetation reconstruction using the δ 13 C org values may provide an erroneous estimate of %C 3 -C 4 plants. The δ 13 C n-alkane values can refine the interpretation of bulk δ 13 C org data due to its resistance to biodegradation. Result of linear regression analysis shows that the δ 13 C C29 and δ 13 C C31 values of the Chachi core sediments exhibit a strong and significant positive correlation (R 2 = 0.66, p < 0.05). The large spread in the δ 13 C C29 and δ 13 C C31 values ( Fig 3A) shows substantial variations in the vegetation composition. The δ 13 C n-alkane values are not only governed by the vegetation composition but also by physiological changes against moisture stress and δ 13 C value of atmospheric CO 2 (δ 13 C atmospheric CO2 ) [42,47,48]. The results from the Indian subcontinent show that δ 13 C values of C 3 plants change only −0.4 ‰ for every 100 mm increase in the precipitation amount [49]. Therefore, ca. 2350 mm change in the annual precipitation would be required to explain the observed spread in the δ 13 C C31 values (Table 2). Thus, we exclude moisture stress on the δ 13 C n-alkane values as the estimated precipitation change in the region seems unrealistic.
The δ 13 C atmospheric CO2 value has been declining rapidly in the recent decades as a consequence of the Suess effect (http://cdiac.ornl.gov/trends/co2/iso-sio/iso-sio.html). The modern δ 13 C atmospheric CO2 value (−8.2 ‰) is 1.7 ‰ lower compared to that in the Mid-Late Holocene period [47], and accordingly the end-member δ 13 (Table 1). In the core sediments, the δ 13 C C31 values correspond to 13% to 70% abundance of C 4 plants, while δ 13 C C29 values indicate 5% to 38% abundance of C 4 plants (Fig 3C). Chemotaxonomic investigations showed that C 4 grasses in the savannah region synthesize more C 31 , whereas C 29 is the dominant homologue in C 3 dicots (S1 Fig) [50,51]. The δ 13 C C29 and δ 13 C C31 values in the core sediments are believed to represent two distinct organic matter sources. Accordingly, in the present study, it has been assumed that δ 13 C C29 values of the sediment samples reflect the changes in the abundance of C 3 trees, whereas δ 13 C C31 values indicate the changes in %C 4 grasses. The measured δ 13 C C31 values provide higher estimation of C 4 plants compared to both δ 13 C org and δ 13 C C29 values (Fig 3C, S1 Table). Therefore, it can be suggested that the δ 13 C C31 values would be a more suitable proxy to infer vegetation change in the grassland setting.
The δ 13 C C31 value-based calculation indicates ca. 20% higher abundance of C 4 plant during phase-2 compared to phase-1 ( Table 4). Estimate of lower %C 4 plant till 2500 cal yr BP is concurrent with the pollen and phytolith records which showed the dominance of woody taxa (i.e., C 3 plants) during this interval [13]. The δ 13 C n-alkane values from the Luna core sediments for the last 1000 cal yr BP also shows high %C 4 plants in the Banni region. The trend in the δ 13 C n-alkane values corresponds well with the vegetation record from central India [42].

δD C29 value as a proxy for paleoprecipitation
The δD n-alkanes values from plant leaf wax have been increasingly used for the reconstruction of paleohydrological conditions [52][53][54]. The δD variability in leaf n-alkanes are mostly explained in terms of the changes in the δD values of regional meteoric water [54][55][56][57][58][59]. In addition, plant functional types and photosynthetic pathways can also affect the δD n-alkane values [60][61][62][63]. In the Chachi core sediments, the δD C29 values differ by -14 ‰ to 22 ‰ to the δD C31 values of the same sample. Difference sources of n-alkane can be assumed to explain this offset [64] as net apparent hydrogen fractionation between source water (i.e., MAP) and lipid (ε MAP-n-alkane ) of C 3 dicots is ca. 30 ‰ more positive compared to that of C 4 grasses [65].
Results from Banni grassland also show that ε MAP-C29 value for C 3 plant is −113±25.1 ‰ while for C 4 plant the value is −156±28.9 ‰ (Table 1). Similar differences have been inferred from C 31 homologue as well. The difference in water absorption systems in plant forms plays an important role that cause more negative δD values in n-alkanes from grasses than those from woody plants. The grasses (C 4 ) in the open ecosystem utilize water from deeper water sources (due to greater rooting depth of perennial grasses) compared to trees (C 3 ) in woodland setting as woody plants draw most of their water from soil surface through radial and shallow roots [65]. The different water-usage strategies of C 3 and C 4 plants may lead to a significant offset in the ε MAP-n-alkane (as well as δD n-alkane ) values and required to be addressed when the vegetation changes are large (as in the case of the present study). In this study, the proportion of C 4 plants for each time-point has been calculated from the δ 13 C C29 and δ 13 C C31 values of the sediment samples (Table 3). Using the δ 13 C n-alkane based estimates of vegetation changes, the weighted averaged apparent fractionation factor between δD n-alkane and the δD precipitation (ε n-alkaneprecip ) has been estimated using the following equation; ε nÀ alkaneÀ precip ¼ A � ε nÀ alkaneÀ precipðC 3 plantsÞ þ ð100 À AÞ � ε nÀ alkaneÀ precipðC 4 plants Þ Where, A is the %C 3 plants. The weighted averaged ε n-alkaneprecip values for each time-point have been presented in Fig 3D. Before the estimation of δD precipitation values, it is required to calculate the hydrogen isotopic fractionation during organic matter transport from leaf-wax to soil. Using the above-mentioned equation, the δD n-alkane value of modern leaf would be −125 ‰. On the other hand, the δD C29 value of the core-top sediment is −78 ‰, indicating incorporation of n-alkane into soil resulted in ca. 47 ‰ increase of the δD C29 value. Accordingly, the δD n-alkane values for each time-point have been corrected [66]. The δD value of regional precipitation has been calculated using the following equation [67]; The correlation analyses have been conducted between δ 13 C and δD values of both the homologues for further interpretation of paleohydrological conditions. Results of the linear regression analysis have shown strong and significant positive correlation (R 2 = 0.58, p < 0.05, Fig 5) between δD C31 and δ 13 C C31 values. The vegetation change as primary driver of hydrogen isotopic variability of C 31 homologue can result in negative correlation between δ 13 C C31 and δD C31 values. In contrast, the positive correlation indicates that the vegetation change has not influenced the hydrogen isotopic variations in C 31 homologue. It has been further observed that the correlation between δD and δ 13 C values for C 29 homologue is weaker and insignificant (R 2 = 0.15, p = 0.13, Fig 5) suggesting that the δD values of this homologue are not influenced by the change from C 3 plants to C 4 grass [53]. Therefore, the δD n-alkane values of both the homologues can be used as a surrogate to understand the paleohydrological conditions. In this study, the further discussion is based on the δD C29 values. The δD precipitation(C29) values found to vary between 10 ‰ to −60.2 ‰ (Fig 3C).
In the low latitude tropical region, δD precipitation values are negatively related to the precipitation amount (i.e., amount-effect) and this inverse correlation is often used to explain the variations in the δD n-alkanes values [50,51,68,69]. Instrumental record and back-trajectory computation of air-parcel show that precipitation in the Banni grassland is strongly seasonal (Fig 1D and 1E) with the dominant moisture contribution received from Arabian Sea. The seasonal difference in the net advection of moisture is reflected in the δD precipitation values (Fig 4B). Seasonal precipitation amount peaks during ISM months and coincides with low δD precipitation values, which indicates more negative δD C29 value of the sediment samples are representatives of high monsoonal precipitation events and vice-versa.

Paleohydrological changes in Banni grassland
More negative δD C29 values till 2500 cal yr BP suggest an enhanced monsoonal precipitation in Banni. Little fluctuations (max. 10 ‰) show that the region had experienced minor changes in the precipitation amount during phase-1. Relatively stable and high precipitation during this phase accompanied by higher %C 3 plants (Fig 3C). After 2500 cal yr BP, the δD C29 values gradually shift towards more positive values, and the aridification peaked around 1500 cal year BP. Regional aridity is also marked with the higher abundance of C 4 grasses during this interval (Fig 3C). A negative excursion (ca. 10 ‰) after 1000 cal yr BP shows a slight increase in precipitation amount in the Banni region towards the present. Overall, it can be suggested that the Banni region had experienced a gradual shift from more mesic to more arid condition since the last 4600 cal yr BP. The δD C29 value based record is consistent with the interpretation made from δ 18 O shell values in Banni core sediments [12].

Critical role of fire, CO 2 and precipitation on grassland stability
In general, low CO 2 condition and frequent fire events promote higher %C 4 grasses at the expense of C 3 woody plant cover [70]. According to the CO 2 crossover model [71], Banni with temperature > 30˚C, the threshold atmospheric CO 2 level favoring C 4 plants over C 3 plants should be below 300 ppmv. The pCO 2 reconstruction for the studied period showed that atmospheric CO 2 level increased from 268 ppmv to 282 ppmv, a range favorable for the expansion of C 4 plants. However, the trend of C 4 plant expansion in Banni does not mimic the paleo-pCO 2 curve and exhibits an anti-phase relationship in between 2500 to 1500 cal yr BP (S3 Fig). Therefore, pCO 2 level alone cannot be attributed to the changing vegetation composition in Banni for the last 4600 cal yr BP.
The present study, in combination with published records, has attributed the increasing % C 4 plants to the feedback mechanism between aridity, fire and vegetation composition. Increasing fire events (evidenced from Charcoal accumulation rate (CHAR)) from 2500 to 1500 cal yr BP helped to reduce C 3 woody plant cover and favored the expansion of C 4 grasses (S3 Fig). These fire events also promoted an arid understory microclimate and resulted in further drying in Banni. Continuous aridification helped fuel drying and led to more frequent fire events. In a nutshell, these effects have been enhanced by biogeophysical feedbacks; well-developed grasslands can favour the expansion of grasses by increasing light availability, reducing humidity and keeping high temperatures, all of which promote the growth of C 4 plants and are capable of maintaining the grassland setting.
Expansion of C 4 grasses can also impact the isotopic composition of the regional precipitation. Both IsoGSM2 model predictions and δD n-alkane value reveal extremely high precipitation δD values from Banni region. The present study suggests the following mechanism to explain this distinct isotopic character: the transpiration of grasses is higher compared to forest trees [72,73]. Although transpiration is a non-fractionation process, it indeed enriches the atmospheric vapor in heavier isotopes [73] and resultant precipitation has more positive isotopic composition.

Regional comparison of climate data
The variations in precipitation in Banni grasslands for the last 4600 cal yr BP correspond well with published records from the ISM realm (Fig 6). The enhanced ISM precipitation in Banni during ca. 4600-2500 cal yr BP was synchronous with pollen and phytolith based investigations from Pariyaj lake in western India [74]. The geochemical proxies measured from the sediment core of an active mudflat of Diu Island in western India also indicate warm and humid conditions between 4105 and 2640 cal yr BP [75]. The results from Banni grassland are also corroborated by the geochemical and palynological analyses of relict mudflat from southern Saurashtra coast with wet climatic conditions and simultaneous occurrence of marginally high sea-level between 4710 and 2825 cal yr BP [76]. Likewise, the climatic reconstructions based on lake sediments from Nal Sarovar, central Gujarat [77] and Lunkaransar and Didwana Lakes of Rajasthan, western India [78,79] observed a high lake stand during ca. 4680 and 3500 cal yr BP. The wetter conditions in Banni region during 4600-2500 cal yr BP is in good agreement with the record based on δ 13 C values of sedimentary leaf waxes from the core monsoon zone of India [9], and mineralogical and isotopic investigation on core sediments from Lonar lake, central India [10,80]. This humid phase also matches with the geochemical investigations of core sediments retrieved from continental shelf in eastern India [81].
The onset of aridity in the Banni region after 2500 cal yr BP was in phase with lowering of the sea-level during 2825 and 1835 cal yr BP, as observed from the Saurashtra coast in western India [76]. The results are also in line with the high resolution record from NW India: Nal Sarovar lake [77]; Pariyaj Lake [74], Lunkaransar and Didwana lakes [78,79]. The onset of aridity after 3 ka is also inferred in a tidal terrace sequence at the mouth of the Kharod River in western India [83]. A stalagmite record from peninsular India also indicates an abrupt climate change characterized by the decline of ISM around 2800 yr BP [84]. The findings of drier conditions in Banni region during ca. 2500-1000 cal yr BP is good accordance with the pollen record from NE India [85]; sedimentary leaf waxes from core monsoon zone [9] and geochemical data from eastern India [81]. More positive δD n-alkane values of the Lonar lake sediments (central India) also provide evidences of the Late Holocene arid conditions (Fig 6B). The δ 18 O values of the carbonates from the Tso Moriri lake, northern India also demonstrates a decline in the intensity of ISM precipitation during the Late Holocene (Fig 6C). Our record from Banni grassland is also in line with the climatic reconstruction from northern Bay of Bengal [82] that confirms a gradual decrease in the ISM intensity during the Late Holocene (Fig 6D).
The record from the Banni grasslands matches with the evidence of precipitation changes and vegetation responses in the other tropical grasslands across the globe. For instance, terrestrial ecosystems in the Sahara responded to the weakening of African monsoon during the Mid-Late Holocene by a conspicuous reduction in tropical trees from nearly 4300 cal yr BP to the establishment of current day desert ecosystem by ca. 2700 cal yr BP [86]. Studies from the Manga Grasslands in northeastern Nigeria have also shown that climate was the major determinant of vegetation change during the Mid-Late Holocene with a shift towards more arid conditions until 1000 cal yr BP followed by wetter conditions towards the present [87]. Distinct biostratigraphical and sedimentological evidences of the shift towards drier climatic conditions during the Mid-Late Holocene were reported from multiple sites across Western Africa [88].
Teleconnections governing long-term regional climate variability As various components of the different monsoon systems are interconnected through feedback mechanisms, perturbation of any of those components can affect the global monsoonal circulation [89]. For instance, less boreal summer monsoonal precipitation is linked with the increased Australian-Indonesian summer monsoonal (AISM) precipitation [90]. Therefore, understanding of land-ocean-atmosphere interaction on a global scale requires quantitative data from different monsoon domains. Towards this, climate records from different monsoonal domains (Fig 7) have been compared in this study to understand the role of forcing factors that persist across Hemispheres. The mean δD C29 values of the Chachi core sediments are significantly different for the phase-1 and 2, and this difference is correlated with the mean position of the ITCZ which is driven by the solar insolation [91] (Fig 7D). Monsoonal precipitation in the Northern Hemisphere increases with northward migration of ITCZ, accompanying decreased precipitation in the Southern Hemisphere; and vice versa [92]. Paleoclimatic records showed that the Southern Hemisphere insolation became more seasonal compared to Northern Hemisphere in the Late Holocene. The mean position of ITCZ during the phase-2 is thus postulated as the driving mechanism for the drying trend in ISM and East Asian summer monsoon dominated region and wet condition in the Southern Hemisphere (Fig 7A and 7D).
Increasing ENSO activity couldn also be argued for the increase in the δD C29 values after 2500 cal yr BP. An investigation based on the modern instrumental record suggested that majority of droughts in the ISM dominated regions are linked with the warm phases of ENSO events (i.e., El-Nino) [93]. The modern ENSO regime was established 3000 to 4500 cal yr BP and El-Nino events significantly increased after 2500 cal yr BP [94] concurrent with the marked dry condition in the Banni region (Fig 7E). This pattern in the precipitation correlates well with the grain size data from EL Junco Crater Lake in the Galapagos Islands, inferred to reflect the higher frequency of El-Nino events [95]. Together position of ITCZ and ENSO activity appear to be linked to the inferred aridification in Banni region during the Late Holocene.

Conclusion
Climate-driven changes in grassland ecosystems of the Indian Summer Monsoon realm have been investigated, for the first time using the isotopic composition of biomarkers in lake sediments. After ca. 2500 cal yr BP, a gradual increase in the δ 13 C n-alkane values indicates the dominance of C 4 grasses at the expense of C 3 plants. Difference between the δ 13 C n-alkane values indicates that each n-alkane homologue had different C 3 -C 4 proportional contributions. In the lake sediments, C 29 and C 31 homologues were mainly derived from C 3 plants and C 4 grasses respectively. The present study has used the δD C29 values as a proxy of the isotopic composition of paleoprecipitation. Meteorological and modeling observations show that δD values of modern precipitation in the Banni region are inversely related to the precipitation amount. This 'amount effect' based reconstruction suggests a marked decrease in the ISM precipitation after 2500 cal yr BP. The observed precipitation pattern in the climate record from Banni region corresponds well with published records. Coupled influence of southward migration of ITCZ and higher frequency of warm ENSO events led to the decrease in the ISM precipitation after 2500 cal yr BP and helped in establishing the grassland ecosystems.  Table. The %C 4 plants for the last 4600 cal yr BP has been estimated using bulk δ 13 C org , δ 13 C C29 and δ 13 C C31 values from the Chachi core sediments. Numbers in bold represent outliers. (DOCX) Yoshimura for his support to extract data from IsoGSM2 model. We are thankful to academic editor Prof. Hilton and the reviewers for their critical and thoughtful comments which helped to improve the quality of the manuscript.