Were springline carbonates in the Kurkur–Dungul area (southern Egypt) deposited during glacial periods?

Tufas are terrestrial carbonates that form under open air conditions at ambient temperature from calcium bicarbonate water in streams, rivers and lakes; thus they indicate active spring lines or water table intersection with the ground surface and meteoric water recharge (Pedley 1990; Andrews 2006). Tufa and travertine (similar deposits forming in geothermally warmed springs) have been used to study palaeoclimatic, tectonic and palaeoenvironmental conditions on local, regional and global scales (Chafetz and Folk 1984; Ford and Pedley 1996; Hancock et al. 1999; Arenas et al. 2000; Andrews 2006; Capezzuoli et al. 2010; Özkul et al. 2013; Della Porta et al. 2017a; Toker 2017; Török et al. 2017). Egyptian tufa studies have focused on palaeohydrology, palaeoclimate and hominin dispersals (e.g. Crombie et al. 1997; Smith et al. 2004a, b; Drake et al. 2013). As the sporadic presence of tufa and speleothems indicates, the areas of the Saharan and Arabian deserts were not always as arid as they are today (5–20 mm a⁻¹ rainfall; Burns and Matter 1995; Fleitmann et al. 2003; Osmond and Dabous 2004; Patterson et al. 2005; Cremaschi et al. 2010; Drake et al. 2013; Dramis et al. 2014; Dramis and Fubelli 2015; Nicoll and Sallam 2017; El-Shenawy et al. 2018). Dating and geochemical study of these tufas can elucidate these past humid periods, recording information about the type of recharge, moisture source origins and relationships with larger scale changes in the climate system. In Egypt, previous studies provided important information about tufa and travertine deposits (Butzer 1964; Ahmed 1996; Anwar 2004; Adelsberger and Smith 2010; Smith 2012; Sallam et al. 2018; Sallam and Ruban 2019), and some studies (Crombie et al. 1997; Sultan et al. 1997; Smith 2001; Wanas 2012; Hassan 2014, 2015; Jimenez 2014; Hamdan and Brook 2015; Abotalib et al. 2016; Nicoll and Sallam 2017; Gaber et al. 2018) utilized stable isotopes to reconstruct the temperature or δ¹⁸O of the precipitating water. However, the timing of humid periods compared with forcing (which is expected to underlie change in rainfall in the region) is still debated, and research progress has been challenged by uncertainties owing to assumptions regarding the δ¹⁸O or T values of the precipitating water, and because of kinetic fractionation during tufa deposition. As a result, there are critically important open questions regarding the moisture source, the ambient temperature during deposition and the processes controlling the North African climate throughout the Pleistocene and Holocene.

In the Western Desert in Egypt, several oases (e.g. Kurkur, Dungul, Farafra, Khraga and Dakhla) host travertine and/or tufa deposits (Crombie et al. 1997; Sultan et al. 1997; Kleindienst et al. 1999; Smith et al. 2007; Jimenez 2014; Wanas and Armenteros 2019). This study addresses the sedimentological and geochemical aspects of these tufa systems, including their lithofacies and depositional environments. We focused on the Kurkur, Dungul and Dineigil oases, as well as the Gebel Kalabsha and Gebel El-Digm mountains located between them, and used petrographic, stable carbon and oxygen isotope, clumped isotope and U–Th age data for tufa deposits to (1) date and reconstruct the palaeoclimatic conditions that prevailed in southern Egypt during the Quaternary, (2) understand the palaeoenvironmental conditions of tufa formation and (3) determine the temperature and origin of the tufa-depositing water (the source of precipitation).

The Kurkur–Dungul area is located in southern Egypt along the eastern part of the south Western Desert between latitudes 24° and 23°22'N and longitudes 31° and 32°46'E and includes parts of the Nubian plain (Fig. 1). The study area comprises the Kurkur, Dungul and Dineigil oases as well as Gebel Kalabsha and Gebel El-Digm. Kurkur Oasis is located about 62 km SW of Aswan City in southern Egypt at 23°54'N, 32°19'E. Dungul Oasis (region centred at 23°25'N, 31°36'E) is considerably smaller than Kurkur Oasis. It lies about 60 km to the west of Kurkur Oasis, in a topographically low area in Wadi Dungul. Wadi Dungul collects rainwater from the surrounding plateau surface where dendritic drainage networks are well developed. The water trapped in Dungul feeds springs located in the main depression of the wadi. The study area between Kurkur and Dungul oases comprises two mountains (Gebel Kalabsha to the east and Gebel El-Digm to the west) forming a part of the scarp face of the Sinn El-Kaddab Plateau. Dineigil Oasis is a small basin lying about 7 km south of the Dungul Oasis on a step of the plateau surface near the scarp face.

The area is dissected by numerous deep-seated strike-slip and dip-slip faults, mainly in east–west and north–south directions (Issawi 1968). The east–west-trending faults dissect the limestone plateau of Sinn El-Kaddab forming several horsts and grabens, and some major faults (e.g. Kalabsha fault) extend into the Nubia plain. The north–south-trending faults primarily dissect the sandstone beds of the Nubia Formation, but can also affect the limestone plateau areas (Issawi 1968; Khamies and El-Tarras 2010).

The Kurkur–Dungul area exposes a thick sedimentary succession deposited on a corrugated surface of Precambrian rocks. This succession, ranging in age from the Late Cretaceous to the Quaternary (Fig. 1; supplementary material, Fig. S1) (Issawi 1968; Van Houten et al. 1984; Nicoll and Sallam 2017), is made up of thick fluvio-marine to nearshore marine deposits (Nubia Formation, Coniacian–Santonian; Issawi 1968; Van Houten et al. 1984; Khamies and El-Tarras 2010) that are conformably overlain by nearshore marine gypsiferous shale (Dakhla Shale, Maastrichtian; Said 1961; Issawi et al. 2009). Towards the top, the latter formation conformably passes to shallow marine fossiliferous limestone with sandstone intercalations (Kurkur Formation, Paleocene; Issawi 1968). An unconformity marks the passage to the uppermost marine limestone, chalky limestone and shale (Garra Formation, Paleocene–Early Eocene) and reefal limestone (Dungul Formation, Early Eocene) testifying to the termination of the southward transgression of the Neotethys over Egypt (Issawi 1968).

The Quaternary deposits in the Kurkur–Dungul region were classified by Issawi (1968) as (1) conglomerate sheets, (2) freshwater carbonates (tufa and travertine), (3) freshwater limestone mapped from Gebel El-Digm, (4) calcareous deposits covering some wadi channel banks between Gebel Kalabsha and Gebel El-Digm, (5) mudpans covering the pediplain between the Nile and the Sinn El-Kaddab scarp and (6) sand dunes developed in arid intervals during the late Quaternary.

The Nubian Aquifer extends over a large area including northern Sudan, eastern Libya and western Egypt (Sultan et al. 1997) and two sites have been proposed as recharge regions: the SW mountainous areas and the desert itself under conditions more humid than today (Sanford 1935; Thorweihe 1982). Tufa deposits in the Kurkur–Dungul area were precipitated from springs from perched aquifers above this main Nubian Aquifer system during wet periods when the water table was significantly higher (Nicoll and Sallam 2017). In this scenario, past pluvial episodes recharged the Nubian Aquifer in the southern areas, causing deep groundwaters to move in a NW direction. Consistent with this view, Jimenez (2014) compared tufa geochemistry (⁸⁷Sr/⁸⁶Sr, stable isotopes) with modern groundwater chemistry and suggested a consistent Nubian groundwater source over the last 500 kyr in the Western Desert with some deeply derived component. Retention time for water in the aquifer is at least several hundred thousand years (Gossel et al. 2010; Sherif et al. 2019) reflecting slow rates of flow (0.5–3.5 m a⁻¹), and residence time may be as high as 1.3 myr (Sturchio et al. 2004; Patterson et al. 2005; Jimenez 2014). Throughout the last million years, the hydrological conditions have changed several times in the recharge areas, with mixing of water from different times and different sources (Gossel et al. 2010). Between 300 and 1000 ka, infiltrating precipitation also underwent evaporation and condensation processes, further complicating its isotopic and compositional characteristics (Gossel et al. 2010). Quantitative work by Gossel et al. (2004) and Voss and Soliman (2014) has shown that (1) 5500 years is needed from the beginning of a wet period for an aquifer to recharge completely, (2) groundwater levels begin to decline in depressions 1000 years after termination of wet periods and (3) up to 10 kyr (or even more) is needed for natural groundwater discharge to cease (Abotalib et al. 2019).

Tufa and calcite deposits located at different elevations at five locations (Table S1) in the Kurkur–Dungul area were lithologically classified and sampled (Table S2; Figs S2 and S3). These locations are the Kurkur Oasis (Fig. S4), Gebel El-Digm (Fig. S5), Dineigil Oasis (Fig. S6), Dungul Oasis and Gebel Kalabsha (Fig. S7). From each location, several samples were collected from each lithofacies and from different topographic elevations. Forty-five thin sections, representing a selection of all the identified lithofacies, were prepared and examined under Olympus BX51 light microscopy. This allows the avoidance of secondary cements and diagenesis and the choice of the best samples for geochemical analyses.

Eighty-two stable (oxygen and carbon) isotope analyses from 74 carbonate rock samples (tufa or primary cement ‘calcite’ materials) were determined using an automated carbonate preparation device (Gasbench II) and a Thermo Fisher Scientific Delta Plus XP continuous flow mass spectrometer at the Institute for Geological and Geochemical Research, Budapest, Hungary. Carbonate powders were extracted with a dental microdrill avoiding the mixing of carbonate components and were reacted with 100% phosphoric acid at 70°C. Standardization was conducted using laboratory calcite standards calibrated against the NBS-19 standard. The carbon and oxygen isotopic compositions are expressed in the conventional delta notation against the international standard V-PDB (for δ¹³C and δ¹⁸O). The 1σ reproducibility for both C and O isotope analyses is better than ±0.1‰.

Clumped, carbon and oxygen isotopic compositions of bulk carbonate powder of 10 tufa carbonate and primary calcite cement samples were measured at the University of Washington's IsoLab (USA). The details of the automated vacuum line and sample purification methods used in IsoLab have been described by Burgener et al. (2016). In the present study 6–8 mg of powdered subsample was reacted in a common acid bath at 90°C and cryogenically purified using an automated system. Purified CO₂ was analysed using a multi-collector MAT253 isotope ratio mass spectrometer (Thermo Scientific) configured to measure masses 44−49 inclusive. Reference gas piped into this instrument eliminated the need for reference gas refilling, and ensured sample and standard are balanced throughout a run. The correction for ¹⁷O interference used a value of 0.528 to relate abundances of ¹⁷O and ¹⁸O (Brand et al. 2010), which has been shown to increase the accuracy of Δ₄₇ and eliminate discrepancies among abiogenic Δ₄₇−temperature calibrations (Schauer et al. 2016). Pressure baseline correction (He et al. 2012) was made by measuring the reference gas signal 0.0084 V down-voltage of the peak centre. Samples were converted to the carbon dioxide equilibrium scale (CDES) Δ₄₇ reference frame (Dennis et al. 2011) using a suite of CO₂ samples equilibrated with deionized and isotopically enriched water at 4, 60 and 1000°C. All Δ₄₇ values reflect 90°C acid digestions with the application of an acid fractionation factor to approximate a 25°C reaction. Long-term Δ₄₇ precision based on repeat measurements of an in-house calcite standard is ±0.018‰ (1σ). Sample reproducibility is taken as the standard error calculated using the number of sample replicates and the standard deviation (SD) of sample replicates or the SD of carbonate standards, whichever is larger. Both of these values can be found in the EarthChem archived file (Kele et al. 2020).

From each locality (Fig. 2; Fig. S2) the cleanest samples were selected for dating. Eleven tufa samples were dated with U–Th techniques at the High-Precision Mass Spectrometry and Environmental Change Laboratory (HISPEC), Department of Geosciences, National Taiwan University (Shen et al. 2003, 2012). For each sample, bulk and clean 1–2 g subsample was selected, gently crushed, ultrasonicated and then dried at 50°C on a class-100 bench in a class-10 000 clean room (Shen et al. 2008). The cleanest fragments, 0.1–0.7 g, were picked for U–Th chemistry (Shen et al. 2003). A triple-spike, ²²⁹Th–²³³U–²³⁶U, isotope dilution method (Shen et al. 2002) was used to determine U–Th isotopic and concentration data. Instrumental analyses were carried out by multi-collector inductively coupled plasma mass spectrometry (MC-ICP-MS), using a Thermo Electron Neptune system (Shen et al. 2012). Uncertainties in the U–Th isotopic data were calculated offline (Shen et al. 2002) and include corrections for blanks, multiplier dark noise, abundance sensitivity and contents of the four nuclides in spike solution. Half-lives of U–Th nuclides used for age calculation, relative to AD 1950, have been given by Cheng et al. (2013).

Tufas from paludal and pool environments usually contain significant detrital contamination, and this, combined with low initial U concentrations and short time for ingrowth of radiogenic ²³⁰Th, makes high-resolution U–Th dating of tufa extremely challenging (Garnett et al. 2004). The very low concentrations of authigenic ²³⁰Th mean that even slight variations in the initial ²³⁰Th/²³²Th value of the detrital component can have a large impact on the calculated age (Garnett et al. 2004). In this study age corrections for initial detrital ²³⁰Th content were calculated using an estimated atomic ²³⁰Th/²³²Th ratio of 4 (± 2) × 10⁻⁶; those are the values for a material at secular equilibrium, with a crustal ²³²Th/²³⁸U value of 3.8 and an arbitrary uncertainty of 50%. Isotopic and age errors given are two standard deviations of the mean and two standard deviations, respectively, unless otherwise noted.

Seven main lithofacies (FA1–FA7) were recognized and their description and classification are summarized according to the terminology of Gandin and Capezzuoli (2014) and Della Porta et al. (2017a) (Table S2, Text S-1 and Fig. S8).

The lithofacies are mainly characterized by tabular clotted peloidal mudstone or wackestone (FA1), hemidomic laminated boundstones (FA3) and local micritic dendrite boundstone (FA6). In the marginal portions, lenticular to tabular coated plants boundstones (FA4) and crystalline dendrite boundstone (FA2) become predominant. Intraclastic rudstones or packstones (FA5) are present at the wadi margins and locally intercalated in the more distal portions. Only at Gebel Kalabsha, blocky primary calcite cements are present along local fractures as crusts and druses (FA7).

δ¹³C and δ¹⁸O values range between −9 and −1.1‰ and between −13.3 and −8.5‰ (V-PDB), respectively (Fig. 3; Table S3). The highest δ¹³C values were measured from sites 1 and 2 of Kurkur Oasis and site 5 of Dungul Oasis, and the lowest values were from the calcite samples from Gebel Kalabsha (δ¹³C_average  =  −7.3‰) and a calcite part of a country rock from Gebel Kalabsha (δ¹³C_average  =  −5.3‰). The lowest δ¹³C_tufa values are from the coated plants boundstone tufa of the Dineigil Oasis (δ¹³C_average  =  −5.9‰) (Fig. 3). δ¹³C and δ¹⁸O values of four primary calcite cement samples from Gebel Kalabsha range between −9.0 and −6.3‰ and between −11.8 and −11.3‰, respectively. The rock part and the vein and vug-filling calcite of a country rock sample (CR 1, Fig. 3) of Gebel Kalabsha were also analysed separately (δ¹³C_rock  = −5.2‰ and δ¹³C_calcite  =  −5.3‰; δ¹⁸O_rock  =  −11.9‰ and δ¹⁸O_calcite  =  −12.3‰). Average δ¹³C and δ¹⁸O values of tufa samples from the Kurkur Oasis, Dungul Oasis and Gebel El-Digm are similar, as well as the average δ¹⁸O value of tufa samples from Dineigil Oasis (Fig. 3). Calcite samples from Gebel Kalabsha have values around −11.5‰ and the lowest δ¹⁸O value ( −12.3‰) was measured from the calcite part of a country rock sample (Fig. 3; Table S3). Only the coated plants boundstone tufa of the Dineigil Oasis and the crystal calcite samples of Gebel Kalabsha show deviation from the average tufa values (Fig. 3).

Most of the locations from which more than two samples were analysed show strong positive correlation between the δ¹³C and δ¹⁸O values. At sites 4, 5 and 6 at Kurkur strong correlation (r² > 0.85) exists between the δ¹³C and δ¹⁸O values (Fig. 3) and in case of Kurkur sites 1 and site 3 the r² is +0.58 and +0.7, respectively. From Kurkur site 7 only two samples were measured, and at site 2 no correlation was observed. At Dineigil Oasis the δ¹³C and δ¹⁸O values show slight correlation (r² =  +0.33), similarly to Gebel El-Digm, whereas samples from site 2 of Dungul Oasis show very strong correlation (r² =  +0.98) (Fig. S9).

δ¹³C values of all carbonate samples (Table S3) are in the meteogene range of Pentecost (2005). The empirical equation of Panichi and Tongiorgi (1976) resulted in δ¹³C_CO2 values between −17.9 and −12.3‰ (V-PDB), and the theoretical equation of Bottinga (1968) provided similar values (from −17.5 to −11.0‰) (Table 1).

Δ₄₇ values of 10 selected samples from Gebel Kalabsha, Kurkur, Dungul and Dineigil oases and Gebel El Digm range between 0.658  ±  0.017 and 0.715  ±  0.012‰ (Table 1), corresponding to a temperature range between 36 and 16°C, using the calibration of Petersen et al. (2019). Temperature uncertainties are of the order of 3°C (1 standard error, or c. 6°C, 95% confidence interval). Kurkur 5-2 and Kurkur 6-3 samples were collected from sites where possible kinetic effects could have been present based on the strong correlation between the δ¹³C and δ¹⁸O values (Fig. 3). However, these two samples do not show significantly different temperature values from the other samples. Although the studied sites are located relatively close to each other, there are significant differences in the calculated temperatures (Fig. S10). The highest temperatures are from the fracture-filling crystalline calcite from Gebel Kalabsha (36°C) and from the tufa of Gebel El-Digm (33–34°C), and the lowest temperature (16°C) is from a tufa from Dineigil Oasis.

Nine samples provide reliable U–Th age data from 368 ± 14 to 11.7 ± 1.2 ka (Table 2). From the Kurkur Oasis four samples were dated between 326 ± 14 and 139 ± 11 ka, whereas the two samples from Gebel El-Digm show relatively old ages (368 ± 14 and 258 ± 20 ka). Tufa samples of the Dungul Oasis were dated as 184.4 ± 5.4 and 43.52 ± 0.23 ka, whereas the youngest age of 11.7 ± 1.2 ka is from the Dineigil Oasis. At Dungul Oasis, the Dungul 12-1 sample (43.52 ± 0.23 ka) is located at the lowest elevation (243–250 m above sea-level (a.s.l.)), whereas the Dungul 16-5 sample (184.4 ± 5.4 ka) is located at the highest elevation (302 m a.s.l.) (Fig. 2). At Gebel El-Digm, the sample located at the lowest elevation (Gebel El-Digm 9-6, 310 m a.s.l., 258 ± 20 ka) is younger than the Gebel El-Digm 11-2 sample located at 350 m a.s.l. (368 ± 14 ka). At Kurkur Oasis, the Kurkur 2-2 sample located at the lowest elevation (272 m a.s.l., 139 ± 11 ka) is the youngest, whereas the Kurkur 5-2 sample (350 m a.s.l., 251 ± 4.7 ka) is older. However, the Kurkur 7-1 sample located highest (375 m a.s.l.) is younger than the Kurkur 6-3 sample located at 330 m (a.s.l.) with an age of 326 ± 14 ka.

The lithofacies recognized (see Text S-1) suggest slow sedimentation at ambient temperatures in low-energy environments. These environments include stagnant ponds, pools and shallow lakes, palustrine and/or lacustrine shores, water cascades and at the margin of a low-energy continental carbonate depositional system in dammed areas. Tufa mounds are concentrated along fault planes and seem to be formed by the circulation and emergence of carbonate-rich palaeosprings through fissures and cracks during periods of enhanced groundwater discharge (Nicoll and Sallam 2017). Outgassing of CO₂ from carbonate-rich waters was an important factor in the formation of tufas along the paths of water streams, and particularly at water cascades where tufa deposition continually raised the profile of the stream, creating a positive feedback system (Nicoll et al. 1999; Nicoll and Sallam 2017). The clotted peloidal tufas (Kurkur Oasis, sites 1 and 2) indicate deposition in ponded water areas subjected to wave action and turbulent currents or flowing water in gently sloped or stepped channels (Della Porta 2015). Carbonate-cemented intraclasts and gravels at the wadi margins (Kurkur Oasis, sites 1 and 2) could have been deposited during the beginning of a humid period when surface runoff flushed loose debris into the wadi (Crombie et al. 1997), reflecting aggradation in slow-flowing water environments (dammed areas, pools, shallow lakes). The increasingly humid conditions raised the water table and led to carbonate precipitation in the wadi channel and in shallow lakes and pools at the confluence of wadi branches. Carbonate moulds and casts of mosses (Kurkur Oasis, sites 3, 4, 5 and 6) indicate deposition in stagnant pools or slow-flowing palustrine areas (Pedley 2009; Arenas-Abad et al. 2010), within interchannel areas, lacustrine shores and the parts of terraces that were subjected to periodic water splash (Nicoll and Sallam 2017). The presence of crystalline dendrite boundstone in the proximal parts of some sites (Kurkur Oasis, sites 4 and 6) indicates deposition in fast-flowing, smooth to stepped slopes and rims of pools (Jones and Renaut 2010). Localized ‘shrubby’ facies at these same sites imply a higher mineralization rate than is typical for ambient-temperature tufa sites and/or involvement of microbial mats metabolizing reduced Sulphur (Erthal et al. 2017). Consequently, it is possible that there was some geothermal influence at Kurkur, although this is unlikely to be above 35°C as this would inhibit the mosses and other macrophytes that are abundantly preserved in these materials. Crystal calcite crusts, primary cements and druses at Gebel Kalabsha are derived from local deposition from waters along fractures, and probably occurred below the surface.

There is a poor correlation of Egyptian tufa dating with expected periods of rainfall in the Mediterranean basin derived from the understanding of marine sapropels. Our youngest tufa (Dineigil 8-4 sample, 11.7  ±  1.2  ka, marine isotope stage (MIS) 1–MIS 2 interglacial, Table 2) is probably too old to be ascribed to several humid periods that have been inferred at the start of the Holocene. For instance, this sample is older than sapropel S1 (10.5–6.1 ka) and the high African Monsoon Index of that time (Rossignol-Strick and Paterne 1999; Vaks et al. 2010; Grant et al. 2017) arising from high northern hemisphere summer insolation (Laskar et al. 2004). It is also older than widespread African humidity of the early Holocene (Szabo et al. 1995; Churcher et al. 1999; Kindermann et al. 2006; Geyh and Thiedig 2008; Brookes 2010; Drake et al. 2011; Hamdan and Lucarini 2013; Hamdan and Brook 2015; Kleindienst et al. 2016). Similarly, it is older than the closest phase of growth in speleothems from Hoti cave (10.5–6.2 ka, Burns et al. 2001) and Mukalla cave (10–6 ka, Fleitmann et al. 2011) or early Holcene low δ¹⁸O values of Soreq cave speleothems (Bar-Matthews et al. 2003) (Fig. 4). Late Pleistocene and Holocene carbonates (McCool 2019), pedogenic carbonate concretions (Dal Sasso et al. 2018) and fossil gastropod shells (Williams et al. 2010) dated to 15–5 ka BP by accelerator mass spectrometry (AMS) radiocarbon and ascribed to the African Humid Period are reported from the central and northern Sudan. From the Fezzan Basin (SW Libya) optically stimulated luminescence (OSL) data for lacustrine sediments implied the existence of at least one perennial lake (Palaeolake Shati) between 11 and 9 ka (Armitage et al. 2007; Drake et al. 2008). The global early Holocene ‘travertine’ age data frequency were found by Ricketts et al. (2019) to peak at around 8.6 ka, with ‘travertine’ in that study being synonymous with ‘tufa’ in this study. Again, the Dineigil age is older, placing it outside the global trend to greater humidity at the start of the Holocene.

Similarly, the Dungul 12-1 tufa sample (43.5  ±  0.2 ka, glacial, MIS 3c, Table 2) does not fit to the sapropel events and is in the middle of a dry period in NW and East Africa (Grant et al. 2017), although it does fit one of the main travertine age data frequency peaks (at 41 ka) and occurs within the timeframe of inferred increased monsoon intensity from 40 to 50 ka (Ricketts et al. 2019) (Fig. 4). This period of humidity is not reflected to the further north and east, as Soreq cave speleothems (Bar-Matthews et al. 2003) show high δ¹⁸O values and no dated speleothems are known from the Negev Desert (Vaks et al. 2010), Hoti cave (Oman, Burns et al. 2001) or Mukalla cave (Yemen, Fleitmann et al. 2011). Growth at Wanni Sannur cave (Egypt, El-Shenawy et al. 2018) at this time is also absent, and no north and east Saharan lacustrine deposits are known of this age (Szabo et al. 1995; Geyh and Thiedig 2008). In contrast, growth phases of speleothems from the Susah cave (Libya) were documented from the period immediately before (37.5–33 ka) and after (52.5–50.5 ka) the deposition of the Dungul 12-1 sample (Hoffmann et al. 2016), reflecting greater humidity during MIS 3 to the west, and spatial complexity in the wider region. The probability density function (PDF) analysis of OSL and U–Th dates from humid sites in North Africa combined with palaeohydrological mapping suggest that during MIS 3 and the start of MIS 2 there were three brief periods of enhanced humidity and humidity peaks centred on 76 ka, between 92 and 129 ka (Drake and Breeze 2016). A humid (lacustrine) phase occurred in SW Libya during the Last Interglacial as well (Thiedig et al. 2000; Armitage et al. 2007) and OSL age data of Drake et al. (2011) show evidence for humidity and an interconnected hydrological system during MIS 5 in the Sahara. However, no deposition of this age has been found in the Egyptian tufa systems we discuss here.

From the Kurkur Oasis, the youngest tufa age data (Kurkur 2-2, 139  ±  11 ka, glacial MIS 6a) date to the Penultimate Glacial Maximum, but this age is older than the youngest ‘Greening Period’ (centred at 128.5  ±  1.1 ka), as suggested by El-Shenawy et al. (2018) based on dating of Wadi Sannur cave speleothems (Fig. 4). The age of the Kurkur 2-2 tufa is older than the S5 and significantly younger than the S6 sapropel (Grant et al. 2017). According to the ODP967 wet–dry index (Grant et al. 2017) this period was significantly more humid than today, and that is also reflected in the formation of north and east Saharan lacustrine deposits (Szabo et al. 1995; Geyh and Thiedig 2008), speleothems (e.g. Negev Desert, Vaks et al. 2010) and tufa and travertine (Gaven 1982; Kleindienst et al. 1999, 2008; Smith 2001; Smith et al. 2004b, 2007). There is evidence for humidity during MIS 6 at 135, 154 and 180 ka in the north and east of the Sahara (Drake and Breeze 2016). In Wadi Shati (SW Libya) an extensive Pleistocene lake existed with highstands between 140 and 130 ka (MIS 5d and 5e) (Gaven 1982). Lacustrine terraces with abundant carbonate accumulated during humid episodes at >420, 380–290, 260–205 and 140–125 ka, roughly coincident with the MIS 9, 7 and 5e interglacial periods (Cancellieri et al. 2016).

The age of the Dungul 16-5 sample is 184.4  ±  5.4 ka (glacial MIS 6e, Table 2), which was a period of runoff from North Africa according to Grant et al. (2017) although the age of this tufa falls between the S6 (175.63 ka) and S7 (197.53 ka) sapropels. From this period there are age data from the Hoti and Mukalla caves (Burns et al. 2001; Fleitmann et al. 2011) but there is no overlap with the ‘Greening Period’ (centred at 219.4  ±  7.3 ka) of El-Shenawy et al. (2018) (Fig. 4). There are, however, several papers documenting travertine and tufa deposition at this time (Sultan et al. 1997; Brookes 2010; Kleindienst et al. 2016; Abotalib et al. 2019). δ¹⁸O values of speleothems from the Peqiin cave (Bar-Matthews et al. 2003) are high during this time period, implying aridity to the north of Egypt.

The Kurkur 5-2 sample has an age of 251.1  ±  4.7 ka (glacial MIS 8a, Table 2), which was a rather dry period according to Grant et al. (2017), and is not synchronous with any of the Greening Periods of El-Shenawy et al. (2018) (Fig. 4). This age is also slightly older than the S9 sapropel (241.35 ka, Grant et al. 2017). Dated speleothems from the nearby countries are reported only from the Mukalla cave (Fleitmann et al. 2011); however, lacustrine carbonates were present in the northern and eastern Sahara (Szabo et al. 1995; Geyh and Thiedig 2008).

The Gebel El-Digm 9-6 sample (258  ±  20 ka, glacial MIS 8b, Table 2) does not correspond to any of the sapropels, and Greening Periods and dated speleothems are known only from the Mukalla cave (Fleitmann et al. 2011) (Fig. 4). This time period was dry in North Africa, as shown by Grant et al. (2017). However, lacustrine deposits are reported from the northern and eastern Sahara (Szabo et al. 1995; Geyh and Thiedig 2008), and a tufa sample from Gebel el Agouz was also dated to a similar age by Sultan et al. (1997).

The Kurkur 7-1 sample (283  ±  17 ka, interglacial MIS 9a, Table 2) does fit to a humid period of Grant et al. (2017) (Fig. 4). This timing is incompatible with the age of the S9 and S10 sapropels, and does not fit with the Greening Periods of El-Shenawy et al. (2018) or with humid periods recorded in speleothems (Burns et al. 2001; Vaks et al. 2010; Fleitmann et al. 2011; El-Shenawy et al. 2018). However, east Saharan lacustrine deposits were mentioned by Szabo et al. (1995) and tufa samples from Gebel el Yabisa were dated to this age (Sultan et al. 1997).

The Kurkur 6-3 sample (326  ±  14 ka, interglacial peak, MIS 9e) is slightly younger than the age of S10 sapropel (334.83 ka, Grant et al. (2017) and the related humid period (Fig. 4), and partly overlaps with the Greening Periods and speleothem age data published from the Wadi Sannur cave (El-Shenawy et al. 2018), Negev Desert (Vaks et al. 2010), Hoti cave (Burns et al. 2001), Mukalla cave (Fleitmann et al. 2011) and north Saharan lacustrine deposits (Geyh and Thiedig 2008).

The Gebel El-Digm 11-2 sample (368  ±  14 ka, MIS 11a interglacial, Table 2) does not fit to any of the Greening Periods of El-Shenawy et al. (2018) or to the speleothem age data and sapropel events reported above and it is in a very dry period of Grant et al. (2017) (Fig. 4). However, Geyh and Thiedig (2008) reported lacustrine deposits from the north Sahara and Armitage et al. (2007) reported a humid phase from Libyan Sahara (SW Libya), which was tentatively correlated with MIS 11.

The poor correlation of the Egyptian tufa dating with expected periods of rainfall in the Mediterranean basin (derived from the understanding of marine sapropels described above) is surprising for a system that is expected to be driven homogeneously by insolation-forced migration of the Intertropical Convergence Zone (ITCZ) (Bosmans et al. 2015). However, we find better coherence between tufa phases reported here and tufa, ‘travertine’ and lacustrine deposits further south in eastern Africa. The Kurkur area is among the driest areas on Earth and currently Kurkur Oasis has limited spring discharge (Butzer 1965; Nicoll and Sallam 2017), so we can be confident that these deposits reflect regionally significant decrease in water deficit. Consequently, we are able to consider any period of significant tufa growth to be a direct result of significant increase in the recharge of the aquifer, which combined with the eastern African lake and tufa data requires a large region of increased rainfall. An important caveat provided by Abotalib et al. (2019) is uncertainty about whether aquifer recharge needs to be close to the site of tufa formation, or could reflect rain hundreds of kilometres further south that was transported through the exceptionally large Nubian sandstone aquifer. Because such transport of water would depend on recharge of these aquifers, this could delay formation of spring deposits by thousands of years, significantly altering interpretation of depositional ages. As emphasized by Abotalib et al. (2019), modelling of the Nubian Sandstone aquifer indicates that it may take c. 5.5 kyr from the beginning of a recharge period for the aquifer, and a 1 kyr delay between the onset of arid conditions in the recharge area and significant groundwater level fall has been proposed by Voss and Soliman (2014). Such aquifer recharge dynamics would be expected to delay spring flow inception, and maximum tufa formation, potentially by several thousand years. If the delay between the onset of arid conditions in the recharge area and the timing of maximum tufa formation is consistent through time, dating of tufa sites should therefore exhibit a probability peak with the same characteristic timing delay compared with atmospheric forcing.

We investigate this effect by exploring the timing of the deposits described herein relative to the orbital forcing that is expected to underlie changes in rainfall in the region. The primary driver of rainfall changes in the Sahara region is considered to be northern hemisphere insolation, specifically with excess warming of the northern hemisphere leading to more rainfall in currently arid and hyperarid regions (Grant et al. 2016). If growth of tufa deposits reflects rainfall distal from the site, transferred to a more northerly location and a later time by aquifer transport, tufa timing should be expected to follow peaks in northern hemisphere insolation (Berger et al. 2003).

The delay periods indicated in the upper part of Table 3 are very variable, and some are so large as to almost place tufa formation onto the subsequent northern hemisphere insolation maximum. For example, Dineigil 8-4 actually leads northern hemisphere insolation, with a delay of –1.4  ±  1.2 kyr. Other tufa formation phases have long (e.g. Dungul 12-1) or short delay periods (e.g. Gebel El-Digm 11-2 with a delay of 3.5  ±  14 kyr), although dating uncertainty means that these delays could still be >10 kyr. On average, delay for the new data presented here is c. 8.6 kyr compared with the assumed forcing. Extending the same analysis to the literature review data compiled by Abotalib et al. (2019) is complicated by the use of different dating technologies and incorporation of different types of deposit in their analysis. For instance, playa deposits may arise from local pooling above a perched water table and reflect a shorter rainfall period that may not significantly recharge the aquifer. Restricting the analysis to tufa sites dated by the U–Th method permits a realistic evaluation of apparent delays, and we find the pattern shown in the lower half of Table 3. Again, we find that these delays are very large and very variable (average 6.8 kyr, range –0.3  ±  1.3 to 19.5  ±  2.8 kyr). Even with 2σ error taken into account, three sites show delays >9 kyr, which are difficult to explain even with a long aquifer recharge delay, and a further three sites show delays less than 5 kyr even with 2σ error taken into account, which is more rapid than would be expected. Although it is possible that variable control on timing is arising from a complex interaction between orbital forcing and aquifer recharge dynamics, the long average delay, highly variable delay and the lead over forcing at the site we report with the best constrained date (Dineigil 8-4) are all challenging for this conceptual model to explain.

One solution for this is that the forcing is more complex, and does not arise essentially from precession-forced motion of the ITCZ. Similar to the results and interpretation of Abotalib et al. (2019), our new age data coincide with glacial periods. This may be significant, as recent climate modelling (Di Nezio et al. 2016) suggests that the emergence of the Sunda and Sahul shelves owing to sea-level fall resulted in significant oceanographic and atmospheric reorganization in the Indian Ocean region, which affected rainfall in eastern Africa. This reorganization would result in periods of increased rainfall in eastern Africa during glacial periods when the Indonesian throughflow was significantly reduced. Based on the depth of the majority of the Sahul and Sunda shelf areas, and also the depth of the Karimata Strait (Di Nezio et al. 2016), we estimate that significant reduction of the Indonesian throughflow would occur owing to sea-level fall of 50 m or more below modern levels.

Seven of the nine sites we present have timings that place them within periods with sea-level of at least –50 m, using the eustatic sea-level curve presented by Grant et al. (2014) (Fig. 4). The remaining two have mean dates outside these periods, but uncertainties that extend into low sea-level periods renders them ambiguous. Within the period analysed by Grant et al. (2014), periods with sea-level at least as low as –50 m occur about 53% of the time. Laying the ambiguous sites aside, the probability that seven sites would occur during low sea-level periods and none would occur during high sea-level owing to random sampling is thus c. 1%. Extending the same analysis to the data compiled by Abotalib et al. (2019), we find 10 sites (50%) that occur during low sea-level and a further three (15%) ambiguous sites with uncertainty ranges extending into low sea-level periods. All seven of the remaining sites are of MIS 5e age, and exhibit consistently low apparent delays to precession (an average of 1.0 kyr and a maximum of 3.7  ±  3.6 kyr).

We find it challenging to conclude that a single forcing can underlie this pattern, and suggest that the simplest explanation is that there are at least two causes for flow at these spring sites: wet phases during very warm interglacials (MIS 5e) and during glacial periods arising from different forcing factors. The very short and very consistent delay for the MIS 5e sites suggests that these periods are not being significantly delayed by aquifer recharge processes, implying rather a short distance between the site of rain and the tufa site itself. It is not possible to make a similar evaluation for the glacial mode, but if this period is being forced by a sea-level lowstand Indian Ocean Warm Pool, the rainfall arising from this would be expected to be located to the south of Egypt. Consequently, it is likely that periods of glacial tufa growth are delayed relative to the rainfall.

Gaber et al. (2018) measured δ¹³C and δ¹⁸O values of tufa samples from different elevations at Kurkur, ranging from Pleistocene (older upper level, c. 345 m a.s.l.) to Recent (younger lower level, c. 270 m a.s.l.) and concluded that the old tufa had been developed during a warm pluvial period, whereas the young ones developed in drier periods. Jimenez (2014) obtained ages of 76–246 ka for the lower levels and 514 ka for the higher levels of the travertines at Kurkur, and suggested deposition between 600 ka and 1 Ma as well. Based on the study by Crombie et al. (1997) the youngest spring and lacustrine travertines (70–160 ka) were found in Wadi Kurkur, whereas deposits with ages of 191–220 ka are exposed as linear mounds over fracture zones in ancient wadis and the oldest travertines (> 260 ka) are extensive units on the plateau surface, deposited in paludal and lacustrine environments. Our age data show good correlation (r²  =  0.83) with elevation, supporting the observations of Jimenez (2014) and Gaber et al. (2018) (Table 2; Fig. S11). In almost all cases, the oldest tufa is located at the highest position, whereas the younger samples are at lower positions, suggesting a progressive lowering in the water table with time (Fig. 2). This supports the original suggestion of Szabo et al. (1989) that the water table levels declined in this region through the Pleistocene. Reduction of the intensity of humid episodes throughout the Pleistocene was suggested by Szabo et al. (1995), Smith et al. (2004b) and Brookes (2010), who confirmed a decreasing trend of lacustrine sedimentation in NE Africa over time during the late Quaternary. In addition, Geyh and Thiedig (2008) reported a trend of shrinking lake size through the Quaternary, which was even more pronounced in the Late Pleistocene and Early Holocene (Cancellieri et al. 2016).

Carbon and oxygen isotopic composition of tufas and travertines are useful for understanding the palaeoenvironmental and palaeoclimatological conditions that prevailed during their deposition (e.g. Andrews et al. 2000; Andrews 2006; Gandin and Capezzuoli 2008; Pedley 2009; Cremaschi et al. 2010; Capezzuoli et al. 2014; Teboul et al. 2016; Della Porta et al. 2017a, b; Pla-Pueyo et al. 2017), although the data can be affected by diagenesis. Covariations between the δ¹³C and δ¹⁸O of tufa (Fig. 3) would also probably arise from aridity gradients during deposition (higher balance of evaporation over precipitation would simultaneously enrich δ¹⁸O_water and δ¹³C derived from organic matter; Talma and Netterberg 1983). Heavy carbon isotope values can imply long residence of carbon isotope species in water and equilibration with atmospheric CO₂ (Andrews et al. 1993) and evaporation can induce ¹⁸O and ¹³C enrichment, and thus covarying δ¹³C and δ¹⁸O values in lake systems (e.g. Talbot 1990; Horton et al. 2016). In the case of Kurkur tufas, Nicoll and Sallam (2017) showed petrographic evidence for the lack of diagenetic features typical of a burial regime, and we find no indication of significant alteration in our thin section analysis beyond cementation. Such early diagenesis generally occurs in the same water that the non-cement phases were deposited from, and this very early diagenetic impact on isotope geochemical proxies may be insignificant (Andrews and Brasier 2005; Andrews 2006; De Boever et al. 2017). It therefore seems likely that the isotope values established for these materials are a good indication of the water and carbon available at these sites at the time of deposition. More probably, the correlation between δ¹³C and δ¹⁸O in most of our sampled locations, together with our petrographic data, reflects evaporation, rapid degassing and related kinetic effects during deposition under an arid climate (Kele et al. 2008, 2011), which would have been exaggerated by the shallow pool environments that dominated these systems (see facies analysis in Text S-1). This reflects similar findings by Hassan (2015), who observed significant (r  =    +  0.85) correlation between the δ¹³C and δ¹⁸O values of soil carbonates from the New Cairo Petrified Forest and interpreted it as a result of aridity, Gaber et al. (2018), who found very weak correlation between the δ¹³C and δ¹⁸O values of their tufa samples from Kurkur, and Wanas and Armenteros (2019), who observed moderate covariance in the case of fluvial tufa at Farafra Oasis (Egypt). To conclude, although the positive covariance between the δ¹³C and δ¹⁸O values caused by local disequilibrium conditions is common in tufa samples, their isotopic compositions do contain palaeoenvironmental and palaeoclimatic information (e.g. Pazdur et al. 1988; Andrews et al. 2000; Kano et al. 2007; Pedley 2009; Cremaschi et al. 2010).

The calculated isotopically depleted δ¹³C_CO2 values (Table 1) suggest that the studied carbonates are mostly tufa samples deposited from meteoric waters associated with soil and atmospheric CO₂ characterized by low δ¹³C, with minimal crustal carbon addition (Fig. 3). This is also consistent with their sedimentary facies, which are dominated by features of ambient temperature systems with strong biomediation of precipitation, and even where some indication of warming is noted (as at Kurkur) the temperatures do not seem to have been raised above 30–35°C. As we find no consistent relationship between facies and isotopic characteristics, it seems that precipitation kinetics and fractionation are rather coherent across these systems, and that disequilibrium effects are limited. Exceptions are (1) crystalline dendrite deposits (FA2 facies) showing the lowest δ¹⁸O values (samples are mostly from the probably slightly geothermal Kurkur Oasis, sites 4, 5 and 6), which could be explained by deposition from low δ¹⁸O_water (Table 1); (2) cement samples (Gebel Kalabsha) that show the lowest δ¹³C values (Fig. 3; Table S3) as they were probably less affected by evaporation than the pool samples, whereas their low δ¹⁸O values could be the result of high (36°C) temperature of deposition (Table 1).

Our δ¹³C_tufa values (c. –6 to –1‰) are generally higher than the δ¹³C values of other North African and Middle Eastern speleothems (Susah cave, Libya, Rogerson et al. 2019, –11 to –7‰; Soreq cave, Israel, Affek et al. 2008, –11 to –5‰) and partly overlap with the δ¹³C values of the early Holocene tufa of Tadrart Acacus Mt (SW Fezzan, Libya, –4.8 to 2.6‰, Cremaschi et al. 2010). However, the Egyptian values lie within the range of speleothems from the Negev Desert, Israel (Vaks et al. 2006, 2010, –9 to +1‰). The trend in the speleothems in the eastern Mediterranean climate region reflects trends in aridity, with δ¹³C values between –12 and –8‰ reflecting C3 type Mediterranean vegetation, whereas the more ¹³C-enriched range of the Negev reflects the C3–C4 mixed vegetation of the steppe and semi-desert (Vaks et al. 2006). We conclude that despite the evidence for increased spring flow, the environment in which these tufas were depositing remained an arid semi-desert, and that the rainfall providing the water may not be significantly altering the local landscape.

The youngest dated sample (Dineigil 8-4; Fig. 4) has the lowest δ¹³C value (–5.9‰), which is consistent with the age of deposition being a humid period, reflecting a significant contribution of CO₂ from the soil zone derived from mixed C3–C4 type vegetation. This site also has relatively depleted δ¹⁸O_cc. It is likely that this deposit formed within a landscape more dissimilar to, and less arid than, most of the sites investigated in this study. That finding is surprising, as this deposit formed during late MIS 2, when the ITCZ is expected to be located far to the south of the Sahara. Our isotope data thus reinforce the conclusion from U–Th datings that this part of Egypt experienced important humid periods with different phasing relative to orbital forcing than is found in other systems. Clearly, the glacial eastern Sahara at least for short periods exhibited better water resources than are found in the current interglacial.

The δ¹⁸O values of our dated Egyptian tufa samples (–12.7 to –8.9‰, Table 1) overlap with the most depleted speleothem values reported by Vaks et al. (2006, –11.0‰ < δ¹⁸O < –3.0‰) and by Vaks et al. (2010, –10.5‰ on average) from different caves of the Negev Desert (Israel) and with the values published by Cremaschi et al. (2010) from the Holocene tufa of Tadrart Acacus (Libya, –11.3‰ < δ¹⁸O < –2.1‰) (Fig. 5). Soreq cave (Israel) speleothems have δ¹⁸O values of –11.1‰ < δ¹⁸O < – 5.2‰) (Affek et al. 2008), and similar ranges are reported from both the Mukalla cave (Yemen, –12.3‰ < δ¹⁸O < –2.8‰; Fleitmann et al. 2011) and Hoti cave (–12‰ < δ¹⁸O < –4‰; Burns et al. 2001). In both of those caves, the most depleted values are reported during MIS 5e. A stalagmite from the Wadi Sannur cave (El-Shenawy et al. 2018) has an δ¹⁸O_ave value of –11.6  ±  0.8‰, which lies within the same range. Our δ¹⁸O_mean tufa values are also very close to the δ¹⁸O_mean data for lacustrine tufas from the Dungul region (−9.4‰, Hassan 2014) and for tufas of Kharga Oasis (−9.7‰, Smith et al. 2004a) (Fig. S10).

These datasets are in strong contrast to Susah cave speleothems from Libya (Rogerson et al. 2019), where δ¹⁸O_cc values are considerably higher (–5.3‰ < δ¹⁸O < –3.8‰). Pleistocene carbonates deposited in Libya, central Sahara have higher δ¹⁸O values compared with our δ¹⁸O_tufa values, although these will be modified by evaporation in a closed basin lake (Gaven et al. 1981). Nevertheless, the significant differences in speleothem data strongly imply that two different hydrological provinces are active, one of which is located in central and western northern Africa (‘maghrebian’) and is isotopically enriched, and the other located in Egypt (‘mashriqian’), which is isotopically depleted. The two isotopic provinces co-occur in the Arabia–Levant region, providing exceptionally large variances in records from these regions. It would seem likely that the relatively enriched isotopes recorded at Farafra Oasis and the New Cairo Petrified Forest (Fig. S10) imply that the maghrebian isotopic province extends into northernmost Egypt. Indeed, it is possible that the rainfall driving tufa formation with mashriqian isotopic characteristics is in fact falling considerably further south, and the occurrence of this water in Egypt reflects aquifer transport.

The observation of an isotopic mixing zone in Egyptian palaeoclimate data is not itself new, having first been proposed by Smith et al. (2004a). In that original hypothesis, both westerly and Indian Ocean monsoonal precipitation contributed to the recharge of aquifers. The compilation of data in this study gives strong support to this concept, and reveals that rainfall in the region exhibits a much more complex spatio-temporal structure than has previously been appreciated. Although we support the suggestion of Smith et al. (2004a) that the maghrebian precipitation is probably ‘westerly’, reflecting some Atlantic and considerable Mediterranean source moisture (Rogerson et al. 2019), the occurrence of mashriqian precipitation during MIS 5e and during glacial periods makes a specific identification of this as ‘monsoon’ rain difficult to sustain. It may reflect a mixture of monsoon and Indian Ocean Warm Pool-forced precipitation, which occur with different phasing compared with insolation.

The average of our 10 T(Δ₄₇) values (Table 1) is 29°C, which is higher than the mean annual ground temperature (26.6°C) of Egypt, but similar to groundwater temperatures in the Western Desert (26 to c. 40°C) (Swanberg et al. 1983). Lower temperatures (17.5–26.2°C) were found by Abouelmagd et al. (2014), whereas 10–30°C was inferred by Nicoll and Sallam (2017) for the tufas of the Kurkur region and by Hamdan and Brook (2015) for the Late Pleistocene Eastern Desert tufas (14–20.8°C) and for the Holocene tufas at the same site (21–24°C). Today, there are no thermal waters in the Western Desert, and waters with elevated temperatures are either artesian or pumped well waters arising from great depth (Crombie et al. 1997). The deep source at Kharga and Kurkur oases was supported by the Sr isotopic data of Abotalib et al. (2019), by the ²³⁴U/²³⁸U activity ratio and U concentrations of groundwater samples (Dabous and Osmond 2001), and by noble gas and geochemical analyses of springs and groundwater (Mohammed 2015). Swanberg et al. (1983) reported low (<20°C  km⁻¹) regional temperature gradient in the Kharga, Dakhla, Farafra and Bahariya oases, and wells tapping deep artesian aquifers produce large volumes of water in the 35–43°C range. High temperature values (29–53°C) were measured by Sultan et al. (1997) on fossil groundwaters from wells that sample various horizons in the Nubian Aquifer at Kharga, Dakhla, Farafra and Bahariya oases.

Five out of our nine dated tufa samples were deposited during glacial periods, whereas four samples formed during interglacials. Considering the interglacial samples, the Dineigil 8-4 sample formed during the MIS 1–MIS 2 interglacial with a T(Δ₄₇) value of only 16  ±  4°C, whereas the Kurkur 7-1 sample (MIS 9a), the Kurkur 6-3 sample (MIS 9e) and the Gebel El-Digm 11-2 samples (MIS 11a) formed at 30  ±  1°C, 30  ±  4°C and 33  ±  2°C, respectively. Samples from glacials also show scatter in their T(Δ₄₇) data: Dungul 12-1 sample (MIS 3c) formed from a water at 22  ±  <1°C, but the other samples formed at higher temperatures (Kurkur 2-2, MIS 6a, 32  ±  7°C; Dungul 16-5, MIS 6e, 27  ±  2°C; Kurkur 5-2, MIS 8a, 34  ±  5°C; Gebel El-Digm 9-6, MIS 8b, 34  ±  1°C). The T(Δ₄₇) values of our tufa samples thus do not follow the pattern of glacial–interglacial periods. This is consistent with mild geothermal heating affecting some or all of these sites inconsistently over time. It is worth noting that the facies showing the most clear evidence of mild geothermal influence occurs at Kurkur, which also shows the most consistently high (30–34°C) temperatures in clumped isotope measurements. In the case of the Dungul Oasis the older sample, located at higher elevation, is precipitated from a warmer (27°C) water than the one located at lower elevation (22°C) (Fig. 2). At the other sites (e.g. at Kurkur) there is no relationship between the temperature of deposition and elevation. Thus, although there is a connection between the altitude of the samples and their age, this is actually not followed by the calculated temperatures.

To properly resolve whether there are two precipitation isotopic provinces interacting in this region, we calculated δ¹⁸O_water from δ¹⁸O_cc data using the temperature constraints from clumped isotope thermometry (Table 1). These data indicate that the aquifer waters forming tufa deposits had δ¹⁸O_water values between −9.7 and −6‰ using the Kim and O'Neil (1997) equation (or between −11.9 and −8.3‰ for the Kele et al. (2015) equation). There is considerable uncertainty in these determinations, however, as the average temperature uncertainty is 3°C. When propagated into δ¹⁸O_water alongside uncertainty in δ¹⁸O_cc and calibration (we used a Monte Carlo approach, combining uncertainties in the measurement assuming they are all normally distributed), we find that the 1σ uncertainty is c. 2‰ regardless of calibration equation.

Today, the waters in Egypt can be distinguished on the basis of their stable isotope composition. Modern waters in an alluvial aquifer in northern Egypt are isotopically heavy (δ¹⁸O +2.3 to +3.9‰, Bakri et al. 1992), whereas modern meteoric waters in Sudan have an average value of −2.1‰ (Joseph et al. 1992). Relatively young (<20 ka), shallow groundwaters have δ¹⁸O values between −9.0 and −6.0‰ (Sonntag et al. 1978a, b; Haynes and Haas 1980; Muller and Haynes 1983). Old (>20 ka) Western Desert groundwaters have an average composition between −11.5 and −10.5‰ (Sonntag et al. 1978a, b; Thorweihe 1990; Sultan et al. 1997). Abotalib et al. (2016) analysed groundwater samples from the Bahariya Oasis, tapping the Nubian Aquifer (δ¹⁸O from −11 to −10.3‰), and from Wadi El-Natrun and Wadi El-Farigh areas, tapping the Miocene and Pliocene aquifers (δ¹⁸O from −2.4 to −1.3‰). Consequently, despite the wide uncertainty in our measurement this demonstrates that the springs we have analysed record waters consistent with ‘old Western Desert groundwaters’, not younger groundwater, or older fossil groundwater tapping reservoirs emplaced in the pre-Pleistocene. This assists with placing our findings into regional context, as it demonstrates that we are revealing conditions during the major humid periods that recharged the Western Desert aquifers, and that these conditions are not analogous to the modern climatology.

We come to the same conclusion when comparing reconstructed springwater δ¹⁸O with modern rainwater isotope properties. The δ¹⁸O of the rainwater 15 km north from the New Cairo Petrified Forest (Fig. S10) was −8.2 to +5‰ from 1986–1991 to 2000–2003 (IAEA data, Hassan 2015). The weighted average value of δ¹⁸O for present-day rain is −4.6‰ (OIPC data, Bowen 2008). This is also beyond the range of the values for palaeowater we provide, even given the range of uncertainty we have. This modern water is also more depleted in ¹⁸O than the water recorded in fluid inclusions of MIS 3 age from Susah cave (Rogerson et al. 2019), although the wide range of possible values consistent with the data we present for Egyptian tufa does overlap with that range.

However, bearing in mind the broad uncertainty range in δ¹⁸O_water we report in our analysis and in this discussion, it is not possible to analyse differences between sites or times in detail.

This paper presents the first detailed geochemical study of Pleistocene tufa at Kurkur, Dineigil and Dungul oases, Gebel El-Digm and Gebel Kalabsha (southern Egypt), providing new U–Th ages and direct temperature and δ¹⁸O_water estimations using the clumped isotope method. Based on stable isotopes and petrography the studied carbonates were deposited from meteogene waters associated with soil and atmospheric CO₂ characterized by low δ¹³C values. The observed facies associations (clotted peloidal mudstone/wackestone, crystalline dendrite boundstone, laminated boundstones, coated plants boundstones, intraclastic rudstones/packstones and micritic dendrite boundstones) suggest deposition in stagnant ponds, pools and shallow lakes, palustrine and/or lacustrine shores, and at the margin of a low-energy depositional system in dammed areas. Stable isotopic compositions of the tufa samples are not significantly influenced by the facies. Correlation between the δ¹³C and δ¹⁸O values of tufa can be explained by kinetic effects caused by strong evaporation, in cases where the deposition occurred in shallow lakes or ponds. Facies analysis and clumped isotope geochemistry both suggest some slight geothermal heating at some sites or times, but not to the point that vascular plants were excluded (30–36°C).

The U–Th ages of the tufa are between 368  ±  14 and 11.7  ±  1.2 ka (between MIS 11 and MIS 1) indicating deposition during periods with low sea-level (below –50 m). There is no consistent lag between tufa deposition and sapropel events, humid periods and speleothem deposition around the studied area, or relative to the expected orbital forcing (high northern hemisphere insolation). Rather, we suggest that it is likely that the region experienced spring flow during glacial periods. A potential explanation for this is that during low sea-level, reduced Indonesian throughflow created an Indian Ocean Warm Pool, resulting in anomalous rainfall on its western margin. Tufas are younger towards lower elevations, indicating a progressive lowering of the water table level with time and therefore a long-term (million year) trend towards greater aridity. Comparison with published datasets provides a regional picture that is compatible with this model of secular drying punctuated by glacial-age periods of aquifer recharge, with the exception of MIS 5e, which appears to be the only high sea-level period with substantial tufa development in Egypt.

The relatively high δ¹³C_tufa values (–6.2 and –1.4‰) compared with the δ¹³C of European tufa can be explained by contribution of mixed C3–C4 vegetation. Consequently, despite demonstrating a substantially increased amount of flow from the aquifer compared with today, the tufa deposits seem to have formed in semi-desert conditions comparable with the modern state. This is consistent with the argument by previous researchers (Abotalib et al. 2019) that at least some of these deposits reflect rainfall further south, with water transported through the aquifer displacing the springs both from sites and from times of enhanced rainfall.

The δ¹⁸O_tufa values (–12.7‰ < δ¹⁸O < –8.9‰) are comparable with the values for tufa deposits from elsewhere in southern Egypt, but are incompatible with values for speleothems in Libya and tufa and associated deposits from northern Egypt. Combining data from our study and the literature, we support an argument from previous researchers (Smith et al. 2004a) that Egypt lies at the intersection between two palaeoprecipitation provinces. These provinces display different δ¹⁸O characteristics, with relatively enriched rainfall to the west reflecting Atlantic–Mediterranean sources (‘maghrebian’) and relatively depleted rainfall to the east reflecting an Indian Ocean source (‘mashriqian’). The very large range of isotopes in speleothems recorded from Israel and Arabia may indicate rainfall from both sources within an intersecting zone.

Interpreting calculated δ¹⁸O_water values is complicated by the propagated uncertainty arising from joint measurement of δ¹⁸O and temperature via clumped isotopes. However, this approach demonstrates that the water depositing the tufa we report is inconsistent with modern rainfall in Egypt but is consistent with the water of Pleistocene age in Western Desert aquifers. This supports the proposal that rainfall in the past was not an analogue of that today in terms of moisture source.

Overall, we find that the record of climate in Egypt's tufa is inconsistent with a simple model of palaeoclimate for this region. We find precipitation under both glacial and exceptionally warm interglacial times, but not during interglacials, which exhibit lower global temperature rise. We also suggest that the region is responding to forcing related to the hydrology of the Indian Ocean, as well as to orbital forcing of movement in the ITCZ. Finally, we find that locations of spring formation are not necessarily locations of enhanced rainfall. Evidently, more field research in this region is needed to provide data further constraining this complex emerging picture.

Constructive comments made by J. Andrews and an anonymous reviewer greatly improved the paper.

SK: conceptualization (lead), data curation (lead), formal analysis (equal), funding acquisition (lead), investigation (lead), methodology (equal), project administration (lead), resources (equal), supervision (lead), visualization (equal), writing – original draft (lead), writing – review & editing (lead); ESS: conceptualization (equal), data curation (equal), investigation (lead), visualization (equal), writing – original draft (equal), writing – review & editing (supporting); EC: conceptualization (supporting), formal analysis (supporting), investigation (equal), visualization (equal), writing – original draft (equal), writing – review & editing (equal); MR: conceptualization (equal), data curation (supporting), investigation (supporting), software (equal), supervision (equal), writing – original draft (equal), writing – review & editing (equal); HW: investigation (supporting), writing – original draft (supporting), writing – review & editing (supporting); CS: formal analysis (equal), funding acquisition (supporting), investigation (equal), methodology (equal), writing – review & editing (supporting); MAL: formal analysis (equal), methodology (equal); TY: formal analysis (equal), methodology (equal); AS: formal analysis (equal), methodology (equal), writing – review & editing (supporting); KWH: formal analysis (equal), funding acquisition (supporting), investigation (supporting), methodology (equal), supervision (supporting), writing – review & editing (equal)

S.K. received support by the KH 125584 project of the Nemzeti Kutatási, Fejlesztési és Innovaciós Alap (NKFIH; National Research, Development and Innovation Office, Hungary). Clumped isotope analyses at the Isolab (University of Washington, Seattle, USA) were supported partly by the TraRAS (Travertine Reservoir Analogue Studies) project, and K.H. acknowledges funding from US National Science Foundation grant EAR-1156134. The research was supported by the European Union and the State of Hungary, co-financed by the European Regional Development Fund in the project GINOP-2.3.2-15-2016-00009 ‘ICER’. U–Th dating was supported by grants from the Science Vanguard Research Program of the Ministry of Science and Technology (MOST), Taiwan (108-2119-M-002-012), the Higher Education Sprout Project of the Ministry of Education, Taiwan (108L901001) and the National Taiwan University (109L8926).

All data generated or analysed during this study are included in this published article (and its supplementary information files)

Scientific editing by Philip Hughes

An ORCID id for ESS has been added and the caption of figure 4 has been corrected.