Planktonic oxygen isotope ratios off the Indus delta reveal climate changes with a multi-centennial pacing during the last 6 ka, with the most prominent change recorded at 4.2 ka BP. Opposing isotopic trends across the northern Arabian Sea surface at that time indicate a reduction in Indus river discharge and suggest that later cycles also reflect variations in total annual rainfall over south Asia. The 4.2 ka event is coherent with the termination of urban Harappan civilization in the Indus valley. Thus, drought may have initiated southeastward habitat tracking within the Harappan cultural domain. The late Holocene drought cycles following the 4.2 ka BP event vary between 200 and 800 years and are coherent with the evolution of cosmogenic 14C production rates. This suggests that solar variability is one fundamental cause behind Holocene rainfall changes over south Asia.
 Northern hemisphere records reveal millennial-scale climate changes during the Holocene, which may have been tuned to solar radiation variability [Sirocko et al., 1996; deMenocal et al., 2000; Bond et al., 2001]. The most pronounced changes in northern Africa and western Asia occurred at the mid-late Holocene transition between 5.5 and 4.0 ka cal. BP, where one or several successive shifts towards dryer conditions are well documented [Bar-Matthews et al., 1997; deMenocal et al., 2000; Gasse, 2000]. Around 4.2 ka BP the ancient civilizations of Egypt and Mesopotamia suffered from sustained drought or even collapsed entirely [Weiss et al., 1993; Hassan, 1997; Cullen et al., 2000]. At the same time the Harappan civilization of the Indus valley (Pakistan) transformed from a highly organized urban phase to a post-urban phase of smaller settlements accompanied by a southeastward migration of the population [Possehl, 1997]. However, previous climate records are inconclusive on the timing of south Asian Holocene climate change, the underlying mechanisms of such change, and whether the Harappan civilization decline was the result of climate change [Singh et al., 1990; Possehl, 1997; Enzel et al., 1999; Weiss, 2000].
2. Regional Setting and Chronology
 Due to the effects of the south Asian monsoon a strong zonal precipitation gradient exists across the Arabian Sea (AS) (Figure 1a). The only significant freshwater source to the northern AS was the river Indus, until irrigation and flood control measures dramatically reduced its outflow over the last century [Milliman et al., 1984]. Summer monsoon ocean upwelling, which in the western AS lowers sea surface salinity, does not affect the northern AS or the Gulf of Oman (GO), and low salinity water from the Bay of Bengal does not flow into the northern AS during winter [Levitus et al., 1994]. However, a strong seasonality is imposed on northern AS and GO sea surface temperatures (SST) and mixed layer depth by the reversing monsoon winds and associated air temperatures (Figures 1b and 1c) [Rao et al., 1989].
 Laminated sediment core 63KA was retrieved from a site on the continental margin off Pakistan at 316 m water depth off the formerly active Indus delta (Figure 1a). The chronology of the mid-late Holocene section presented here is based on 36 14C dated single species samples of planktonic foraminifera, typically Globigerinoides sacculifer, but Orbulina universa in three occasions (Figure 2). 14C ages were calibrated by least-squares fitting of a 14C plateau resolved between 5 and 6 ka BP to the 14C calibration record (supplementary note 1)1 [Staubwasser et al., 2002]. The obtained reservoir age of 565 14C years was then used throughout the rest of the core.
3. The 4.2 ka BP drought Event
 The δ18O record of surface dwelling planktonic foraminifer Globigerinoides ruber shows little change in the mid-Holocene, but enhanced variability in the late Holocene with δ18O values between −1.7‰ to −2.1‰ (Figure 2). Between ∼1.0 ka BP and ∼0.4 ka BP detailed observations are somewhat compromised by intermittent absence of G. ruber in the core. The relatively variable late Holocene appears to begin with a (positive) shift to heavier δ18O at 4.2 ka BP. This shift in G. ruber δ18O may reflect any combination of cooler sea surface temperatures (SST) and heavier δ18O of the ambient water due to enhanced salinity [Bemis et al., 1998]. A comparison with the δ18O record (G. ruber) of core M5-422 from the GO shows opposing δ18O trends across the northern AS at the 4.2 ka BP event (Figure 3). Under the dominant monsoon forcing of AS surface temperature (Figures 1b and 1c), opposing δ18O trends across the northern AS are difficult to explain in terms of SST changes. In addition, a late – mid-Holocene alkenone record from the northeastern AS does not indicate significant SST change between 4.5–3.5 ka BP [Doose-Rolinski et al., 2001]. Because of the strong zonal precipitation gradient over the northern AS and the proximity of core 63KA to the Indus delta, an alternative explanation for this contrast is a change in the annual northern AS evaporation - precipitation budget and Indus river discharge. The 0.3‰ heavier δ18O in 63KA after 4.2 ka BP (0.4‰ if the change to more negative, i.e. lighter, values in core M5-422 reflects a warming throughout the northern AS) would amount to a 1.0–1.2‰ higher surface salinity (1.3–1.6‰ accounting for sea surface warming), depending whether linear mixing between the very light Indus water (average δ18O ∼ −10‰) and AS surface water (δ18O ∼ 0.8‰) is assumed or the δ18O - salinity relation for the open AS is used [Mook, 1983; Delaygue et al., 2001].
 Around the 4.2 ka BP shift in 63 KA δ18O the Harappan civilization in the Indus valley transformed from a highly urban urban phase to a rural post urban phase [Possehl, 1997]. In particular, cultural centers, such as the large cities of Mohenjo-Daro and Harappa, were almost completely abandoned while locations in northern India grew in population. The concordance of Harappan habitat tracking with the 4.2 ka Indus discharge event suggests a causal relationship. A possible explanation is that a reduction of the average annual rainfall over the Indus river watershed restricted Harappan farming in the Indus valley and left large city populations unsustainable.
4. Causes of Holocene Monsoon Change
 A reduction in annual rainfall over the Indus watershed must not necessarily be the result of a change in the amount of summer monsoon rain alone. Although Indus discharge is extremely biased towards the summer monsoon season [Milliman et al., 1984], the δ18O of Indus water is lighter than monsoon rainwater and significantly affected by melt water from glaciers and snowfields in the Karakoram and western Himalayas, which are largely fed by winter and spring precipitation [Mook, 1983; Wake, 1989; also The Global Precipitation Climatology Centre, available at http://www.dwd.de/research/gpcc, 1998, hereinafter referred to as GPCC, 1998]. Advanced glaciers in the Karakoram mountains as well as pollen assemblages and lake-levels in the Thar desert suggest that winter/spring rain was also higher than today during the mid Holocene and extended further southwards than at present [Singh et al., 1990; Enzel et al., 1999; Phillips et al., 2000]. It is therefore likely, that Indus discharge was then distributed more evenly over the year, and that an enhanced annual Indus discharge rate led to lighter δ18O values in the northeastern AS. At the time of the Indus discharge reduction at 4.2 ka BP, dust flux from northern Arabia and Mesopotamia also increased (Figure 3), and rainfall in the Eastern Mediterranean was reduced [Bar-Matthews et al., 1997; Cullen et al., 2000]. This regional pattern suggests a significance of the 4.2 ka BP drought event beyond the south Asian summer monsoon regime into the winter rain dominated extra tropics of the northern hemisphere. It is likely, that the drought induced decline of ancient civilizations in Mesopotamia and south Asia at ∼4.2 ka BP is the consequence of altered extra tropical airflow during winter and a change in monsoon seasonality.
 Recently, a relationship between global climate change during the Holocene and the variation in solar radiation has been demonstrated [Neff et al., 2001; Bond et al., 2001]. Solar radiation variability is generally inferred from cosmogenic 14C production records, where enhanced production rates are attributed to lower solar radiation intensity and a reduced shielding of the earth from cosmic particles by the solar magnetic field [Stuiver and Braziunas, 1993]. The coherent quasi-periodic pacing of Indus discharge and 14C production on the multi-centennial band (630–780 years) during the late Holocene following the 4.2 ka BP event suggests a link between solar variability and south Asian climate change (Figure 4, see also supplementary note 2). Minima in the bi-centennially averaged record of 14C production rate (enhanced solar radiation) correspond to heavier δ18O values (reduced Indus discharge/drier conditions over the NW Indian subcontinent), and 14C production maxima (reduced solar radiation) correspond to lighter δ18O (higher discharge/wetter conditions) after the 4.2 ka BP event. Across the 4.2 ka BP event a similar reduction in the 14C production rate can be observed, but there is no apparent similarity between the two records prior to that event between ∼4.5 and 7 ka BP. As such, no full cycle is observable across the event, and the contribution of solar forcing to the Indus discharge reduction at 4.2 ka BP in particular remains somewhat uncertain. However, a well-dated dry spell has been observed in the Thar desert around 4.7 ka cal BP [Enzel et al., 1999], which coincides with a 14C production minimum. Therefore, we cannot rule out that the absence of coherency between solar variability and Indus discharge between ∼4.5 and 7 ka BP may be due to either a sampling artifact or a different δ18O - salinity - SST relationship. Because of a potential inaccuracy of the AS surface 14C reservoir correction of up to 150 years prior to 7 ka [Staubwasser et al., 2002], chronological uncertainty may obscure any present coherency in the early Holocene. However, spectral analysis of the AS 63KA and Oman stalagmite δ18O records suggests a significant contribution of solar forcing to early Holocene south Asian climate variability [Neff et al., 2001; Staubwasser et al., 2002].
 Climate cycles in the ∼700 years band are inherent in other Holocene records from south Asia, but have hitherto not been matched with solar radiation [Wang et al., 1999; Sarkar et al., 2000]. The large-scale significance of this cycle suggests that subtropical Asian climate in general is relatively sensitive to changes in solar radiation. Climate modeling has demonstrated a non-uniform global response of annual surface temperatures to changes in solar radiation, with a relative warming of the east African tropics and the Tibetan Plateau occurring at high solar radiation [Cubasch et al., 1997]. This would have a direct effect on the fundamental cause of the south Asian summer monsoon, i.e. the heat gradient between the warm Tibetan Plateau and the cool southern Indian Ocean. However, a mismatch of amplitudes in the monsoon and solar radiation records (Figure 4) suggests that any such link would be complex. On inter-annual time-scales, changes in boundary conditions due to non-solar causes result in large amplitude variations of the monsoon [Webster et al., 1998]. There is a tendency of a strong south Asian monsoon to be followed by a relatively weak one [Meehl, 1994]. This is part of the large scale tropospheric biennial oscillation, in which tropical/subtropical summer convection over east Africa, south Asia and the Indo-Pacific Ocean is coupled with extra tropical atmospheric flow over Asia in a way that modulates air flow direction, temperature and precipitation over south and central Asia in winter. It is possible that higher levels of solar energy output during the Holocene may have enhanced inter-annual summer monsoon variability, altered winter airflow over south Asia, and changed total annual precipitation over the Indus watershed.
 We thank J. Negendank (GFZ, Potsdam) for additional funding within the KIHZ project of the Helmholtz Gemeinschaft (HGF), and G. Henderson (Oxford) for support. MS was funded by an EC Marie Curie Fellowship.