Climate Dynamics

, Volume 36, Issue 9, pp 1717–1735

Using idealized snow forcing to test teleconnections with the Indian summer monsoon in the Hadley Centre GCM

Authors

    • NCAS-Climate, Walker Institute for Climate System Research, Department of MeteorologyUniversity of Reading
  • J. M. Slingo
    • NCAS-Climate, Walker Institute for Climate System Research, Department of MeteorologyUniversity of Reading
    • Met Office
Article

DOI: 10.1007/s00382-010-0805-3

Cite this article as:
Turner, A.G. & Slingo, J.M. Clim Dyn (2011) 36: 1717. doi:10.1007/s00382-010-0805-3

Abstract

Anomalous heavy snow during winter or spring has long been regarded as a possible precursor of deficient Indian monsoon rainfall during the subsequent summer. However previous work in this field is inconclusive, in terms of the mechanism that communicates snow anomalies to the monsoon summer, and even the region from which snow has the most impact. In this study we explore these issues in coupled and atmosphere-only versions of the Hadley Centre model. A 1050-year control integration of the HadCM3 coupled model, which well represents the seasonal cycle of snow cover over the Eurasian continent, is analysed and shows evidence for weakened monsoons being preceded by strong snow forcing (in the absence of ENSO) over either the Himalaya/Tibetan Plateau or north/west Eurasia regions. However, empirical orthogonal function (EOF) analysis of springtime interannual variability in snow depth shows the leading mode to have opposite signs between these two regions, suggesting that competing mechanisms may be possible. To determine the dominant region, ensemble integrations are carried out using HadAM3, the atmospheric component of HadCM3, and a variety of anomalous snow forcing initial conditions obtained from the control integration of the coupled model. Forcings are applied during spring in separate experiments over the Himalaya/Tibetan Plateau and north/west Eurasia regions, in conjunction with climatological SSTs in order to avoid the direct effects of ENSO. With the aid of idealized forcing conditions in sensitivity tests, we demonstrate that forcing from the Himalaya region is dominant in this model via a Blanford-type mechanism involving reduced surface sensible heat and longwave fluxes, reduced heating of the troposphere over the Tibetan Plateau and consequently a reduced meridional tropospheric temperature gradient which weakens the monsoon during early summer. Snow albedo is shown to be key to the mechanism, explaining around 50% of the perturbation in sensible heating over the Tibetan Plateau, and accounting for the majority of cooling through the troposphere.

Keywords

SnowIndian monsoonSeasonal forecastTibetan PlateauEurasiaHimalaya

1 Introduction

The Asian summer monsoon is relied on by more than a third of the world’s population for the majority of their water resources, for agriculture and, increasingly, industrial uses. While statistical seasonal forecasts are routinely made of the summer monsoon rainfall, consistently incorporating parameters based on ENSO (Krishna Kumar et al. 1995; Rajeevan et al. 2004, 2007), other global predictors are also used, reflecting the potential predictability of tropical rainfall variability implied by slowly varying surface boundary conditions (Charney and Shukla 1981). Historically, Himalayan snow cover during winter was used to infer details of monsoon rainfall in the subsequent summer. Blanford (1884) suggested an inverse relationship between such snow cover and the subsequent monsoon rains, especially in northern India. Subsequently Eurasian snow cover averaged over the much larger 0–180°E, 20–90°N region was included in statistical forecasts following correlation analysis by Parthasarathy and Yang (1995). However direct measures of snow have been removed from the latest India Meteorological Department power regression model (Rajeevan et al. 2007), perhaps related to decadal variability in the strength of the snow–monsoon teleconnection (e.g. Liu and Yanai 2002; Robock et al. 2003) or decadal changes in snow depth over the Tibetan Plateau (Zhang et al. 2004). Despite several observational and modelling studies the mechanisms involved are poorly understood, relating to several complicating factors, namely the region involved and the presence or absence of ENSO. Any mechanism may also depend on how snow is measured, whether weight, depth, or fraction of area covered.

Several studies use often limited observations to find correlations between regional snow cover and subsequent Asian summer monsoon rainfall, and to infer details of the mechanism involved. For the Himalaya/Tibetan Plateau region, this mechanism essentially fits the Blanford hypothesis (as named in Fasullo 2004), that persisting heavy snow reduces the upward surface sensible heating in spring and hence reduces the deep heating of the troposphere above the Tibetan Plateau. In consequence, the reversal of the meridional tropospheric temperature gradient essential for the monsoon onset (Li and Yanai 1996) is reduced, weakening the Asian monsoon, particularly in north India. Fasullo (2004) made a composite analysis of 1967–2001 NSIDC NH-EASE (National Snow and Ice Data Center Northern Hemisphere Equal-Area Scalable Earth grid) snow cover data for spring (MAM) covering the whole of Eurasia and its effect on Indian monsoon rainfall. Regions neighbouring northern India were noted to feature robust negative correlations, in support of the Blanford hypothesis. Negative correlations with snow were also found further north, although Fasullo (2004) suggested no mechanism.

Earlier observational studies were limited by short data records. Hahn and Shukla (1976) estimated fractional snow cover from satellite photographs over 9 years and saw negative correlations between DJFM Eurasian snow (south of 52°N only) and subsequent monsoon rainfall. Bamzai and Shukla (1999) later analysed 22 (9) years of satellite-derived snow cover (depth) data to show that inverse correlations between snow cover and monsoon rainfall anomalies existed only for western Eurasia, although their composites show consistent anomalies to extend east of this region. Such correlations are particularly significant when snow anomalies persist from winter (DJFM) into spring (AM). Contrary to other work, they found no significant relationship between Himalayan snow cover and the monsoon, despite this region showing amongst the largest interannual variations in snow.

Some studies based on the second release of the Historical Soviet Daily Snow Depth (HSDSD-II) station measurements also noted negative correlations between winter/spring snow cover and Indian monsoon rainfall. During the period 1957–1994, Dash et al. (2005) found strong negative correlations emanating from west Eurasia (35–65°N, 25–70°E) and positive correlations from east Eurasia (70–140°E at the same latitude). Highlighting the non-linearity of the snow–monsoon relationship, they found 57% of heavy snow events in the west were followed by deficient Indian summer monsoon rainfall, against only 24% of light snow events followed by heavy rain. They describe a mid-latitude circulation mechanism affecting the upper level monsoon easterlies in late spring. A similar result was also obtained by Kripalani and Kulkarni (1999). Singh and Oh (2005) also noted the west–east (negative–positive) dipole in correlations with monsoon rainfall using HSDSD-II, describing a mechanism in which the anomalous persistence of snow cover into spring delays heating of the Eurasian continent, leading to a weakened thermal low and consequent weak monsoon westerlies over India, and perturbations to the low-level jet over East Asia. Instead, Ye and Bao (2005) use mean JFM snow depth from 1950 to 1995 HSDSD-II data to show reasonable positive correlations between Eurasian snow and Indian rainfall measured using the Global Gridded Monthly Precipitation data. The HSDSD-II data have no coverage of the Himalaya/Tibetan Plateau region however.

The El Niño-Southern Oscillation (ENSO) has strong negative correlations with the Asian summer monsoon (e.g. Webster and Yang 1992, among many others) however it can also influence snow distribution over the Eurasian landmass via its effect on the zonal flow (Ferranti and Molteni 1999). Therefore any snow–monsoon interaction may be an indirect effect of ENSO, and some works are careful to take ENSO into account. In his stratified analysis, Fasullo (2004) considers anomalous monsoon years under ENSO and ENSO-neutral conditions, to demonstrate that the Blanford hypothesis is much more robust when ENSO is absent. Moreover, during ENSO years the sign of the correlation reverses, reflecting the dominance of ENSO forcing over land surface characteristics in perturbing the monsoon. Other authors make no allowance for ENSO (e.g. Kripalani and Kulkarni 1999; Bamzai and Shukla 1999, likely due to their short data record). In some cases, a significant reduction in sample size is found when removing ENSO effects (Dash et al. 2005). Singh and Oh (2005) show clear evidence of the classic El Niño-horseshoe pattern in a composite difference of Indo-Pacific SSTs between high and low snow years (their Fig. 4). Robock et al. (2003) used the NOAA (National Oceanic and Atmospheric Administration) northern hemisphere snow cover data to show a negative (positive) relationship between Eurasian (Tibetan) snow cover and All-India monsoon rainfall, however virtually all of their composited years were known ENSO events. In their observed study of the impact of Eurasian snow on the East Asian monsoon, Wu and Kirtman (2007) note that whilst ENSO and Tibetan Plateau snow cover compete to affect Indian monsoon rainfall, they act together to increase precipitation over southern China.

Models allow a more targeted and cleaner analysis of snow–monsoon connections and Fasullo (2004) provided a good review of early modelling studies, noting with only one exception that the monsoon response to elevated snow in central or southern Eurasia is negative. Using the UGAMP (UK Universities Global Atmospheric Modelling Programme) GCM, Dong and Valdes (1998) showed robust negative correlations between south Eurasian snow mass and Indian monsoon rainfall, delaying the onset through soil moisture and evaporation processes. Several modelling studies also assess the role of ENSO carefully. Corti et al. (2000) found that their leading mode of snow variability, consisting of a dipole between Himalaya/Tibetan Plateau and north/west Eurasia (their Fig. 2a, and similar to Fig. 5 here), was perturbed strongly by ENSO activity during the immediately preceding winter. This confirmed their hypothesis that the mode was of dynamical origin and caused by the effect of tropical SST anomalies on the large-scale circulation. Corti et al. (2000) showed the effects of nearby snow (i.e., the Himalaya) to dominate unless a strong El Niño is present, in which case the large-scale circulation anomaly prevails, leading to negative correlations with more remote Eurasian snow. By comparing atmosphere-only GCM ensembles forced by observed and average climatologically varying SSTs, Becker et al. (2001) reach much the same conclusion, that Indian rainfall is detrimentally influenced by local snow forcing through hydrology and thermodynamic mechanisms in the absence of ENSO. Meanwhile, monsoon dynamics undergo a negative influence from west Eurasian snow. In ensemble AGCM experiments comparing monsoons following the 1982/83 (El Niño) and 1983/1984 (La Niña) winters, Ferranti and Molteni (1999) show clear evidence for enhanced (reduced) snow over north/west Eurasia following El Niño (La Niña) related to changes in the circulation. Subsequently, weakened westerlies over south Asia and a southward shift of the subtropical jet in the South China Sea highlight a dynamically weakened monsoon.

Potentially local and remote regions of snow forcing can influence the monsoon, although as yet it is unclear which region should dominate in both GCMs and the real world. Teleconnections from the more remote north Eurasia region are also ill-understood. This study aims to assess the impact of snow forcing from local and remote regions separately on the subsequent Indian summer monsoon, determine which region dominates, and under what mechanism. The Hadley Centre models used, their validation and the experimental method are outlined in Sect. 2. In Sect. 3, we perform a composite analysis from a long integration of HadCM3, the coupled version of the Hadley Centre model, and also consider the dominant modes of snow variability in some other coupled models. Results from atmosphere-only experiments in HadAM3 and further sensitivity tests are described in Sect. 4. Discussion and conclusions are listed in Sect. 5. To avoid ambiguity, snow forcing regions are referred to as HimTP meaning Himalaya/Tibetan Plateau and WNEur for the west northern mid-latitudes of Eurasia in the remainder of this manuscript.

2 Methodology

2.1 The model and its validation

Monthly mean atmospheric and surface data from a 1050-year pre-industrial control integration of the UK Met Office HadCM3 coupled model were obtained from the NERC British Atmospheric Data Centre (BADC) for this study. HadCM3 (Pope et al. 2000; Gordon et al. 2000) was one of the first coupled ocean-atmosphere models capable of integrating for several hundred years without requiring artificial ocean surface heat-flux corrections to counteract climate drift (Johns et al. 2003) and has been widely used as a global climate model in both the Third and Fourth Assessment Reports of the Intergovernmental Panel for Climate Change (IPCC). The ocean model is solved on a 1.25° grid on 20 vertical levels while the atmosphere has a horizontal resolution of 3.75° × 2.5° in longitude and latitude respectively, on 19 vertical levels. The atmospheric component of this model (HadAM3) was used for the ensemble experiments described in the next section, with a vertical resolution of 30 levels offering a more realistic response of the atmospheric circulation to SST forcing in the tropics (Spencer and Slingo 2003). When referring to both the coupled (HadCM3) and atmosphere-only (HadAM3) models in general, the term Hadley Centre model will be used to avoid confusion.

Both HadCM3 and HadAM3 feature the Met Office Surface Exchange Scheme (MOSES), an interactive land surface model (Cox et al. 1999) featuring a Penman-Monteith boundary layer and hydrology and a four-layer model for soil moisture and temperature. Soil moisture is capable of melting and freezing, allowing for a more realistic simulation of surface temperatures, and in the interactive vegetation canopy, evaporation is dependent on stomatal resistance, itself related to temperature and CO2 (Pope et al. 2000). As Cox et al. (1999) remark, snow acts to make the land surface brighter and smoother, impacting on the radiative, turbulent and ground heat fluxes, as well as insulating the soil beneath. Gridbox surface albedo varies between the snow-free value and a deep-snow value itself dependent on temperature, thus providing a simple representation of snow metamorphosis. A fixed snow density is assumed (ρsnow = 250 kg m−3).

2.1.1 Model mean state

In a comparison of the climatological snow depth (expressed as soil water equivalent, SWE) with ECMWF ERA-40 reanalysis, Putt (2008) shows HadCM3 to possess a reasonable seasonal evolution in the northern hemisphere. Over Eurasia, the main biases are a weakened maximum in snow depth over central Russia during spring, perhaps as much as −30%, and deficient snow over east Siberia during winter and spring.

In the absence of a reliable long-period dataset of snow depth that covers the entire domain of interest, to validate the model’s climatological evolution of snow with observations more closely we compare with NSIDC NH-EASE snow cover fields (Armstrong and Brodzik 2005) on a 25 km equal-area grid as in Fasullo (2004). As the Hadley Centre model does not output fractional snow cover directly, we use a technique suggested by Qu and Hall (2007) in their multi-model comparison. For instances of snow depth Sd ≥ 60 kg m−3, full cover is assumed (Sc = 1). Below this threshold, the fractional snow cover in a grid cell decreases linearly to zero, Sc = Sd/60 for 0 < Sd < 60 kg m−3. Qu and Hall (2007) validated this conversion in five models that output both snow depth and cover information and demonstrated only 1–2% differences in predicted and outputted snow cover fields in those models. Furthermore, near-perfect correlations (>0.925) were noted between timeseries of the output and predicted cover. Figure 1 shows the climatological evolution of this snow cover proxy in the HadCM3 coupled control run in comparison with NSIDC NH-EASE observed average snow cover for 1966–2005 during late winter and spring. Over the January to May period shown, HadCM3 shows a remarkably good representation of the position of the snow edge and its seasonal cycle, particularly over the Eurasian continent. This suggests the Hadley Centre model is well-representing the atmospheric and surface processes responsible for snow formation and melt. An assessment of the springtime snow albedo feedback, largely due to the dependence of surface albedo on temperature variation (ΔαsTs), reveals HadCM3 to have a mid-range value among the CMIP3 models (Qu and Hall 2007). This snow albedo feedback is important for the rate at which snow melts through the season, and hence the seasonal cycle. Note that in the model data, the much lower resolution (∼300 km vs. 25 km in these observations) necessarily allows a less distinct snow edge to be defined. A minor deficiency is the low snow cover in a narrow band extending from Tibet through Mongolia during late spring. Elsewhere, May snow cover recedes too rapidly from Alaska and northern Quebec, Canada. In early winter (December, not shown) there is evidence that the snow line has reached southern eastern/central Europe too early, although this will have no bearing on the results presented here.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-010-0805-3/MediaObjects/382_2010_805_Fig1_HTML.gif
Fig. 1

Climatology of snow cover fraction expressed as a percentage in (top) NSIDC NH-EASE observations averaged over 1966–2005 and (bottom) the HadCM3 coupled control run from January to May. Note the differing grid sizes of observational (25 km) and model (∼300 km) data

Variance in the spring (FMA) season of interest is compared between NSIDC NH-EASE observations and the model proxy in Fig. 2. In the northern part of the Eurasian landmass, the predominant signal lies at the edge of the snow-pack, reflecting interannual variations in the rate of melting. The spatial distribution of this signal is well represented in HadCM3, with a strong centre over eastern Europe, extending eastwards into Kazakhstan, Mongolia and the Russian Pacific coast. There is a negative bias in variance in this latter region, although a positive bias exists further north in eastern Siberia. Over the HimTP region, HadCM3 understates the strong variance somewhat, or more precisely, the area of maximum variance is reduced compared to the observed dataset. The pronounced seasonality in the variance band extending across eastern Europe and central Eurasia is well captured by the coupled model. These two figures depicting the mean state evolution of the snow-pack over the Eurasian continent and its variance suggest the Hadley Centre model is well suited to this study.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-010-0805-3/MediaObjects/382_2010_805_Fig2_HTML.gif
Fig. 2

Variance of spring (FMA) snow cover fraction in (top) NSIDC NH-EASE observations over 1966–2005 and (bottom) the HadCM3 coupled control run in the northern hemisphere

2.1.2 Forcings of the snow distribution

A thorough analysis of the remote forcings affecting snow distribution over the Eurasian continent is beyond the scope of this study, however we briefly summarize some of the main influences in Fig. 3. One of the largest forcings is the North Atlantic Oscillation, which Robock et al. (2003) show to have clear strong negative correlations with snow cover over north-western Europe. We use the principal component timeseries of the leading EOF of DJF sea-level pressure over the Atlantic (20–80°N, 90°W–40°E) as a measure of the NAO, although this makes little difference to using the station-based index suggested by Hurrell (1995). Strong negative correlations are noted with FMA snow in north-west Europe (Fig. 3a), reaching as high as −0.4 and consistent with Robock et al. (2003). Robock et al. (2003) also showed weak positive correlations over Tibet in late spring, and these are also found in HadCM3.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-010-0805-3/MediaObjects/382_2010_805_Fig3_HTML.gif
Fig. 3

Correlations between HadCM3 snow amount in spring (FMA) with various climate indices over the 1050-year control run: a winter NAO index, b winter Niño-3 index, c autumn Indian Ocean dipole index (SON). Positive (negative) contours solid (dashed), interval 0.1

Authors such as Shaman and Tziperman (2005) have demonstrated relationships between ENSO and enhanced snowfall over the Tibetan Plateau via a stationary wave mechanism, leading to persistent snow depth anomalies even during boreal summer (their Fig. 1; albeit using a short satellite record). Clifford et al. (2009) have also demonstrated strong positive correlations between a DJF ENSO index and concurrent SWE over the HimTP region in a different 545-year control run of HadCM3. This correlation is reproduced in this 1050-year control run (Fig. 3b). There is some observational evidence to suggest that the Indian Ocean dipole may play an even more important role than ENSO in generating winter/spring snow anomalies over the Tibetan Plateau region (Yuan et al. 2009), although we note also that widespread SST warming in the Indian Ocean in the late 1970s has been implicated in increasing snow depth over the Tibetan Plateau on decadal timescales (Zhang et al. 2004). Spencer et al. (2005) have previously shown HadCM3 to be capable of simulating the seasonality of the Indian Ocean dipole (including the timing of its onset, peak and decay) and associated equatorial wind and thermocline behaviour. Correlations between winter/spring snow amount in HadCM3 and the IOD peak during the previous autumn (SON, using the index defined by Saji et al. 1999) also show positive results over the Tibetan Plateau (Fig. 3c), consistent with observations.

This range of evidence suggests the Hadley Centre model is capable of simulating some of the key observed forcings of snow variation over the Eurasian continent.

2.1.3 The Indian monsoon

HadCM3 offers a good representation of the seasonal cycle of monsoon rainfall (see, e.g. Turner and Slingo 2009), which is among the best in state-of-the art coupled GCMs (Annamalai et al. 2007), although the south-westerly winds of the Somali Jet are too strong (Turner et al. 2005). Although interannual variability is reasonable in this model, the teleconnection with ENSO is slightly weak and phased incorrectly, such that equatorial Pacific SSTs during spring feature the largest inverse correlations with the monsoon (Turner et al. 2005; Fig. 15) rather than during boreal summer. Indeed there is some influence of ENSO during the previous winter due to deficiencies in the spring-predictability barrier, which Peings and Douville (2010) note relates to the unusually strong influence of ENSO on climate variability in this and several other coupled models.

2.2 AGCM experimental design

To test the impact of springtime snow in different regions on the subsequent Asian summer monsoon, ensemble integrations of the land-atmosphere component of the Hadley Centre model (HadAM3) have been performed. The model is run using a seasonal cycle of climatological SST and sea-ice forcing in order to completely remove the effects of ENSO on both the monsoon itself and preceding snow cover, and any deficiencies in these links. The snow forcing conditions are described below.

2.2.1 Snow forcing initial conditions

The snow forcing is derived from spring (FMA) averaged snow depth conditions taken from the 1050-year HadCM3 coupled control run. Based on the regions of high snow depth variability during spring (Fig. 5), indices are designed to reflect variations over WNEur (30–110°E, 50–65°N) and HimTP (67.5–100°E, 27.5–40°N). To examine the influence of snow forcing that is strong yet within the realistic range of variability seen in the model, composites are generated for each of these indices separately, where snow amount exceeds ±2σ from the area mean. Outside of the index region, snow depths are set to climatological FMA values. High and low snow forcing experiments are performed for both regions:
  • HimTP ± 2σ (named HimTPpos, HimTPneg),

  • WNEur ± 2σ (WNEurpos, WNEurneg).

The forcing anomalies are shown in Fig. 4. A control ensemble of 32 members is also integrated.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-010-0805-3/MediaObjects/382_2010_805_Fig4_HTML.gif
Fig. 4

Snow forcings for HimTP and WNEur experiments based on composites of spring (FMA) snow forcing exceeding ±2σ for each region from the 1050-year HadCM3 coupled control run. Snow forcings are shown as anomalies to FMA climatological average, units are kg m−2

2.2.2 Spin-up procedure and ensemble set-up

Previous studies have used a range of methods for initialising snow forcing experiments. Some authors applied their snow forcing initial conditions during winter or spring, in combination with other (climatological) inputs, and allowed the atmosphere or land surface to freely evolve subsequently (e.g. Dash et al. 2006). This method may be disadvantageous as the response to suddenly imposed snow forcing may not be entirely realistic, and part of the signal seen in the subsequent monsoon may involve some spin-up component. Others (e.g. Douville and Royer 1996) ensure a lengthy spin-up procedure is used such that all components of the imposed snow forcing, other land surface parameters and the atmosphere are in balance before the spring to summer evolution is considered. Becker et al. (2001) took initial atmosphere and land/ocean surface conditions from observed data (in pairs of ENSO years with strong snow signals) and started ensemble integrations from early November of year 0. Their snow-forcing experiments then began in early April, using start conditions obtained from their spin-up integrations.

In these experiments, we utilise the spin-up period suggested by Becker et al. (2001). Atmospheric and land initial conditions from 1 November in an existing control integration of HadAM3 are used, and the model is integrated for six months. However, to ensure that the desired snow forcings are present from early spring so that their influence on the subsequent Asian summer monsoon can be examined, snow depth is updated to the desired forcing condition each hour during the spin-up, replacing any existing snow depth. From this initial integration, daily land-atmosphere restart data are taken from 15 March to 16 April inclusive (in the 30-day month common to many GCMs, this gives us an ensemble size of 32 members). These are re-designated as 1 April and the 32-member ensemble is integrated for a further 8 months to 1 December, during which snow is no longer constrained and is free to melt. Such an ensemble size will provide robust statistics with which to test any signal resulting from the lower boundary forcing. As explained above, climatological SST forcings are used throughout the experiment.

3 Results of coupled model analysis

The 1050-year control integration of the HadCM3 coupled model provides a large number of samples to examine snow–monsoon relationships, even if taking the effects of ENSO into account. Figure 5a shows the leading mode of variability in spring (FMA) snow depth as simulated by this control run. The domain of the area-weighted empirical orthogonal function (EOF) analysis is as indicated by the bounds of the figure. This mode explains approximately 28% of the interannual variability and is well separated from lower-order modes, which all explain less than 6% of the variation (not shown). Variability is concentrated most strongly in a dipole covering the Himalaya/Tibetan Plateau and a zonal band across much of mid-latitude Eurasia, although skewed heavily towards HimTP. This leading mode raises the question of how negative correlations to monsoon rainfall can exist from both the HimTP region and from further north in Eurasia. Such a north–south dipole is also clear in the observed snow cover record (e.g. Bamzai and Shukla 1999, particularly the April–May season, their Fig. 7). Other modelling studies have also shown a similar dipole: the leading EOF of March snow depth from ensemble integrations of the ECMWF spectral model used by Corti et al. (2000) in the PROVOST project shows some similarity to Fig. 5a, with opposite signs over the Himalayan/Tibetan Plateau and a broad zonal band across Eurasia. In a later study using the ECMWF IFS model at T63 horizontal resolution, Becker et al. (2001) used an index of west Eurasian snow depth in March to show opposite anomalies over HimTP. We note that the regions of interest depicted in this EOF are also regions with strong and realistic climatological snow cover (Fig. 1) suggesting that EOF analysis is not inappropriate in this case. In addition, the anomaly over HimTP features a south-westward protrusion at its western edge (see Figs. 4, 5a) roughly over the south-west Asia region identified by Fasullo (2004) as important for perturbing Indian monsoon rainfall. This feature is apparent despite the relatively coarse resolution of this model. If the domain used to compute the EOFs is restricted to north of 40°N as in Fig. 5b (thereby neglecting data from the HimTP region) then east–west patterns of snow co-variation across mid-latitude Eurasia begin to emerge in HadCM3. This may explain some observational results (e.g. Dash et al. 2005; Singh and Oh 2005; Zhao and Moore 2004) that note east–west dipoles in Soviet station data only, and reflect the data sparcity below 40°N in HSDSD-II. The mode in Fig. 5b explains only 8% of the variance and is indistinct from higher-order modes, again highlighting the dominance of snow variability over HimTP on the Eurasian landmass.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-010-0805-3/MediaObjects/382_2010_805_Fig5_HTML.gif
Fig. 5

Leading mode (EOF-1) of variability in spring (FMA) snow amount calculated over (a) 0–180°E, 20–80°N and (b) the same domain restricted to north of 40°N in the 1050-year HadCM3 coupled control run. Percentage variance explained by EOF-1 in each case is also shown. Red solid (blue dotted) shading indicates positive (negative). The −1 < 0 < 1 interval is blanked out for clarity. Zero contour is dashed. Contours are placed at ±1, 2, 4, 8, 16, 32, 64 kg m−2. In (a) the distribution is very strongly skewed toward the negative, i.e., signals of interannual variation are strongest over the Himalaya/Tibetan Plateau region

To compare HadCM3 briefly against other state-of-the-art coupled GCMs, we calculate leading EOFs from monthly snow depth data in all available years of pre-industrial control integrations from the World Climate Research Program (WCRP) 3rd Coupled Model Intercomparison Project (CMIP3). (This database was prepared for the IPCC AR4 and is held at the Program for Climate Model Diagnostics and Intercomparison—PCMDI.) However, for brevity, we limit the study to only four models which satisfy criteria in Annamalai et al. (2007) of (a) successful simulation of the seasonal cycle of Indian summer monsoon rainfall, like HadCM3, and (b) ability to represent the sign, strength and timing of the monsoon–ENSO teleconnection. HadCM3 failed the latter test owing to its monsoon–ENSO teleconnection being slightly weak and peaking too early in April/May (Turner et al. 2005). In Fig. 6, the gfdl_cm2_0 (500 years of data), gfdl_cm2_1 (500), mpi_echam5 (506) and mri_cgcm2_3_2a (350) models show very similar leading modes of variation in spring snow depth to HadCM3, although with minor differences over mid-latitude Eurasia. Those four models also show strong skewness toward spring snow depth variability over HimTP.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-010-0805-3/MediaObjects/382_2010_805_Fig6_HTML.gif
Fig. 6

Leading mode (EOF-1) of variability in spring (FMA) snow amount calculated over 0–180°E, 20–80°N in four IPCC coupled models from the CMIP3 database. Percentage variance explained by EOF-1 in each model is also shown. Red solid (blue dotted) shading indicates positive (negative). The −1 < 0 < 1 interval is blanked out for clarity. Zero contour is dashed. Note that the distribution is very strongly skewed toward the negative, i.e., signals of interannual variation are strongest over the Himalaya/Tibetan Plateau region

To demonstrate possible impacts of snow forcing from the Eurasian landmass on the Asian summer monsoon in the coupled version of the Hadley Centre model, we devise composite evolutions based on the previously defined snow indices. The composites are based on strong minus weak spring (FMA) snow conditions in the HimTP and WNEur regions, followed by weak minus strong rainfall conditions measured over all-India during summer (JJAS). Only those years in which snow indices fall outside of ±1σ from the mean and monsoon rainfall indices fall outside of ±0.5σ may enter the composite. In addition, we seek a simple way of removing the effects of ENSO and this is achieved by removing from the composite any year in which MAM Niño-3 SST anomalies fall outside of ±0.5σ from the mean (in observations the strongest negative monsoon-ENSO correlation occurs during July–September however in HadCM3 it is during spring, Turner et al. 2005). From these definitions we plot the evolution of various fields from spring through to summer in Fig. 7 for the HimTP and WNEur indices respectively. In both cases, heavy snow anomalies over the chosen index region co-exist with large anomalies of opposite sign over the other region, reflecting the EOF-1 pattern of Figs. 5a and 6. These dipoles exist despite any constraint on snow in the other pole. For a given index region, anomalies are most widespread and coherent over that region. Highlighting the strong variability over HimTP, Fig. 7b suggests that even when a snow index based on WNEur is used, snow anomalies are larger over HimTP. In April, the surface temperature anomalies strongly reflect the local snow forcing, with an inverted Ω pattern of warm–cold–warm anomalies in response to heavy snow over the Tibetan Plateau in Fig. 7c, or a much more zonal pattern of temperature anomalies following the coherent heavy snow anomalies over mid-latitude Eurasia (Fig. 7d). At the mid-troposphere (500 hPa), geopotential height anomalies are consistent with surface temperature perturbations. During the preceding January, the height field over the Atlantic sector (not shown) indicates some evidence for NAO+ behaviour in the HimTP composite, and NAO− behaviour in the WNEur case. These are consistent with observed correlations with European snow and Fig. 3a, although these composites are by no means perfect representations of the NAO. In the upper level flow (200 hPa), whilst showing marked differences over the Tibetan Plateau, Fig. 7e and f both show anomalous westerlies at Indian latitudes. Such anomalies lead to weakened upper-level easterlies over India and are consistent with reduced heating in the atmospheric column over India during the monsoon onset, assuming a first baroclinic mode in the vertical (Webster and Yang 1992). The HimTP index composite (Fig. 7e) shows more obvious strengthening of the jet following heavy snow, particularly over the Arabian peninsula, whereas in the WNEur case (Fig. 7f) the response is somewhat more complex. Finally, both composites feature similar precipitation anomaly patterns over India during JJAS (Fig. 7g, h), together with a weakened Somali Jet in the lower troposphere and anticyclonic anomalies in the Bay of Bengal, weakening the monsoon trough.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-010-0805-3/MediaObjects/382_2010_805_Fig7_HTML.gif
Fig. 7

Anomaly composite evolutions based on heavy minus light snow forcing followed by weak minus strong All-India rainfall during ENSO-neutral conditions in the 1050-year HadCM3 coupled control integration. Left panels show composites based on Tibetan Plateau/Himalaya spring (FMA) snow index, right panels are based on spring north/west Eurasia snow index. Rows represent: a, b snow amount in FMA × (10 kg m−2); c, d surface air temperature (shaded, °C) and 500 hPa geopotential height (contoured, interval 5, negative solid, positive dashed) during April; e, f 200 hPa winds (5 m s−1 unit vector) during May; g, h precipitation (shaded, mm/day) and 850 hPa winds (2 m s−1 unit vector) during JJAS

These composites thus demonstrate that the Hadley Centre model is capable of simulating weak (strong) monsoon events following heavy (light) snow over both forcing regions, using a simple method to remove the effects of ENSO. That variance in springtime snow cover is slightly weak over HimTP in HadCM3 suggests that the response of the monsoon and other atmospheric features to snow over this region may be underestimated. Hence we now pursue experiments with the model’s land-atmosphere component, HadAM3, in order to determine the dominant region of forcing in the absence of ENSO and further elaborate on the mechanism involved.

4 HadAM3 AGCM ensemble experiments

Results from the ensemble experiments with the HadAM3 AGCM are analysed in turn, depicted as composite evolutions of differences between the strong and weak forcings outlined in Sect. 2.2.1.

4.1 HimTP results

Figure 8 shows ensemble mean differences between the HimTPpos and HimTPneg AGCM experiments for various atmospheric and surface fields as they evolve through late spring and early summer (April, May, June). Firstly considering snow amount in Fig. 8a–c, there is clear evidence for the imposed strong minus weak forcing anomaly in April. Interestingly, despite relaxation to climatological conditions outside of the HimTP region prior to 1 April, the April mean snow amount shows that an anomaly of opposite sign has been induced in the Eurasian mid-latitudes (roughly 30–70°E). This anomaly has diminished by May. The positive HimTP snow depth anomaly is still present however, and even persists into June, albeit over a smaller region. Now examining surface temperature differences in April and May, Fig. 8d, e initially shows anomalies similar to those in Fig. 7 for the HimTP index, namely strong surface cooling in a large region of South Asia surrounding the Himalaya and Tibetan Plateau, and significant warming anomalies to the northeast and northwest. In May, these more remote surface temperature signals have diminished, leaving a dominant cooling signal over South Asia. Subsequently in June, a cool anomaly remains over a small region of the Himalaya. Now considering the tropospheric circulation, Fig. 8g shows that the 500 hPa geopotential height anomalies over the Eurasian continent during April closely match the coupled model composites shown earlier. These height anomalies, including a low centred over South Asia, show a strong barotropic response to surface temperature. Following in May, significant (at the 95% level although not indicated in the figure) but weakened low anomalies remain over HimTP together with a weak significant high over Mongolia. Upper tropospheric winds in May (Fig. 8h) reveal strong cyclonic anomalies situated over HimTP, which represent weakening of the upper level Tibetan anticyclone during the monsoon onset period, indicative of reduced heating over the Tibetan Plateau. Upper level westerly anomalies at Indian latitudes also persist into June (not shown). Finally, the low-level flow depicts a weakened Somali Jet in June (Fig. 8i) which may reduce the supply of moisture for the monsoon in India. To the east, anomalous westerlies reach landfall on the Indochina peninsula, while cyclonic anomalies bring easterlies to central China.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-010-0805-3/MediaObjects/382_2010_805_Fig8_HTML.gif
Fig. 8

Ensemble mean differences in various atmospheric and surface fields between HimTPpos and HimTPneg HadAM3 ensemble experiments in April (left), May (centre), June (right). Top snow amount (kg m−2), middle surface temperature (°C), bottom April and May 500 hPa geopotential height (m) and 200 hPa winds (m s−1), June 850 hPa winds (m s−1). Stipples on surface temperature indicate significance at the 95% level

To determine the impact on the monsoon itself, Fig. 9 shows ensemble mean differences in precipitation for June to August. Over India in June, strong negative anomalies (generally 2 mm day−1, and 6 mm day−1 in places) spread over central states across the monsoon trough to the Bay of Bengal and further north to the Himalayan foothills. These anomalies are most significant over Gujarat and the northern states, and the drying helps explain the increased surface temperature over India during June (Fig. 8f). Southeast Asia sees increases of around 4 mm day−1 significant at least above the 90% level, and in northern China anomalous divergence has led to a zonal band of drying. Wet anomalies in the East China Sea relate to the cyclonic anomaly there. The other region of significant signal lies over the western equatorial Indian Ocean where strong low-level convergence leads to wetter conditions. In the subsequent monsoon months, the rainfall response over India is much reduced (and significant only at the 80% level, not shown) off Mumbai and Kolkata, consistent with observed studies which noted that local snow forcing affects mainly the early monsoon period (e.g. Fasullo 2004).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-010-0805-3/MediaObjects/382_2010_805_Fig9_HTML.gif
Fig. 9

Ensemble mean precipitation difference between HimTPpos and HimTPneg HadAM3 ensemble experiments in June, July, August. Units are mm day−1. Open (solid) stipples indicate significance at the 90% (95%) levels

Thus we have shown here that the early Indian summer monsoon in HadAM3 suffers a detrimental response to strong snow forcing over HimTP. That the HadCM3 coupled model, from which the snow forcing conditions were derived, shows too little variance in snow cover over HimTP suggests this result may be an underestimate of the effects of heavy HimTP snow forcing on Indian monsoon rainfall.

4.2 WNEur results

We describe similar analyses from the WNEurpos and WNEurneg experiments as indicated in Fig. 10. The initial analysis suggests an analogous response to the HimTP experiments. Strong snow depth anomalies spread in a zonal band from Scandinavia to around 105°E, together with an induced response of negative anomalies over HimTP in April. The Eurasian mid-latitude snow anomaly melts from the south but persists into May, however by June, the HimTP negative anomaly still exists (the dominant contribution to this is the heavy HimTP snow anomaly in the WNEurneg experiment). Thus through to the monsoon onset the strongest snow forcing appears likely to come from the HimTP region, even in the WNEur experiments. In surface temperature, the April response is rather complex, with a zonal band of cooling over mid-latitude Eurasia, together with significant warm anomalies to the south and west. By May, significant anomalies remain only at Indian longitudes, with cooling over Eurasia but significant warming centred over the Tibetan Plateau. As in the HimTP experiments, mid-tropospheric geopotential heights (Fig. 10g, h) follow a barotropic response to surface temperature anomalies, yielding a significant high north of India in May. This is associated with strengthening of the Tibetan anticyclone during the monsoon onset. Winds at 850 hPa display a strengthened Somali Jet during May (not shown) and June, and evidence for cyclonic anomalies over the monsoon trough (Fig. 10i). In consequence, monsoon precipitation during June is significantly wetter over all of India, by up to 4 mm day−1 (Fig. 11).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-010-0805-3/MediaObjects/382_2010_805_Fig10_HTML.gif
Fig. 10

As Fig. 8 but showing the ensemble mean difference between WNEurpos and WNEurneg ensemble experiments in HadAM3

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-010-0805-3/MediaObjects/382_2010_805_Fig11_HTML.gif
Fig. 11

As Fig. 9 but showing ensemble mean precipitation difference between WNEurpos and WNEurneg ensemble experiments in HadAM3

HadAM3 thus responds in an unusual manner to strong spring snow forcing over the WNEur region, whereby an induced anomaly of opposite sign over HimTP persists longer than the initial forcing, and subsequently influences the Indian monsoon during early summer as outlined in Sect. 4.1. Thus even in the absence of ENSO and its influence on the zonal flow (Corti et al. 2000), snow depth EOF-1 (Fig. 5) dominates in the forced AGCM framework. The bias toward low snow climatological conditions over central and eastern Siberia, and too rapid snow-melt outlined in Putt (2008) may contribute to this, however Qu and Hall (2007) found the land surface model used here to have a mid-range snow-albedo feedback in the seasonal cycle when compared against other GCMs, suggesting that the land surface components of other current models may not necessarily perform better than this one.

It is worth noting that the pattern of June rainfall anomalies in Fig. 11, and indeed in Fig. 9 (though of opposite sign), bears a striking resemblance to the regional pattern demonstrated in observations by Fasullo (2004) (his Fig. 7b) following strong spring snow cover over southwest Asia (the southwest Himalaya and Pakistan). These consist of significant negative correlations over central to eastern north India and positive correlations over the southern tip of peninsular India and the detached northeastern states. Even these latter features are represented in the HadAM3 composites here.

4.3 Sensitivity tests and mechanism

In order to further examine the effects of the dominant HimTP snow forcing region on the Asian summer monsoon, and provide more details of the mechanism involved, more extreme sensitivity tests are carried out. Instead of the coupled model-derived snow forcings used earlier, which may incorporate the effects of model bias, here we use forcings of HimTP1000 (1,000 kg m−2, equivalent to 4 m snow depth) and HimTPzero (0 kg m−2). These highly idealised scenarios, while unrealistic (particularly zero snow cover over the Himalaya during current climate conditions), provide a larger response. The experiments are spun-up in a similar fashion to HimTPpos and HimTPneg outlined in Sect. 2.2.2. The response of the atmosphere through spring to early summer (not shown) is in qualitative agreement with the earlier Fig. 8 although larger in amplitude. Here we show the ensemble-mean difference composite of Asian summer monsoon precipitation in Fig. 12. The figure shows a large and significant response over India during June, with precipitation reduced by 6 mm day−1 compared to the HimTPzero experiment. Notably, the detrimental effect on precipitation spreads well north into the Himalaya. Significant positive rainfall anomalies exist in the north and west Indian Ocean. In these experiments, signals are stronger over East Asia when compared with HimTPpos minus HimTPneg in Fig. 9. In response to observed decadal-timescale increases in spring snow depth over the Tibetan Plateau, Zhang et al. (2004) have noted strong increases in summer (JJA) precipitation over the Yangtze valley in central China, and reduced rainfall to the north and south. In Fig. 12 however the consequence of strong HimTP snow forcing is a zonal band of drying spreading east from northern China, and a wet band emanating from southern China which relates to a strong cyclonic anomaly north of the Philippine Sea (not shown), consistent with Wu and Kirtman (2007). In July the dry anomaly retreats to the far north of India while the zonal bands over east Asia move northward slightly. By August the main significant dry anomalies are in the South China and Philippine Seas. Discrepancies may arise with the result of Zhang et al. (2004) owing to a difference in forcing. Those authors found the primary forcing to be concurrent decadal-timescale warming over the Indian Ocean and attributed snow increases over the Tibetan Plateau as a response.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-010-0805-3/MediaObjects/382_2010_805_Fig12_HTML.gif
Fig. 12

As Fig. 9 but showing the precipitation difference between HimTP1000 and HimTPzero ensemble experiments in HadAM3

The results here suggest that even when strong highly idealized initial conditions are used, the snow-forced perturbations to the Indian summer monsoon occur only during the early season. Effects on the East Asian and Western North Pacific summer monsoons are potentially longer lasting and thus of greater significance to society.

To analyse the mechanism of strong Himalayan snow forcing on the Asian monsoon and to test the Blanford hypothesis we now consider the evolution of various surface fluxes. Figure 13 shows the ensemble mean evolution and a measure of the ensemble spread of daily mean snow depth, surface net radiation, land snow-melt heat flux and surface sensible heat flux in all HimTP experiments and the control ensemble in HadAM3. These are averaged over the HimTP region (67.5–100°E, 27.5–40°N). The standard (Fig. 13a) and sensitivity tests (Fig. 13b) are shown separately for clarity. A Student’s t test has been used to examine statistical significance although this is not shown in the figure; mentions of significance refer to the 95% level. In the control experiment, snow averaged over the HimTP region melts by around 1 June, and by late June in the HimTPpos experiment (Fig. 13a). In HimTP1000 however, full melting does not occur until late July. The effects of delayed snow melt are clearly seen in the reduced surface net radiation (significant until May in HimTPpos and until late June in HimTP1000). The surface net radiation is a function of increased reflected shortwave radiation but is offset by a smaller reduction in longwave radiated up from the surface (not shown). The MOSES land surface scheme diagnoses the heat flux used in snow melt at each time step and it is clearly evident that the heavy snow experiments require more energy to melt the heavy snow, significantly exceeding control values until July in HimTPpos and until August in HimTP1000. In both reduced HimTP snow experiments, significant reductions in snow-melt heat flux persist until mid-June. Finally, the impact on sensible heat emanating from HimTP is substantial, with increases during the weak snow forcing experiments (significant until at least July), and a very strong decrease in upward sensible heating in HimTP1000 (Fig. 13c). This strongly supports the Blanford hypothesis. The comparison between HimTPpos and control (Fig. 13a) shows little difference in sensible heating however (periods of significant difference occur only during late-April/May and May/June) and in this case reduced upward longwave radiation from the surface is acting to cool the atmosphere (not shown).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-010-0805-3/MediaObjects/382_2010_805_Fig13_HTML.gif
Fig. 13

Evolution of ensemble mean daily snow amount (kg m−2), net surface radiation (W m−2, downward positive), land snow-melt heat flux (W m−2) and sensible heat flux (W m−2) in all HimTP experiments in HadAM3, averaged over the HimTP region (67.5–100°E, 27.5–40°N). Experiments are plotted separately for clarity: a shows the standard experiments HimTPpos and HimTPneg, b shows sensitivity tests HimTP1000, HimTPzero and HimTP1000sfa. See later in this section for a description of HimTP1000sfa. The ensemble spread (standard deviation of the 32 members) is shown for each curve by the shaded bands

The effect of reduced upward sensible heat and longwave fluxes is confirmed by considering the impact on the evolution of temperatures through the troposphere at Asian longitudes, shown as differences for both pairs of HimTP experiments and the WNEur pair in Fig. 14. Substantial cooling occurs over 600–200 hPa at 15–40°N during most of April in the standard HimTP experiments, of up to 2.5°C initially. Such differences are larger (up to 4°C) when comparing the HimTP sensitivity tests, and they last much later into the summer, to late August. In the WNEurpos minus WNEurneg composite, tropospheric temperatures show clear warming, consistent with the induced low snow forcing over HimTP in these experiments and the Blanford hypothesis.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-010-0805-3/MediaObjects/382_2010_805_Fig14_HTML.gif
Fig. 14

Ensemble mean difference in tropospheric temperature (600–200 hPa) at Asian longitudes (40–100°E): (left) HimTPpos minus HimTPneg (middle) HimTP1000 minus HimTPzero, (right) WNEurpos minus WNEurneg ensemble experiments in HadAM3. Positive (negative) contours dashed (solid). Zero contour is dotted. Units are °C

Perturbations to the large-scale meridional temperature contrast through the deep troposphere are essential for onset and maintenance of the Asian summer monsoon (e.g. Li and Yanai 1996). The reversal of this gradient in particular is associated with the monsoon onset. Xavier et al. (2007) devised a useful metric of this reversal, defined as the evolution of vertically integrated tropospheric (600–200 hPa) temperature difference between northern and southern regions at Asian longitudes (40–100°E), termed ΔTT. The northern and southern regions are at 5–35°N and 15°S–5°N respectively. For clarity, we show curves for the HimTP sensitivity tests and their ensemble spread only in Fig. 15. This shows clear evidence for a weakened north–south tropospheric temperature gradient in the HimTP1000 ensemble. In addition, Xavier et al. (2007) define the onset date at the point which the tropospheric temperature gradient changes sign, and the length of the rainy season (LRS) as the interval between the two sign changes (we note that the monsoon onset in HadAM3 has been shown to occur around 1 week early, Martin et al. 2000). Figure 15 gives clear indications that the onset is significantly delayed by around two weeks in HimTP1000 compared to HimTPzero, with clear implications for monsoon rainfall as seen earlier in Fig. 12. The impact of snow forcing is felt until around early August when the area-averaged depth curves begin to converge and statistical significance becomes less clear. When comparing the effects of the much weaker HimTPpos snow forcing against HimTPzero (not shown), the effects at the onset time are much less pronounced (and significant only at the 80% level) while greater significance is found throughout May. However the HimTPpos result may be understated due to the weak bias in snow variance over HimTP outlined earlier and shown in Fig. 2. The difference in ΔTT between the two HimTP sensitivity tests during the early monsoon is slightly larger than the El Niño–La Niña composite difference shown in Xavier et al. (2007). However, while snow forcing from the preceding winter/spring clearly decays by boreal summer, the impact of ENSO on the Indian monsoon is felt strongly throughout ENSO growth (ENSO events peak in boreal winter following the monsoon season). Thus the impact of ENSO forcing on ΔTT in Xavier et al. (2007) is noted even during the monsoon withdrawal. Additionally, once the monsoon onset has passed, latent heat release from convection is thought to maintain the monsoon (Sperber et al. 2000), making any impact of snow forcing on surface sensible heating redundant beyond the onset period.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-010-0805-3/MediaObjects/382_2010_805_Fig15_HTML.gif
Fig. 15

Ensemble mean evolution of the tropospheric (600–200 hPa) temperature gradient in HimTP1000 and HimTPzero experiments in HadAM3 after Xavier et al. (2007). Difference computed over northern (5–35°N) minus southern (15°S–5°N) regions at 40–100°E. The ensemble spread (standard deviation of the 32 members) is shown for each curve by the light blue (HimTP1000) and pink (HimTPzero) bands. Dates when the ensemble mean difference between the two curves are significantly different at the 95% level using a Student’s t test are marked with thick red lines on the time axis

In order to test the Blanford mechanism relating HimTP snow and reduced heating above it in more detail, we repeat the HimTP1000 sensitivity experiment by replacing the albedo value for deep snow with snow-free albedo over the HimTP forcing region only. HadAM3 snow-free albedo (sfa) is dependent on the vegetation and soil types present in the grid box but is time-invariant. In HimTP1000sfa, the average albedo over the HimTP region is much reduced at 0.20 compared with 0.67 in HimTP1000. Thus the same snow mass is present at 1 April at the commencement of the ensemble integration, but with much lower albedo. The HimTP1000sfa curves in Fig. 13 reveal much more rapid snow-melt, and additional heat flux melting the snow initially. Strongly increased surface net radiation relative to the other experiments is a consequence of much reduced reflected shortwave radiation, and sensible heating anomalies over HimTP reach only half as much as in HimTP1000. These findings confirm that albedo plays a significant role in the impact of HimTP snow forcing on the Asian summer monsoon. However, slight tropospheric cooling still occurs in HimTP1000sfa (but around 1°C relative to HimTPzero, or 3.5°C warmer than HimTP1000, and lasting only until mid-June, not shown) suggesting that reduced upward longwave radiation caused by the insulating effect of snow also plays a minor role. Weak land-atmosphere coupling in HadAM3 (between soil-moisture and any atmospheric response as in Lawrence and Slingo 2005) suggests that soil moisture-evaporation feedbacks may not play such an important role in this model.

5 Discussion

In this study we have examined the effects of spring snow forcing on the Asian summer monsoon in the Met Office Hadley Centre model. In an extended control integration of the coupled version of the model (HadCM3) we have shown the strongest mode of variability in spring snow amount to be a north-south mode with opposite signs over west/north mid-latitude Eurasia (WNEur) and the Himalaya/Tibetan Plateau (HimTP; although dominated by this latter region), similar to other coupled models which simulate the monsoon and ENSO well. Composite analysis has shown that the HadCM3 coupled model can simulate weak monsoon conditions following strong spring snow forcing over either the WNEur or HimTP regions in the absence of ENSO, in common with observational studies in the literature. In order to test the dominant forcing region on the monsoon in the Hadley Centre model, the HadAM3 land-atmosphere component forced by climatological SST and sea-ice at the lower boundary was used to run large ensemble integrations for 8 months, avoiding the effects of ENSO. When forced by snow anomalies over the HimTP region, HadAM3 acts as suggested by the Blanford hypothesis in that increased spring snow depth leads to protracted snow melt, reducing the surface sensible heat and upward longwave fluxes over the Tibetan Plateau, and hence heating the troposphere to a lesser degree. This reduces or delays the reversed meridional temperature gradient during late spring and consequently delays the onset of the monsoon (Li and Yanai 1996), thus reducing Indian rainfall during June. In southern China a zonal band of increased precipitation is noted persisting into July, particularly under idealized forcing conditions. This is consistent with Duan and Wu (2005) who noted weakened rainfall in southern China in association with increased sensible heating over the Tibetan Plateau. While the bias towards low variance in spring snow cover over HimTP in the coupled version of the Hadley Centre model may suggest underestimated variability of Indian monsoon rainfall in response to HimTP snow in the coupled framework, the idealized prescribed snow anomalies in the atmosphere-only ensemble experiments suggest a robust response unaffected by this bias. The effect of snow cover on surface albedo over HimTP is found to be very important to the Blanford mechanism, contributing around 75% of the tropospheric cooling at Indian longitudes in sensitivity tests.

These findings have underlined the intensification of the Asian monsoon climate induced by Tibetan Plateau heating (Wu et al. 2007) and the importance of even small heating anomalies for determining the Asian monsoon flow (Liu et al. 2007).

Results are less clear when forced by snow anomalies over the WNEur region, however. Strong anomalies of opposite sign over HimTP outlast the WNEur forcing and ultimately perturb the monsoon in the same way as outlined above. The dominance of the HimTP region in this model may relate to systematic errors in the snow distribution over central-east Siberia, where the model underestimates snow depth in spring and undergoes too rapid melting (Putt 2008). This rapid melting is despite a very reasonable seasonal snow-albedo feedback (Qu and Hall 2007) and reasonable simulation of the snow edge as seen in our assessment of the snow cover. Alternatively, the distinct seasonality (south to north migration during spring) in the variance of snow cover across Eurasia shown in Fig. 2 makes WNEur rather more complex than the HimTP region, where the position of maximum variance changes little through the season. A more complex experimental design, where the idealized snow forcing is varied spatially through the spin-up period may be better in this case.

Previous studies linking remote (not HimTP) snow with the Asian monsoon have done so in the presence of ENSO (e.g. Becker et al. 2001). Such a mechanism has yet to be elucidated under neutral ENSO conditions here or in other studies. For this reason, and the possible weak bias in Siberian snow in this model, a co-ordinated multi-model study would be advantageous to solving this issue. This could compare the impact of idealized spring snow forcing in a variety of different regions in Eurasia, in order to test the impact of individual and common model biases on any teleconnection that is found. Such experiments would be run in the AGCM framework, as in this study, to avoid contamination by ENSO.

The results here have shown that heavy snow forcing over the Himalaya and Tibetan Plateau region can cause significant detriment to the early Indian summer monsoon. Assuming such results could be repeated in other state-of-the-art GCMs, these findings would potentially be of use in a revised statistical forecast model for the Indian monsoon and its onset. A potential difficulty may lie in the availability of reliable data for a large enough portion of the domain in question. The role of ENSO must also be considered carefully owing to its dominance in many models (Peings and Douville 2010). Fasullo (2004) have suggested a conceptual model in which ENSO alters the snow–monsoon relationship. The experiments here could be expanded upon, perhaps by introducing idealized ENSO forcing in the AGCM framework in conjunction with snow forcing. Alternatively, the effects of coupling with the ocean surface could be examined.

Acknowledgments

A. G. Turner was funded by the EU-ENSEMBLES project and NCAS-Climate, a NERC collaborative centre. Computing resources for running the Hadley Centre model were provided by HPCx and subsequently HECTOR. The 1050-year HadCM3 control run was obtained from the NERC British Atmospheric Data Centre (BADC). We acknowledge the modelling groups, the PCMDI and the WCRP’s Working Group on Coupled Modelling (WGCM) for their roles in making available the WCRP CMIP3 multi-model dataset. Support of this dataset is provided by the Office of Science, U.S. Department of Energy. The authors wish to thank Susanna Corti (the Editor), J. Fasullo, and a further anonymous reviewer for their constructive comments that greatly improved this manuscript. A. G. Turner wishes to thank Jon Vincent for computational support in running the ensemble experiments on HECTOR.

Copyright information

© Crown Copyright 2010