More than a century ago, Blanford suggested that an inverse relationship existed between summer rainfall over Northwest India and snow cover in the western Himalaya. Recently it has been found that there is a positive correlation between Tibetan snow cover and Indian summer monsoon rainfall (IMR), a result opposite to that of Blanford. In this paper, we attempt to reconcile these contradictory observations through the analysis of spatial and temporal variability of Tibetan snow cover and its relationship with the Indian summer monsoon. We show that there exists an east-west dipole-like correlation pattern between snow cover over the Tibetan plateau and IMR that underwent a change in sign around 1985. We argue that variability in the Tibetan plateau monsoon is responsible for the spatial and temporal variability in the relationship between Tibetan snow cover and the Indian summer monsoon.
 The Tibetan Plateau is one of the most prominent topographic features on the Earth's surface covering a large area from central to western Asia. Along its southern flank are the Himalayan Mountains with the most extreme changes in elevation anywhere in the world. This plateau plays an important role in the Asian monsoon system by acting as a heat source during the summer and a heat sink during the winter [Ye, 1981]. The Asian monsoon consists of several components such as Indian monsoon, East Asian monsoon and western North Pacific monsoon [Wang and LinHo, 2002]. It has been suggested that there also exists a separate relatively shallow Tibetan plateau monsoon (TPM), best defined at 600 hPa surface, that results from the thermal contrast between the plateau and surrounding lower elevation land [Tang and Reiter, 1984; Tang et al., 1998]. Observations indicate that there exists an opposing tendency in precipitation between the northwest and southeast Tibetan plateau due to the influence of the plateau monsoon [Tang et al., 1998].
 The Indian summer monsoon is one of the most important components in Asian summer monsoon system. More than a century ago, Blanford  suggested that an inverse relationship existed between summer rainfall over Northwest India and the mean snow accumulation from January to May in the western Himalaya. In recent years, numerous studies have re-examined the relationship between Indian summer monsoon rainfall and snow cover over Eurasia including the Himalayan area [Hahn and Shukla, 1976; Bamzai and Shukla, 1999; Kripalani and Kulkarni, 1999; Liu and Yanai, 2002; Robock et al., 2003]. These studies have shown that the winter and spring snow cover over western Eurasia (eastern Eurasia) is negatively (positively) correlated to the subsequent Indian summer monsoon. In addition, some have argued that there is positive correlation between the snow cover over Tibetan plateau and IMR [Bamzai and Shukla, 1999; Robock et al., 2003], a result opposite to that of Blanford. Also, it has been found that the relationship between snow cover over the western Himalaya and IMR has recently undergone a change in sign [Kripalani et al., 2003a]. The negative correlation between snow and summer monsoon rainfall can be interpreted as being the result of two basic mechanisms: the snow-albedo feedback and snow-hydrological effects [Yasunari et al., 1991]. The first mechanism occurs when excessive snow cover, as a result of its high albedo, reduces the solar radiation absorbed at the surface and leads to a decrease in the surface temperature. The second mechanism occurs when the melting of an anomalous snow mass results in surface cooling and higher surface pressure. Through both mechanisms, anomalously low surface temperatures over Tibet result in a weakened subsequent Asian summer monsoon.
 In this paper, we will show that there exists an east-west dipole-like pattern in the correlation between snow cover over the Tibetan plateau and IMR. Around 1985, there was a sign change in the sense of the spatial dipole-like pattern over Tibetan plateau. We attribute this dipole-like pattern and its decadal variability to the existence of the plateau monsoon.
2. Data and Snow Cover Indices
 To explore the spatial and temporal variability in the relationship between snow cover over the Tibetan plateau region and IMR, we use a high-resolution snow cover dataset [Armstrong and Brodzik, 2002] derived from surface and satellite observations over the period 1972–2000. Snow cover is of course a binary field and as a result, there exists a climate signal only in regions where the field exhibits large inter-annual or inter-seasonal variability. The Tibetan plateau is near the southern limit of snow cover in the northern hemisphere and thus is a region where the field contains useful climate information. For this work, monthly mean snow cover data on a .5 deg by .5 deg was calculated from this dataset. As an index of IMR, we use the All Indian Summer Monsoon Rainfall time series [Parthasarathy et al., 1992]. Over the period from 1871–2002, the correlation coefficient between IMR and the Northwest Indian monsoon rainfall, the region investigated by Blanford, is 0.81 suggesting that a strong connection exists between the two. To explore the circulation associated with snow cover variability and with variability associated with the TPM, we use monthly mean data from the NCEP reanalysis [Kalnay et al., 1996].
 Based on previous work and our analysis of correlation coefficients (CCs) between snow cover and IMR, we produced independent snow cover indices averaged over two domains that constitute the Tibetan plateau (Figure 1): the Eastern Plateau (EP), which includes the central and eastern Tibetan plateau with a mean elevation of approximately 4000 m; the Western plateau (WP) which contains the western Tibetan plateau and West Asian mountains with elevations ranging from several hundreds of meters in its northwestern corner to about 4000 m in its southeastern region. The western Himalaya, which was the focus of Blanford , are situated in the southeastern corner of the WP domain (Figure 1). Bamzai and Shukla  used a much wider region, which was referred as the ‘Himalayan’ domain, to characterize Tibetan snow cover. In terms of our definitions, it covered almost all of both the EP and WP domains. In this paper, we exclude southern Himalaya because its extreme height decouples the snow cover in the region from that in adjoining regions of the plateau [Thompson et al., 2000; Zhao and Moore, 2002]. The CC between the two snow cover indices over the full period 1972–2000 for which we have data is very small. This indicates that the snow cover in the two domains is independent of each other. In what follows, we examine if this independence extends to their correlation with IMR.
3. A Spatial Dipole-Like Pattern in the Relationship Between Tibetan Snow Cover and IMR
Table 1 gives the CCs between IMR and the two snow cover indices identified above. Focusing on the first column, which shows the CCs over the period 1972–2000, we see that IMR is negatively correlated with the WP snow cover index which is consistent with the result of Blanford . However, IMR is positively correlated with the EP snow cover index, which is consistent with those of Bamzai and Shukla  as well as Robock et al. . This suggests that there exists a spatial east-west dipole-like pattern in the correlation between the snow cover and IMR over the Tibetan plateau.
Table 1. Correlation Coefficients Between IMR and the Mean Snow Cover Indices (January–May)a
Note that EP = Eastern Plateau, WP = Western Plateau. Bold correlation coefficients are significant at 90% level while the bold underlined ones are significant at 95% level.
 We note that the CCs between IMR and snow cover in the two regions (EP and WP) are not significant statistically over the entire period. In time series of the mean winter-spring snow cover indices in both regions, there exists a period of low snow cover around 1985 (not shown). In addition, there appears to be a reversal in the sign of the CC between each of the snow cover indices and IMR across 1985. This is confirmed in the final two columns of Table 1 where we have computed the CCs between the two snow cover indices and the IMR for the period pre and post 1985.With the exception of the EP snow cover post-1985, all the correlations are statistically significant at the 90 or 95% level. We take this reversal as evidence of the existence of a spatial dipole-like pattern in the relationship between Tibetan snow cover and IMR that exhibits decadal timescale variability.
4. Links to the Tibetan Plateau Monsoon
 We hypothesize that the spatial dipole-like pattern and its decadal variability identified above are the result of existence of the TPM. Due to the plateau's role as a seasonal heat source/sink, the circulation over the plateau tends to be anti-cyclonic during the winter and cyclonic during the summer in the planetary boundary layer [Tang et al., 1998]. We therefore produced a Tibetan plateau monsoon index (TPMI) from NCEP reanalysis by calculating the difference between mean winter (DJF) and mean summer (JJA) 600 hPa geopotential height averaged over the domain 30–35°N and 80–100°E. Similar results were also obtained at the 500 hPa level consistent with the results of Reiter and Gao  who showed that the TPM was best observed on the 600 hPa level. High values of the index indicate a strong thermal contrast between winter and summer over the plateau and vice-versa. Our TPMI, which is shown in Figure 2, indicates the presence of turning points around 1967 and 1985. The former represents a shift in the TPM from a strong to a weak regime while the latter is from a weak to a strong regime. We note that the post 1985 regime is not very strong due to a weaker summer circulation. Similar turning points were identified by Tang et al.  using the sea level pressure field. We tested the significance of both regime shifts by means of a resampling technique based on the generation of synthetic time series with the same, assumed, AR(1) statistics as the TPMI [Gershunov and Barnett, 1998]. We find that the magnitudes of transition between the first period (1948–1967) and the second period (1968–1985) and between the second and third periods (1986–2000) are both statistically significant at the 95% confidence level. It should be noted that IMR has similar decadal changes but the transition years are not exactly the same [Kripalani et al., 2003b]. This indicates that in general, a strong TPM regime corresponds to an above normal IMR regime and vice versa.
Figure 3 shows the regressions of winter (DJF) and summer (JJA) 500 hPa geopotential height and horizontal wind fields against the TPMI over the period of 1949–2000. As discussed above, the regressions indicate that a high value of the TPMI corresponds to the presence of an anti-cyclonic high during winter and a cyclonic low during summer over the plateau region. Throughout the rest of Asia, the regressions imply that the TPM is positively correlated with the East Asian and Indian summer monsoons. This provides an explanation as to why a strong TPM regime is usually associated with an above normal IMR regime.
 With regard to Tibetan snow cover, we calculated the CCs between the TPMI and the snow cover index for the sum of the EP and WP domains. Unlike what occurred with respect to the IMR, the sign of the CC was the same in both domains. It was +0.43 for the period 1972–1985 and −0.22 for 1986–2000. This suggests that the albedo-hydrological mechanism between plateau snow and summer circulation is responsible for the connection between the winter and summer plateau monsoons. Lower than normal snow cover over the plateau resulting from stronger TPM regime due to anomalous anti-cyclonic subsidence will lead to an intensified summer monsoon regime while higher than normal snow cover over the plateau resulting from weaker TPM regime due to anomalous cyclonic ascending motions will lead to a weakened summer monsoon through the feedback mechanism.
 To further investigate the effects of the TPM on the links between snow cover and IMR, we show in Figure 4 the regression of the 500 hPa winter-spring (Jan–May) mean geopotential height and wind fields against the snow cover indices. Figure 4 clearly shows that the anomalous circulation over the plateau associated with snow cover in the EP is cyclonic while that associated with snow cover in the WP is anti-cyclonic. These circulations imply that above average snow cover in either region of the plateau is associated with southerly moisture transport into that region. Therefore, in addition to the dominant albedo and hydrological feedback mechanisms that act to correlate the winter and summer plateau monsoons, the shift of the TPM regime resulted in a change in the regional circulation from one that favored moisture transport towards the eastern plateau before 1985 to one that favored transport towards the western plateau after 1985.
 In this study, we have identified an east-west spatial reversal in the correlation between Tibetan snow cover and the rainfall associated with the Indian summer monsoon. This dipole-like pattern displays evidence of decadal variability with a sign change occurred around 1985. The identification of this spatial reversal clarifies the confusion that exists between the early work of Blanford  and more recent work [Bamzai and Shukla, 1999; Robock et al., 2003]. In particular, the early work focused on the western Himalaya, which is included in what we refer to as the Western plateau (WP). In contrast, more recent work focused on what we refer to as the Eastern Plateau (EP) or a blend of both the EP and WP [Bamzai and Shukla, 1999; Robock et al., 2003].
 The relative shortness of the snow cover time series makes it difficult to make a quantitative assessment of decadal timescale variability in Tibetan snow cover. However, through the consideration of the Tibetan plateau monsoon, we find that the correlations between snow cover over the eastern and western regions of the Tibetan plateau and IMR are of opposite sign. Furthermore, there was a reversal in the sign of these correlations that occurred around 1985. We attribute these reversals to the statistically significant transition from a weak to a strong plateau monsoon. We attribute the consistency between strong winter TPM and strong summer TPM to the albedo-hydrological feedback mechanism between snow cover and summer monsoon. As we have noted, the summer TPM has weakened since 1970s indicating that it may also be modified by other factors such as changes in the Asian regional Hadley circulation (H. Zhao and G. W. K. Moore, Trends in the regional Hadley and Walker circulations as expressed in precipitation records from Asia and Africa during the latter half of the 20th century, submitted to Journal of Climate, 2003) or in the ENSO-Indian summer monsoon relationship [Kumar et al., 1999].
Kripalani and Kulkarni  also found the existence of a dipole pattern in the relationship between the Eurasian snow depth and IMR. Furthermore, Saito et al.  found that there was a significant change occurred in the mid 1980s in the relation between Eurasian snow cover and Arctic oscillation (AO). Therefore, there is a need to explore to what extent the TPM influences the global atmospheric circulation.
 This research was funded by the Natural Sciences and Engineering Research Council of Canada. The snow cover dataset was provided by the National Snow and Ice Data Center. The All Indian Summer Monsoon Rainfall time series was developed by the Indian Institute of Tropical Meteorology. The authors would like to thank Dr. Xiaodong Liu, Hendra Adiwidjaja and Xieqiong Dong for their assistance.