1. Introduction
Western disturbances (WDs) are synoptic-scale (or α-mesoscale) cyclonic perturbations in the subtropical westerly jet stream (Dimri and Chevuturi 2016), noted in particular for their ability to bring extreme winter precipitation and flooding to Pakistan and northern India (Mooley 1957; Rangachary and Bandyopadhyay 1987; Lang and Barros 2004; Hunt et al. 2018c), where they are responsible for a significant fraction of the annual rainfall (Yadav et al. 2012). They exist predominantly as midlatitude vortices in the mid-to-upper troposphere, propagating eastward (Mull and Desai 1947), and either originating as extratropical cyclones or developing as frontal systems over Eurasia (Dimri and Chevuturi 2014). A majority of WDs (known as “active” disturbances) also exhibit notable synoptic conditions at the surface, chiefly a fall in temperature and pressure (Dimri 2004).
WDs have been the subject of a number of modeling case studies. The first of these studies (Ramanathan and Saha 1972; Chitlangia 1976) showed that even early, simple models were capable of producing good approximations of the synoptic dynamics and movement of WDs. More recently, the focus of such studies has been on improvement of associated precipitation forecasts (Das 2005; Semwal and Giri 2007; Dimri 2012; Semwal and Dimri 2012; Patil and Kumar 2016, 2017), as well as the relative importance of data assimilation (Rakesh et al. 2009; Dasgupta et al. 2004) and sensitivities to parameterization schemes and ancillaries (e.g., orography, land surface classification) (Thomas et al. 2014; Dimri and Chevuturi 2014; Thomas et al. 2018).
These recent studies indicate that WDs can be well simulated in models; however, these are typically high-resolution, regional models with a WD or progenitor already in the initial conditions. Thus, we do not know how well represented WDs are in GCMs,1 despite it being an important question to answer if we are to eventually consider how storms in the South Asian region are affected by future climate forcings.
At the time of writing, there exists no study that explicitly tracks western disturbances in a global climate model (or group thereof); however, a number of proxies have been used to assess how well their impact on the region is simulated in GCMs. Tiwari et al. (2014) showed that the magnitude of winter precipitation over northern India across five GCMs was generally underestimated, although the distribution of extreme events was well captured. Their experiments, however, were seasonal forecasts with a 1-month lead time, and thus not strictly free-running. Conversely, Palazzi et al. (2015) demonstrated that CMIP5 models exhibit a substantial positive winter precipitation bias over the Hindu Kush Himalaya region. Ridley et al. (2013) used a weather-regime based argument to suggest that the patterns of surface pressure associated with WDs are well simulated in a regional climate model. Objective feature tracking techniques have been used successfully in climate models for tropical cyclones (Camargo 2013), extratropical cyclones (Zappa et al. 2013a,b), anticyclones (Purich et al. 2014), and tropical depressions (Serra and Geil 2017; Sandeep et al. 2018). Western disturbances have a similar length scale to such systems and should, therefore, be as easily tracked—as indeed they have been in reanalysis data (Cannon et al. 2016; Hunt et al. 2018b).
Models from phase 5 of the Coupled Model Intercomparison Project (CMIP5) are known to exhibit significant biases in summer precipitation over the Indian subcontinent (e.g., Levine et al. 2013; Sperber et al. 2013; Meher et al. 2017; Akhter et al. 2017), and the few studies on winter precipitation in the same context suggest that there are generally positive biases—particularly, as we have seen, over the Himalayan foothills (Palazzi et al. 2015). Given that there is some bias over wintertime northern India, and assuming that western disturbances bring most of the seasonal rainfall to the area, can the differences in their simulated and observed behavior account for this?
In this study we will adapt the tracking algorithm of Hunt et al. (2018b) to assess the behavior of WDs in the CMIP5 models, how and why these differ from observed features, and what the sources of intermodel variability are. This will be the first comprehensive tracking and assessment of WDs in CMIP5 global climate models.
In section 2 we outline the tracking algorithm and data sources used in this study; in section 3 we compare spatial and temporal distributions, both among CMIP5 models and against reanalysis; in section 4a we explore the causes of intermodel variability in WD frequency; in section 4b we examine what causes the differences between simulated WDs and observed ones; in section 5 we investigate the relationship between simulated WDs and precipitation; and, finally, we conclude in section 6.
2. Methods and data
a. Global climate models
For this study, all 23 freely accessible CMIP5 models (Taylor et al. 2012) for which 6-hourly wind data were available were used. Temperature data were not required, as assumptions about the thermal structure of western disturbances are not made prior to tracking. Where possible, the r1i1p12 ensemble member was chosen as the representative of each model. The exception was EC-EARTH, for which, due to data availability reasons, the member r9i1p1 was used. Although most models have historical runs extending from 1850 to 2005, we have chosen to use only the period for which all models have available data, namely 1950–2005. The historical experiments of all models used here are forced with observed natural and anthropogenic contributions.
b. Reanalysis data
The European Centre for Medium-Range Weather Forecasts (ECMWF) interim reanalysis (ERA-Interim, herein ERA-I; Dee et al. 2011) outputs data on 6-hourly time steps, over 37 pressure levels (cf. 60 model levels), of which 27 are between 1000 and 100 hPa. It has a spatial resolution of T255, corresponding to approximately 80 km at the equator, and spans from 1979 to the present day. Data from ships, buoys, satellites, and sondes are assimilated. A full catalogue of WD tracks in ERA-I has already been produced (Hunt et al. 2018b) and is freely available online.3 In this study, we will compare ERA-I WD tracks with CMIP5 WD tracks, although we must recompute the former so that the methodology is consistent (see section 2d). To do so, we use vorticity data for the whole output period (1979–2017) at 6-hourly intervals.
c. Observational data
The Asian Precipitation–Highly Resolved Observational Data Integration toward Evaluation of Water Sources (APHRODITE; Yatagai et al. 2009, 2012) is a gridded, gauge-based precipitation product, available at daily time steps and a resolution of 0.25°, covering the period 1951–2007. In terms of continuity and gauge density, it is one of the best precipitation products available over South Asia (Prakash et al. 2015) and performs well against satellite-based products (Guo et al. 2015). We will use this as our “observed” precipitation when comparing real to modeled WD rainfall attribution and when looking at winter precipitation biases in the region.
d. Tracking WDs in CMIP5 data
Objective feature-based (i.e., parcel-following) tracking of western disturbances has recently been performed on reanalysis data with a view to exploring the structure and variability (Hunt et al. 2018b). We use that algorithm here, with a few modifications. For the reader’s convenience, the entirety of the algorithm, including the necessary changes for use with CMIP5 output, is given below.
Compute the 6-hourly relative vorticity at 500 hPa. In previous work with reanalysis data, the 450–300-hPa mean vorticity was used; however, CMIP5 6-hourly output is available only (for this part of the troposphere) at 500 and 250 hPa. Sensitivity tests carried out with reanalysis data indicated that 500 hPa was a suitable replacement, the cost being a slightly shorter average track. Since we are not particularly interested in the genesis/lysis regions, this is an acceptable compromise.
Truncate the vorticity field at a spectral resolution of T63 (~200 km at the equator). We shall call this quantity ξ. There are several advantages to this step: first, the orography of the region generates some noise in the midtropospheric vorticity field through gravity wave production, but spectral truncation filters this out due to its comparatively small spatial scale; second, this resolution is coarser than most CMIP5 models and thus quickly eradicates effects caused by improved resolving power. We state the caveat that resolved physical processes in the model, as well as the upstream Zagros Mountains, are still affected by resolution and as a result there will be components of vorticity tendency sensitive to the underlying grid scale that will not be mitigated by this preprocessing.
Locate all local maxima in ξ subject to some radius δ, such that a point is considered a local maximum if no points with a distance δ have a greater value of ξ. We shall call this set of local maxima
For each
(i) To group the candidate points into tracks: for each
(ii) The efficacy of this step can be increased by introducing the concept of a background velocity, important when considering the high-frequency, high-velocity nature of WDs. Here this is done by biasing the search radius using the contemporaneous wind field; for example, in a wind field
We also hold the tracks in memory for one time point, looking for a candidate in time point
These resulting tracks are then filtered three times. First, “stubs” of length shorter than two days are rejected. Second, tracks that do not pass through Pakistan or northern India, defined as 20°–36.5°N, 60°–80°E (see Fig. 1), are rejected as not of interest to this study. Third, tracks whose geneses are east of their lyses and thus do not propagate eastward are rejected. Finally, disturbances with a genesis east of 60°E are filtered out; this serves to remove contamination from midtropospheric cyclones that spin up during the summer monsoon as well as expunging any vortices that might arise from Hindu Kush and Himalaya lee cyclogenesis.
The values of δ and
WD tracks for two of the CMIP5 models used in this study. CMCC-CM has the longest mean track (7058 km), and FGOALS-g2 the shortest (5815 km). Blue contours mark track point densities of 0.1, 0.5, 1, and 2 yr−1 (100 km)−2, respectively. The domain through which all tracks must pass is marked by the red box.
Citation: Journal of Climate 32, 7; 10.1175/JCLI-D-18-0420.1
For illustration, the tracks of the two models with the shortest and longest mean track lengths (FGOALS-g2 and CMCC-CM, respectively) are given in Fig. 1.
3. Frequency and spatial distribution
a. Frequency
We start by comparing the WD frequencies of each model for both winter [December–March (DJFM)] and summer [June–September (JJAS)] seasons.4 Frequency is computed on a daily basis—from which coarser statistics can be derived—by counting the number of tracks intercepting the Pakistan–northern India domain (given in red in Fig. 1).
These are given as violin plots in Fig. 2, with results from ERA-I for comparison. In winter, the CMIP5 multimodel mean of 30.4 per season is 0.64 standard deviations above the ERA-I frequency of 26.9 per season; in summer, the CMIP5 multimodel mean (MMM)5 of 4.8 per season is 1.17 standard deviations above the ERA-I frequency of 2.9 per season. Comparing the shapes of the distributions by looking at higher-order moments, the ERA-I variances are not significantly different from the distribution of individual model variances for either season. However, the skewness does significantly differ in both cases: for winter and summer respectively in ERA-I, the skewnesses are 0.66 and 0.42, compared with MMM values of 0.05 and 0.35. In other words, the models tend to underestimate the relative thickness of the right tail of the seasonal frequency distributions. For both seasons, the mean intramodel variance6 and intermodel variance were not statistically separable, and we thus cannot make any confident claims on how consistently WDs are represented across the CMIP5 models; only that when compared to reanalysis, the models do generally well except for a slight, though not universal, overestimation of frequency (three models underestimate summer frequency and five underestimate winter) and an inability to capture the asymmetries of the seasonal histograms.
Violin plots denoting the estimated density functions for annual frequencies of western disturbances in each of the 23 CMIP5 historical experiments used in this study, separated by season: DJFM in blue, and JJAS in red. Bounds indicate the extrema, with the central ticks indicating the means. The equivalent functions for ERA-I are given at the top.
Citation: Journal of Climate 32, 7; 10.1175/JCLI-D-18-0420.1
b. Spatial distribution
Figure 3a shows the distribution of WD track genesis points in ERA-I and the difference with the CMIP5 historical MMM; these are each computed by applying a spherical kernel density estimate to the set of all track geneses. The overestimation in WD frequency shown in Fig. 2 is again clear, but the spatial pattern is otherwise generally well represented, except for a slight underestimation of the eastward and southward extents. Extending this to all track points, in Fig. 3b we see again that although the spatial extent is generally well represented, the signal from the frequency overestimate is clear. Furthermore, there is a significant eastward extension of the tracks in CMIP5 compared with ERA-I, indicating that the GCM WDs tend to penetrate more deeply across the subcontinent and into the central Himalaya. Most of the intermodel variance in these fields is controlled by variation in frequency rather than location.
Spherical kernel density estimates [yr−1 (5° spherical cap)−1] for (a) track genesis points and (b) all track points. CMIP5 multimodel mean densities are given in colored contours and the difference (CMIP5 minus ERA-I) in line contours.
Citation: Journal of Climate 32, 7; 10.1175/JCLI-D-18-0420.1
c. Relationship with modes of internal variability
It has been shown that western disturbances tend (though not exclusively) to feed off baroclinic instabilities (Hunt et al. 2018a), hence their typically baroclinic structure (Hunt et al. 2018b). A simple way to indicate this is demonstrated in Fig. 4—we take the “baroclinic angle,” that is to say the local angle between the density and pressure isosurfaces, which can be computed directly using 
Interannual correlation coefficients between the monthly mean baroclinic angle at (a) 500 and (b) 250 hPa and the monthly western disturbance frequency, computed using ERA-I for DJFM. The reanalysis data were coarsened to match the available vertical resolution from the CMIP5 output, so that calculations based on vertical gradients were consistent. Stippling indicates where the correlation coefficient was significant at the 90% confidence level. Note that the color scales differ between here and Fig. 5.
Citation: Journal of Climate 32, 7; 10.1175/JCLI-D-18-0420.1
The value of the correlation coefficient between baroclinic vorticity tendency and WD frequency in ERA-I is given in Fig. 4, computed interannually using data from the winter months (DJFM, 1979–2016). This is computed at two levels: 500 hPa (Fig. 4a) and 250 hPa (Fig. 4a). At 500 hPa, there is a large area of significantly positive correlation, centered over the Persian Gulf; at 250 hPa this is reduced in size and intensity and is translated poleward, over Iran. This corroborates previous findings that WDs lean poleward with height, that they have a vorticity maximum near 500 hPa, and that they seem to intensify rapidly on approach to the Hindu Kush and Karakoram (Hunt et al. 2018b).
Figures 5a and 5b show the CMIP5 (historical) multimodel mean values of the interannual correlation coefficient at 500 and 250 hPa respectively. These are computed in the same way as they were for ERA-I for each model, except for different dataset lengths, with the mean of the results given here. The results are broadly the same as in Fig. 4, albeit slightly reduced in magnitude: a large area of significantly positive correlation located roughly over the northern Arabian Sea at 500 hPa, which at 250 hPa shrinks and migrates poleward (and westward). These similarities strongly imply two key results: first that the structures of WDs in CMIP5 models are close to those in reanalysis, although this cannot be demonstrated directly due to the limited vertical resolution of the output from the former; and second that the processes governing development and subsequent intensification are also well represented. Note also that at 500 hPa (Figs. 4a and 5a), the areas of highest correlation are collocated with the regions of highest track genesis density in Fig. 3a.
As in Fig. 4, but instead showing the multimodel mean of correlations computed for each CMIP5 model. Stippling indicates where at least half of the models had a correlation coefficient that was significant at the 90% confidence level. Computed for DJFM.
Citation: Journal of Climate 32, 7; 10.1175/JCLI-D-18-0420.1
4. Causes of intermodel variability
a. Resolution
Horizontal resolution has been shown to be an important control in tropical cyclone frequency and intensity in GCMs (Roberts et al. 2015), although work on CMIP3 models has suggested that this is likely due to the local vorticity tendencies rather than larger-scale progenitors (Walsh et al. 2013). The relationship of model resolution with frequency or intensity of mesoscale or synoptic-scale systems in the vicinity of the Indian subcontinent has not been explored in depth, although it has recently been shown that higher model resolution leads to increased intensity and track length of monsoon depressions in an NWP framework (Hunt and Turner 2017).
The intermodel relationship between resolution and tracked WD frequency is shown in Fig. 6. There is a clear correlation between the two: increasing resolution is associated with a significant increase in WD frequency; for example, a linear regression suggests that decreasing the effective grid length from 3° to 1° will raise the average frequency of simulated WDs from 45 to 65 per year. The correlation coefficient across all WDs with resolution (i.e., the red lines in Fig. 6) is −0.68. Figure 6a shows how this relationship changes for subsets of events whose geneses are west of 50° and 20°E respectively. The relative slope (i.e., gradient over absolute value) does not vary significantly across the three categories, and thus the null hypothesis that resolution has no bearing on upstream genesis longitudes cannot be rejected.
Mean WD frequency as a function of resolution, subset by thresholds in (a) genesis longitude and (b) peak intensity, computed using 500-hPa truncated vorticity. Effective grid length is defined as the geometric mean of the longitude and latitude spacings. Popular resolutions are provided for reference on the additional abscissa. Values for ERA-I are denoted by crosses for each case.
Citation: Journal of Climate 32, 7; 10.1175/JCLI-D-18-0420.1
In contrast, Fig. 6b shows that there are strong variations in slope between the three threshold categories, when using peak intensity.7 Here, the trend line gradients do not decrease relatively with increasing thresholds. Instead, they suggest that higher-resolution models are more likely to spawn higher intensity WDs. As previously mentioned, decreasing the effective grid length from 3° to 1° will raise the annual frequency of all WDs by about 40%, but will raise the frequency of the strongest WDs—those which reach a midtropospheric vorticity exceeding 10−4 s−1—by about 450%. This result is similar to that previously found for tropical cyclones, indicating that there are preferential increases in the most intense storms with resolution (e.g., Roberts et al. 2015).
The relationship between resolution and peak intensity can be quantified further by looking at the probability density functions in Fig. 7. The CMIP5 models, whose individual PDFs are given by the solid colored lines, show a marked shift with increasing resolution: both the mean and median increase by over 50% across the range, although, perhaps more importantly, the right tail exhibits a strong sensitivity to resolution. This suggests that higher-resolution models are capable of regularly simulating WDs with intensities 3–4 times higher than the median, which is very different from the statistics of ERA-I WDs.
Probability density functions for peak WD intensity—measured using the 500-hPa relative vorticity, as in Fig. 6b—for each model (solid) and ERA-I (dashed), colored by model resolution.
Citation: Journal of Climate 32, 7; 10.1175/JCLI-D-18-0420.1
It is clear that resolution is responsible for a great deal of the intermodel variance in WD frequency; now, we consider what mechanism might cause increased resolution to spawn more WDs. It has been shown previously that the location of the subtropical westerly jet is the strongest control on the frequency of WDs incident on India (Hunt et al. 2018b), so we start by regressing mean model boreal winter (December–March) zonal wind at 200 hPa against model resolution (measured using the aforementioned grid spacing), and this is shown in Fig. 8. Recalling that a positive coefficient would imply a strengthening of winds under a coarsening of resolution, we note therefore that in lower-resolution models the jet appears slightly weaker (though not significantly so) over central and northern India, as well as upstream. Conversely, there are significantly stronger upper-level winds north of 30°N in lower-resolution models, across almost the whole continent.
Correlation of model mean DJFM 200-hPa zonal wind with model grid spacing (as defined in Fig. 6). The climatological winter jet axis (i.e., latitude of greatest 200-hPa zonal wind) computed using ERA-I is given by the green line. Stippling indicates regions where the correlation is significant at the 90% confidence level.
Citation: Journal of Climate 32, 7; 10.1175/JCLI-D-18-0420.1
This is most readily interpreted as the jet being climatologically wider in GCMs with a low resolution. To see whether or not this is the case, some mean winter jet statistics were computed for each of the CMIP5 models, as well as ERA-I. These are given in Fig. 9, and show the relationships between model resolution and jet thickness, location, and core speed, computed over 30°–80°E; we see that at a lower resolution the jet is markedly wider, as well as being slightly stronger and positioned at a slightly lower latitude, which is in agreement with Lu et al. (2015). The correlation coefficients for thickness and speed are significantly different from zero at the 90% confidence level (although the latter is not at 99%). With a correlation coefficient of 0.4, the jet thickness explains more variability than either of the other metrics suggested, consistent with Fig. 8.
Mean DJFM subtropical westerly jet statistics for CMIP5 historical experiments and ERA-I, computed between 30° and 80°E, at 200 hPa, using zonal wind speed. (top) Jet thickness, defined as the mean meridional distance between isotachs of 30 m s−1 (red); (middle) mean latitude of the zonal wind speed centroid, i.e., the center of the jet (blue); (bottom) mean core speed, i.e., the mean of the highest wind speed at each longitude (yellow). Given for each are the trend lines and correlation coefficient for the CMIP5 values.
Citation: Journal of Climate 32, 7; 10.1175/JCLI-D-18-0420.1
b. Biases
It has been shown both here and in previous literature (Hunt et al. 2018a) that the isopycnal–isobaric angle (shortened here to baroclinic angle) is an important upstream predictor of winter WD frequency in both reanalysis and GCMs. A logical extension to this is to see if there is a bias in the baroclinic angle that can account for the general positive bias in winter WD frequency in GCMs (e.g., Fig. 2). In this section, CMIP5 and ERA-I climatologies are computed for a common time period (1979–2005) to ensure that resulting comparisons are robust. Figure 10 shows the biases in the winter (DJFM) climatologies of the baroclinic angle, given as a percentage error with respect to the reanalysis. At 500 hPa (Fig. 10a), there is a negative bias over most of Asia, except for a zonal belt between about 15° and 20°N. In that belt, the bias becomes positive, almost reaching 20% over the northern Arabian Sea and northern India. Conversely, at 250 hPa (Fig. 10b), there is a negative bias (i.e., the GCMs typically make the angle too small) over the entire region of interest, exceeding a 20% error over much of the domain, although substantially smaller upstream of the Hindu Kush.
Percentage error in climatological DJFM baroclinic angle in the CMIP5 multimodel mean, compared with ERA-I, at (a) 500 and (b) 250 hPa.
Citation: Journal of Climate 32, 7; 10.1175/JCLI-D-18-0420.1
At both levels, the areas of most positive bias (excepting some areas of the equatorial ocean at 250 hPa) are found, with similar size and shape, in the same locations as the genesis maxima of Fig. 3a and the correlation maxima of Fig. 5. This suggests that because the baroclinic angle is substantially overestimated in the GCM midtropospheres, CMIP5 models have a positive bias in winter WD frequency.
Climatological winter values of p(jet) at 250 hPa for (a) ERA-I and (b) the CMIP5 MMM. Stippling indicates where the correlation coefficient between monthly means of p(jet) and the associated monthly WD frequencies is significantly different from zero at the two-tailed 90% confidence level, and for CMIP5 where this is true in at least half of the models. In each case, p(jet) is computed on daily data before the relevant means are taken.
Citation: Journal of Climate 32, 7; 10.1175/JCLI-D-18-0420.1
On the balance of evidence presented here, we conclude that the strongest cause of the positive bias in WD frequency and intensity is a positive upstream bias in midtropospheric baroclinic vorticity tendency.
5. Precipitation
As discussed in the introduction, some substantial biases in winter precipitation are known to affect this region in CMIP5 models (Palazzi et al. 2013, 2015), particularly over the Himalayan foothills. Figure 12a shows the ratio of climatological winter rainfall in CMIP5 models (1950–2005) to observed (APHRODITE; 1951–2007). We have presented in this way, as opposed to an absolute difference, because of the large precipitation maximum along the Himalayan foothills, wherein a small fractional change of relative unimportance could dwarf much large fractional changes elsewhere.
Bias in climatological winter (DJFM) precipitation, computed as the logarithm of the ratio of CMIP5 MMM precipitation and (a) APHRODITE gridded gauge data or (b) ERA-I forecast precipitation. Data over oceans are not defined in APHRODITE and are thus not given here. The 2000-m smoothed isohypse is given in black.
Citation: Journal of Climate 32, 7; 10.1175/JCLI-D-18-0420.1
It is clear that across almost all of India and northern Pakistan, there is a marked wet bias, reaching as much as a factor of 3 in Gujarat. Even poorer is the overestimate of seasonal precipitation over the Tibetan Plateau (note that this is where the APHRODITE gauge density is lowest), which is a known issue (Su et al. 2013) and beyond the scope of this study. There is a small area of dry bias, too, in southern Pakistan and Afghanistan.
To complete this discussion, we also consider the relationship between ERA-I and CMIP5 precipitation (see Fig. 12b). We see that only some of the biases from Fig. 12a persist, notably the slight positive bias over Gujarat and the significant wet biases at the edge of the Tibetan Plateau. The cause of the former is not clear, but the latter is almost certainly due to inadequate representation of the Tibetan orography at the relatively coarse resolutions of GCMs and reanalyses. It is interesting to note that the positive precipitation bias over the center of the Tibetan Plateau in CMIP5 models does not persist in ERA-I, suggesting that the aforementioned representation problem is leading to some larger-scale dynamical bias.
To further explore the relationship between model resolution and precipitation, Fig. 13 shows the correlation coefficient between model grid spacing (as defined in Fig. 6 and associated text) and climatological DJFM precipitation. Where the correlation value is positive, increasing the model grid spacing will result in increased precipitation (and vice versa for model resolution); we expect WD-caused precipitation to fall into this category, because—as we have seen—increased model resolution leads to more populous and more intense WDs. There is a substantial tongue of negative correlation over the head of the Arabian Sea and toward the Hindu Kush, leading into a maximum in the western foothills of the Himalaya. Conversely, across the windward edge of the Himalayas, there is a band of significant positive correlation (i.e., increasing model resolution acts to reduce climatological precipitation). It is not clear what causes this, but it is possible the enhanced precipitation upstream—perhaps due to improved resolution of the Himalayan front, increased WD activity, or both—results in less precipitable moisture over the southern Tibetan Plateau. The roles played by moisture transport and larger-scale dynamics are significant in WD-generated precipitation (Cannon et al. 2016; Hunt et al. 2018c) and, although beyond the scope of this manuscript, should be the subject of future work.
Correlation coefficient between model grid size and climatological winter (DJFM) precipitation. Model grid size is defined as in Fig. 6, i.e., the geometric mean of longitudinal and latitudinal grid spacings. Stippling indicates where the correlation coefficient is significantly different from zero at the 90% confidence level. The 2000-m smoothed isohypse is given in black.
Citation: Journal of Climate 32, 7; 10.1175/JCLI-D-18-0420.1
So, to what extent can we attribute these biases to misrepresentation of western disturbances? We have seen already that these GCMs tend to overestimate WD frequency by an average of about 15% in the winter, and further that they tend to drastically overestimate the intensity, particularly in the tail. We can start to explore this relationship by using simple attribution plots.
Figure 14a shows the climatological winter (DJFM) precipitation that can be attributed to western disturbances using a naïve radius-of-influence approach with tracks from ERA-I that are described in section 2d. Simply, we take a fixed radius of influence, 800 km, derived from Fig. 11 of Hunt et al. (2018b); then precipitation occurring at a point on a day in which a WD passes within this distance is attributed to that WD. For the CMIP5 case, in areas of northwest India and Pakistan, over 60% of the winter precipitation occurs in the vicinity of a western disturbance. Of this, about 85% is provided by the strongest half of systems.
Fraction of climatological winter (DJFM) precipitation that can be explained by western disturbance activity. Computed using (a) APHRODITE daily gridded gauge data (1951–2007) and (b) CMIP5 precipitation data (1950–2005). The attribution (see text) is computed for each model before the mean is taken.
Citation: Journal of Climate 32, 7; 10.1175/JCLI-D-18-0420.1
We compute the attribution in the same way for CMIP5 models before taking the overall mean as shown in Fig. 14b. The two attribution maps demonstrate a striking similarity. Although the CMIP5 MMM is generally smoother in form, it too has a maximum of almost 70%, and indicates the strong influence of WDs on winter rainfall in Pakistan and northern India. Perhaps surprisingly, given the bias shown in Fig. 12, the attribution is well represented along the Himalayan foothills and into the Tibetan Plateau.
What we can deduce from this, therefore, is that the misrepresentation of WDs is responsible for most of the northern Indian winter wet bias in CMIP5 models. We cannot state with certainty, however, whether this is due to the intensity bias or the frequency bias, since more intense WDs are correlated with heavier associated precipitation in observations (Hunt et al. 2018b).
Figure 15a shows the partial correlation between monthly precipitation and monthly WD frequency, holding mean intensity constant. As one would intuitively expect, there is a widespread positive trend; more WDs means more rainfall. The correlation coefficient peaks over the foothills of the western Himalaya and is significant almost everywhere, until the sign of the relationship changes to negative over the Bay of Bengal and some parts of the Tibetan Plateau. This pattern is largely in agreement—as it ought to be, to a first-order approximation—with Fig. 14b.
Multimodel mean partial correlation coefficient between winter (DJFM) monthly precipitation and (a) monthly WD frequency and (b) mean monthly peak WD intensity; in each instance holding the other variable constant. Red contours on each indicate the CMIP5 MMM winter precipitation (mm day−1). Stippling indicates where more than half the models indicate a significance exceeding 95%.
Citation: Journal of Climate 32, 7; 10.1175/JCLI-D-18-0420.1
Figure 15b shows the partial correlation between monthly precipitation and monthly mean (peak) intensity, while holding WD frequency constant. Here, the relationship pattern is more striking; there is a dipole whose positive peak is situated at the southwest corner of the Tibetan Plateau and whose negative peak is spread across the Bay of Bengal and Arabian Sea near the south of the peninsula. This implies that during months when the models produce stronger WDs, there is more precipitation being generated over the Himalayas and Tibetan Plateau (and to a lesser extent, over the Hindu Kush and Karakoram). This relationship is partially corroborated by previous work, which shows that stronger western disturbances produce heavier precipitation (Hunt et al. 2018b), and that extreme precipitation events in winter in this region are strongly dependent on meridional moisture flux (Hunt et al. 2018c).
The overall positive bias appears to be related to the overestimation of WD frequency in CMIP5 GCMs, whereas the meridional gradient of that bias (i.e., that it is more positive toward the Himalayan massif) seems to be due to the general overestimation of WD intensity.
6. Conclusions and summary
An assessment of the behavior of western disturbances (WDs) over 23 CMIP5 models was carried out, with the foci of the investigation being the sources of intermodel and intramodel variability as well as multimodel mean biases against reanalysis. WDs were objectively tracked in the historical (1950–2005) runs of each model, as well as the ERA-Interim reanalysis for comparison, using 500-hPa relative vorticity and a domain filter.
Marked variability was found in the climatological WD frequencies between models. These frequencies are strongly anticorrelated with model grid spacing (i.e., a higher-resolution model tends to produce more WDs), with evidence indicating that this is due to coarser models producing a wider subtropical jet that tends to carry disturbances (which are embedded in its northern flank) too far north of India. An alternative explanation is that at lower resolutions the interaction of the jet with the Hindu Kush/Karakoram orography—which is partially responsible for the generation/spinup of WDs—is increasingly poorly represented.
On average, CMIP5 models tend to slightly overestimate the frequency of WDs. Furthermore, models with higher resolution tend to generate higher-intensity WDs. This effect is particularly pronounced in the tail of the distribution, where the highest-resolution models can occasionally create WDs with intensities far higher than those tracked in ERA-I.
Upstream baroclinic vorticity tendency (baroclinic angle) has previously been shown to be an important contribution to both the genesis and intensification of WDs (Hunt et al. 2018a). Here it has been shown that the spatial covariance of this parameter with downstream WD frequency is represented well in the multimodel statistics in both the mid- and upper troposphere. This indicates that the process by which models are spinning up western disturbances is accurate compared to reanalysis, and hence that their gross structures likely follow suit; however, we cannot probe this directly in the multimodel database, since multiple vertical levels are not provided at 6-hourly frequency.
There is, however, a large positive bias in midtropospheric baroclinic angle over much of the region where it is significantly correlated with WD frequency. Such a prominent bias is not found in other fields to which WD genesis is sensitive (e.g., proxies of jet location and strength), which indicates that this bias is likely the source of the models’ propensity to overestimate WD frequency and intensity.
The most important characteristic of western disturbances, from an impacts point of view, is the winter precipitation associated with them. Using a simple radius-threshold attribution method, it was shown that WDs bring over 70% of the climatological winter precipitation to much of northern India and Pakistan, and that the spatial attribution pattern is very similar in the CMIP5 MMM climatology. This attribution fraction is not homogeneous, however: it is over 40% across much of the Hindu Kush and Himalayan foothills, and as high as 30% even through into West Bengal. These values indicate just how important a component orographic forcing is in the context of WD precipitation.
Moreover, there exists a marked winter wet bias over South Asia in these GCMs, which could thus, in theory, be linked to the WD frequency bias. Using a novel correlation technique, it was shown that in general this bias can indeed be explained by overestimated WD frequency and that its spatial variability was more likely explained by simulated WDs having too high an intensity.
One obvious avenue for future work is the exploration of WD characteristics in future climate scenarios; it has been shown here that WDs are sufficiently well represented in CMIP5 models to allow such analysis, and there now exists a tracking framework upon which to base it. However, future studies should be aware of the shortcomings found in this work, most importantly biases in frequency and intensity.
Further work should also seek to frame the results of section 5 in the context of large-scale dynamics and thermodynamics; these, as well as synoptic-scale moisture transport, have been shown to be important contributors to the precipitation caused by western disturbances (Cannon et al. 2016; Hunt et al. 2018c). Such work should also seek to attribute precipitation on an event-by-event basis, rather than the statistical approximation we have used, which would allow much deeper analysis of the problems presented here. Furthermore, a detailed analysis of the role of orography and lee cyclogenesis is required to fully understand the impact model resolution has on circulation in this region.
Acknowledgments
KMRH, AGT, and LCS are funded by the JPI-Climate and Belmont Forum Climate Predictability and Inter-Regional Linkages Collaborative Research Action via NERC Grant NE/P006795/1. We acknowledge the World Climate Research Programme’s Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups for producing and making available their model output. For CMIP the U.S. Department of Energy’s Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. We thank two anonymous reviewers for their suggestions which have greatly improved the clarity of this manuscript.
REFERENCES
Akhter, J., L. Das, and A. Deb, 2017: CMIP5 ensemble-based spatial rainfall projection over homogeneous zones of India. Climate Dyn., 49, 1885–1916, https://doi.org/10.1007/s00382-016-3409-8.
Camargo, S. J., 2013: Global and regional aspects of tropical cyclone activity in the CMIP5 models. J. Climate, 26, 9880–9902, https://doi.org/10.1175/JCLI-D-12-00549.1.
Cannon, F., L. M. V. Carvalho, C. Jones, and J. Norris, 2016: Winter westerly disturbance dynamics and precipitation in the western Himalaya and Karakoram: A wave-tracking approach. Theor. Appl. Climatol., 125, 27–44, https://doi.org/10.1007/s00704-015-1489-8.
Chitlangia, P. R., 1976: Mean model of western depression. Indian J. Meteor. Geophys., 87, 157–162.
Das, S., 2005: Mountain weather forecasting using MM5 modelling system. Curr. Sci., 88, 899–905.
Dasgupta, M., S. Das, and R. Ashrit, 2004: MM5 3D-Var data assimilation and forecast system over Indian subcontinent: Results from recent experiments. Fifth WRF/14th MM5 Users’ Workshop, NCAR, Boulder, CO, sgi200.ust.hk/mm5/workshop/ws04/Session5/Gupta.Das.pdf.
Dee, D. P., and Coauthors, 2011: The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553–597, https://doi.org/10.1002/qj.828.
Dimri, A. P., 2004: Impact of horizontal model resolution and orography on the simulation of a western disturbance and its associated precipitation. Meteor. Appl., 11, 115–127, https://doi.org/10.1017/S1350482704001227.
Dimri, A. P., 2012: Wintertime land surface characteristics in climatic simulations over the western Himalayas. J. Earth Syst. Sci., 121, 329–344, https://doi.org/10.1007/s12040-012-0166-x.
Dimri, A. P., and A. Chevuturi, 2014: Model sensitivity analysis study for western disturbances over the Himalayas. Meteor. Atmos. Phys., 123, 155–180, https://doi.org/10.1007/s00703-013-0302-4.
Dimri, A. P., and A. Chevuturi, 2016: Western disturbances—Structure. Western Disturbances—An Indian Meteorological Perspective, Springer, 1–26.
Guo, H., S. Chen, A. Bao, J. Hu, A. S. Gebregiorgis, X. Xue, and X. Zhang, 2015: Inter-comparison of high-resolution satellite precipitation products over Central Asia. Remote Sens., 7, 7181–7211, https://doi.org/10.3390/rs70607181.
Holton, J. R., and G. J. Hakim, 2012: An Introduction to Dynamic Meteorology. 5th ed. Academic Press, 552 pp.
Hunt, K. M. R., and A. G. Turner, 2017: The effect of horizontal resolution on Indian monsoon depressions in the Met Office NWP model. Quart. J. Roy. Meteor. Soc., 143, 1756–1771, https://doi.org/10.1002/qj.3030.
Hunt, K. M. R., J. Curio, A. G. Turner, and R. Schiemann, 2018a: Subtropical westerly jet influence on occurrence of western disturbances and Tibetan Plateau vortices. Geophys. Res. Lett., 45, 8629–8636, https://doi.org/10.1029/2018GL077734.
Hunt, K. M. R., A. G. Turner, and L. C. Shaffrey, 2018b: The evolution, seasonality, and impacts of western disturbances. Quart. J. Roy. Meteor. Soc., 144, 278–290, https://doi.org/10.1002/qj.3200.
Hunt, K. M. R., A. G. Turner, and L. C. Shaffrey, 2018c: Extreme daily rainfall in Pakistan and north India: Scale interactions, mechanisms, and precursors. Mon. Wea. Rev., 146, 1005–1022, https://doi.org/10.1175/MWR-D-17-0258.1.
Lang, T. J., and A. P. Barros, 2004: Winter storms in the central Himalayas. J. Meteor. Soc. Japan, 82, 829–844, https://doi.org/10.2151/jmsj.2004.829.
Levine, R. C., A. G. Turner, D. Marathayil, and G. M. Martin, 2013: The role of northern Arabian Sea surface temperature biases in CMIP5 model simulations and future projections of Indian summer monsoon rainfall. Climate Dyn., 41, 155–172, https://doi.org/10.1007/s00382-012-1656-x.
Lu, J., G. Chen, L. R. Leung, D. A. Burrows, Q. Yang, K. Sakaguchi, and S. Hagos, 2015: Toward the dynamical convergence on the jet stream in aquaplanet AGCMs. J. Climate, 28, 6763–6782, https://doi.org/10.1175/JCLI-D-14-00761.1.
Meher, J. K., L. Das, J. Akhter, R. E. Benestad, and A. Mezghani, 2017: Performance of CMIP3 and CMIP5 GCMs to simulate observed rainfall characteristics over the western Himalayan region. J. Climate, 30, 7777–7799, https://doi.org/10.1175/JCLI-D-16-0774.1.
Mooley, D. A., 1957: The role of western disturbances in the production of weather over India during different seasons. Indian J. Meteor. Geophys., 8, 253–260.
Mull, S., and B. Desai, 1947: The origin and structure of the winter depression of Northwest India. Tech. Rep. 25, India Meteorological Department, 18 pp.
Palazzi, E., J. von Hardenberg, and A. Provenzale, 2013: Precipitation in the Hindu-Kush Karakoram Himalaya: Observations and future scenarios. J. Geophys. Res. Atmos., 118, 85–100, https://doi.org/10.1029/2012JD018697.
Palazzi, E., J. von Hardenberg, S. Terzago, and A. Provenzale, 2015: Precipitation in the Karakoram-Himalaya: A CMIP5 view. Climate Dyn., 45, 21–45, https://doi.org/10.1007/s00382-014-2341-z.
Patil, R., and P. P. Kumar, 2016: WRF model sensitivity for simulating intense western disturbances over North West India. Model. Earth Syst. Environ., 2, 82, https://doi.org/10.1007/s40808-016-0137-3.
Patil, R., and P. P. Kumar, 2017: WRF model sensitivity to initial conditions and study of some important features associated with intense western disturbances over NW India. Int. J. Eng. Tech. Sci. Res., 4 (6), www.ijetsr.com/images/short_pdf/1498748261_ieted651A_etsr.pdf.
Prakash, S., and Coauthors, 2015: Seasonal intercomparison of observational rainfall datasets over India during the southwest monsoon season. Int. J. Climatol., 35, 2326–2338, https://doi.org/10.1002/joc.4129.
Purich, A., T. Cowan, W. Cai, P. van Rensch, P. Uotila, A. Pezza, G. Boschat, and S. Perkins, 2014: Atmospheric and oceanic conditions associated with southern Australian heat waves: A CMIP5 analysis. J. Climate, 27, 7807–7829, https://doi.org/10.1175/JCLI-D-14-00098.1.
Rakesh, V., R. Singh, D. Yuliya, P. K. Pal, and P. C. Joshi, 2009: Impact of variational assimilation of MODIS thermodynamic profiles in the simulation of western disturbance. Int. J. Remote Sens., 30, 4867–4887, https://doi.org/10.1080/01431160902980332.
Ramanathan, V., and K. R. Saha, 1972: Application of a primitive equation barotropic model to predict movement of “western disturbances”. J. Appl. Meteor., 11, 268–272, https://doi.org/10.1175/1520-0450(1972)011<0268:AOAPEB>2.0.CO;2.
Rangachary, N., and B. K. Bandyopadhyay, 1987: An analysis of the synoptic weather pattern associated with extensive avalanching in Western Himalaya. Int. Assoc. Hydrol. Sci. Publ., 162, 311–316.
Ridley, J., A. Wiltshire, and C. Mathison, 2013: More frequent occurrence of westerly disturbances in Karakoram up to 2100. Sci. Total Environ., 468–469, S31–S35, https://doi.org/10.1016/j.scitotenv.2013.03.074.
Roberts, M. J., and Coauthors, 2015: Tropical cyclones in the UPSCALE ensemble of high-resolution global climate models. J. Climate, 28, 574–596, https://doi.org/10.1175/JCLI-D-14-00131.1.
Sandeep, S., R. S. Ajayamohan, W. R. Boos, T. P. Sabin, and V. Praveen, 2018: Decline and poleward shift in Indian summer monsoon synoptic activity in a warming climate. Proc. Natl. Acad. Sci. USA, 115, 2681–2686, https://doi.org/10.1073/pnas.1709031115.
Schiemann, R., D. Lüthi, and C. Schär, 2009: Seasonality and interannual variability of the westerly jet in the Tibetan Plateau region. J. Climate, 22, 2940–2957, https://doi.org/10.1175/2008JCLI2625.1.
Semwal, G., and R. K. Giri, 2007: Precipitation simulation of synoptic scale systems over western Himalayan region using Advanced Regional Prediction System (ARPS) model. Mausam, 58, 471–480, metnet.imd.gov.in/mausamdocs/15842_F.pdf.
Semwal, G., and A. P. Dimri, 2012: Impact of initial and boundary conditions on simulations of western disturbances and associated precipitation. Nat. Hazards, 64, 1405–1424, https://doi.org/10.1007/s11069-012-0304-4.
Serra, Y. L., and K. Geil, 2017: Historical and projected eastern Pacific and Intra-Americas Sea TD-wave activity in a selection of IPCC AR5 models. J. Climate, 30, 2269–2294, https://doi.org/10.1175/JCLI-D-16-0453.1.
Sperber, K. R., H. Annamalai, I.-S. Kang, A. Kitoh, A. Moise, A. G. Turner, B. Wang, and T. Zhou, 2013: The Asian summer monsoon: An intercomparison of CMIP5 vs. CMIP3 simulations of the late 20th century. Climate Dyn., 41, 2711–2744, https://doi.org/10.1007/s00382-012-1607-6.
Su, F.-G., X.-L. Duan, D.-L. Chen, Z.-C. Hao, and L. Cuo, 2013: Evaluation of the global climate models in the CMIP5 over the Tibetan Plateau. J. Climate, 26, 3187–3208, https://doi.org/10.1175/JCLI-D-12-00321.1.
Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485–498, https://doi.org/10.1175/BAMS-D-11-00094.1.
Thomas, L., S. K. Dash, and U. C. Mohanty, 2014: Influence of various land surface parameterization schemes on the simulation of western disturbances. Meteor. Appl., 21, 635–643, https://doi.org/10.1002/met.1386.
Thomas, L., S. K. Dash, U. C. Mohanty, and C. A. Babu, 2018: Features of western disturbances simulated over north India using different land-use data sets. Meteor. Appl., 25, 246–253, https://doi.org/10.1002/met.1687.
Tiwari, P. R., S. C. Kar, U. C. Mohanty, S. Kumari, P. Sinha, A. Nair, and S. Dey, 2014: Skill of precipitation prediction with GCMs over north India during winter season. Int. J. Climatol., 34, 3440–3455, https://doi.org/10.1002/joc.3921.
Walsh, K., S. Lavender, E. Scoccimarro, and H. Murakami, 2013: Resolution dependence of tropical cyclone formation in CMIP3 and finer resolution models. Climate Dyn., 40, 585–599, https://doi.org/10.1007/s00382-012-1298-z.
Yadav, R. K., K. R. Kumar, and M. Rajeevan, 2012: Characteristic features of winter precipitation and its variability over northwest India. J. Earth Syst. Sci., 121, 611–623, https://doi.org/10.1007/s12040-012-0184-8.
Yatagai, A., O. Arakawa, K. Kamiguchi, H. Kawamoto, M. I. Nodzu, and A. Hamada, 2009: A 44-year daily gridded precipitation dataset for Asia based on a dense network of rain gauges. SOLA, 5, 137–140, https://doi.org/10.2151/sola.2009-035.
Yatagai, A., K. Kamiguchi, O. Arakawa, A. Hamada, N. Yasutomi, and A. Kitoh, 2012: APHRODITE: Constructing a long-term daily gridded precipitation dataset for Asia based on a dense network of rain gauges. Bull. Amer. Meteor. Soc., 93, 1401–1415, https://doi.org/10.1175/BAMS-D-11-00122.1.
Yianilos, P. N., 1993: Data structures and algorithms for nearest neighbor search in general metric spaces. Fourth ACM-SIAM Symp. on Discrete Algorithms (SODA), Austin, TX, SIAM, 311–321.
Zappa, G., L. C. Shaffrey, and K. I. Hodges, 2013a: The ability of CMIP5 models to simulate North Atlantic extratropical cyclones. J. Climate, 26, 5379–5396, https://doi.org/10.1175/JCLI-D-12-00501.1.
Zappa, G., L. C. Shaffrey, K. I. Hodges, P. G. Sansom, and D. B. Stephenson, 2013b: A multimodel assessment of future projections of North Atlantic and European extratropical cyclones in the CMIP5 climate models. J. Climate, 26, 5846–5862, https://doi.org/10.1175/JCLI-D-12-00573.1.
By GCMs here, we explicitly mean those without the benefit of data assimilation or external forcing, such as reanalyses.
That is, first realization (r), first initialization (i), first physics setup (p).
WDs are typically at their most numerous and most intense during the winter months, whereas those that do occur during the summer months, while unusual, can be extremely devastating due to their constructive interaction with the monsoon.
To clarify, we define multimodel mean as the simple average of the first ensemble member (unless otherwise stated) for all valid models.
By this, we mean the interannual variance for a given model.
Here, we define peak intensity as the largest value of T5–T63 spectrally bandpassed 500-hPa relative vorticity achieved by a western disturbance when over Pakistan or India. This has been shown to be a good proxy for rainfall (Hunt et al. 2018b).
Conventionally this is computed at 200 hPa, but wind speeds are not available for model outputs at that level at the required temporal resolution.