PRECIPITATION CHARACTERISTICS OF THE EURASIAN ARCTIC DRAINAGE SYSTEM

Transcription

1 INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 23: (2003) Published online 6 August 2003 in Wiley InterScience ( DOI: /joc.941 PRECIPITATION CHARACTERISTICS OF THE EURASIAN ARCTIC DRAINAGE SYSTEM MARK C. SERREZE* and ANDREW J. ETRINGER Cooperative Institute for Research in Environmental Sciences (CIRES), Campus Box 449, University of Colorado, Boulder, CO , USA Received 17 September 2002 Revised 4 June 2003 Accepted 4 June 2003 ABSTRACT This study examines characteristics of precipitation over the major watersheds of the Eurasian Arctic drainage system over the period In addition to the Ob, Yenisey and Lena (the three largest drainage systems), we examine the combined Kolyma Indigirka in eastern Eurasia. Each basin exhibits approximately symmetric mean annual cycles of monthly total precipitation and daily event size, with winter minima and July maxima. These are strikingly similar to the annual cycles of total column water vapour (precipitable water), which fundamentally reflects the control on saturation vapour pressure by temperature. Effective precipitation mechanisms exist in all seasons. However, because of the long distance from strong moisture sources (continentality), precipitation tends to follow the seasonality in column water vapour. An effective contrast is presented for the Iceland sector. Here, the annual cycle of precipitation is tied not to the seasonality in column water vapour, but to the stronger precipitation-generating mechanisms in winter. Hence, the annual cycles of precipitation and column water vapour in this region oppose each other. Mean winter precipitation over the Eurasian watersheds is primarily driven by a modest convergence of water vapour. Whereas precipitation peaks in summer, the mean flux convergence exhibits a general minimum (negative in the Ob). Summer precipitation is hence primarily associated with surface evaporation. A strong role of convection is supported from consideration of static stability, the fairly weak spatial organization of precipitation totals and results from prior studies. On daily time scales, the largest basin-averaged precipitation events, for both summer and winter, are allied with synoptic-scale forcing. This is seen in relationships with cyclone frequency, and patterns of 500 hpa height, vertical motion and the 700 hpa vapour flux. The relative frequency of four 500 hpa synoptic types captures the basic time series structures of precipitation. Copyright 2003 Royal Meteorological Society. KEY WORDS: precipitation; evaporation; convection; cyclones; hydro-climatology; reanalysis; Arctic 1. INTRODUCTION While current Arctic research is wide ranging, both within and across disciplines, the freshwater budget has emerged as a dominant integrating theme. The Arctic Ocean is largely land locked. Although containing only 1% of the world s ocean water, it receives the discharge from four of the world s largest rivers, the Ob, Yenisey, Lena and Mackenzie. About 11% of global river discharge flows into the Arctic Ocean (Shiklomanov et al., 2000). River discharge helps to maintain a low-salinity surface layer in the Arctic Ocean, allowing sea ice to form readily. The high albedo of the ice cover and its ability to decouple the relatively warm ocean from the cold overlying atmosphere during winter work together to reinforce the role of the Arctic region as the Northern Hemisphere heat sink. Through the temperature albedo feedback, changes in the extent of sea ice and terrestrial snow cover are expected to amplify the response of the Arctic to global climate warming. The Arctic Ocean s primary freshwater sink is represented by the export of sea ice and low salinity water through Fram Strait and into the North Atlantic. Deepwater production in the Greenland Iceland Norwegian * Correspondence to: Mark C. Serreze, Cooperative Institute for Research in Environmental Sciences (CIRES), Campus Box 449, University of Colorado, Boulder, CO , USA; serreze@kryos.colorado.edu Copyright 2003 Royal Meteorological Society

2 1268 M. C. SERREZE AND A. J. ETRINGER and Labrador seas, crucial to maintaining the thermohaline circulation, appears sensitive to modest freshening of the upper ocean (Aagaard and Carmack, 1989). Observational and modelling evidence indicate that such freshening events can be driven by changes in the Fram Strait outflow (Walsh and Chapman, 1990; Häkkinen, 1993; 1999), in which river discharge plays a potentially important, albeit largely unresolved, role. Peterson et al. (2002) document a general increase in annual Siberian river discharge since the advent of routine monitoring in the 1930s. Serreze et al. (2003) recently addressed terrestrial aspects of the Arctic hydrologic system, including variability in monthly precipitation P, evaporation E and P E, links with atmospheric variability, and relationships with river discharge. The effort used gridded fields of monthly P compiled from station records, P E calculated via aerological methods (Cullather et al., 2000), E obtained as a residual and discharge records at gauging stations closest to the river mouths. To summarize, annual precipitation over the Arctic terrestrial drainage is low to modest. For the water year, basin averages range from 533 mm in the Ob to 403 mm in the Lena. Precipitation over most of the drainage has a cold season minimum. Over parts of Eurasia and Canada, monthly totals are less than 10 mm. Most winter-half precipitation is stored as snow, which melts during spring, resulting in a strong runoff pulse. Precipitation is generally greatest in July, but because of high evaporation rates, little of the summer precipitation is available for runoff. Estimated July evaporation is somewhat higher for the Ob and Yenisey (75 mm and 70 mm respectively) in comparison with the other basins (62 mm). Variability in monthly precipitation for each watershed is positively correlated with P E and the vertically integrated vapour flux convergence. However, the mean annual cycles of precipitation and P E are in rough antiphase. For some regions, such as the Ob, summer averages of P E are negative. For large domains centred over each watershed, typically 25% of July precipitation is associated with water that evaporates and falls in the same domain. This recycling points to a significant effect of the land surface on the summer hydrologic regime. Except for the Lena, year-to-year variability in runoff is weakly related to variability in P and P E, reflecting system storages. The stronger relationships in the Lena basin reflect the continuous underlying permafrost, which simplifies the runoff process (Serreze et al., 2003). The present paper builds upon the Serreze et al. (2003) paper through a closer focus on attributes of the mean annual cycle and the characteristics of large precipitation events. We examine the Ob, Yenisey, Lena and combined drainages on the Kolyma Indigirka (K I). Our study uses daily station precipitation data for the former Soviet Union (FSU), a gridded monthly precipitation climatology and fields from atmospheric reanalysis. The four watersheds, defined from a digital river network (Vorosmarty et al., 2000), are shown in Figure 1 along with the network of precipitation monitoring stations across Eurasia (see Section 2.1) and generalized domains for synoptic analysis. To help set the stage, Figure 2 plots basin-mean daily precipitation over two years, 1980 and 1981, along with the mean annual cycles (based on long-term monthly means). As in most subsequent analyses, Figure 2 is based on grouping all stations lying within each respective basin that passed tests for missing data and available record length (described shortly). Immediately obvious are the distinct winter minima and summer maxima of precipitation in each basin. The time series illustrate pronounced high-frequency variability in basinaveraged precipitation. The large precipitation events are those that primarily determine total monthly and annual precipitation; therefore, they are especially important in understanding the time mean and variability of river discharge. The characteristics of such large precipitation events are examined through synoptic analysis and compositing approaches. 2. DATASETS 2.1. Precipitation Analysis of daily precipitation makes use of archived station time series for the FSU assembled by Groisman and Rankova (2001). The complete archive contains 2187 stations, with records for some available as far back as The most recent data are for 1998, although there is significantly less coverage after This is due to the loss of many stations after the breakup of the FSU and the difficulty of obtaining updates. We

3 EURASIAN ARCTIC PRECIPITATION CHARACTERISTICS 1269 Figure 1. Location map for Eurasia showing the location of the major watersheds examined in this study (shaded), simplified watershed domains used for synoptic analysis and the distribution of stations reporting daily precipitation. From west to east, the watersheds are the Ob, Yenisey, Lena and Kolyma Indigirka use records for the years , which provides for a long period of overlap with reanalysis fields (see Section 2.2). The daily data are instrumentally homogeneous, but not bias adjusted (Groisman and Rankova, 2001). Bias adjustments relate primarily to gauge undercatch of blowing snow. This is a significant problem in cold, windy environments (Groisman et al., 1991; Goodison et al., 1998; Yang et al., 2001). Adjustments of daily data are difficult because the information on winds during precipitation events is generally not available. This problem should be kept in mind when interpreting our results. The station density is by no means uniform (Figure 1). Coverage is best in the Ob and is fairly sparse over the high Arctic and eastern Eurasia. We use only those records with long and relatively complete records. A complete month was taken as one with fewer then three missing daily records. If more than three daily records were missing for a given month and year, all daily records for that month and year were set to missing. If, for a given month, the first check resulted in a record of less than 20 years in length, then the station was rejected. This left a total of 579 stations in the Eurasian Arctic drainage. The Ob, Yenisey, Lena and K I contain, respectively, 206, 87, 56 and 5 stations (a total of 354). The remaining stations lie in other watersheds of the drainage. Recognizing the problem of precipitation measurement biases, some analyses also make use of the Legates and Willmott (1990) long-term monthly climatology. This global product (land and ocean), provided as grid point values at a resolution, includes a best faith adjustment of gauge biases. The issue of bias adjustments has, at times, been contentious and there is little community agreement regarding appropriate techniques.

4 1270 M. C. SERREZE AND A. J. ETRINGER Figure 2. Basin mean daily precipitation for the years (solid lines) and monthly averages over the period (dotted lines) 2.2. NCEP reanalysis Circulation analyses use data from the National Centers for Environmental Prediction National Center for Atmospheric Research (hereafter NCEP) reanalysis (Kalnay et al., 1996; Kistler et al., 2001). These are available six-hourly and as daily averages from the Climate Diagnostics Center (CDC) in Boulder, Colorado. Most NCEP data are provided on a grid. The NCEP reanalysis represents more than 50 years (continually updated) of global atmospheric analyses and forecasts. The effort involves recovery and assembly of numerous atmospheric datasets that are quality controlled and assimilated with a constant ( frozen ) assimilation and forecast system. Outputs from operational systems (used for routine weather prediction) contain pseudo-climate signals (jumps) due to frequent changes in these systems. Reanalysis is intended to eliminate this problem. However, inhomogeneities can still be present due to changes in the amount and quality of assimilation data. Prior to 1958, the frequency of rawinsonde reports in the Arctic is very low. Rawinsonde coverage increased after 1958, and again in the early 1970s. Satellite data (temperature and humidity profiles) began to be incorporated in the 1970s. Starting in 1979, drifting buoy data began to provide regular reports of surface pressure over the Arctic Ocean. A general distinction can be made between variables in the NCEP archives (Kistler et al., 2001). Type A variables are strongly influenced by observations. Examples are geopotential height, u and v winds on pressure levels and sea-level pressure (SLP). Type B variables are those for which observed data directly affect the value, but where the model also has a strong influence. An example is humidity on pressure levels.

5 EURASIAN ARCTIC PRECIPITATION CHARACTERISTICS 1271 Fields of type A and B variables are generally referred to as analyses. Type C variables are those for which no observations directly affect the value (termed forecasted, or predicted variables). Examples are precipitation and radiation fluxes. Our study makes sole use of analyses (e.g. 500 hpa height) and datasets generated from analyses. The latter include total column water vapour (otherwise known as precipitable water), the vertically integrated vapour flux convergence and a record of extratropical cyclone characteristics. Recent applications of the vapour flux convergence (and with adjustment of storage changes P E) to Arctic hydrologic studies include Cullather et al. (2000), Rogers et al. (2001) and Serreze et al. (2003). The cyclone dataset was generated from an algorithm applied to six-hourly NCEP SLP fields for the Northern Hemisphere. Cyclones are identified using a series of search patterns, testing whether a grid-point SLP value is surrounded by grid point values at least 1 hpa higher than the central point being tested. Cyclones are tracked by comparing system positions on subsequent six-hourly charts using a nearest-neighbour algorithm. Cyclogenesis is taken at the first appearance of a closed (1 hpa) isobar. Earlier versions of the algorithm are described by Serreze (1995) and Serreze et al. (1997). Recent applications include the Serreze et al. (2001) examination of the relationships between the summer Arctic frontal zone and patterns of cyclogenesis and storm tracks and the McCabe et al. (2001) examination of the dataset for trends in Northern Hemisphere cyclone frequency and intensity. The algorithm is being used at CDC to generate updated maps of Northern Hemisphere cyclone tracks Basic characteristics 3. THE MEAN ANNUAL CYCLE We begin by looking more closely at the mean annual cycles of precipitation. Figure 3 shows monthly mean precipitation totals and Figure 4 provides monthly means of the daily precipitation event size (the mean of all non-zero precipitation reports). Both are based on the Groisman and Rankova (2001) dataset. The most striking aspect of these figures is the approximate symmetry of the annual cycles around July maxima and winter minima. The decline in mean daily precipitation and event size from July through October is somewhat slower than the increase from April through July. This is most evident for mean monthly precipitation in the Ob. The annual cycles depicted in the bias-adjusted Legates and Willmott (1990) climatology (not shown) are fundamentally the same, but with some differences with regard to the magnitude of the summer peaks and consistently higher cold-season precipitation. For example, January precipitation for the Ob in Figure 3 is 20 mm, compared with the Legates and Willmott (1990) adjusted value of 30 mm. Figure 3. Monthly means of total monthly precipitation

6 1272 M. C. SERREZE AND A. J. ETRINGER Figure 4. Monthly means of daily average precipitation event size (from all non-zero precipitation reports) Figure 5. Monthly means of precipitable water, integrated from the surface to the top of the atmosphere What drives this general symmetry about July? We argue that the basic answer lies with the inherent temperature control on atmospheric water vapour content. To help build this argument, Figure 5 depicts the mean annual cycles of precipitable water over each basin. Precipitable water W is the vertical integral of specific humidity from the surface to the top of the atmosphere. Figure 5, and all those that follow based on gridded fields, represent averages of grid-point values falling within each basin, weighted by the latitude dependency of the area represented by each grid point. Like the mean monthly precipitation and event size, precipitable water exhibits approximately symmetric annual cycles about July maxima, at which time the temperatures are at their highest. Note how the higher precipitation for all months in the Yenisey and Lena compared with the K I are reflected in the precipitable water. By contrast, whereas the Ob shows the highest precipitable water for all months, summer precipitation is lower than for either the Yenisey or Lena. It is instructive to examine a contrasting region. In keeping with a high-latitude focus, a good example is provided by the Iceland sector. Figure 6 displays the mean annual cycles of precipitation and precipitable water for this sector, computed as means over the area enclosing N and E. Precipitation

7 EURASIAN ARCTIC PRECIPITATION CHARACTERISTICS 1273 Figure 6. Monthly means of total precipitation and precipitable water for the Iceland region (60 65 N, E). Precipitation is from the Legates and Willmott (1990) climatology, which includes bias adjustments for gauge undercatch is based on the Legates and Willmott (1990) climatology. The mean January temperature in this region is near the freezing point. This is, of course, very high in comparison with the latitude, and manifests a largely ice-free ocean and horizontal heat transports along the North Atlantic cyclone track. Oceanic influences and extensive cloud cover keep summer temperatures fairly low (8 10 C). This seasonal temperature contrast is observed in mean precipitable water. However, in sharp contrast to the Eurasian Arctic watersheds, the annual cycles of precipitation and precipitable water oppose each other. Mean precipitable water in this area for January is about 7 mm (quite low given the temperature), compared with about 14 mm in July. However, precipitation for the same months is about 110 mm and 40 mm, respectively. The summaries in Table I help to interpret these results. Here, we show mean January and July components of the atmospheric moisture budget. The atmospheric budget can be written as W T = E P Q where W is precipitable water as defined earlier, E is evaporation, P is precipitation and Q is the vapour flux divergence, with Q representing the vertically integrated vapour flux. To provide the most direct comparison, P for each region in Table I is computed from the Legates and Willmott (1990) monthly climatology. The flux divergence is computed from vertically integrated NCEP moisture fluxes over the period at each grid cell. NCEP archives the monthly means of the vertically integrated zonal and Table I. Mean January and July moisture budget components (mm) for the Iceland sector and the Eurasian Arctic watersheds January July Q E P W/ T Q E P W/ T Iceland Ob Yenisey Lena Kolyma Indigirka

8 1274 M. C. SERREZE AND A. J. ETRINGER meridional vapour flux components based on the averages of six-hourly fluxes on sigma levels, simplifying the calculations. Evaporation is estimated as a residual from the other measured terms. It must be acknowledged that ocean values from the Legates and Willmott (1990) climatology (also used in Figure 6) are open to some question, as they are based on empirical relationships between observed precipitation totals and current weather synoptic codes at coastal and island sites, which are then applied to synoptic weather reports onboard ships. The flux convergence estimates are sensitive to mass imbalances and errors in both the wind vectors and humidity. Of particular note, NCEP has a poor formulation of the horizontal moisture diffusion, causing local moisture convergence, which is seen as blotchy patterns of the flux convergence and forecasted precipitation associated with topographic features (Cullather et al., 2000). This is mitigated by averaging the flux divergence over large areas, as done here. The residual estimate of E, of course, contains the errors in the flux divergence, P and W. These caveats must be kept in mind. It is evident from Table I that the monthly mean tendency in precipitable water for both January and July is quite small, and to a good approximation can be ignored. As such, the moisture budget equation for these months can be simplified to state that the mean precipitation must equal the sum of the vapour flux convergence (or the negative divergence) and evaporation. Caution should be exercised in the interpretation. The relationship does not imply large-scale saturation of the atmosphere, such that any additional input of water vapour will cause condensation and precipitation. The correct interpretation is that horizontal flux convergence or evaporation (or a combination of both) must be associated with uplift mechanisms inducing local saturation, precipitating the water vapour back out so as to keep the atmospheric storage constant. The flux convergence as depicted in the NCEP reanalysis relates primarily to synoptic-scale processes. A typical situation that would promote horizontal convergence and uplift (uplift of both moisture converged horizontally and evaporated from the surface) is the passage of a synoptic-scale low. An effective uplift mechanism operating at local scales, is, of course, convection. Convergence and uplift could also occur locally due to orographic effects. As evaluated in this framework, the high January precipitation in the Iceland sector is supported by roughly equal contributions from the vapour flux convergence and evaporation. The large flux convergence reflects the close proximity of strong Atlantic moisture sources, and frequent eddy activity associated with the Icelandic Low and North Atlantic storm track. The high evaporation rates relate to strong winds and vertical humidity gradients, the latter fostered by the contrast between the primarily ice-free ocean surface and fairly cold overlying atmosphere. As is evident from Table I, the fact that precipitable water is itself low over the Iceland sector during January has little bearing on the precipitation that can fall. The much lower precipitation in July is driven almost entirely by the vapour flux convergence. The flux convergence, however, is much smaller than in January, consistent with seasonal weakening of the North Atlantic cyclone track. The weak contrasts between the ocean and atmosphere and relatively slack winds keep evaporation small. Hence, although precipitable water is at its maximum in July, precipitation is fairly low, as the mechanisms to generate precipitation are comparatively weak at this time. In sharp contrast to the Iceland sector, the Eurasian watersheds are isolated from strong moisture sources. More simply, they are characterized by continentality. Turning back to Table I, winter evaporation rates over the watersheds are very small. The precipitation that falls is hence primarily related to the vapour flux convergence. But there are no nearby moisture sources to tap from. In comparison with the Iceland sector the January convergences must be fairly small, with the low resulting precipitation allied with the inherent low moisture content of the atmosphere. Note that, compared with the Yenisey and Lena, larger January convergences are found for the Ob and K I. This follows as these two basins are closest to (respectively) Atlantic and Pacific moisture sources and the primary North Atlantic and East Asian cyclone tracks. The higher precipitable water over the basins in July provides the potential for higher precipitation. Except for the Lena, where January and July values are similar, the flux convergences are even smaller in July, with net divergence in the Ob. Based on this consideration only, we would expect (except for the Lena) summer precipitation to be lower than for winter, as observed in the Iceland sector. But in contrast to the Iceland sector, the potential for higher July precipitation is realized because strong surface heating promotes much higher evaporation rates. This, in turn, points toward convection as an important uplift mechanism. Although

9 EURASIAN ARCTIC PRECIPITATION CHARACTERISTICS 1275 July evaporation is highest in the Ob, this basin does not have the highest July precipitation because of the compensating effects of the mean flux divergence Closer examination of precipitation mechanisms We expand the above analyses through fuller examination of the annual cycles. A useful start is to address seasonality in the spatial organization of precipitation totals. If synoptic-scale forcings dominate, then precipitation can be expected to be organized over large spatial scales. If convection is important, then precipitation should exhibit more local variability. Station siting can complicate this relationship. Synopticscale effects on precipitation could be modified by local orographic uplift and rain shadow effects. Conversely, although convective activity is generally seen at local scales, it is often organized on larger scales by the synoptic environment. With this understanding, we approach the problem by examining month-to-month differences in inter-station correlations of precipitation. Correlations were calculated between time series ( ) of monthly total precipitation between every pair of stations in the Eurasian Arctic drainage. The correlations were averaged with respect to station separation within bins of 50 km, up to 1000 km. For a monthly inter-station correlation to be computed, at least 20 years of shared records were required between each station pair. To have a sufficient number of cases, the Eurasian drainage is examined as a whole. Figure 7 gives results for the four mid-season months. For nearby stations (within 50 km of each other), the time series for each month correlate at better than 0.8. The correlations, of course, fall off with distance, but lower correlations are consistently found for July, the month of highest precipitation. For stations separated by 100 km, inter-station correlations for January, April and October lie between about 0.70 and 0.80, compared with 0.60 for July. For stations separated by 400 km the inter-station correlation for July is about 0.30, compared with for the other months. When the remaining months are inspected (not shown) one sees a clear transition to lower correlations from April into July, and a transition to successively higher correlations from July into October. These results point to a contribution by convective activity in summer, at least in the more southerly parts of the drainage. It should come as no surprise from Figures 8 and 9 that the summer maxima in precipitation and evaporation are allied with summer minima in mean tropospheric stability (taken as 850 hpa minus 500 hpa temperatures) and maxima in the surface air temperature, taken from temperatures at the lowest NCEP model level. We have little direct information on convective precipitation amounts. As summarized by Lydolph (1977), mean thunderstorm days over Eurasia for July range from less than two north of 70 N to about seven over the central part of the continent. Other evidence comes from our earlier study (Serreze et al., 2003). There, Figure 7. Inter-station correlations between monthly precipitation time series binned according to station separation for the four mid-season months. Results are based on all stations lying within the Eurasian Arctic drainage

10 1276 M. C. SERREZE AND A. J. ETRINGER Figure 8. Monthly means of the difference between 850 and 500 hpa temperatures, taken as an index of lower-tropospheric static stability Figure 9. Monthly means of temperature at the lowest NCEP model level (0.995 sigma) it was shown that, on the scale of these large watersheds, about 25% of July precipitation is associated with water that evaporates in each domain and precipitates back in the same domain. As summarized by Serreze and Hurst (2000), both the NCEP and European Centre for Medium Range Forecasts ERA-15 reanalyses depict convective precipitation in summer even over high Arctic latitudes. Although the NCEP forecasts in particular contain serious biases (precipitation is much too high), the model evidence lends further weight to the importance of convection. Insight regarding the relative importance of evaporation and the flux convergence in the moisture budget has already been provided in Table I. To examine the issue further, we computed by month the fraction of precipitation (from the Legates and Willmott (1990) climatology) contributed by (a) the sum of evaporation and the tendency in precipitable water and (b) the sum of the flux convergence and the tendency in precipitable water. From Figure 5, it is obvious that the tendency in precipitable water is small in winter and summer and largest in the shoulder seasons. However, at most, the tendency represents about 15% of the mean precipitation. Figures 10 and 11 show very clearly how evaporation dominates in the warm months, and how the flux convergence dominates in the cold months. Depending on the basin, evaporation dominates variously from April May through to August September. Building on Table I, from June through to August

11 EURASIAN ARCTIC PRECIPITATION CHARACTERISTICS 1277 Figure 10. Monthly means of the ratio between evaporation and the sum of precipitation and the tendency of precipitable water Figure 11. Monthly means of the ratio between the vertically integrated vapour flux convergence and the sum of precipitation and the tendency of precipitable water the evaporation exceeds precipitation in the Ob (net drying of the surface) and the flux convergence is negative. This is readily seen in the ratios just discussed. It is useful to examine the annual cycles of the flux convergence directly (Figure 12). The choice of January and July in Table I was intended to illustrate seasonal contrasts in precipitation. As is evident in Figure 12, these two months do not provide the strongest contrasts in the flux convergence. Whereas the Ob, Yenisey and K I show general cold-season maxima and warm-season minima, the Lena is characterized by a fairly distinct double minimum (July and February) and maximum (May and October). It is reasonable to expect relationships between seasonality of the vapour flux convergence and the characteristics of cyclone activity. This is examined by aggregating monthly statistics of cyclone activity over regions of 20 latitude by 40 longitude, centred roughly over each watershed (Figure 1). These regions are large enough so that a cyclone within the box boundaries could be expected to impact on precipitation within the watershed. Figure 13 shows, for each simplified region, the monthly means of cyclone days, cyclogenesis days and cyclone intensity. Results are based on the period , using the six-hourly cyclone dataset.

12 1278 M. C. SERREZE AND A. J. ETRINGER Figure 12. Monthly means of the vertically integrated vapour flux convergence Cyclone intensity is the local Laplacian of SLP at the cyclone centre. This is a better measure of intensity than cyclone central pressure, which can be influenced by the background pressure field. The values in Figure 13(c) are simply the mean intensities of all cyclones found in the region by month. Cyclone days was determined by counting the number of times a cyclone was found in each box by month over the period , and then dividing each monthly total by the total number of analysis times for that month (33 years the number of days per month four six-hourly charts per day). Cyclogenesis days are calculated similarly. The results for cyclone days have been screened to eliminate weak systems, taken as those with a cyclone intensity of less than 10 6 hpa m 2. The general conclusion that can be drawn is that the flux convergence is associated with the effects of both the cyclone frequency and the intensity of the systems themselves. The obvious characteristic of the annual cycles of strong cyclone days are the general maxima in spring and autumn, and the minima in summer and winter. The annual cycles of cyclone intensity, by contrast, exhibit more distinct summer minima and cold-season maxima. A plot like Figure 13(a) based on all cyclones regardless of intensity (not shown) reveals more of an early summer maximum and a slow decline through autumn. This relates to the seasonal pattern of within-region cyclogenesis (Figure 13(b)). Summer cyclogenesis is especially pronounced over eastern central Eurasia, seen in the June and July peaks over the Lena and K I, respectively. This is argued to be associated with enhanced baroclinicity along the Arctic frontal zone, which develops in response to differential heating between the snow-free land and cold Arctic Ocean, and vorticity production in the lee of the Verkhoyansk range (Serreze et al., 2001). It is obvious, however, that many of the summer systems having their origins in these watersheds are rather weak. One expects, however, that the relatively weak, albeit plentiful, summer systems promote local moisture convergence and uplift, masked by the use of regional means, and provide environments more favourable to convective activity Introduction 4. PRECIPITATION VARIABILITY When the issue of mid- or high-latitude precipitation variability arises, examination typically turns to synoptic-scale influences. For the Eurasian Arctic drainages, conversion of water vapour into precipitation during the cold season is primarily via the vapour flux convergence, pointing to synoptic-scale entrainment and uplift of moisture. Summer, by contrast, sees contributions by both the flux convergence and evaporation, but with strong dominance of the latter. There is ample evidence of a strong role of convection in summer.

13 EURASIAN ARCTIC PRECIPITATION CHARACTERISTICS 1279 Figure 13. Mean monthly (a) cyclone days, (b) cyclogenesis days and (c) cyclone intensity. Cyclone intensity is based on the Laplacian of SLP at the cyclone centers and has units of 10 5 hpa m 2. Cyclone days is based on systems with an intensity of at least 10 6 hpa m 2

14 1280 M. C. SERREZE AND A. J. ETRINGER However, drawing from mid-latitude observations, convection will often be enhanced/organized by the synoptic-scale environment. Hence, for either season a focus on the synoptic scale seems appropriate. The Serreze et al. (2003) study examined hydrologic variability over the Eurasian drainages on monthly time scales. They showed that time variability in monthly precipitation for the Ob, Yenisey and Lena are positively correlated with monthly P E. In general, the squared correlations exceed 0.40 and are often much higher. This is true even of summer. Variations of monthly P and P E are, in turn, linked to characteristic atmospheric circulation patterns. For the Ob, a strong influence of the Urals trough is indicated (Serreze et al., 2003). The goal here is to understand the characteristics of the largest basin-averaged daily precipitation events better, such as those shown in Figure 2. Some immediate insight into synoptic controls is provided in Table II, which contrasts winter (December February, DJF) and summer (June August, JJA) cyclone statistics based on all days within each season (over the period ) with those computed for days when the daily basinaverage precipitation was in the top 20% of the distributions. The top 20% consists of 595 cases for winter and 608 for summer. Some of the precipitation events seen in Figure 2 are included in this analysis. As before, the cyclone statistics were computed over the simplified domains shown in Figure 1. Cyclone (cyclogenesis) frequency for the high precipitation days represents the fraction of all six-hourly charts corresponding to high precipitation days for which the centre of a strong cyclone (cyclogenesis event) was present in the basin. Strong cyclones were defined earlier. A number of 0.30 for cyclone frequency (for example) means a strong cyclone was present in the basin for 30% of the days. Mean intensity is simply based on the average intensity of these cyclones. Results based on all days (i.e. with no stratification by precipitation) were computed similarly. For winter, the frequency of strong cyclones for the high precipitation cases exceeds the long-term average by roughly a factor of two. In the Ob, strong cyclones were present for over half of the high precipitation days, compared with 33% for the Lena. Much higher percentages (e.g. 83% for the Ob) are obtained when both strong and weak cyclones are considered. High precipitation days are, in turn, associated with a somewhat higher frequency of cyclogenesis. However, there is little difference in cyclone intensity. Although for summer there is still a higher frequency of cyclone and cyclogenesis activity compared with mean conditions, the contrast is not nearly as strong as for winter. The following sections examine relationships between large basin-averaged daily precipitation events and fields of the mid-tropospheric (500 hpa) horizontal flow, vertical motion at 500 hpa, moisture fluxes at 700 hpa and surface cyclone activity. Analyses are conducted separately for each basin for winter and summer through a compositing scheme applied to the same top 20% of basin-average precipitation days used in Table II The synoptic typing scheme The simplest approach to examine characteristics of the mid-tropospheric circulation is to simply average fields of 500 hpa height for the identified high precipitation days in each basin. However, such an approach could mask important differences in circulation characteristics associated with different precipitation events. In recognition, we apply the Kirchhofer classification scheme to daily 500 hpa height data over the watersheds Table II. Winter (DJF) and summer (JJA) cyclone frequency, cyclogenesis frequency and cyclone intensity (10 5 hpa m 2 ) for days when basin-average precipitation is in the top 20% of the distribution and (in parentheses) for all days Winter Ob Yenisey Lena Kolyma Indigirka Summer Ob Yenisey Lena Kolyma Indigirka Cyclone days 0.52 (0.28) 0.38 (0.23) 0.33 (0.16) 0.43 (0.21) 0.34 (0.25) 0.28 (0.20) 0.43 (0.30) 0.44 (0.27) Cyclogenesis 0.07 (0.05) 0.09 (0.06) 0.07 (0.04) 0.07 (0.04) 0.09 (0.07) 0.09 (0.08) 0.10 (0.09) 0.10 (0.08) days Intensity 14.2 (13.2) 13.2 (12.8) 13.0 (12.1) 15.7 (14.8) 9.8 (10.0) 9.8 (9.7) 11.3 (10.7) 11.1 (10.4)

15 EURASIAN ARCTIC PRECIPITATION CHARACTERISTICS 1281 for the top 20% of precipitation days. This is a map correlation approach first developed by Lund (1963) and altered by Kirchhofer (1973). Key and Crane (1986) and El-Kadi and Smithson (1992) provide useful summaries. Yarnal (1984) provide Fortran code, used here with slight modifications. The procedure is based on evaluating the similarity between pairs of gridded map fields of the selected variable. Grid point values are first normalized by subtracting the mean for the grid point and dividing by the grid-point standard deviation. The similarity between map pairs is first evaluated as S = n (Z ai Z bi ) 2 i=1 where S is the total Kirchhofer score, Z ai is the normalized value at the ith grid point on day a, andz bi is the normalized value at the ith grid point on day b. A low total score connotes overall similarity between the two map patterns. However, even with a low total score, the map pairs could be quite different in a specific sector. This is addressed through considering the similarity not only with the total map score, but also in conjunction with scores computed by row and column (zone scores). The three scores are calculated for every pair of maps. The three scores are then compared with user-selected threshold values. If the total and zone scores are smaller than the selected threshold values, then the map pairs are taken as similar. After all pairs of map patterns have been compared in this manner, the pattern day to which most other pattern days is similar is designated as the Keyday for type 1. Keyday 1 is then removed from the data set, as are the days associated with it. The pattern day with the next highest frequency of similar pattern days is designated the Keyday for type 2. Keyday 2 is then removed from the data set, along with the days associated with it. The process continues, defining the Keyday for types 3, 4, 5, etc. This proceeds until all of the remaining days have less than a minimum number of similar days required per type, as selected by the analyst. These remaining days are taken to be unclassified. This is followed by re-classification. This is necessary because, in the initial classification, days were removed that were actually more correlated with a different Keyday. A correlation is calculated between each day in the record and each Keyday. Each day becomes a member of a certain Keyday if it is more similar to that Keyday than with all the other Keydays. The Keydays are then ranked with respect to the number of members. The one with the most members becomes the new Keyday 1. The investigator decides how many Keydays to retain in the analysis. In our case, four are retained. The re-classification scheme is run again to classify all sample days with only the specified number of Keydays. A map-pattern catalogue results, in which each sample day becomes a Keyday, a member of a Keyday or an unclassified day. The Keydays are again ranked by number of members and become key map-patterns one through four. The Kirchhofer technique has a number of subjective elements. If the similarity thresholds are set too low then one gets numerous well-defined synoptic types that occur infrequently. Conversely, if the threshold is set too high then fewer synoptic type are obtained, which contain large internal variability. Altering the size of the domain, as well as the sample size, will influence the results. Yarnal (1987) stresses the need for a large sample size in the initial classification, tests with different thresholds to restrict the number of types to a manageable number while classifying the highest possible percentage of days, and performing several runs to optimize the solution. Each of the four regions analysed here span 20 latitude by 40 longitude, and include 153 NCEP grid points. These are the same simplified domains (Figure 1) used for the cyclone analysis. They are large enough to detect synoptic features in the 500 hpa flow, but small enough to be specific to the region examined. We use daily averaged NCEP fields. The similarity thresholds were determined in a trial- and -error fashion. Following Yarnal (1987), the program was run several times with different threshold values. The threshold was altered for each region and season to maximize the number of days similar to the first four key map patterns. The final threshold choice was determined as the most highly correlated threshold value in which no less than 75% of the sample was classified in the first four map patterns. The correlation coefficients range from to 0.85.

16 1282 M. C. SERREZE AND A. J. ETRINGER 4.3. Results from the synoptic analysis A composite approach is employed by averaging the 500 hpa fields for each Keyday and the days associated with it. To place the regional circulation in each basin in a larger spatial context, the composites were computed for the entire Eurasian region from E and N.Thesamedayswerealsousedtoconstruct composite mean fields of the frequency of upward vertical motion (omega) at 500 hpa and the vector vapour flux at 700 hpa. 500 hpa is near the mean level of non-divergence where vertical motion is maximized. 700 hpa tends to be close to the level that vapour fluxes are strongest. As part of this framework, from the days of high (top 20%) precipitation for each basin and season, we extracted the location of all cyclone centres (with no screening for intensity) over the same domain used in the 500 hpa composites. The cyclone centres were then binned as counts in equal-area grid cells of 250 km 250 km. Table II simply says that (especially for winter) high precipitation in a basin tends to be associated with more cyclone activity in the general vicinity of the watershed. The maps extend this analysis by providing more information on the preferred location of the systems. The winter 500 hpa height, vertical velocity and moisture flux composites for each basin appear in Figure 14, with the cyclone maps appearing in Figure 15. As with the summer maps that follow, the frequency of upward vertical motion is plotted for regions where it exceeds 60% (i.e. at least 60% of the daily means had upward motion). The moisture flux vectors are plotted at every 2.5 latitude and 5 longitude. Depending on the basin, the winter synoptic type 1 accounts for between 36 and 43% of all high precipitation days. There is strong similarity between the type 1 composite and the mean 500 hpa height fields composited over all top 20% precipitation days (not shown). The composites for the Ob (Figure 14(a)) reveal the obvious influence of a 500 hpa trough. In each composite, a high frequency of upward vertical motion is found in the region ahead of the trough, where cyclonic vorticity advection tends to increase with height. Each composite shows a relatively strong influx of moisture into the region of preferred uplift. The vapour flux patterns, however, are quite different between the composite synoptic types. Types 2 and 4 depict strong moisture fluxes from the southwest. Type 3, by contrast, shows the flux to be from the northwest. This manifests the varying longitude of the trough axes. In turn, cyclone activity associated with high precipitation days tends to be centred roughly over the simplified watershed domain (Figure 15). This is consistent with the results in Table II. For the Yenisey (Figure 14(b)), there is again a good relationship between high precipitation, upward motion and relatively strong moisture fluxes into the region of uplift. The relationship with surface cyclone activity is not as well expressed as for the Ob. The relationship between the frequency of vertical motion and 500 hpa height for this basin is also not as clear as for the Ob. For synoptic types 1 and 3, the upward vertical motion is best expressed along the descending leg of the 500 hpa ridge. Further analysis shows the regions of uplift for these types as collocated with regions exhibiting a high frequency of positive geostrophic temperature advection at 700 hpa (not shown). This is also true for types 1, 3 and 4 in the Lena basin (Figure 14(c)). For the K I (Figure 14(d)), the four synoptic types capture the influence of a strong Pacific-side trough, associated with closed height contours variously centred over Kamchatka and the Sea of Okhotsk in types 1, 3 and 4. The 500 hpa height patterns are very much in accord with the location of primarily upward vertical motion and the maxima in surface cyclone activity. The surface cyclone activity is most pronounced southeast of the watershed, near the Kamchatka Peninsula, associated with the East Asian cyclone track. The moisture flux patterns for types 1, 2 and 4 clearly illustrate the importance of the influx of Pacific moisture sources associated with these systems. For each basin, the precipitation anomaly was calculated associated with the days corresponding to each of its four primary synoptic types. To allow for a more direct comparison, these were expressed as a percentage of long-term daily summer means. For the most part, the anomalies for each type within each basin are of similar magnitude. As an exception, type 2 for the Lena is associated with a much larger anomaly (263%) compared with the other three types for this basin (range %). In contrast to the other types, the region of predominantly upward motion for type 2 is roughly centred over the basin. Results for summer are provided in Figures 16 and 17. Synoptic type 1 accounts for 23 49% of all high precipitation days. As for winter, high precipitation in each basin tends to be associated with a 500 hpa trough feature, but with the expected weaker height gradients compared with winter. Again, there are generally clear

17 EURASIAN ARCTIC PRECIPITATION CHARACTERISTICS 1283 (a) (b) Figure 14. Winter (DJF) composite 500 hpa height fields (m) for the four dominant circulation types associated with days of high basin-averaged precipitation for the (a) Ob, (b) Yenisey, (c) Lena and (d) K I. Regions where the frequency of upward vertical motion at 500 hpa exceeds 60% of days is also shown, along with vapour fluxes at the 700 hpa level. The shaded regions are the simplified basins over which the Kirchhofer typing scheme was applied. Also provided beside each panel is the percentage of days represented by each synoptic type relationships between the location of the 500 hpa troughs, the frequency of upward vertical motion and relatively strong vapour fluxes into the region of upward motion. The pattern of the vapour flux again varies widely between synoptic types. Like winter, cyclone activity associated with high precipitation over the Ob is

18 1284 M. C. SERREZE AND A. J. ETRINGER (c) (d) Figure 14. (Continued) roughly centred over the basin. Recall that, for winter, cyclone activity associated with high precipitation over the K I is found over the ocean. By contrast, summer sees a very pronounced area of cyclone activity focused over the basin itself. The location of peak activity corresponds closely to the preferred area of cyclogenesis depicted by Serreze et al. (2001) associated with the summer Arctic frontal zone. For all composites, the southern part of the map, corresponding to the Tibetan Plateau, shows a very high frequency of activity. This

19 EURASIAN ARCTIC PRECIPITATION CHARACTERISTICS 1285 Figure 15. Winter (DJF) distribution of cyclone centres over the Eurasian continent and surrounding oceans based on days of high precipitation for each basin (see text for further explanation) feature is largely fictitious, and is associated with strong surface heating and problems of reducing surface pressures to sea level. Precipitation anomalies associated with the four different synoptic types in a given basin are again generally similar. The sharpest contrast is in the K I between types 2 and 4, associated with precipitation anomalies of 323% and 248% of normal. The larger anomaly for type 2 is seen in terms of higher frequency of upward vertical motion and the strikingly stronger vapour fluxes into the region. This is consistent with the strong 500 hpa trough, compared with the weak cutoff feature in type 4. Type 2 is only composite for this watershed with a strong flux of moisture from the Pacific. Figure 18 summarizes time series of total winter and summer precipitation along with counts of the four Keyday types and the unclassified days, using the Yenisey as an example. Also listed are the Spearman rank correlations between the precipitation time series and the counts of each synoptic type, unclassified types, and types 1 4 taken together. Given that the circulation types are constructed on the basis of high precipitation days, there is an obvious general agreement between basin-averaged seasonal precipitation and the summed occurrence of the four circulation types. For the Yenisey, the correlation is 0.89 in winter and 0.64 in summer. The lower correlation in summer relates to the effects of local convection not captured in synoptic-scale analyses and the higher frequency of unclassified days. Similar results characterize the other basins (not shown). From the example show here, it is evident that although Keyday 1 is the most common synoptic type, the other types do have a comparatively strong influence for individual years. For example, the high precipitation for winter 1979 was associated with a high frequency of types 1 and 2 only. Note also the relationship between the extremely dry summer of 1987 and the virtual absence of key synoptic types.

20 1286 M. C. SERREZE AND A. J. ETRINGER (a) (b) Figure 16. Summer (JJA) composite 500 hpa height fields (m) for the four dominant circulation types associated with days of high basin-averaged precipitation for the (a) Ob, (b) Yenisey, (c) Lena and (d) K I. Regions where the frequency of upward vertical motion at 500 hpa exceeds 60% of days is also shown, along with vapour fluxes at the 700 hpa level. The shaded regions are the simplified basins over which the Kirchhofer typing scheme was applied. Also provided beside each panel is the percentage of days represented by each synoptic type 5. SUMMARY AND CONCLUSIONS This study examines precipitation characteristics over the major watersheds of the Eurasian Arctic drainage system over the period In addition to the Ob, Yenisey and Lena, which represent the three largest

21 EURASIAN ARCTIC PRECIPITATION CHARACTERISTICS 1287 (c) (d) Figure 16. (Continued) drainages, we examine the combined K I in eastern Eurasia. The analysis complements the effort of Serreze et al. (2003) through more focused assessments of mean annual cycles and the synoptic characteristics of large daily precipitation events. A key feature of each basin is the approximate symmetry of the mean annual cycles of monthly mean precipitation and event size about July maxima and winter minima. The same basic symmetry is seen

22 1288 M. C. SERREZE AND A. J. ETRINGER Figure 17. Summer (JJA) distribution of cyclone centres over the Eurasian continent and surrounding oceans based on days of high precipitation for each basin (see text for further explanation) in the mean annual cycles of precipitable water. Effective precipitation mechanisms exist in all seasons. However, because of distance from strong moisture sources (continentality), the amount of precipitation that can be generated tends to be strongly linked to the inherent temperature control on atmospheric water vapour content. A striking contrast is presented for the region around Iceland. Here, the annual cycle of precipitation is determined not by seasonality in column water vapour, but by stronger precipitation-generating mechanisms in winter than in summer. Hence, the annual cycles of precipitation and precipitable water oppose each other. Precipitation events over the Eurasian Arctic watersheds exhibit the strongest spatial organization during the cold season, pointing to synoptic-scale controls. Weaker spatial organization is seen in summer, pointing to a combination of synoptic-scale processes and local convection. These contrasts are consistent with annual cycles of evaporation, the vertically integrated vapour flux convergence, cyclone activity, vertical stability and precipitation recycling. Large daily precipitation events for both winter and summer (basin-averaged events in the top 20% of the distributions) are typically found ahead of well-defined 500 hpa troughs, where the vertical motion is predominantly upward. In general, there is a clear pattern of strong moisture fluxes into the region of upward motion. For either season, the relative frequency of four primary 500 hpa synoptic types captures the basic time series structure of basin-averaged precipitation. Some final comments are warranted regarding inter-basin correlations of precipitation and linkages with large-scale patterns of atmospheric variability. As part of our study, we computed correlation matrices between the time series ( ) of basin-averaged precipitation in each of the four watersheds. This was done for each month using both the daily basin values and monthly basin totals. Daily inter-basin correlations are quite weak none exceeds It is frequently observed that baroclinic waves associated with high precipitation

23 EURASIAN ARCTIC PRECIPITATION CHARACTERISTICS 1289 (a) (b) Figure 18. Winter (a, DJF) and summer (b, JJA) time series of the number of the four dominant 500 hpa circulation types and unclassified types (top panel) and the basin-averaged total precipitation (bottom panel) for the Yenisey. Also provided are the Spearman rank correlations between the precipitation time series and the counts of each of the four synoptic types and unclassified types, and the combined counts of the four synoptic types in a given basin will be associated with precipitation a day or two later in the next downstream basin. Not surprisingly, correlations in daily precipitation between immediately neighbouring basins tend to be somewhat stronger (but still rather weak) at lags of 1 to 2 days. Stronger instantaneous correlations between neighbouring basins emerge for monthly totals. Positive relationships are evident between the Yenisey and Lena for autumn and winter months. Significant correlations range from 0.35 (March) to 0.52 (October). Time series for these neighbouring basins are also modestly correlated during July (0.40). Also of note is the 0.62 correlation between monthly time series in the Ob and Yenisey for February. An interesting inverse relationship exists between precipitation in the Ob and the Lena (western and eastern Siberia) during summer. Using data over the period , Fukutomi et al. (2003) find an inverse correlation of In one mode of the relationship there is a cyclonic circulation anomaly over the Ob (associated with high precipitation) and an anticyclonic anomaly over the Lena (associated with low precipitation). When precipitation is high in the Lena and low in the Ob, the circulation anomaly structure reverses. This seesaw-like alternation between wet and dry regimes in the two basins occurs on a time scale of 6 to 8 years. The underlying cause of this summer teleconnection is, at present, not well understood. Our analysis confirms this inverse relationship in summer precipitation, but the correlation based on the longer record of is lower ( 0.44). There is a great deal of interest in climate variability associated with the Arctic oscillation (AO), which is very similar to the North Atlantic oscillation (NAO), and its recent upward trend during the winter season (Thompson and Wallace, 1998, 2000). The positive mode of the AO is associated with positive anomalies in winter precipitation over parts of the Eurasian Arctic drainages as well as over Alaska (Thompson et al.,