Synoptic forcing of precipitation in the Mackenzie and Yukon River basins

Transcription

1 INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 0: (00) Published online 8 April 009 in Wiley InterScience ( DOI: 0.00/joc.96 Synoptic forcing of precipitation in the Mackenzie and Yukon River basins Elizabeth N. Cassano a * and John J. Cassano a,b a Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, CO 8009, USA b Department of Atmospheric and Oceanic Sciences, University of Colorado, Boulder, CO 8009, USA ABSTRACT: The relationship between near-surface atmospheric circulation, as characterized by sea level pressure patterns, and precipitation in the Mackenzie and Yukon River basins is presented. A synoptic climatology of sea level pressure patterns based on daily sea level pressure anomalies from the ERA0 reanalysis dataset was created using the method of self-organizing maps. This objective analysis identified all major near-surface atmospheric circulation patterns in the region and illustrated the change in dominant circulation patterns throughout the seasons, with strong Aleutian low patterns dominant in the winter and patterns characterized by low pressure over land areas and the Beaufort/Chukchi Seas in the summer. These synoptic patterns were then related to daily precipitation in the Mackenzie and Yukon River basins. The largest daily precipitation values, for both the Mackenzie and Yukon basins, were associated with patterns that occur most frequently in the summer, likely associated with increased frequency of cyclones and convective events that occur over land in that season. During winter, the largest positive precipitation anomalies were along the coastal mountain range in southeastern Alaska associated with Aleutian lows bringing warm, moist flow from the south resulting in upslope flow on the windward side of these mountains. These patterns were responsible for many of the large precipitation events in the winter in the Mackenzie basin. The largest precipitation events in the winter in the Yukon basin occurred with patterns that have a low pressure centre to the southwest of the basin. This synoptic pattern results in southerly flow advecting moisture into the basin to the west of the higher topography which bounds much of the southern boundary of the Yukon watershed. Copyright 009 Royal Meteorological Society KEY WORDS Precipitation; ERA0 Reanalysis; Self-organizing maps; Synoptic climatology Received 7 August 007; Revised February 009; Accepted March 009. Introduction Freshwater is an important component of the hydrologic cycle in the Arctic. The top layer of the Arctic Ocean is relatively fresh, which helps to create and maintain the sea ice in the region. There are three major sources of freshwater into the Arctic Ocean: river runoff, net precipitation falling into the Arctic Ocean itself, and relatively fresh water advected into the Arctic basin via the Bering Strait. Of these three sources of freshwater, river runoff is the largest (Serreze et al., 006). River runoff is strongly controlled by net precipitation over the river basin. Therefore, it is important to understand the characteristics of precipitation over the major river basins in the Arctic and the synoptic controls on them. The focus of this paper is the precipitation characteristics of the Mackenzie and Yukon River basins, and the relationship between precipitation and synoptic patterns in these areas. The Mackenzie River in northwestern Canada is one of the largest contributors of river runoff to the Arctic Ocean (Aagaard and Carmack, 989). The basin covers approximately 0% of Canada s land mass and includes * Correspondence to: Elizabeth N. Cassano, Cooperative Institute for Research in Environmental Sciences, University of Colorado, 6 UCB, Boulder, CO 8009, USA. ecassano@cires.colorado.edu three large lakes and three deltas (Stewart et al., 998). The northern section of the Rocky Mountains is located on the western edge of the basin and includes some peaks reaching heights over 00 m (Figure ). The Yukon River basin encompasses much of central Alaska and extends into the western edge of Canada. Mountains are located on the northern, eastern, and southern edges of the basin, with lower elevations on the west and southwest portions of the basin near the Bering Sea. The freshwater from the Yukon River drains into the northern Bering Sea, which is then advected northward through the Bering Strait and into the Arctic Ocean (Serreze et al., 00). Net precipitation and river runoff have strong seasonal controls. Maximum precipitation in the Mackenzie basin (Bjornsson et al., 99) and central Alaska (Fleming et al., 000) occurs in the summer. However, winter precipitation contributes to water storage and spring runoff, which is the peak time of year for river runoff (Lammers et al., 00), and is therefore an important component of the hydrological cycle of the basins (Lackmann et al., 998). Bjornsson et al. (99) examined precipitation from several stations in the Mackenzie basin and found a seasonal cycle of precipitation, with precipitation peaking in the late summer, decreasing through the early winter, Copyright 009 Royal Meteorological Society

2 SYNOPTIC FORCING OF PRECIPITATION IN THE MACKENZIE AND YUKON BASINS 69 Figure. Analysis domain with topography. The Yukon River basin is outlined with the thick black solid line, and the Mackenzie River basin is outlined with the black dashed line. and remaining low until spring at which time precipitation increases rapidly. They also related precipitation in the basin to large-scale circulation and found a significant correlation between precipitation in the basin and synoptic variability in the North Pacific storm track in the autumn, winter, and spring. They found that more frequent and stronger storms in the North Pacific and western Canada were correlated with above-average precipitation in the Mackenzie basin. Lackmann and Gyakum (996) studied the large-scale circulation associated with significant precipitation events in the Mackenzie basin. They found cyclogenesis in the lee of the Rocky Mountains associated with a strong Aleutian low particularly in the autumn, winter, and spring to be associated with significant precipitation events. Lackmann et al. (998) performed a case study of a significant winter precipitation event in the Mackenzie basin. They found that the source of moisture for the precipitation was the Pacific Ocean and this moisture was transported across the Rocky Mountains on the western edge of the basin. Moisture below the 800 hpa level was blocked by the mountains; therefore, moisture transport between the 700 and 800 hpa levels was an important component of the moisture for the precipitation event. They also found that lee cyclogenesis and a frontal boundary over the central portion of the basin had provided the forcing for the precipitation. In this paper, the method of self-organizing maps (SOMs) is used to create an objective synoptic climatology of sea level pressure (SLP) patterns over northwestern North America and adjacent regions. The SLP patterns identified in this climatology are then related to precipitation in this area with a focus on the Mackenzie and Yukon River basins. Section describes the data used in this analysis and the analysis methodology. The SOM-based SLP synoptic climatology and the relationship between the SLP patterns and precipitation are presented in Section. A summary of this research and concluding remarks are given in Section.. Data and methodology.. Self-organizing maps The field of synoptic climatology provides a powerful method to study the climate of a region by stratifying large volumes of data (daily or higher temporal resolution fields of the atmospheric state) into a small number of categories on a physically meaningful basis. Such an approach provides important information on the synoptic events that control the local climate, such as precipitation, which may be hidden by monthly or seasonal mean fields (Barry and Perry, 00; Hanson et al., 00). An important step in this type of analysis is developing a robust classification scheme that can be applied to large volumes of data. Barry and Perry (00), and references therein, provide a detailed overview of synoptic climatology and its applications. Here we use the method of SOMs (Kohonen, 00) to create a synoptic climatology for northwestern North America and adjacent regions. Cassano et al. (006b) discuss the utility of SOMs for synoptic climatology studies in greater detail. Copyright 009 Royal Meteorological Society Int. J. Climatol. 0: (00)

3 660 E. N. CASSANO AND J. J. CASSANO The SOM technique employs a neural network algorithm that uses unsupervised learning to determine generalized patterns in data. This technique reduces the dimensions of large data sets by grouping similar data records together and organizing them into a two-dimensional array, referred to as a map (note that the terms map and SOM will be used interchangeably below). As a result large, multidimensional data sets are reduced to more easily interpreted forms. Used in this way the SOM algorithm may be considered a clustering technique, but unlike other clustering techniques, the SOM method does not need a priori decisions on data distribution, and is instead trained when processing the data itself. Further, the resulting map is organized such that similar synoptic patterns are located in the same portion of the map. The SOM method is not sensitive to the resolution, quality, or time period of the data per se. The patterns that emerge from the SOM will depict the range of conditions represented in the input data space. Therefore, the usage of the SOM method, as is the case with any statistical method, is dependant on the quality of the data to reproduce realistic results. In this application the SLP field from the ERA0 reanalysis (described in more detail in the next section) is likely an accurate representation of reality, and thus is adequate for our intended purpose of classifying the near-surface synoptic circulation patterns. The SOM is a useful tool in these types of studies and our application because we can obtain additional details of the synoptic climatology of an area compared to that obtained from other methods such as seasonal averages. As we are considering precipitation, which is inherently a synoptic, and shorter, time-scale phenomenon the use of monthly or seasonal averages of precipitation and circulation would likely mask details relevant to the understanding of the forcing for the observed precipitation. Kohonen (00) provides a detailed description of the SOM algorithm and Hewitson and Crane (00) provide additional information on the application of the SOM technique to climate data. Several other studies have also applied SOMs for atmospheric analyses (e.g. Cavazos, 999, 000; Michaelides et al., 00; Crane and Hewitson, 00; Reusch et al., 00a,b; Cassano et al., 006a,b; Lynch et al., 006)... Data The 0 + year European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis (ERA0) data (Uppala et al., 00) were used for this analysis for the time period of September 97 through August 00. The reanalysis data analysed were gridded daily fields of SLP and precipitation. The ERA0 data were interpolated to a polar stereographic grid with a resolution of 0 km. The original data are available every 6 h, so the SLP data for each day s four 6-hourly time periods were averaged to create daily SLP data. The daily precipitation was calculated by summing the precipitation values from each 6 hourly period per day to obtain a daily amount matching the same time period as the SLP data. The precipitation was then averaged over each river basin to obtain the basin averaged daily precipitation. Figure shows the domain of the analysis, as well as the outlines of the Mackenzie and Yukon River basins. This domain was chosen to ensure that the full range of synoptic patterns which impact the basins were represented. SLP anomalies were used as the basis for the synoptic climatology because the SLP gradients, rather than absolute values of SLP, are responsible for determining the near-surface circulation and therefore are of most interest in our analysis. SLP anomalies were calculated for each day by subtracting the mean SLP over the analysis domain for a day from the grid point values of SLP for the same day. In addition, SLP values from locations with elevation of greater than 00 m were filtered out of the analysis because of errors associated with the reduction of surface pressure to SLP for high elevation locations (Wallace and Hobbs, 977; Sangster, 987). Serreze et al. (00) evaluated northern high-latitude precipitation output from ERA0 and performed a comparison between this data and observations. They found ERA0 captured the large-scale patterns of precipitation and the depiction of interannual variability. Specifically for the Mackenzie basin they found high correlations between ERA0 and observed precipitation for January, April, and October, but lower correlations in July. Betts et al. (00) found that the bias and spinup of ERA0 precipitation over the Mackenzie basin changed over the analysis period, the magnitude of which decreased considerably after the 960s... Data analysis The SOM algorithm identifies a user specified number of patterns (also referred to as nodes) in the training data. The training data used were gridded fields of daily SLP anomalies over northwestern North America and adjacent regions (Figure ) for the time period September 97 through August 00. For this study a total of SLP patterns were identified by the SOM algorithm, resulting in a 7 SOM (Figure ). This SOM size, and thereby the number of patterns, was selected after evaluating SOMs of varying sizes, with both more and less patterns. Different initializations and sizes of SOMs identified the same general synoptic patterns that characterize the area. It was found that the 7 SOM identified all major SLP patterns in this region and illustrated a range of details in the basic synoptic circulation patterns that aided in the interpretation of the relationship between circulation patterns and precipitation events. SOMs of this size have been found suitable for synoptic climatology studies as the SOM compactly displays the major circulation patterns and storm tracks (Hewitson and Crane, 00; Cassano et al., 006b; Lynch et al., 006), while still distinguishing important patterns in the data, such as varying intensity and positions of high and low pressure centres. Once the SOM algorithm has identified the patterns in the training data, the final step in the analysis is Copyright 009 Royal Meteorological Society Int. J. Climatol. 0: (00)

4 SYNOPTIC FORCING OF PRECIPITATION IN THE MACKENZIE AND YUKON BASINS 66 Figure. Self-organizing map (SOM) of daily sea level pressure (SLP) anomaly patterns for September 97 August 00. Blue shading indicates negative anomalies and red shading indicates positive anomalies. The colour scale represents a range of SLP anomalies from to + hpa. The contour interval is hpa. Numbers along the top and left sides of the SOM are the node numbers. the association of each daily field of SLP with a single pattern on the SOM. The SLP data from each day were compared to each pattern on the SOM to determine which pattern they most closely match. The squared difference between the daily SLP and the node SLP is used as the measure of the similarity for this analysis. This results in a list of days associated with each SOM node. This list of days associated with each node can then be used to determine how frequently each node occurs (node frequency of occurrence). The statistical significance of the node frequency of occurrence was determined as nodes whose frequencies differ from that expected if each node were to occur equally frequently (/ 00 or.86%) at the 9% confidence interval. The confidence interval was calculated by assuming a random, binomial distribution of frequencies throughout the nodes and is given by p ±.96 p( p) / n where p is the probability a daily sample would map to any node (/N where N is the number of nodes), and n is the number of daily samples. The list of days associated with each SOM node can also be used to relate other fields to the SOM SLP patterns. For this study, average daily precipitation, both on a grid point basis over the entire analysis domain and over the Mackenzie and Yukon basins, was calculated for each SLP pattern on the SOM. These values were determined by summing the precipitation for all days that were associated with each SOM pattern and then calculating the average precipitation for that particular pattern.. Results.. Synoptic climatology of northwestern North America Figure shows the SLP anomaly SOM created from the ERA0 data. The numbers along the top and left of the SOM correspond to the node numbers. The SLP patterns on the right portion of the SOM are dominated by an Aleutian low with differing strengths and locations, and high pressure over northern Canada, northern Alaska, and the Beaufort/Chukchi Seas. In the bottom right corner are Aleutian lows centred in the Gulf of Alaska. Moving to the upper right corner of the SOM, the Aleutian low migrates westward over the Aleutian Islands. Moving across the upper portion of the SOM from right to left, the centre of the low pressure systems moves progressively westward and northward until it is located Copyright 009 Royal Meteorological Society Int. J. Climatol. 0: (00)

5 66 E. N. CASSANO AND J. J. CASSANO in the Beaufort/Chukchi Seas in the upper left corner of the SOM. Patterns in the lower left portion of the SOM are characterized by low pressure located southwest of the Aleutian Islands and also over the Canadian archipelago with high pressure over the Chukchi Sea and Siberia. The SLP patterns shown in the SOM represent the major nearsurface circulation patterns expected in this region and are in agreement with previous climatologies (Overland and Hiester, 980; LeDrew, 98, 98; Serreze and Barry, 00), indicating that the SOM algorithm used in this manner is able to identify the dominant SLP patterns. Interpretation of the SOM shown in Figure is aided by a Sammon map. Sammon mapping is used to represent how each node in the original data space is related to each other and provides an indication of the degree of similarity between the nodes of the SOM (Figure ). The Sammon mapping algorithm fits the data points (nodes) onto a two-dimensional space such that the Euclidian distances between the nodes approximate the corresponding distances in the original data space (Sammon, 969). This is done to illustrate how the nodes are related to each other and provides a simple graphical way to identify which patterns are most (or least) similar. Each point on the Sammon map represents one of the nodes of the SOM [in the same orientation as in Figure, as will be the case for the remainder of the figures in the paper relating to the SOM, e.g. (, ) in the upper left corner corresponds to (, ) in the SOM] and the distance between the points is representative of the Euclidian distance between the nodes. Nodes in the lower central portion of the SOM [nodes (, ), (, ), (, ), (, )] are most similar to adjacent nodes, as indicated by their close proximity on the Sammon map, while nodes along the right portion of the SOM are most dissimilar from adjacent nodes, as indicated by the larger distances between the nodes on the Sammon map. Moving from left to right, the Euclidian distances are largest across the top row of the SOM, consistent with the large difference in SLP patterns evident in this row of the SOM in Figure (the low pressure area moves from the northern portion of the domain to the southern portion). The distance across the bottom row of the SOM is much smaller as a result of the greater similarity of the patterns in this row, which are all characterized by low pressure, of varying intensity, located in the North Pacific. The nodes along the right edge of the SOM are more dissimilar than nodes along the left edge of the SOM. On the right edge of the SOM all of the patterns are characterized by strong Aleutian low pressure centres of varying position. Slight changes in the position of these large SLP anomalies result in large differences between adjacent nodes. On the left side of the SOM the SLP anomalies are much smaller, and as a result the differences between adjacent nodes are smaller. Analysis of how each of the days map to the SOM allows us to determine how frequent each of the patterns is, and how the occurrences of the synoptic patterns vary from season to season. Figure shows the frequency of occurrence of each of the synoptic patterns identified by the SOM. In this figure, nodes that have frequencies of occurrence significantly greater (less) than.86% (i.e. the node frequency that would occur if each node occurred equally frequently) are highlighted by dark (light) shading. The most frequent patterns represent differing strengths and locations of the Aleutian low (right and bottom portions of the SOM) and patterns with a Beaufort/Chukchi low or low pressure over Siberia, Alaska, or Canada (upper left corner and centre of the leftmost column of the SOM). Less frequent patterns are progressively weaker and westward shifted Aleutian lows (centre of the SOM) and low pressure in the North Pacific south and west of the Aleutian Islands, as well as high pressure in the Beaufort/Chukchi Seas (lower left corner of the SOM). (7,) (,) (7,) (,) Figure. Sammon map of the SOM shown in Figure. Copyright 009 Royal Meteorological Society Int. J. Climatol. 0: (00)

6 SYNOPTIC FORCING OF PRECIPITATION IN THE MACKENZIE AND YUKON BASINS Figure. Frequency of days that map to each SOM node annually. The dark (light) grey shading indicates positive (negative) statistical significance as discussed in the text. Each number on the figure represents percentage frequency of daily occurrence on an annual basis for that particular node. Analysis of node frequencies on a seasonal basis provides further insight into the synoptic climatology of this region (Figure ). Each bar in the node plots in Figure represents the node frequency of occurrence for an individual season, with frequencies significantly greater (less) than.86% coloured by black (grey) bars. The leftmost bar in each plot represents node frequency of occurrence for December, January, and February (DJF). The most frequently occurring patterns during DJF are strong Aleutian low patterns (right portion of the SOM), and patterns associated with low pressure systems centred to the southwest of the Aleutian Islands and over the Canadian archipelago (lower left corner of the SOM). Patterns occurring infrequently during the winter are weaker, westward shifted Aleutian lows and Beaufort/Chukchi lows (centre to upper left portion of the SOM). This is in agreement with previous climatologies (Overland and Hiester, 980; Rodionov et al., 00). The winter node frequencies can be contrasted with those for the summer months (June, July, and August; JJA) (second bar from the right in Figure ). The most frequently occurring summer patterns are low pressure systems in the Beaufort/Chukchi Seas, broad low pressure over land areas, and weak low pressure systems in the Bering Sea (mid to upper left corner of the SOM). The patterns with broad low pressure over land areas are indicative of increased cyclonic activity, especially along the Arctic frontal zone (Serreze et al., 00), that occurs in the summer compared to winter and is associated with increased convection that occurs in these areas during the summer (Serreze and Hurst, 000; Serreze et al., 00). In general, Aleutian low patterns, which are common during DJF, occur less frequently during JJA, and the strongest Aleutian low patterns located in the rightmost column of the SOM do not occur at all. Comparison of node frequencies for DJF and JJA highlights the transition of circulation patterns from winter to summer, as the frequency distribution of synoptic patterns during JJA is an almost mirror image of the DJF frequency distribution. The synoptic patterns most frequently occurring during the transition seasons of spring and autumn bridge the gap between the winter and summer patterns. Spring (March, April, May; MAM) frequencies are represented by the second bar from the left in Figure. Patterns that occur most frequently during MAM are moderately strong low pressure systems centred in the Bering Sea to lows centred in the Gulf of Alaska (along the fifth column of the SOM). Also frequent at this time of year are weak to moderately strong Gulf of Alaska centred low pressure systems [along the mid-bottom row of the SOM, nodes (, ) though (6, )]. Patterns which occur infrequently are weak low pressure over the western portion of the domain to Beaufort/Chukchi lows (upper left portion of the SOM) and strong Aleutian lows (the rightmost column of the SOM). Node frequency of occurrence for the autumn months (September, October, and November; SON) shows a transition between summer and winter type patterns (rightmost bar in each plot in Figure ). The most frequent patterns are Beaufort/Chukchi lows and low pressure over Alaska and Canada [nodes (, ) and (, )] and moderately strong Gulf of Alaska centred Aleutian lows [nodes in the lower, right portion of the SOM, especially nodes (, ) and (6, )]. Infrequent patterns are the strongest Aleutian lows (along the rightmost Copyright 009 Royal Meteorological Society Int. J. Climatol. 0: (00)

7 66 E. N. CASSANO AND J. J. CASSANO Figure. Frequency of days that map to each SOM node seasonally. Each bar represents the percentage frequency of daily occurrence for a particular season. Bars that are coloured in black (grey) indicate positive (negative) statistical significance as discussed in the text. column of the SOM) and low pressure in the Bering Sea (centre/left portion of the SOM). Frequency analysis with respect to large-scale climate circulation modes [here focused on the Arctic Oscillation (AO; Thompson and Wallace, 998) and the Pacific Decadal Oscillation (PDO; Mantua and Hare, 00)] was performed for DJF when these modes are most pronounced. The difference in DJF frequencies between positive and negative AO regimes shows a decrease in strong Aleutian low pressure systems when the AO is positive compared to when it is negative. During a positive AO regime, higher pressure (e.g. more mass) is located in the sub-arctic and lower pressure (e.g. less mass) is located in the central Arctic. Therefore, during a positive event, there is more mass present to fill in the Aleutian low, and thereby fewer strong Aleutian lows occur (Figure, right side of SOM). This same analysis but for the PDO shows that patterns along the right side of the SOM (strong Aleutian lows) are more frequent during a positive PDO regime, while the left two-thirds of the SOM are less frequent. This is not unexpected as a positive PDO regime is characterized by stronger Aleutian lows (Mantua and Hare, 00)... Synoptic forcing of precipitation... General characteristics The synoptic forcing for precipitation was evaluated by mapping the daily precipitation anomalies from ERA0, on the same grid as the SLP data, to each pattern of the SOM. The precipitation anomalies were calculated as the difference between the node averaged daily precipitation and the mean precipitation (at each grid point in the analysis domain) for the period of the ERA0 data for each node (Figure 6). Blue shading indicates positive anomalies and yellow to red shading indicates negative anomalies. Strong Aleutian lows (lower right corner of the SOM) are associated with positive precipitation anomalies along the coastal range in southeast Alaska and along the mountains of western British Columbia, consistent with the counterclockwise flow around the Aleutian low creating upslope flow impinging on the mountains. The utility of the SOM analysis for exploring the relationship between the synoptic circulation patterns and associated precipitation is demonstrated by examining the rightmost column of the SOM. Here, the ability of the SOM analysis to identify subtle differences in the circulation patterns, such as slight changes in the position and/or Copyright 009 Royal Meteorological Society Int. J. Climatol. 0: (00)

8 SYNOPTIC FORCING OF PRECIPITATION IN THE MACKENZIE AND YUKON BASINS 66 intensity of the low pressure centres, is critical as these slight changes in circulation result in significant changes in the precipitation distribution. For patterns with a Gulf of Alaska centred Aleutian low (lower right corner of the SOM), the maximum precipitation is located in southeast Alaska and along the coast of British Columbia, consistent with the onshore, upslope flow associated with this synoptic pattern. Moving from the bottom to the top of the rightmost column of the SOM (Figure ), the Aleutian low shifts to the west. With this shift in the position of the Aleutian low is a westward shift in the location of the strongest onshore flow and an associated westward shift along the southern Alaskan coast of the area of maximum precipitation (Figure 6). Another illustration of the physically consistent relationship between the SOM defined circulation patterns and the precipitation anomalies is demonstrated by considering the right half of the bottom row of the SOM. The patterns in this portion of the SOM, moving from left (, ) to right (7, ), represent progressively stronger Gulf of Alaska centred Aleutian lows (Figure ). The precipitation associated with these patterns becomes progressively greater moving from left to right (Figure 6), consistent with the increasing intensity of the low pressure centre and coincident stronger onshore flow impinging on the mountains creating upslope precipitation. In general, the synoptic patterns represented in the rightmost three columns of the SOM are characterized by positive precipitation anomalies along the southern coast of Alaska and/or the coast of British Columbia and negative precipitation anomalies in the interior of Canada and Alaska. Consistent with the inferred upslope forcing in the areas of greatest precipitation identified with these synoptic patterns, the ERA0 data also indicate upward vertical velocity in these areas at both the 700 and 00 hpa levels (not shown). This distribution of precipitation is consistent with the varying position and intensity of the Aleutian lows depicted in Figure and the interaction of this flow with the regional topography. In contrast to the precipitation patterns associated with varying strengths and locations of the Aleutian low (right side of the SOM), patterns characterized by low pressure over land areas and the Beaufort/Chukchi Seas are associated with positive precipitation anomalies over the interior of Canada and Alaska and the Beaufort/Chukchi Seas, with negative precipitation anomalies over the Gulf of Alaska and adjacent coastlines (Figure 6; upper left portion of the SOM). Analysis of other ERA0 fields Figure 6. Daily precipitation anomalies mapped to the SOM. Blues are positive anomalies and reds and yellows are negative anomalies. The contour interval is 0.0 cm/day. The numbers along the top and left sides of the SOM are the node numbers. Copyright 009 Royal Meteorological Society Int. J. Climatol. 0: (00)

9 666 E. N. CASSANO AND J. J. CASSANO indicates that these patterns are also associated with higher levels of upper level (700 hpa) relative humidity over land areas, particularly Alaska and even more so over the Brooks Range in northern Alaska (not shown). Along the leftmost column of the SOM, broad upper level troughing at upper levels (i.e. 700, 00, and 00 hpa) associated with the low pressure located in the Beaufort/Chukchi Seas (not shown) is present over land areas. Additionally, these patterns are associated with upward vertical velocity at 00 and 700 hpa over land areas, particularly over Alaska and the southern Mackenzie basin (not shown). These features are consistent with the surface circulation represented by the SLP patterns. These SLP patterns are more frequent during JJA. In addition, during JJA, evaporation is greater than in the winter, making more moisture available for local precipitation recycling, which contributes to the positive precipitation anomalies in these areas (Walsh et al., 99). Calculations of the ratio of local to total precipitation for the Mackenzie basin show a maximum in June with a ratio of approximately %, and in general the highest values occurring May through August (Serreze et al., 00). Transpiration from vegetation during the growing season and evaporation from unfrozen lakes during the warm season contribute to the moisture available for local precipitation recycling (Szeto et al., 007). Dynamically these positive precipitation anomalies can be attributed to an increase in cyclonic activity (Serreze and Barry, 00) especially along the Arctic frontal zone, which occurs over land areas in the summer (Serreze et al., 00) as indicated by the low pressure over the land areas in these nodes, and to increased convective precipitation over land (Serreze and Hurst, 000). Positive precipitation anomalies in the southern and southwestern portion of the domain (Figure 6) are associated with low pressure in the Gulf of Alaska and low pressure in the North Pacific (lower left corner of the SOM; Figure ). Evaluation of additional fields from ERA0, particularly of upper level height fields, shows a strong relationship between the upper level flow and the SLP patterns (not shown). The spatially consistent relationship between circulation, as represented by SLP patterns of the SOM, and precipitation provides a useful framework for considering the forcing of precipitation over large areas such as the Mackenzie and Yukon watersheds.... Mackenzie River basin Node averaged daily precipitation was calculated for both the Mackenzie and Yukon River basins to link the synoptic patterns shown in the SOM to precipitation in these specific regions. Figure 7(a) shows the node averaged daily precipitation in the Mackenzie River basin. The maximum daily precipitation values are associated with widespread positive anomalies over much of the basin (Figure 6) and occur in patterns that occur most frequently during JJA (Figure ). The patterns associated with the maximum daily precipitation values are low pressure over the Mackenzie basin [node (, )] and (a) (b) (c) Figure 7. Precipitation mapped to the SOM for the Mackenzie basin (a) annually, (b) for DJF, and (c) for JJA. Each number on the figure represents the average daily precipitation in centimetre per day that occurs on an annual or seasonal basis for that particular node. Contouring/shading in the figure is for ease of visualization. patterns with high pressure to the north of the Mackenzie basin and easterly flow in the basin with broad low pressure over the basin [nodes (, ) and (, )]. Minimum precipitation values are associated with patterns with widespread negative precipitation anomalies over much of the Mackenzie basin (Figure 6) and with patterns that occur most frequently during DJF (Figure ). These patterns are characterized by a strong Aleutian low and broad high pressure over the Mackenzie basin (right portion of the SOM). The lowest values are associated Copyright 009 Royal Meteorological Society Int. J. Climatol. 0: (00)

10 SYNOPTIC FORCING OF PRECIPITATION IN THE MACKENZIE AND YUKON BASINS 667 with westward shifted Aleutian lows (upper right corner of the SOM). For these patterns the Aleutian low is centred far enough to the west so that there is little onshore flow directed towards the Mackenzie basin or dynamic forcing that would lead to significant precipitation events. As the Aleutian low shifts further east (along the centre and bottom of the right side of the SOM) the counterclockwise flow around the low produces flow that is southerly/southwesterly into basin. Before entering the basin, this flow first encounters the coastal range, which produces upslope precipitation there and in general little moisture is advected over the Rocky Mountains into the basin. However, some moisture advected over the mountains at higher levels can contribute to significant precipitation events in the basin (Lackmann et al., 998). Precipitation anomalies on the entire analysis grid mapped to the SOM for the winter (not shown) do provide some evidence for this as there are some positive precipitation anomalies in the Mackenzie basin for patterns in the lower right corner of the SOM associated with strong, Gulf of Alaska centred low pressure systems (see Section. for additional discussion of these precipitation events). Given the strong seasonal signal in precipitation (Bjornsson et al., 99) it is instructive to consider the basin averaged precipitation for individual seasons. During DJF, the node averaged daily precipitation over the Mackenzie basin is small compared with that during other seasons (Figure 7(b)). This is due to the colder temperatures at this time of year, and a reduction in the amount of moisture in the air. During the winter months the maximum precipitation in the Mackenzie basin is associated with strong Gulf of Alaska centred low pressure systems bringing relatively warm, moist air into the basin from the south (bottom right corner of the SOM). The maximum node averaged winter daily precipitation is associated with a pattern with high pressure located to the north of the basin resulting in easterly flow into the basin [node (, )], This creates upslope flow on the eastern side of the mountains on the western edge of the basin. A plot of the spatial winter precipitation anomalies (not shown) indicates a local maximum along the eastern slopes of these mountains, with positive precipitation anomalies throughout the basin. Some higher precipitation values are associated with Beaufort/Chukchi lows bringing westerly flow into the basin, broad low pressure over the basin, or a moderately strong low pressure centred to the northeast of the basin bringing north/northwesterly flow into the basin (along the left column of the SOM). These patterns do not occur frequently during the winter (Figure ), but are more frequent during SON, so that the relatively large precipitation values associated with these synoptic patterns are due to some relatively warm, and thus moist, early winter events which add considerably to the total for these patterns as they occur so infrequently in the winter. Minimum node averaged wintertime daily precipitation values are associated with patterns that have moderate to strong low pressure system centred over the Aleutian Islands, while high pressure is dominant over the Mackenzie basin [central/right portion of the SOM nodes (, ) to (6, )]. As discussed above the position of the Aleutian low further west in these patterns (compared to the position of the Aleutian low in the lower right corner of the SOM) results in relatively weak onshore flow towards the Mackenzie basin and generally negative precipitation anomalies over the basin. In contrast to DJF, JJA is the season with the highest precipitation values of the year (Figure 7(c)). The maximum node averaged precipitation value is associated with broad low pressure over the basin [node (, )]. The low pressure over the basin is associated with convergent low level flow (due to frictional effects) which can act as a triggering mechanism for convection. It is likely that convective events are responsible for the large node averaged daily precipitation values for this SLP pattern. Other patterns associated with large precipitation amounts are patterns with broad low pressure over the basin [nodes (, ) to (, )] and some with weak high pressure to the north and northwest of the basin bringing easterly/northeasterly upslope flow into the basin (nodes in the lower left corner and third column of the SOM). Patterns with low pressure present over the basin, as stated above, are indicative of the generally convective nature that occurs over land areas in the summer in the Arctic (Serreze and Hurst, 000) and contributes to large precipitation values in these areas. High evaporation values present during the summer likely contribute moisture for the convection. Minimum precipitation values are associated with strong westward shifted Aleutian lows (upper right portion of the SOM). Note that the node averaged daily precipitations for the nodes along the rightmost column are all zero as these patterns do not occur during JJA (Figure ). The pattern of average daily precipitation mapped to the SOM for MAM shows the transition between DJF and JJA (not shown). Minimum values are associated with westward shifted Aleutian lows (upper right corner of the SOM) and low pressure located southwest of the Aleutian Islands and over the Canadian archipelago (lower left corner of the SOM). The largest value is associated with a pattern with low pressure over the basin and a weak high north of the basin bringing weak easterly flow into the basin [node (, )]. In general, values are greater than those for DJF, associated with the warmer conditions and greater moisture content of the atmosphere at this time of the year. For SON, the average daily precipitation begins to decrease compared with the JJA values (not shown), as the atmosphere begins to cool and the moisture content of the atmosphere decreases. Maximum daily precipitation values are associated with patterns with low pressure over the basin and high pressure to the north bringing easterly/northeasterly flow into the basin (upper left portion of the SOM). Minimum values are associated with moderate to strong Aleutian low patterns and high pressure over the basin (central to upper right portion of the SOM). Copyright 009 Royal Meteorological Society Int. J. Climatol. 0: (00)

11 668 E. N. CASSANO AND J. J. CASSANO... Yukon River basin The distribution of node averaged daily precipitation mapped to the SOM for the Yukon River basin (Figure 8(a)) is quite similar to that of the Mackenzie basin (Figure 7(a)). The maximum daily precipitation values are associated with a pattern with broad low pressure over the basin with a weak high to the north, resulting in easterly/northeasterly flow into the basin [node (, )] and a pattern characterized by broad low pressure over the basin [node (, )]. In general, the SLP patterns associated with large daily precipitation values are those which occur most frequently during JJA (Figure ) and are patterns with broad low pressure over the basin and also patterns with a Beaufort/Chukchi low bringing westerly flow into the basin (Figure ). Focusing on the seasonal node averaged daily precipitation, values for DJF are greater over the Yukon basin (Figure 8(b)) compared to the Mackenzie basin (Figure 7(b)). The largest value is associated with a synoptic pattern with a moderate low pressure system located north of the Aleutian Islands bringing southerly flow into the western portion of the basin [node (, )]. This southerly flow is located to the west of the highest terrain along the southern edge of the basin (Figure ). With the position of this southerly flow moisture from the Pacific Ocean is not blocked by higher terrain and is advected into the basin. In addition to this moisture, the proximity of this low pressure centre to the basin results in strong dynamic forcing and large daily precipitation amounts for this SLP pattern. Other patterns associated with large precipitation are Beaufort/Chukchi lows [node (, )] and patterns with low pressure in various locations to the west of Alaska bringing southerly to southwesterly flow into the basin [nodes (, ) to (, ); (, ) to (, ); (, )]. Minimum precipitation values are associated with strong Aleutian lows (upper right corner of the SOM) and patterns with high pressure over Alaska with a low pressure system centred to the southwest of the Aleutian Islands (nodes along the left portion of the bottom row of the SOM). For strong Aleutian low patterns, the nearsurface flow is southeasterly into the basin and therefore, encounters the high terrain of the Alaska Range. The high terrain blocks the transport of moisture into the basin, and the southeasterly flow across the barrier leads to downslope flow into the basin, which suppresses precipitation production. For patterns with high pressure over Alaska and a low pressure system centred to the southwest of the Aleutian Islands, the near-surface flow implied by the SLP pattern is southeasterly, again resulting in moist Pacific air entering the basin only after passing over the high terrain on the southern edge of the basin. In addition, for both of these SLP patterns the low pressure centres are located far to the southwest of the basin and do not provide any dynamic forcing that would lead to significant precipitation events. These patterns are also characterized by relatively high pressure over the Yukon basin, further limiting the potential for precipitation production. For JJA, the maximum node averaged daily precipitation is associated with broad low pressure over the (a) (b) (c) Figure 8. Precipitation mapped to the SOM for the Yukon basin (a) annually, (b) for DJF, and (c) for JJA. Each number on the figure represents the average daily precipitation in centimetre per day that occurs on an annual or seasonal basis for that particular node. Contouring/shading in the figure is for ease of visualization. basin [node (, )] (Figure 8(c)). Other patterns associated with large precipitation values are those with Beaufort/Chukchi lows bringing westerly flow into the basin [nodes (, ), (, ), and (, )], and patterns with broad low pressure over the basin and weak highs in the Beaufort or Chukchi Seas [nodes (, ) to (, )]. As discussed above for the Mackenzie basin, patterns that have low pressure over the land areas are associated with increased cyclonic activity and low level convergence which can Copyright 009 Royal Meteorological Society Int. J. Climatol. 0: (00)

12 SYNOPTIC FORCING OF PRECIPITATION IN THE MACKENZIE AND YUKON BASINS 669 trigger convection. In addition high evaporation rates during JJA provide moisture for local precipitation recycling. This combination of factors results in large daily precipitation amounts for these SLP patterns. Minimum precipitation values are associated with strong Aleutian lows (right portion of the SOM). These patterns occur very infrequently during JJA, and the strongest Aleutian low patterns (the rightmost column of the SOM) do not occur at all. As discussed for the DJF daily precipitation, the strong Aleutian low patterns are not conducive to large daily precipitation amounts because of blocking of moist low level flow from the Pacific Ocean by the Alaska Range. Average daily precipitation values during MAM are of a similar magnitude to those of DJF with a similar precipitation distribution across the nodes as the daily precipitation over the entire year shown in Figure 8(a) (not shown). The maximum node averaged precipitation value is associated with a synoptic pattern with a weak high to the northeast of the basin and broad low pressure across the basin [node (, )]. Other patterns with large precipitation values are those with a low pressure to the west of Alaska bringing southerly flow into the basin and those with broad low pressure over the basin. Minimum precipitation values are associated with strong Aleutian lows (upper right corner of the SOM) and low pressure located southwest of the Aleutian Islands and over the Canadian archipelago (left portion of the bottom row of the SOM). Average daily precipitation mapped to the SOM for SON has a similar distribution of daily precipitation on an annual basis as shown in Figure 8(a) (not shown). The maximum value is associated with broad low pressure in the basin with a weak high in the Beaufort/Chukchi Seas [node (, )] Other synoptic patterns similar to this one are also associated with large precipitation values as are Beaufort/Chukchi low patterns. Minimum precipitation values are associated with westward shifted Aleutian lows (upper right corner of the SOM)... Synoptic pattern based analysis of annual precipitation The previous section discussed the node averaged daily precipitation associated with each synoptic pattern in the SOM. To understand the contribution of each SLP pattern to the total annual precipitation it is necessary to consider not only the node averaged daily precipitation but also the frequency of occurrence of each node. A synoptic pattern may have large average daily precipitation associated with it, but if that pattern does not occur very frequently, it would not contribute much to the annual total. Conversely, if a synoptic pattern does not have a high average daily precipitation associated with it but it occurs frequently, those values could add up to a significant contribution to the total annual precipitation. For each node in the SOM, its contribution to annual precipitation was calculated as N = P F 6 day/year where N is the node contribution to annual precipitation (Figure 9), P is the average daily precipitation associated with that node (Figures 7(a) and 8(a)) and F is the frequency of occurrence of that node (Figure ). The analysis below discusses the contribution of each synoptic pattern to the total annual precipitation. A similar analysis could be conducted on a seasonal basis using the seasonally averaged node daily precipitation and the seasonal node frequencies. As for the node averaged daily precipitation (Figure 7(a)), maximum values of each pattern s contribution to annual precipitation for the Mackenzie basin are associated with Beaufort/Chukchi lows and patterns with broad low pressure over the basin (upper left corner of the SOM) (Figure 9(a)). These nodes have both large daily averaged precipitation amounts (Figure 7(a)) and high frequency of occurrence (Figure ), particularly during JJA. Another local maximum is associated with moderately strong Gulf of Alaska centred low pressure systems [nodes (, ) and (, )]. These patterns occur frequently (Figure ), particularly during the shoulder seasons (MAM and SON; Figure ), but have lower node averaged daily precipitation values than those in the upper left corner of the SOM (Figure 7(a)). The patterns making (a) (b) Figure 9. Synoptic pattern contribution to annual precipitation for the (a) Mackenzie basin, and (b) Yukon basin. Each number on the figure represents the synoptic pattern contribution to annual precipitation in centimetre per year for that particular node. Contouring/shading in the figure is for ease of visualization. Copyright 009 Royal Meteorological Society Int. J. Climatol. 0: (00)