Abstract

Projections of twenty-first-century precipitation for seven Swiss river basins are generated by linking high-resolution (2 km × 2 km) radar-estimated precipitation observations to a global climate model (GCM) via synoptic weather patterns. The use of synoptic patterns characterizes the effect of changes in large-scale circulation, or dynamic effects, on precipitation. In each basin observed total daily precipitation received during advective synoptic patterns is shown to be dependent on the basin’s general topographic aspect. Across all basins convective synoptic patterns follow the same trend in total daily precipitation with cyclonic patterns consistently producing a larger amount of precipitation than anticyclonic patterns. Identification of synoptic patterns from a GCM for the twenty-first century [Community Climate System Model, version 3.0, (CCSM3)] shows increasing frequency of anticyclonic synoptic patterns, decreasing frequency of cyclonic patterns, and constant frequency of advective patterns over Switzerland. When coupled with observed radar-estimated precipitation for each synoptic pattern, the changes in synoptic pattern frequencies result in an approximately 10%–15% decrease in decadal precipitation over the course of the twenty-first century for seven Swiss river basins. The study results also show an insignificant change in the future (twenty-first century) probability of exceeding the current (2000–08) 95th quantile of total precipitation. The lack of a trend in exceeding the 95th quantile of precipitation in combination with a decreasing trend in total precipitation provides evidence that dynamic effects will not result in increased frequency of heavy precipitation events, but that heavy precipitation will account for a greater proportion of total precipitation in Swiss river basins by the end of the twenty-first century.

1. Introduction and background

Precipitation is a primary component of the hydrologic cycle and has significant implications for freshwater supply and storage, as well as flooding. Quantitative precipitation forecasts (QPFs) and quantitative precipitation estimates (QPE) are critical inputs to hydrologic forecasts because of the direct connection between the spatial and temporal distribution of precipitation and resulting runoff. The Mesoscale Alpine Programme (MAP) Demonstration of Probabilistic Hydrological and Atmospheric Simulation of Flood Events (D-PHASE) is an example of a collaborative atmospheric and hydrologic sciences effort to incorporate QPE and QPF into hydrologic forecasts for the European Alps (Ranzi et al. 2007; Zappa et al. 2008). A primary goal of MAP D-PHASE is operational forecasting of Alpine flood events through improvement of single to multiday hydrologic forecasts. Forecasts with short time horizons (hours to days) are typically used for tactical actions such as warning and minimization or prevention of floods, similar to the objectives of MAP D-PHASE. However, hydrologic forecasts with longer time horizons are also important for water resources management (Table 1; adapted from Webster et al. 2010). Long-term forecasts (seasonal, annual, and even decadal) provide a basis for strategic water resource planning, such as development of management policy or investment in infrastructure.

Table 1.

Hydrologic forecast periods are listed with applicable time horizons and typical usage in water resource management (adapted from Webster et al. 2010).

Hydrologic forecast periods are listed with applicable time horizons and typical usage in water resource management (adapted from Webster et al. 2010).
Hydrologic forecast periods are listed with applicable time horizons and typical usage in water resource management (adapted from Webster et al. 2010).

As the time horizon for QPF is extended it becomes important to consider precipitation climatology. Long-term hydrologic forecasting has traditionally been based upon a climatology that includes observed historical precipitation distribution with an assumption of climate stationarity over time. However, it is now realized that as climate changes it is not valid to base water resource planning on static, historical precipitation distributions (Milly et al. 2008; Gilleland and Katz 2011; Hirsch 2011). It has become critical to develop an understanding of how precipitation may be affected in a changing climate, so that water management strategy may adapt, if necessary.

So how can we generate long-term precipitation forecasts at spatial resolution relevant to river basin hydrologic forecasts in a changing climate? Various methods of downscaling have been applied to this problem. Current methods for downscaling regional precipitation distribution are categorized as dynamical or statistical (Schmidli et al. 2007, and references within). Dynamical downscaling uses regional climate models (RCMs) to directly obtain high-resolution atmospheric data at the desired grid spacing. For example, Smiatek et al. (2009) ran multiple RCMs for the European Alps with spatial resolution ranging from approximately 5 to 20 km. The study concluded that lack of high-resolution precipitation observations prevented model validation and limited the use of models for evaluating trends in future precipitation due to climate change (Smiatek et al. 2009). In statistical downscaling, lower-resolution data, either from observations or numerical models, are extrapolated or interpolated, as required, to deduce precipitation distributions at the desired spatial scale. Statistical downscaling of precipitation data over complex alpine terrain is often accomplished by applying functions to account for spatial variations in temperature, elevation, or aspect to convert low-resolution data into a high-resolution grid. For example, Machguth et al. (2009) calculated inputs to glacier mass balance for the Swiss Alps by downscaling RCM precipitation from ~18-km grid spacing to 100-m resolution via application of topography-based temperature and precipitation lapse rates. A spatial resolution of 100 m was chosen because variation in orographic precipitation significant to glacier mass balance occurs at a scale from less than one to a few kilometers (Machguth et al. 2009). Model results were compared to observations at 14 weather stations, and it was found that errors in model output, including errors due to localized precipitation variability, were significant enough that the model did not provide reliable glacier mass balances (Machguth et al. 2009). Therefore, validation of both statistical- and dynamical-downscaled models has proven challenging, either because of model inaccuracy in reproducing climate effects that occur at small spatial scales or because of unavailability of high -resolution observations for comparison. As a result, long-term QPFs relevant to river basin hydrological forecasts remain highly uncertain.

What if we develop a long-term QPF by direct association of high-resolution observations with a climate model? This approach provides an alternative to downscaled models for obtaining a long-term QPF and is investigated in this study by coupling high-resolution precipitation observations (~2 km) with a coarser-resolution (~100 km) global climate model (GCM). The result is a 90-yr precipitation outlook at a spatial scale relevant to river basins and catchments that takes into account the effects of changes in global circulation due to climate change. High-quality multiyear radar data are used to estimate future precipitation throughout the twenty-first century for seven major river basins in Switzerland. Previously, the high spatial and temporal resolution of operational radar has been leveraged in development of immediate term precipitation forecasts used in hydrologic “nowcasting” (Germann et al. 2009; Panziera and Germann 2010). Now, we apply the spatial and temporal resolution advantages of radar-estimated precipitation to a long-term QPF.

In this paper, synoptic weather patterns are used as the link between precipitation observations and GCM representation of future climate since the spatial distribution of total daily precipitation has been shown to be related to the associated synoptic weather situation (Lin et al. 2001; Plaut and Simonnet 2001; Rudari et al. 2004; Rudolph et al. 2011). By using future changes in the occurrence of synoptic patterns to quantify future changes in precipitation the dynamic effect of climate change on future precipitation is evaluated. Thermodynamic effects are not addressed in this study; however, it is recognized that a warmer atmosphere is likely to have greater water content (via the Clausius–Clapeyron equation) that can combine with dynamic effects to impact precipitation distribution (Trenberth et al. 2003; Seager et al. 2010).

2. Data and methods

Figure 1 outlines the process used in this paper to calculate expected precipitation for Swiss river basins in the twenty-first century. The following four datasets are used in this analysis:

  1. daily synoptic weather pattern (DATA1 in Fig. 1) between 1948 and 2008 as determined by MeteoSwiss (Swiss Weather Service) per Schüepp’s classification method (Schüepp 1979);

  2. reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis Interim (ERAi) dataset for 2000–08 (DATA2 in Fig. 1);

  3. surface precipitation estimated by MeteoSwiss operational radar network for 2000–08 (DATA3 in Fig. 1); and

  4. output from the National Center for Atmospheric Research (NCAR) Community Climate System Model Version 3.0 (CCSM3) for Intergovernmental Panel on Climate Change (IPCC) scenario A1B for 1948–2099 (DATA4 in Fig. 1; CCSM3 experiment b030.040e, atmosphere postprocessed data, daily averages, version 1; Collins et al. 2006; Solomon et al. 2007) . The IPCC Fourth Assessment Report (AR4) projects scenario A1B will result in a mid-to-upper range global surface temperature increase as compared to other IPCC scenarios for the years 2090–99 as compared to 1980–99 (Alley et al. 2007, 13–14).

Observational and model data from three time frames are referenced in this analysis: 1) historical, referring to the years 1948–99; 2) current, referring to 2000–08; and 3) future, referring to 2010–99.

Fig. 1.

Process for using synoptic weather patterns to couple radar-estimated precipitation with GCMs to determine future precipitation.

Fig. 1.

Process for using synoptic weather patterns to couple radar-estimated precipitation with GCMs to determine future precipitation.

a. Observed synoptic weather patterns

Synoptic-scale weather patterns are identified using Schüepp’s synoptic weather classifications (DATA1 in Fig. 1), which are described in detail by Schüepp (1979), Wanner et al. (1998), and Stefanicki et al. (1998). Schüepp’s weather classification system provides daily categorization of the synoptic pattern in the central Alps. Daily Schüepp classifications generated by MeteoSwiss for the Swiss Alpine area are used in this study for the historical and current timeframes (1948–99 and 2000–08). Schüepp’s classification is a subjective method because it is based on predefined weather patterns. Many different circulation-type classification (CTC) methods are available for classification of synoptic-scale weather patterns. Schiemann and Frei (2010) evaluated the skill of 71 CTC systems in predicting precipitation over the Alpine area and found the Schüepp system in the top 15 out of 71 compared CTC systems for skill in prediction of precipitation occurrence. The CTCs that outperformed the Schüepp system were all automated classification schemes, whereas Schüepp classifications are performed manually. The Schüepp classification system is used for this study because of its long data record dating back to 1948 and specific tailoring to the Alpine region. Although as noted in Schiemann and Frei (2010) MeteoSwiss is in the process of replacing manual Schüepp weather pattern classification with an automated CTC system. However, the general method described in this study for linking high-resolution observations to large-scale circulation from a GCM is readily adaptable to other synoptic classification systems.

The Schüepp system defines 40 synoptic weather patterns, which are grouped into three main classes and eight subclasses (Wanner et al. 1998). The main three classes of the Schüepp system (convective, advective, mixed) are determined by the strength of the surface pressure gradient across a 444-km diameter centered in Switzerland at 46.5°N, 9°E. The convective type is characterized by lack of surface pressure gradient. The advective classification is associated with the existence of a gradient in surface pressure larger than 5 hPa across the diameter. The mixed classification results when surface and 500-hPa pressure gradients are at odds (strong gradient at one level while weak at the other). The main classes are further divided into the following eight subclasses: i) convective—high (CH), flat (CF), and low (CL) (based 500-hPa geopotential height relative to the annual mean); ii) advective—north (AN), south (AS), east (AE), and west (AW), based on the direction of the 500-hPa geostrophic wind; and c) mixed (MIX)—no further subclassification. Schüepp’s eight subclasses (CH, CF, CL, AN, AS, AE, AW, and MIX) are used in this analysis for classification of weather patterns.

b. Observed precipitation distribution

Data from the MeteoSwiss operational radar network for the years 2000–08 are the basis of the observed precipitation distribution used in this analysis (DATA3 in Fig. 1). The radar network includes three C-band Doppler weather radars located at Lema, Albis, and La Dole (Fig. 2). Each of the radars scans 20 elevations every 5 min monitoring radar reflectivity up to a range of 230 km. MeteoSwiss composites reflectivity from the three radars to estimate surface precipitation in a data product called RAIN. RAIN provides a 30-min running average of surface precipitation rate (in units of mm h−1) on a 2 km × 2 km Cartesian grid. Thorough descriptions of the Swiss radar network and data products are found in Joss et al. (1998), Germann and Joss (2004), Germann et al. (2006, 2009), and Panziera and Germann (2010).

Fig. 2.

Location of Swiss radars (stars with names) and river basins (numbered) relative to 800 m above mean sea level topography lines. Basins are shaded gray to differentiate from topography lines.

Fig. 2.

Location of Swiss radars (stars with names) and river basins (numbered) relative to 800 m above mean sea level topography lines. Basins are shaded gray to differentiate from topography lines.

For this study, a precipitation day is defined as a day having two successive occurrences at a 30-min interval where radar-estimated precipitation rate at the surface is greater than 0.5 mm h−1. The magnitude of the precipitation day threshold and the requirement for sequential 30-min measurements above the threshold differentiate it from the limit of 1 mm day−1 used by previous studies to segregate wet and dry days (Groisman and Knight 2008; Zolina et al. 2010). Then, the Schüepp’s classification for each day of 2000–08 is used to determine the probability of precipitation at each pixel given a certain synoptic classification, P(Precip | Classk), where k ranges from one to eight and represents the number of Schüepp classifications used,

 
formula

The amount of precipitation received on a precipitation day, Precipd (mm h−1), at a specific pixel 2 km × 2 km is calculated as

 
formula

where RR is the hourly radar-estimated precipitation rate. Precipd is then associated with the daily synoptic weather classifications from 2000–08 (CALC2 in Fig. 1) to obtain a distribution of Precipd at each pixel for each synoptic classification. A mean value of Precipd is calculated for each synoptic classification at each pixel, (CALC3 in Fig. 1). The calculation of overall daily mean precipitation resulting from each synoptic classification, , accounts for the probability that precipitation occurs:

 
formula

The distribution of Precipd is also used to determine the observed 95th quantile (Q95obs) of daily precipitation (all synoptic classes over the years 2000–08) and the conditional probability of exceeding Q95obs given a specific synoptic classification, P(Q95obs | Classk).

c. Future occurrence of synoptic weather patterns

In this study, changes in large-scale circulation are evaluated by quantifying trends in probability of occurrence of daily synoptic patterns, P(Classk), over time. This requires identification of daily synoptic patterns for the future time period (years 2010–99). Learning vector quantization (LVQ) is the neural network algorithm used in this analysis for identification of Schüepp’s synoptic patterns in CCSM3 for a future climate scenario. The LVQ method uses supervised learning in which data with known classifications are used to classify unknown data (Kohonen 1997). In this case, daily synoptic classifications are known for the current time period (2000–08) but unknown for the future (2010–99).

The first step in LVQ pattern identification is to provide definition for each Schüepp’s classification using known data. Within the LVQ method the definition of each pattern is known as a codebook vector. As stated in section 2a, Schüepp’s synoptic classifications are based on surface pressure and 500-hPa geopotential height. For this study, daily 1000- and 500-hPa geopotential height (Z1000 and Z500, respectively) data are paired with known daily Schüepp’s classifications to create codebook vectors for each synoptic pattern. Specifically, Z1000 and Z500 from daily 12Z ERAi (1.5° × 1.5° lat–lon; DATA2 in Fig. 1) for the years 2000–08 are sorted based on daily Schüepp’s classifications to define each Schüepp’s synoptic classification in terms of Z1000 and Z500 (CALC1 in Fig. 1, note that this is the same time span as radar dataset).

The next step in the LVQ pattern matching algorithm is classification of unknown data. In this step the objective is to obtain daily Schüepp’s classification of CCSM3 data for future climate (CALC4 in Fig. 1). Daily Z1000 and Z500 values from CCSM3 are compared to the codebook vectors for each Schüepp’s classification using a nearest-neighbor approach to determine the best classification for each day in the CCSM3 output. CCSM3 directly provides Z500; however, Z1000 is not a direct output from the model and must be calculated from equivalent sea level pressure (SLP). SLP from CCSM3 [p1 in Eq. (4) below] is converted to Z1000 (Z2) via the simplified hypsometric equation:

 
formula

where scale height H = 7.5 km; p2 = 1000 hPa; and surface height Z1 = 0 m.

The LVQ pattern-matching algorithm requires spatial alignment between data points represented in the codebook vector and data points in the unknown dataset to be classified. However, the ERAi and CCSM3 data are not initially spatially aligned. ERAi data are on a 1.5° × 1.5° latitude–longitude grid and the CCSM3 data are approximately 1.4° × 1.4°. Therefore, both sets of data are interpolated to an Equal Area Scalable Earth grid (EASE grid) with 100-km grid spacing (Brodzik and Knowles 2002). The EASE grid interpolation provides the additional benefit of distributing the data such that the density of points is equivalent at all latitudes rather than being overweighted in northern latitudes as occurs in a standard latitude–longitude grid. Following EASE grid interpolation, data points within ±9° latitude–longitude of 46.5°N and 7.5°E were selected for inclusion in the pattern-matching algorithm.

After determining daily synoptic classifications for 2000–99 from the CCSM3 data, the occurrence of each classification is assessed over time (CALC5 in Fig. 1). The probability of occurrence of each classification, P(Classk), is calculated for each decade of 2000–99:

 
formula

Time-based trends, or lack thereof, in decadal P(Classk) identified from the CCSM3 data provide indication of the effect of climate on large-scale circulation as modeled by CCSM3.

LVQ codebook generation and subsequent pattern identification in CCSM3 is a source of uncertainty in the values of P(Classk) reported for the twenty-first century. The uncertainty arises because the LVQ method utilizes random sampling from the dataset of ERAi geopotential fields associated with each synoptic classification to optimize generation of the codebook used for classification of unknown data (Kohonen 1997). Therefore, different iterations of codebook generation result in variation in the frequency of patterns identified from CCSM3. Uncertainty associated with pattern identification has been incorporated into the results by performing 100 repetitions of the codebook generation and pattern identification process to report mean and standard deviation of synoptic pattern frequency as identified from CCSM3. The variation in synoptic pattern frequency reported here should not, however, be interpreted as evaluation of uncertainty inherent to CCSM3 geopotential height fields as this is not explicitly quantified in this study.

d. Future precipitation for Swiss river basins

To obtain a precipitation distribution for specific river basins, it is necessary to spatially relate pixels in the radar-estimated precipitation data to geographic boundaries of river basins. Coordinates for vertices of polygons that form the outlines of Swiss river basins (generally >1000 km2), and the subcatchments (approximately 30–50 km2) within each basin, were provided by the Hydrology Group, Institute of Geography, University of Bern. The borders of Swiss river basins relative to 800 MSL topography are shown in Fig. 2. To relate the basin borders to the radar-estimated precipitation rate data, the outline of each basin’s subcatchments are compared to the 2 km × 2 km grid of the radar data. Pixels in the precipitation rate data are identified as within a specific subcatchment when greater than 50% of the area represented by a pixel is within the subcatchment border.

Future mean precipitation at each pixel is calculated by

 
formula

where, as previously, k represents the eight Schüepp’s subclassifications. Future mean precipitation for an entire river basin (; CALC6 in Fig. 1) is calculated by summing mean precipitation over all pixels contained within the boundaries of a basin. The future probability of occurrence of each synoptic pattern P(Classk) is the only term in Eq. (6) that varies; therefore, only changes in large-scale circulation (dynamic effects) affect the calculation of future precipitation. is held constant at the value determined from the current time period (2000–08). The assumption that remains constant likely removes potential thermodynamic effects on precipitation such as increased atmospheric water content, changes in atmospheric moisture flux, and changes in soil moisture, which all may affect the average amount of precipitation associated with a specific synoptic pattern. The appendix provides a further analysis of precipitation projection based on changes in synoptic pattern frequencies on the basis of Swiss rain gauge data.

Similar to the expected precipitation calculation, the future occurrence of heavy precipitation is also based on changes in P(Classk). P(Q95f) is used to denote the future probability of exceeding Q95obs and is calculated as follows:

 
formula

Again, only P(Classk) in Eq. (7) is changing for future decades, so changes in the occurrence of synoptic patterns over time, as identified from CCSM3, determine if any variation occurs in P(Q95f) as compared to P(Q95obs).

3. Results and discussion

a. Identification and verification of observed synoptic patterns

ERAi Z1000 and Z500 for the eight Schüepp’s synoptic patterns averaged over 2000–08 are shown in Fig. 3. The geopotential height data represented in these maps were used to generate codebook vectors for identifying patterns in CCSM3 output (CALC1 in Fig. 1). Table 2 shows the frequency of occurrence and number of days that were averaged for each classification to obtain Fig. 3. During 2000–08 convective synoptic patterns were most frequent, followed by advective patterns. All eight of the Schüepp’s classifications were observed a significant number of times, ranging from 135 to 940 daily occurrences. The years 2000–08 were chosen so that the synoptic pattern codebook definitions are temporally aligned with the radar data used as the basis for the precipitation distribution. As shown in Fig. 3 convective patterns (CH, CF, and CL) exhibit a minimal gradient in Z1000 across Switzerland, whereas geopotential height gradients are more prominent in the advective patterns (AN, AE, AS, and AW). Additionally, the cardinal direction (north, south, east, or west) in the name of each advective pattern is generally consistent with the direction of the 500-hPa geostrophic wind, that is, the 500-hPa geostrophic wind for the AN classification has a northerly component. Overall, the geopotential height contours in Fig. 3 appear as expected based on the Schüepp subclassification definitions. Therefore, Fig. 3 provides qualitative verification that the ERAi geopotential height data and the daily Schüepp classifications were accurately paired to generate the LVQ codebook vectors.

Fig. 3.

Geopotential heights at 1000 (color coded) and 500 hPa (solid black lines at 4-hPa intervals) averaged over 2000–08 for the main Schüepp’s weather classifications: CH, CF, CL, MIX, AN, AE, AS, and AW. Plots are based on ERAi dataset.

Fig. 3.

Geopotential heights at 1000 (color coded) and 500 hPa (solid black lines at 4-hPa intervals) averaged over 2000–08 for the main Schüepp’s weather classifications: CH, CF, CL, MIX, AN, AE, AS, and AW. Plots are based on ERAi dataset.

Table 2.

Observed P(Classk) for each Schüepp’s synoptic classification and number of days each classification occurred over 2000–08 (from MeteoSwiss daily Schüepp’s classifications). It is noted that the total number of days (3184 days) used in this analysis is less than the actual number of days between 2000–08 (3288 days) due to omission of some days because of radar data quality or availability.

Observed P(Classk) for each Schüepp’s synoptic classification and number of days each classification occurred over 2000–08 (from MeteoSwiss daily Schüepp’s classifications). It is noted that the total number of days (3184 days) used in this analysis is less than the actual number of days between 2000–08 (3288 days) due to omission of some days because of radar data quality or availability.
Observed P(Classk) for each Schüepp’s synoptic classification and number of days each classification occurred over 2000–08 (from MeteoSwiss daily Schüepp’s classifications). It is noted that the total number of days (3184 days) used in this analysis is less than the actual number of days between 2000–08 (3288 days) due to omission of some days because of radar data quality or availability.

We base the calculation of future precipitation on the occurrence of synoptic patterns in CCSM3 data. Therefore, it is important to verify our identification of Schüepp’s classifications from CCSM3. Verification is provided by comparing the occurrence of synoptic patterns as identified from CCSM3 to the observed occurrence of synoptic patterns over a known time period. The comparison establishes a bias between model and observations for the occurrence of each synoptic pattern:

 
formula

Model P(Classk) in Eq. (8) is the frequency that each Schüepp’s classification occurs in daily CCSM3 data for the historical period 1948–99 as identified using LVQ. Observed P(Classk) is the frequency that each Schüepp’s synoptic pattern occurred during 1948–99 per daily classifications provided by MeteoSwiss. The bias accounts for errors introduced during synoptic pattern identification from CCSM3 and also the ability of CCSM3 to represent the occurrence of observed historical synoptic patterns over Switzerland. The over 50-yr time span used as the basis for bias calculation ensures longer-term climate cycles that impact large-scale atmospheric circulation, such as El Niño/La Niña–Southern Oscillation (ENSO) and the North Atlantic Oscillation (NAO), with periods of 6–12 years, are adequately captured in the comparison.

The biases for the eight Schüepp’s classifications range from 0.20 for AE to 1.92 for AW (Table 3). Z1000 and Z500 from CCSM3 output for 1948–99 show similar spatial patterns for the eight Schüepp’s classifications as in Fig. 3 (not shown) indicating that model data are properly classified by the LVQ method. Since synoptic patterns appear to be correctly identified in the model data, the bias appears to be primarily attributed to the ability of the model to reproduce the occurrence of synoptic patterns at the frequency with which they were observed over 1948–99. The source of the biases between modeled and observed synoptic pattern frequencies was further investigated by using the LVQ method with the codebook developed from 2000–08 ERAi data to identify Schüepp’s classifications in 1980–99 ERAi data. The biases for identification of synoptic patterns from the 1980–99 ERAi ranged from 0.73 for AE to 1.11 for CH, a smaller range than for the CCSM3 data. This indicates that the LVQ-identified occurrence of synoptic patterns in ERAi is more representative of observations than the occurrence of synoptic patterns in CCSM3.

Table 3.

Bias between P(Classk) in observations (from MeteoSwiss daily Schüepp’s classifications) and model (as identified in CCSM3 by LVQ) for Schüepp’s synoptic weather classifications over the years 1948–99.

Bias between P(Classk) in observations (from MeteoSwiss daily Schüepp’s classifications) and model (as identified in CCSM3 by LVQ) for Schüepp’s synoptic weather classifications over the years 1948–99.
Bias between P(Classk) in observations (from MeteoSwiss daily Schüepp’s classifications) and model (as identified in CCSM3 by LVQ) for Schüepp’s synoptic weather classifications over the years 1948–99.

Comparison of current (2000–08, Table 2) and historical (1948–99, Table 3) synoptic pattern frequencies shows the current time span has a decreased frequency of advective patterns and an increased frequency of convective patterns. The observed frequency of advective patterns [observed P(Classk) in Tables 2 and 3] decreased from 0.42 (1948–99) to 0.36 (2000–08) while the frequency of convective patterns increased from 0.52 (1948–99) to 0.57 (2000–08). The observation that advective patterns are decreasing in frequency as convective patterns are increasing is a result similar to previously reported trends (Stefanicki et al. 1998; Rudolph et al. 2011). The decrease in advective pattern frequency is relatively uniform across the four advective subclassifications (AN, AE, AS, and AW) as the occurrence of each pattern decreased by 1%–2%. However, the overall change in occurrence of convective patterns is not equally distributed among subclassifications CL, CF, and CH. Contrary to the overall increased frequency of convective patterns, the occurrence of CL declined by 1%. CF frequency increased by 2%, but the main contribution to the observed increase in frequency of convective patterns is the 5% increase in the occurrence of CH. The increased occurrence of CH indicates that anticyclonic patterns over Switzerland were more common during 2000–08 than in the previous 50 years (1948–99). The increased occurrence of CH also implies warming in the atmosphere between the surface and 500 hPa. By definition, days classified as CH have Z500 in the upper 25th quantile of historical Z500 values. Increased average temperature between the surface and 500 hPa would lead to a corresponding shift toward higher Z500, and therefore, more days classified as CH.

b. Current precipitation for each synoptic pattern

Figure 4 shows mean radar-estimated daily precipitation for each synoptic classification, , received over the years 2000–08 for seven of the nine Swiss river basins (CALC3 in Fig. 1). The Inn and Adige basins are not included in the analysis because topography significantly affects radar visibility and accuracy of radar-estimated precipitation for these basins. For each basin, the largest value of mean precipitation occurs during advective weather patterns having geostrophic winds, and therefore moisture flux, directed toward the Alps. For example, in the Reuss basin located on the north side and close to the main crest of the Alps, moisture flux advected from the north and west (AW and AN classifications) results in the greatest daily mean values of precipitation (>5 mm). The AN and AW classifications produce the largest mean values of daily precipitation for the Reuss basin because northerly and westerly winds generate orographic lift on the predominantly northwest aspects of this basin (the Schüepp system does not explicitly define an advective northwest classification). This is supported by previous work that found the AN classification results in the highest daily mean value of precipitation, and the AW pattern results in the highest 95th quantile of daily precipitation for the northern Alps (Rudolph et al. 2011). Similarly, Fig. 4 indicates that southerly flow experienced during AS patterns produces the most precipitation in the south-facing Ticino and Adda basins. This agrees well with Sodemann and Zubler (2010) who showed that because of the orographic barrier presented by the Alpine crest the North Atlantic is the main moisture source for the Northern Alps, and the Mediterranean is the main moisture source for the Southern Alps. Also notable in Fig. 4 is the observation that all basins follow a similar trend for the convective patterns. Among the convective patterns, the CL pattern produces the most precipitation, followed by CF, and then CH. The CL pattern is associated with low pressure centered over Switzerland, CF has midtropospheric zonal flow, and CH is associated with high pressure (Fig. 3).

Fig. 4.

Radar-estimated daily precipitation () for 7 Swiss river basins over 2000–08 by Schüepp’s synoptic weather classification. Boxes indicate mean values, connected bars indicate interquartile range (25th–75th quantile), and diamonds indicate 10th and 90th quantiles.

Fig. 4.

Radar-estimated daily precipitation () for 7 Swiss river basins over 2000–08 by Schüepp’s synoptic weather classification. Boxes indicate mean values, connected bars indicate interquartile range (25th–75th quantile), and diamonds indicate 10th and 90th quantiles.

In addition to each synoptic classification resulting in different mean precipitation totals, the synoptic classifications also have varying probabilities of producing heavy precipitation (Fig. 5). The probability of exceeding Q95obs for each synoptic classification shows a similar pattern as mean daily precipitation. The basins in the northern and western Alps generally have a higher P(Q95obs | class) for AN and AW patterns, while the basins that are primarily south-facing have higher P(Q95obs | class) for the AS pattern. Exceptions to this are the Rhone, Aare, and Rhein basins that lie to the north and west of the main Alpine crest, yet have elevated P(Q95obs | class) for the AS pattern. In the Aare basin, the AS pattern results in the largest P(Q95obs | class). Therefore, although moisture advected from the west (AW pattern) results in the highest values of mean daily precipitation for the north- and west-facing Rhone, Aare, and Rhein basins, the relatively high values of P(Q95obs | class) for the AS pattern provide evidence of the importance of the Mediterranean as a moisture source for heavy precipitation events in these basins.

Fig. 5.

Conditional probability of exceeding 95th quantile of daily radar-estimated precipitation in 7 Swiss river basins for the years 2000–08 given Schüepp’s daily synoptic classification (classification abbreviations as in Fig. 3).

Fig. 5.

Conditional probability of exceeding 95th quantile of daily radar-estimated precipitation in 7 Swiss river basins for the years 2000–08 given Schüepp’s daily synoptic classification (classification abbreviations as in Fig. 3).

c. Future synoptic pattern frequency and precipitation

Figure 6 shows mean and standard deviation for decadal frequency of Schüepp’s synoptic pattern classifications P(Classk) over the twenty-first century as identified from CCSM3 output. The mean and standard deviation values reported in Fig. 6 result from 100 replications of LVQ codebook generation and pattern identification (CALC1 and CALC4 in Fig. 1). The time windows used to define decades were also considered as a source of variation in the decadal synoptic pattern frequencies. Plots of annual synoptic pattern frequency (not shown) qualitatively reveal that the decadal trends identified in Fig. 6 are persistent, even on an annual basis, throughout the 100-yr period without substantial oscillation. Therefore, it was deemed unnecessary to perform further analysis on pattern frequency with varying decadal time boundaries.

Fig. 6.

Decadal frequency of Schüepp’s synoptic pattern classifications over the twenty-first century for A1B climate scenario as identified from CCSM3 output. Decades are labeled as 2000 for the years 2000–09, 2010 for the years 2010–19, etc. P(Classk) has been adjusted by the bias for each synoptic pattern that appears in Table 3. Mean (boxes) and three standard deviations (bars) result from 100 replications of LVQ codebook generation and pattern identification (CALC1 and CALC4 in Fig. 1). Classes where lines are present (CL, CF, and CH) have linear slopes significant at 95% confidence (as in Table 4).

Fig. 6.

Decadal frequency of Schüepp’s synoptic pattern classifications over the twenty-first century for A1B climate scenario as identified from CCSM3 output. Decades are labeled as 2000 for the years 2000–09, 2010 for the years 2010–19, etc. P(Classk) has been adjusted by the bias for each synoptic pattern that appears in Table 3. Mean (boxes) and three standard deviations (bars) result from 100 replications of LVQ codebook generation and pattern identification (CALC1 and CALC4 in Fig. 1). Classes where lines are present (CL, CF, and CH) have linear slopes significant at 95% confidence (as in Table 4).

The CCSM3 scenario A1B data show increasing frequency of convective high (CH), decreasing frequency of convective flat (CF), and slightly decreasing frequency of convective low (CL) patterns over the twenty-first century (Fig. 6 and Table 4). The increased occurrence of the CH pattern and decreased occurrence of the CL pattern in CCSM3 data for 2000–99 indicate a continuation of trends in convective pattern frequencies that were identified over 1948–2008 as discussed in section 3a. The advective pattern frequencies in CCSM3 remain relatively constant as none of the advective patterns indicate a significant trend. This differs from the observation of decreased advective pattern occurrence over 1948–2008. However, it is noted that the observed decrease in occurrence of advective patterns over 1948–2008 was relatively small (1%–2%). Also, three (AN, AE, and AS) of the four advective subclassifications have negative slopes over 2000–99 in CCSM3, although none are significant at 95% confidence (Table 4). Also consistent with 1948–2008 the mixed pattern shows a slight tendency toward increased frequency for 2000–99 in CCSM3 data, but the trend is not significant at 95% confidence.

Table 4.

Sign of slope and coefficient of determination (R2) of linear fit to frequency vs decade data appearing in Fig. 5. Bold font indicates patterns with slopes that are significant at 95% confidence.

Sign of slope and coefficient of determination (R2) of linear fit to frequency vs decade data appearing in Fig. 5. Bold font indicates patterns with slopes that are significant at 95% confidence.
Sign of slope and coefficient of determination (R2) of linear fit to frequency vs decade data appearing in Fig. 5. Bold font indicates patterns with slopes that are significant at 95% confidence.

The identified changes in pattern frequency combined with mean precipitation for each pattern subsequently result in a projected 10%–15% decline in total precipitation for Swiss river basins by the decade 2090–99 as compared to 2000–09 (Fig. 7). In addition to basinwide total precipitation, Fig. 7 also shows the range of total precipitation for each basin’s subcatchments. The high-resolution nature of the radar-estimated precipitation data used as a basis for this study enables calculation of future precipitation at the subcatchment (30–50 km2) and even pixel (4 km2) scale should this information be useful as an input to hydrologic models. Standard deviation of decadal precipitation for each basin (Fig. 7) is based on the combination of variance in radar-estimated mean daily precipitation for each synoptic pattern (Fig. 4) and variance due to synoptic pattern identification (Fig. 6). As previously stated in section 2c, the uncertainty present in CCSM3 geopotential height fields has not been quantified for this study and is, therefore, not a component of standard deviations shown in Figs. 7 and 8.

Fig. 7.

Decadal total precipitation over the twenty-first century normalized to 2000–09 for each Swiss river basin. Decades labeled as in Fig. 6. Individual values of normalized precipitation for each basin’s subcatchments are noted as triangles, overall basin average is noted as a diamond. Bars indicate three standard deviations around the basin mean. Standard deviation is calculated from the combined variance of radar estimated daily precipitation (Fig. 4) and synoptic pattern identification (Fig. 6). Kendall–Thiel linear fit is applied to the overall basin averages. In all basins the slope of the linear fit is significant with 95% confidence.

Fig. 7.

Decadal total precipitation over the twenty-first century normalized to 2000–09 for each Swiss river basin. Decades labeled as in Fig. 6. Individual values of normalized precipitation for each basin’s subcatchments are noted as triangles, overall basin average is noted as a diamond. Bars indicate three standard deviations around the basin mean. Standard deviation is calculated from the combined variance of radar estimated daily precipitation (Fig. 4) and synoptic pattern identification (Fig. 6). Kendall–Thiel linear fit is applied to the overall basin averages. In all basins the slope of the linear fit is significant with 95% confidence.

Fig. 8.

Decadal future probability over 2000–99 (indicated by diamonds) of exceeding the 95th quantile of daily precipitation established for 2000–08 for each basin. Decades labeled as in Fig. 6. Connected bars indicate three standard deviations around P(Q95obs) and are calculated from the variance of synoptic pattern identification (Fig. 6). A Kendall–Thiel linear fit is shown for basins where slope is significant at 95% confidence.

Fig. 8.

Decadal future probability over 2000–99 (indicated by diamonds) of exceeding the 95th quantile of daily precipitation established for 2000–08 for each basin. Decades labeled as in Fig. 6. Connected bars indicate three standard deviations around P(Q95obs) and are calculated from the variance of synoptic pattern identification (Fig. 6). A Kendall–Thiel linear fit is shown for basins where slope is significant at 95% confidence.

A Kendall–Thiel linear fit applied to decadal precipitation in each basin results in negative slopes for all basins that are significant with 95% confidence. A linear fit is applied only for testing significance of the trend in precipitation and is not intended to imply that precipitation is expected to decrease linearly over time. The pattern frequency changes that have the greatest impact on total precipitation are an increase in the CH pattern which produced the least precipitation in all river basins over 2000–08 and decreases in both the CL and CF patterns. Currently (2000–08) the CL and CF patterns both produce more precipitation than the CH pattern and some advective patterns (Fig. 4). The respective changes in frequency combined with mean precipitation for each classification result in a reduction in total precipitation due to dynamic changes in synoptic scale circulation. The increased frequency of the CH pattern and associated decrease in precipitation for Swiss river basins is consistent with IPCC AR4 that finds an expected decrease in precipitation in central Europe due to increased anticyclonic flow (Solomon et al. 2007). Seager et al. (2010) studied 15 GCMs used in IPCC AR4 and also found dynamic changes have a negative effect on mean zonal precipitation minus evaporation (PE) at 45°N latitude as a result of poleward movement of storm tracks due to Hadley cell expansion.

Although total precipitation decreases as a result of changes in synoptic pattern frequencies, a corresponding change in the probability of heavy precipitation is not evident (Fig. 8). The future probability of exceeding Q95obs does not appear to change significantly because of the dynamic effects over the twenty-first century (Fig. 8). A Kendall–Thiel linear fit was applied to the data in Fig. 8, and only the Limmat and Adda basins have slopes that are significant at 95% confidence. Even for these basins, the probability of exceeding Q95obs only declines from 5.0% in 2000–09 to 4.7%–4.8% by the decade of 2090–99. The absence of a dynamic effect on heavy precipitation is due to the fact that the relative differences among P(Q95obs | class) for the various synoptic patterns is less than the relative differences in mean precipitation resulting from each pattern. A constant probability of heavy precipitation while total precipitation is declining implies that heavy precipitation events will account for a greater overall proportion of total precipitation in the twenty-first century, a trend that has, in fact, already been observed in historical precipitation observations for Europe (Klein Tank and Können 2003).

The effect of changes in atmospheric moisture content on precipitation is not quantitatively addressed in this study. However, Trenberth et al. (2003) suggest that precipitation intensity may increase at the same rate as the warming-induced increase in moisture content of the atmosphere, or approximately 7% K−1, as set by the Clausius–Clapeyron equation. Simulations from the World Climate Research Programme’s (WCRP’s) Coupled Model Intercomparison Project phase 3 (CMIP3) indicate that column water vapor and surface specific humidity change with global warming at rates of ~7.4% K−1 (range 6%–12% K−1) and 5.9% K−1 (range 5%–12% K−1), respectively (O’Gorman and Muller 2010). Furthermore, Muller et al. (2011) find that precipitation extremes follow the rate of increase in surface water vapor concentration under global warming. The Met Office Hadley Centre and Climatic Research Unit Global Surface Humidity dataset (HadCRUH) shows that surface specific humidity in the European Alpine area increased over 1973–2003 at a rate of 0.1–0.2 g kg−1 decade−1 (Willett et al. 2008). Therefore, although our analysis indicates that the probability of dynamically forced heavy precipitation events will remain constant, the precipitation rates experienced during these events may increase due to thermodynamic effects. An intensification of heavy precipitation events would effectively shift the precipitation distribution and result in increased probability of exceeding Q95obs.

In contrast to the 10 000 + radar pixels that represent Swiss river basins in MeteoSwiss radar data, only three grid points (latitude 46.92°N and longitudes 7.03°, 8.44°, and 9.84°E) represent the area of Switzerland in CCSM3. CCSM3 reports large-scale and convective precipitation; these are summed to calculate total precipitation. Although we do not quantitatively assess uncertainty in the GCM precipitation output, Hawkins and Sutton (2011, hereafter HS11) conclude that GCM precipitation has three sources of uncertainty: 1) internal variability, or random variations in long-term climate parameters within a single model; 2) model uncertainty from variations between different models; and 3) scenario uncertainty due to uncertainty in radiative forcing. Furthermore, HS11 find the relative contributions of the three sources of variability change as the model projects further into the future. Internal variability is the largest source of uncertainty during the first 1–3 decades. Model uncertainty becomes the largest contributor to uncertainty in GCM precipitation output projections beyond 30 years. Scenario uncertainty has minimal contribution to GCM precipitation uncertainty (HS11).

No trends in total, large-scale, or convective precipitation are evident in the precipitation output from CCSM3 in the general region of Swiss river basins over the twenty-first century (Fig. 9). Therefore, although CCSM3 shows changes in decadal frequencies of synoptic patterns, specifically convective patterns, it does not indicate changes in total decadal precipitation. This may be attributed to thermodynamic effects counteracting dynamic effects, or it may simply be due to the fact that convective precipitation occurs at spatial scales much less than the resolution of the GCM, particularly over mountainous terrain. However, as evidenced in this study, the frequency of convective synoptic patterns influences precipitation distribution over Swiss river basins. This highlights the importance for computational models to accurately resolve convective precipitation under varying synoptic conditions to adequately represent future precipitation trends.

Fig. 9.

Total decadal precipitation over Swiss Alpine area indicated by direct CCSM3 output for scenario A1B: total precipitation (large diamonds are average of three CCSM3 grid points, small diamonds are individual CCSM3 grid points), large-scale precipitation (squares), and convective precipitation (triangles).

Fig. 9.

Total decadal precipitation over Swiss Alpine area indicated by direct CCSM3 output for scenario A1B: total precipitation (large diamonds are average of three CCSM3 grid points, small diamonds are individual CCSM3 grid points), large-scale precipitation (squares), and convective precipitation (triangles).

4. Conclusions

This paper describes one of the first studies to combine radar-estimated precipitation observations in mountainous terrain with a GCM to describe the effects of climate change on future precipitation. Synoptic weather patterns have been used as a link between radar data and GCM output to determine expected trends in precipitation for Swiss river basins. This approach to understanding the impact of climate change on precipitation is unique because it provides an additional perspective to downscaled climate models for locations where sufficient radar data are available. The use of radar-estimated precipitation observations with high spatial resolution (2 km × 2 km) enables development of river basin–specific analyses that account for terrain and localized climate variation, yet are not reliant on interpolation of sparse data.

Association of synoptic patterns to radar-based precipitation data reveals locational-specific differences in the amount of precipitation expected for each synoptic pattern. Swiss river basins located in the northern and western Alps receive more precipitation when moisture is advected from the north or west, and moisture advected from the south increases precipitation totals for basins in southern Switzerland. Convective synoptic patterns, defined by minimal surface pressure gradient, consistently follow the same trend for expected precipitation across all basins with an anticyclonic pattern resulting in the least amount of precipitation and the cyclonic pattern producing the most precipitation. It is the difference in precipitation received from the various convective patterns that has the greatest implication for future precipitation in Swiss river basins.

Future precipitation has been quantified within this study by identifying future changes in the frequency of synoptic patterns while holding constant the amount of precipitation expected from each pattern. For IPCC climate scenario A1B the results of this study indicate that dynamic changes in atmospheric circulation over the twenty-first century contribute to decreases of approximately 10%–15% in total decadal precipitation for Swiss river basins. Scenarios other than A1B would likely produce varying results. The reduction in precipitation is primarily attributed to increased probability of convective high, or anticyclonic patterns, positioned over Switzerland. Since the anticyclonic pattern is observed to produce less precipitation than other synoptic patterns, an increase in its frequency results in a trend of reduced precipitation for Swiss river basins. The probability of heavy precipitation or total daily precipitation that exceeds the 95th quantile established for 2000–08, does not appear to be affected by dynamic changes over the twenty-first century. Therefore, total precipitation is expected to decrease, but heavy precipitation events account for a larger proportion of the total.

It is noted that previous work has shown thermodynamic effects on precipitation may counterbalance dynamic effects at midlatitudes (Seager et al. 2010). Thermodynamic effects from potential changes in water content are not addressed here, so this presents an opportunity for future work. A suggested approach to quantifying thermodynamic effects is to determine the dependence of the observed precipitation distribution on specific humidity, atmospheric moisture flux, and soil moisture. These parameters, either derived from a GCM or other model, could then be used as an additional conditional variable for a long-term QPF to assess thermodynamic effects.

Acknowledgments

This research was supported by NSF Grant AGS-0937035 and the University of Colorado at Boulder Department of Atmospheric and Oceanic Sciences. The author would like to thank MeteoSwiss for providing operational radar and rain gauge data products; the European Centre for Medium-Range Weather Forecasts (ECMWF) for providing free internet access to ERA Interim reanalysis data; the Hydrology Group, Institute of Geography, and University of Bern for providing Swiss river basin coordinates; Matthew Higgins of the University of Colorado for information and IDL code related to EASE grid interpolation; and three anonymous reviewers for valuable comments that led to the improvement of this manuscript. This research uses data provided by the Community Climate System Model project (www.ccsm.ucar.edu), supported by the Directorate for Geosciences of the National Science Foundation and the Office of Biological and Environmental Research of the U.S. Department of Energy.

APPENDIX

Synoptic Patterns as Precipitation Predictor: Rain Gauge Basis

A modified version of the method used to project twenty-first-century precipitation for Swiss river basins based on radar-estimated precipitation data (Fig. 1) was also applied to rain gauge data to evaluate the use of synoptic patterns as a precipitation predictor with the assumption that mean precipitation received during each synoptic pattern remains constant. First, precipitation recorded at 63 Swiss rain gauges for the eight main Schüepp synoptic classifications: AN, AE, AS, AW, CH, CF, CL, MIX (as described in section 2a) was evaluated for the 10-yr period of 1989–98. P(Precip | Classk) was determined following Eq. (1) where precipitation days are defined as days when the rain gauge recorded at least 1 mm of total precipitation. For this case, , as appears in Eq. (3), is the mean daily precipitation received at each rain gauge during days when precipitation occurred for each synoptic classification. The observed frequency of each synoptic pattern over 1999–2008 P(Classk) was used in Eq. (6) to predict total precipitation for 1999–2008. This allows comparison of observed precipitation received at each rain gauge to precipitation predicted by synoptic pattern frequency and the mean precipitation distribution for each synoptic pattern (Fig. A1). Figure A1 shows observed versus predicted total precipitation for the 10-yr period of 1999–2008. The data are distributed around the y = x line and have a mean value of predicted – observed (pred – obs) equal to −441 mm with a 95% confidence interval of −1655 to 773 mm. Therefore, the hypothesis that predicted precipitation equals observed precipitation (pred − obs = 0) cannot be rejected, so no bias is found between predicted and observed precipitation. However, performing the analysis on the basis of persistence shows similar results. For the persistence prediction, observed total precipitation over 1989–98 is used as a predictor for precipitation during 1999–2008. For persistence, the mean value of pred – obs is equal to −588 mm with a 95% confidence interval of −1763 to 587 mm. Therefore, for this case, persistence from the previous decade is also an unbiased precipitation predictor. The similarity in results for the synoptic pattern–based prediction and persistence prediction may be attributed to lack of significant changes in synoptic pattern frequency between the decades used for prediction and observation (Table A1). If the prediction and observation periods contained more substantial differences in synoptic pattern frequencies, it is possible that the outcomes would differ, both for the synoptic pattern-based prediction, as well as the persistence prediction. However, the main point of performing the rain gauge–based analysis is to provide an indication of the extent that mean precipitation associated with synoptic patterns varies over decadal time scales. Although, we cannot claim it remains constant in this case, we do not find evidence that it changes enough over the 20-yr period of 1989–2008 to induce bias in the synoptic pattern–based decadal precipitation prediction. One notable difference between the rain gauge analysis and the radar-based analysis described in the main text is that rain gauges provide point measurements of precipitation whereas the radar data are spatially integrated over river basins. Since precipitation in complex terrain typically exhibits high spatial variation, it is likely that the spatial integration incorporated in the radar-based approach reduces variability in the difference between observed and predicted precipitation.

Fig. A1.

Observed versus predicted total precipitation for total precipitation over the 10-yr period of 1999–2008 based on rain gauge data from 63 Swiss rain gauges. The dashed line is y = x.

Fig. A1.

Observed versus predicted total precipitation for total precipitation over the 10-yr period of 1999–2008 based on rain gauge data from 63 Swiss rain gauges. The dashed line is y = x.

Table A1.

Frequency of Schüepp’s synoptic weather classifications P(Classk) for the 10-yr periods 1989–99 and 1999–2008, and the frequency difference between the two periods (Δ).

Frequency of Schüepp’s synoptic weather classifications P(Classk) for the 10-yr periods 1989–99 and 1999–2008, and the frequency difference between the two periods (Δ).
Frequency of Schüepp’s synoptic weather classifications P(Classk) for the 10-yr periods 1989–99 and 1999–2008, and the frequency difference between the two periods (Δ).

REFERENCES

REFERENCES
Alley
,
R.
, and
Coauthors
,
2007
:
Summary for policymakers
.
Climate Change 2007: The Physical Science Basis,
S. Solomon et al., Eds., Cambridge University Press, 1–18
.
Brodzik
,
M. J.
, and
K. W.
Knowles
,
2002
:
EASE-Grid: A versatile set of equal-area projections and grids
.
Discrete Global Grids, M. F. Goodchild and A. J. Kimerling, Eds., National Center for Geographic Information and Analysis. [Available online at http://www.ncgia.ucsb.edu/globalgrids-book/ease_grid/.
]
Collins
,
W. D.
, and
Coauthors
,
2006
:
The Community Climate System Model, version 3 (CCSM3)
.
J. Climate
,
19
,
2122
2143
.
Germann
,
U.
, and
J.
Joss
,
2004
:
Operational measurement of precipitation in mountainous terrain
.
Weather Radar: Principles and Advanced Applications, P. Meischner, Ed., Physics of Earth and Space Environments, Springer-Verlag, 52–77
.
Germann
,
U.
,
G.
Galli
,
M.
Boscacci
, and
M.
Bolliger
,
2006
:
Precipitation measurement in a mountainous region
.
Quart. J. Roy. Meteor. Soc.
,
132
,
1669
1692
.
Germann
,
U.
,
M.
Berenguer
,
D.
Sempere-Torres
, and
M.
Zappa
,
2009
:
REAL—Ensemble radar precipitation estimation for hydrology in a mountainous region
.
Quart. J. Roy. Meteor. Soc.
,
135
,
445
456
.
Gilleland
,
E.
, and
R. W.
Katz
,
2011
:
New software to analyze how extremes change over time
.
Eos, Trans. Amer. Geophys. Union, 92, doi:10.1029/2011EO020001
.
Groisman
,
P. Ya.
, and
R. W.
Knight
,
2008
:
Prolonged dry episodes over the conterminous United States: New tendencies emerging during the last 40 years
.
J. Climate
,
21
,
1850
1862
.
Hawkins
,
E.
, and
R.
Sutton
,
2011
:
The potential to narrow uncertainty in projections of regional precipitation change
.
Climate Dyn.
,
37
,
407
418
.
Hirsch
,
R. M.
,
2011
:
A perspective on nonstationarity and water management
.
J. Amer. Water Resour. Assoc.
,
47
,
436
446
.
Joss
,
J. S.
, and
Coauthors
,
1998
:
Operational Use of Radar for Precipitation Measurements in Switzerland
.
Hochschulverlag AG, 108 pp
.
Klein Tank
,
A. M. G.
, and
G. P.
Können
,
2003
:
Trends in indices of daily temperature and precipitation extremes in Europe, 1946–99
.
J. Climate
,
16
,
3665
3680
.
Kohonen
,
T.
,
1997
:
Self-Organizing Maps
.
2nd ed. Springer, 426 pp
.
Lin
,
Y. L.
,
S.
Chiao
,
T. A.
Wang
,
M. L.
Kaplan
, and
R. P.
Weglarz
,
2001
:
Some common ingredients for heavy orographic rainfall
.
Wea. Forecasting
,
16
,
633
660
.
Machguth
,
H.
,
F.
Paul
,
S.
Kotlarski
, and
M.
Hoelzle
,
2009
:
Calculating distributed glacier mass balance for the Swiss Alps from regional climate model output: A methodical description and interpretation of the results
.
J. Geophys. Res.
,
114
,
D19106
,
doi:10.1029/2009JD011775
.
Milly
,
C.
, and
Coauthors
,
2008
:
Stationary is dead: Whither water management
.
Science
,
318
,
573
574
.
Muller
,
C. J.
,
P. A.
O’Gorman
, and
L. E.
Back
,
2011
:
Intensification of precipitation extremes with warming in a cloud-resolving model
.
J. Climate
,
24
,
2784
2800
.
O’Gorman
,
P. A.
, and
C. J.
Muller
,
2010
:
How closely do changes in surface and column water vapor follow Clausius–Clapeyron scaling in climate change simulations?
Environ. Res. Lett.
,
5
,
025207
,
doi:10.1088/1748-9326/5/2/025207
.
Panziera
,
L.
, and
U.
Germann
,
2010
:
The relation between airflow and orographic precipitation on the southern side of the Alps as revealed by weather radar
.
Quart. J. Roy. Meteor. Soc.
,
136
,
222
238
.
Plaut
,
G.
, and
E.
Simonnet
,
2001
:
Large-scale circulation classification, weather regimes, and local climate over France, the Alps, and western Europe
.
Climate Res.
,
17
,
303
324
.
Ranzi
,
R.
,
M.
Zappa
, and
B.
Bacchi
,
2007
:
Hydrological aspects of the Mesoscale Alpine Programme: Findings from field experiments and simulations
.
Quart. J. Roy. Meteor. Soc.
,
133
,
867
880
.
Rudari
,
R.
,
D.
Entekhabi
, and
G.
Roth
,
2004
:
Terrain and multiple-scale interactions as factors in generating extreme precipitation events
.
J. Hydrometeor.
,
5
,
390
404
.
Rudolph
,
J. V.
,
K.
Friedrich
, and
U.
Germann
,
2011
:
Relationship between radar-estimated precipitation and synoptic weather patterns in the European Alps
.
J. Appl. Meteor. Climatol.
,
50
,
944
957
.
Schiemann
,
R.
, and
C.
Frei
,
2010
:
How to quantify the resolution of surface climate by circulation types: An example for Alpine precipitation
.
Phys. Chem. Earth
,
35
,
403
410
.
Schmidli
,
J.
,
C. M.
Goodess
,
C.
Frei
,
M. R.
Haylock
,
Y.
Hundecha
,
J.
Ribalaygua
, and
T.
Schmith
,
2007
:
Statistical and dynamical downscaling of precipitation: An evaluation and comparison of scenarios for the European Alps
.
J. Geophys. Res.
,
112
,
D04105
,
doi:10.1029/2005JD007026
.
Schüepp
,
M.
,
1979
:
Klimatologie der Schweiz, Band III: Witterungsklimatologie
.
Schweizerische Meteorologische Anstalt, 93 pp
.
Seager
,
R.
,
N.
Naik
, and
G. A.
Vecchi
,
2010
:
Thermodynamic and dynamic mechanisms for large-scale changes in the hydrological cycle in response to global warming
.
J. Climate
,
23
,
4651
4668
.
Smiatek
,
G.
,
H.
Kunstmann
,
R.
Knoche
, and
A.
Marx
,
2009
:
Precipitation and temperature statistics in high-resolution regional climate models: Evaluation for the European Alps
.
J. Geophys. Res.
,
114
,
D19107
,
doi:10.1029/2008JD011353
.
Sodemann
,
H.
, and
E.
Zubler
,
2010
:
Seasonal and interannual variability of the moisture sources for Alpine precipitation during 1995–2002
.
Int. J. Climatol.
,
30
,
947
961
.
Solomon
,
S.
,
D.
Qin
,
M.
Manning
,
M.
Marquis
,
K.
Averyt
,
M. M. B.
Tignor
,
H. L.
Miller
Jr.
, and
Z.
Chen
, Eds.,
2007
:
Climate Change 2007: The Physical Science Basis
.
Cambridge University Press, 996 pp
.
Stefanicki
,
G.
,
P.
Talkner
, and
R. O.
Weber
,
1998
:
Frequency changes of weather types in the Alpine region since 1945
.
Theor. Appl. Climatol.
,
60
,
47
61
.
Trenberth
,
K. E.
,
A.
Dai
,
R. M.
Rasmussen
, and
D. B.
Parsons
,
2003
:
The changing character of precipitation
.
Bull. Amer. Meteor. Soc.
,
84
,
1205
1217
.
Wanner
,
H.
,
E.
Salvisberg
,
R.
Rickli
, and
M.
Schüepp
,
1998
:
50 years of Alpine weather statistics (AWS)
.
Meteor. Z.
,
7
,
99
111
.
Webster
,
P. J.
, and
Coauthors
,
2010
:
Extended-range probabilistic forecasts of Ganges and Brahmaputra floods in Bangladesh
.
Bull. Amer. Meteor. Soc.
,
91
,
1493
1514
.
Willett
,
K. M.
,
P. D.
Jones
,
N. P.
Gillett
, and
P. W.
Thorne
,
2008
:
Recent changes in surface humidity: Development of the HadCRUH dataset
.
J. Climate
,
21
,
5364
5383
.
Zappa
,
M.
, and
Coauthors
,
2008
:
MAP D-PHASE: Real-time demonstration of hydrological ensemble prediction systems
.
Atmos. Sci. Lett.
,
9
,
80
87
.
Zolina
,
O.
,
C.
Simmer
,
S. K.
Gulev
, and
S.
Kollet
,
2010
:
Changing structure of European precipitation: Longer wet periods leading to more abundant rainfalls
.
Geophys. Res. Lett.
,
37
,
L06704
,
doi:10.1029/2010GL042468
.