1. Introduction
Useful monthly weather forecasts are of great societal and economical value. A great benefit of high-quality weather predictions beyond two weeks is expected from a number of industrial sectors such as agriculture, energy, transportation, and water resource management. An efficient early warning of extreme weather events with sufficient lead time can significantly reduce the loss of life and property.
Short-range forecasting of day-to-day changes in weather up to about a week has been routinely produced for several decades, where relatively skillful predictions are usually made from numerical weather prediction (NWP) model integrations starting from initial conditions obtained from observations and sophisticated data assimilation methods. In contrast, forecasting seasonal to interannual changes depends strongly on the slowly evolving components of the earth system [e.g., sea surface temperature (SST) and sea ice]. The subseasonal time scale, related to monthly forecasts, falls in the gap between the short-range and seasonal forecast time scales. Historically this gap received little attention. Forecasting on this time scale is challenging, because on one hand this time scale is relatively long so that initial errors grow and forecast errors become large, but on the other hand, it is too short for anomalies in other climate components (e.g., SST anomalies) to take effect.
Interest on subseasonal prediction has increased enormously recently. This is partly driven by its obvious potential societal and economical benefit. Recent studies have stressed the importance of advancing subseasonal prediction (e.g., Brunet et al. 2010; Shapiro et al. 2010), and indicated that there is the potential of predictability across all time scales (e.g., Hoskins 2013). Studies on sources of predictability have further advanced the prospects of subseasonal forecasting, especially that related to the Madden–Julian oscillation (MJO; e.g., Madden and Julian 1971; Waliser et al. 2003; Vitart et al. 2007; Lin et al. 2010a). The European Centre for Medium-Range Weather Forecasts (ECMWF) started monthly forecasting in 2004 (Vitart 2004). Currently most of the World Meteorological Organization (WMO) Global Producing Centers are producing operational subseasonal forecasts.
The Canadian Meteorological Centre (CMC) started its first generation of dynamical seasonal prediction in September 1995. One of the products was the 30-day-averaged temperature anomaly over Canada. This long-range forecasting system was based on the first phase of the Historical Forecasting Project (HFP) (Derome et al. 2001), where 20-member ensemble forecasts of 3 months were produced using two Canadian atmospheric models with persistent anomalies in SST (i.e., a two-tier system). The system was upgraded in 2003 to use two additional atmospheric models and the ensemble size was increased to 40 with the forecast range increased to 4 months (e.g., Lin et al. 2008; Kharin et al. 2009). Output of the first month was used to produce the 30-day-averaged temperature anomaly forecast over Canada. The two-tier seasonal forecasting system was replaced by the Canadian Seasonal-to-Interannual Prediction System (CanSIPS) in December 2011, which is a 20-member ensemble 12-month forecast of two coupled atmosphere–ocean–land climate models developed at the Canadian Centre for Climate Modelling and Analysis (CCCma) (Merryfield et al. 2013). Again the only forecast product for the first month is the 30-day-averaged temperature anomaly over Canada. Although there is some useful skill, there are a few shortcomings with the seasonal forecast systems when doing forecasts for the first month. The systems were designed for seasonal forecasts where the target products are seasonal mean anomalies of atmospheric conditions, thus less attention was paid to the detail of the atmospheric initial condition. In the CMC’s early two-tier seasonal forecasting system, the ensemble members were generated by 12-h lagged initial conditions, while in CanSIPS a simple nudging was used. In addition, the seasonal forecast systems have a very low resolution, for example, the horizontal resolution of the CanSIPS atmospheric models is T63. Such resolution may be enough for seasonal time scale, but is too low to capture scale interactions of intraseasonal variability. The atmospheric model of CanSIPS has a poor representation of the MJO (Lin et al. 2008), which is an important source of skill for monthly forecasts.
In this paper, we introduce the new monthly forecasting system that replaced CanSIPS in July 2015 to produce the operational monthly forecasts at CMC. This system is based on the Global Ensemble Prediction System (GEPS) (e.g., Charron et al. 2010; Houtekamer et al. 2014), by extending the lead time of the ensemble medium-range weather forecasts out to 32 days once a week (Gagnon et al. 2013, 2014b). Although it is still a two-tier system (i.e., an uncoupled system with specified SST and sea ice conditions), it likely captures most of the major sources of predictability on the subseasonal time scale. Compared to CanSIPS, the GEPS-based monthly forecasts take advantage of the increased model resolution and improved initialization, leading to improved forecast skill.
Section 2 describes the GEPS-based monthly forecasting system in detail. Both the components of real-time forecast and hindcast are introduced. In section 3, as a proof of concept, 20-member hindcasts of GEPS are generated starting from 1 January and 1 July over the 26-yr period from 1995 to 2010. Comparisons of forecast skill with CanSIPS over the same period are discussed. The forecasts of the past 2 yr (2014 and 2015) are verified in section 4 for the winter and summer half-years. In section 5, in order to understand the source of skill in winter over North America, the 18-yr four-member hindcasts are analyzed in terms of the influence of the MJO. A summary and discussion are given in section 6.
2. Description of the GEPS-based monthly forecasts
a. Real-time forecast
The GEPS has been used operationally at CMC to produce medium-range ensemble forecasts since 1996 (Houtekamer et al. 2009; Charron et al. 2010). As for most operational forecast systems, the GEPS has been updated regularly. The most recent update occurred in December 2015 (Gagnon et al. 2015). The atmospheric model used is the Canadian Global Environmental Multiscale (GEM) model (Côté et al. 1998; Girard et al. 2014). The current GEPS has a horizontal grid spacing of 0.45° × 0.45°, and 40 vertical levels. The GEPS is run twice daily out to 16 days with 20 perturbed members and one control member. The initial conditions are produced with the ensemble Kalman filter (EnKF; Houtekamer et al. 2009, 2014), which receives observations that are background checked and bias corrected by the Global Deterministic Prediction System (GDPS; Buehner et al. 2015). Different members of the GEPS have different model configuration perturbations (multiparameterization physics). They also make use of stochastic perturbations of physics tendencies and stochastic energy back scattering (Charron et al. 2010). Land properties are initialized with the real-time CMC analysis. Once a week (Thursday at 0000 UTC), the GEPS forecasts are extended to 32 days, that make the real-time component of the monthly forecast.
In consideration of the SST evolution during the forecast period, the SST anomaly of the previous 30 days is persisted throughout the integration. As GEM reads SST values at the midmonth and performs a linear interpolation during the integration, the 30-day-averaged anomaly is added at the target date (i.e., 32 days into the future). This SST value is then linearly interpolated to midmonth and provided to the GEM model. The sea ice cover is adjusted in order to be consistent with the SST change [details in Gagnon et al. (2014b)].
b. Hindcast
Model drift is a common problem for long-range predictions. As the model integration is extended, systematic errors build up that gradually contaminate the forecast. Reducing systematic errors is fundamental for improving forecast skill on the subseasonal and seasonal time scales. However, this involves tremendous efforts in model development that include improved understanding and representation of physical processes especially those on subgrid scales. It is unlikely that the problem of model drift and systematic error will go away in the foreseeable future. Part of the systematic errors can be corrected through statistical methods based on discrepancies between historical forecasts and past observations. Such a postprocess is referred to as calibration that adjusts the ensemble forecast and improves forecast skill of the forecasting system. Therefore, the hindcast or reforecast is a component of the monthly forecast system, which is as important as the real-time forecasts.
With the GEPS model, hindcasts are produced in the 18-yr period from 1995 to 2012 for the same date of the forecast. Since December 2015, the hindcast is extended to a 20-yr period from 1995 to 2014. Ideally the model configuration and initialization strategies in the hindcast should remain exactly the same as the real-time forecast. However, several factors (e.g., limit of computational resources and lack of reanalysis for the atmosphere and land surface) prevent this from happening. Also the whole hindcast procedure has to be redone each time the system is changed. This is a severe constraint for medium-range ensemble prediction systems since they are usually upgraded at a frequency of once every 12–18 months.
Hagedorn et al. (2008) have described a more flexible reforecast procedure adopted by ECMWF. Only part of the actual operational system is run in the past weekly to build the historical dataset. It was found that it is more beneficial to have more years than more members (Hagedorn et al. 2008). Therefore, in the previous ECMWF’s monthly forecasting system a reforecast set of 5 members over 18 past years are generated (Vitart 2004). They now have increased that numbers to 11 members for the past 20 yr twice per week (see ECMWF 2015). Here we apply a similar configuration of the reforecast that consists in running 4 ensemble members over 32 days for the 18-yr period of 1995–2012 (20-yr period of 1995–2014 since the December 2015 update). Four different model configurations are chosen for different years in order for each of the 20 model configurations to be included as often as the others. The control member is not used in the reforecast.
The starting date is determined by the calendar day of the current forecast of Thursday. This gives 72 reforecasts [i.e., 4 (members) times 18 (yr) per forecast date (the number increases to 80 since December 2015 with 20 yr of hindcast)]. To make the sample size larger and the model statistics more stable and reliable, we group together five reforecast dates spanning over five consecutive weeks centered at the current forecast date. The statistics from this grouped dataset will then be applied to all forecasts of the current forecast date as in Hagedorn et al. (2008). One drawback of this approach, however, is that the maximum and minimum of the annual cycle of model climatology could be underestimated, which may influence the forecast during those two weeks. In the majority of cases, this improves the statistics. In operations, each day of the week (except on Thursdays when two dates are done), reforecasts with four members for three past dates are produced. The reforecast for a given calendar day are run five weeks in advance. This is done during low traffic hours on the supercomputer. Then, at the end of the week, these are summed up to give all reforecast samples corresponding to the Thursday real-time forecast of the week. The issue of frequent model update is thus addressed by this on-the-fly approach of reforecasts.
As CMC does not have its own reanalysis, the reforecasts are initialized using ERA-Interim reanalyses (Dee et al. 2011) for the atmospheric fields. Random isotropic perturbations (Houtekamer et al. 2009) are added to the reanalysis fields to create four different initial conditions. The atmospheric perturbations are homogeneous and isotropic as in Gauthier et al. (1999). Only the streamfunction and the unbalanced temperatures are perturbed here and in the EnKF (Houtekamer et al. 2009). These perturbed fields are then transformed to wind, temperature, and surface pressure.
To initialize the land surface fields in the reforecast, the CMC Surface Prediction System (SPS; Carrera et al. 2010) was run offline forced by the near-surface atmospheric and precipitation fields of the ERA-Interim reanalyses for more than 30 yr (Gagnon et al. 2014a). This gives a land surface condition that is consistent with the atmospheric initial condition (i.e., ERA-Interim reanalysis), as well as the land process scheme of the GEM model.
The SSTs and sea ice cover used in the hindcast also come from the ERA-Interim database. The same approach of persisting SST anomalies and evolving sea ice extent accordingly as in the real-time forecast is applied in the hindcast, although in the GEPS operational real-time forecast the CMC operational analysis (Brasnett 2008) is used.
3. Comparisons with CanSIPS
As mentioned in the introduction section, before the implementation of the GEPS-based monthly forecasting, CMC was doing monthly predictions using the first-month output of CanSIPS. To evaluate whether the GEPS-based system has a better forecast skill than the previous system, a comparison of performance is made between the two systems over the same reforecast period, which is the 26-yr period from 1985 to 2010. Forecasts starting from 1 January and 1 July are performed and evaluated to represent the winter and summer seasons. For this specific purpose, the GEPS is run with 20 members, the same member size as CanSIPS. The hindcasts of CanSIPS starting from 1 January and 1 July for the same 26 yr were previously performed and are ready to use.
The verification is done against the ERA-Interim reanalyses. We first present the forecast skill of the 30-day-averaged 2-m air temperature (T2m) anomaly over Canada, which is the only forecast product at CMC currently. Serial correlations over the 26 cases are calculated separately for the January and July forecasts, and the results are shown in Fig. 1. As can be seen, both GEPS and CanSIPS have significant correlation skill in January and July, and the distributions are similar. The GEPS-based forecast performs slightly better in southern Canada in January. For the forecasts starting from 1 July, GEPS is clearly better than CanSIPS.
Temporal correlation skill of 30-day-averaged T2m: (a) CanSIPS January, (b) CanSIPS July, (c) GEPS January, and (d) GEPS July. The value 0.33 corresponds to the 0.05 significance level. The two numbers in the legend box are averaged correlation (×100), and percentage of area that is statistically significant at the 0.05 level. The persistence forecast skill from the previous 30-day mean anomaly averaged over Canada is 0.18 and 0.32 for January and July, respectively.
Citation: Monthly Weather Review 144, 12; 10.1175/MWR-D-16-0138.1
Forecasts in terms of three categories (i.e., above normal, near normal and below normal) are also evaluated by the percent correct score. The criterion used to define the categories is 0.43 times the interannual standard deviation, so that the three categories have an equal probability to occur. Illustrated in Fig. 2 is the percent correct of 30-day-averaged T2m by GEPS and CanSIPS for the January and July forecasts. A similar conclusion as for the correlation skill can be made, indicating that the GEPS-based monthly forecast outperforms that of the CanSIPS.
Percent correct score of 30-day-averaged T2m anomaly forecasts based on three categories (above normal, near normal, and below normal): (a) CanSIPS January, (b) CanSIPS July, (c) GEPS January, and (d) GEPS July. The two numbers in the legend box are averaged value, and percentage of area that is over 46% corresponding to the statistical significance level of 0.10 according to a binomial test.
Citation: Monthly Weather Review 144, 12; 10.1175/MWR-D-16-0138.1
Verification of the forecasts week by week is also performed for T2m. The skill of T2m in summer mainly comes from the first two weeks, whereas in winter skill can be seen for all four weeks. Again the skill of GEPS is better than that of CanSIPS. A detailed discussion of T2m skill for more cases in the GEPS real-time forecast will be provided in the next section.
Forecast skill of precipitation is evaluated against the Climate Prediction Center Merged Analysis of Precipitation (CMAP; Xie and Arkin 1997). Although the monthly forecast skill of precipitation over Canada is weak, the GEPS-based forecast performs better than that of CanSIPS (Fig. 3).
Temporal correlation skill of 30-day-averaged precipitation rate: (a) CanSIPS January, (b) CanSIPS July, (c) GEPS January, and (d) GEPS July. The value 0.33 corresponds to the 0.05 significance level. The two numbers in the legend box are averaged correlation (×100), and percentage of area that is statistically significant at the 0.05 level.
Citation: Monthly Weather Review 144, 12; 10.1175/MWR-D-16-0138.1
4. Verification
The CMC monthly forecasting has been integrated into GEPS and run routinely since 5 December 2013. Here we report some verification results for the real-time forecasts during the 2-yr period of 2014 and 2015. In addition to the current 30-day mean forecasts, it is planned that products of weekly averages are produced. The averages are performed for calendar weeks of Monday to Sunday. As the monthly forecasting starts from Thursday at 0000 UTC, the four weekly forecasting periods correspond to days 5–11, 12–18, 19–25, and 26–32, respectively. To verify the monthly forecast, the daily averaged data of T2m and 500-hPa geopotential height (Z500) of the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) global reanalysis (Kalnay et al. 1996) are used, which are independent analysis data from those used to initialize the hindcast (i.e., ERA-Interim). A total of 105 forecast cases were conducted during these 2 yr. The analysis is performed separately for the winter half-year (November–April) and summer half-year (May–October), with 52 and 53 cases, respectively. The daily averaged values of the NCEP–NCAR reanalysis and the forecast are averaged for the four 7-day periods of forecasts (i.e., days 5–11, 12–18, 19–25, and 26–32). The verification is performed on the weekly mean data. The forecast anomaly is the departure from the model climatology of the hindcast over the 18-yr period of 1995–2012, whereas the anomaly of the observation is calculated as the departure from the climatology of the NCEP–NCAR reanalysis over the same 18-yr period.
We first present the anomaly correlation and root-mean-square error (rmse) for the ensemble mean 500-hPa geopotential height over the Northern Hemisphere from 20° to 80°N. These scores are commonly used for verification of short- to medium-range weather prediction. Here they would give an overall skill for the monthly forecast, although they are probably not the best scores for assessing forecasts longer than 10 days as discussed in Vitart (2004). As shown in Fig. 4, the skill is better in winter than in summer. By week 2 or days 12–18, the anomaly correlation drops to 0.46 (0.27) in winter (summer). The dashed curves in Fig. 4 are anomaly correlation (top panel) and rmse (bottom panel) of a persistence forecast (i.e., by persisting the observed anomaly of the week before week 1). The model forecast is clearly better than persistence for all lead times. The green curves in the rmse plot (Fig. 4 bottom) are those for climatology forecasts. In winter, the model forecast error matches that of a climatology forecast slightly after week 3 (i.e., days 19–25), whereas in summer it does so slightly after week 2 (i.e., days 12–18). This indicates that some useful skill can be expected beyond week 2, especially in the winter seasons.
(top) Anomaly correlation and (bottom) root-mean-square error (rmse) of weekly averaged 500-hPa geopotential height over the Northern Hemisphere from 20° to 80°N. The solid black curve represents the average of 52 winter cases, whereas the solid red curve is the average of 53 summer cases. The black and red dashed curves are for persistence forecasts in winter and summer, respectively. The green solid and dashed curves in the bottom panel are for climatology forecast in winter and summer, respectively.
Citation: Monthly Weather Review 144, 12; 10.1175/MWR-D-16-0138.1
The above scores represent an average over the Northern Hemispheric extratropical region. To see the spatial distribution of the forecast skill, Figs. 5 and 6 illustrate the temporal correlation between the ensemble mean Z500 anomaly and the observed Z500 anomaly from week 1 to week 4 for the winter and summer seasons, respectively. The colored areas are statistically significant at the 0.05 level according to the Student’s t test. When doing the significance test for a serial correlation, the number of degrees of freedom or effective sample size has been reduced because of the autocorrelation, following Bretherton et al. (1999). Considering that the Northern Hemisphere average 1-week lagged correlation for Z500 is 0.36 in the winter and 0.29 in the summer (see Fig. 4a), a conservative number of 40 is used for the effective sample size for both the winter and summer seasons. As can be seen from Fig. 5 in winter, the area of significant correlation skill becomes small from week 3. In the northwest North Pacific, some skill remains up to week 4. Over northeastern and southwestern North America, significant skill can be found in week 3 (days 19–25). For the summer seasons (Fig. 6), however, the correlation skill drops rapidly, and little forecast skill can be observed beyond week 2.
Correlation skill of weekly averaged Z500 for the winter seasons from weeks 1 to 4. Yellow (orange) areas represent those where the correlation is statistically significant at a 0.05 (0.01) level according to the Student’s t test. The contour interval is 0.1. Red contours represent positive correlations and blue contours represent negative correlations.
Citation: Monthly Weather Review 144, 12; 10.1175/MWR-D-16-0138.1
As in Fig. 5, but for summer.
Citation: Monthly Weather Review 144, 12; 10.1175/MWR-D-16-0138.1
Shown in Fig. 7 is the correlation skill of T2m forecast in North America for the 52 cases of winter. With the increase of lead time, the area of significant skill decreases, but a large part of eastern North America has a significant forecast skill of T2m until week 4. This indicates that a useful temperature forecast is possible in this area beyond week 2. It is likely that the skill in eastern North America at such lead times is related to tropical forcing of the MJO, as will be discussed in detail in the next section. Another region of interest is near the southwestern coast of the United States, where significant forecast skill of T2m is observed all the way through week 4. Illustrated in Fig. 8 is the percent correct score of weekly averaged T2m anomaly forecasts over North America in winter, based on three categories (above normal, near normal, and below normal). The three categories have an equal probability (33%) to occur, and are defined with the criterion of 0.43 times the standard deviation among all members. A similar conclusion as the correlation skill can be made, indicating that the monthly forecasting system has a significant forecast skill of T2m in the eastern North America up to week 4. The skill distribution of T2m forecast in the North American continent for the summer seasons is different from that in winter. As seen in Fig. 9, in week 2 (days 12–18), some correlation skill is found in the region of Canadian Prairies. Some skill in week 3 is observed in the central United States. In week 4, the skill is limited to the southeastern United States. The three-category anomaly T2m forecast gives a more promising performance (Fig. 10) than the correlation skill, although the distributions of these two scores are similar. Relatively high percent correct score is found in the West Coast, the central United States, and the Southeast for all four weeks. It is possible that the temperature forecast skill in some of these regions is related to soil moisture anomalies in the initial condition (e.g., Koster et al. 2011). In northeastern North America, however, little skill can be found beyond week 2 for the summer forecasts, consistent with the correlation skill.
The correlation skill of weekly averaged T2m for the winter for weeks 1–4. Yellow (orange) areas represent those where the correlation is statistically significant at a 0.05 (0.01) level according to the Student’s t test. The contour interval is 0.1. Red contours represent positive correlations and blue contours represent negative correlations.
Citation: Monthly Weather Review 144, 12; 10.1175/MWR-D-16-0138.1
The percent correct score of weekly averaged T2m anomaly forecasts in winter based on three categories (above normal, near normal, and below normal). The contour interval is 10%. Yellow and orange areas represent those where the score is statistically significant at the 0.05 and 0.01 levels, respectively, according to a binomial test.
Citation: Monthly Weather Review 144, 12; 10.1175/MWR-D-16-0138.1
As in Fig. 7, but for summer.
Citation: Monthly Weather Review 144, 12; 10.1175/MWR-D-16-0138.1
As in Fig. 8, but for summer.
Citation: Monthly Weather Review 144, 12; 10.1175/MWR-D-16-0138.1
In summary, significant forecast skill beyond week 2 is evident in eastern North America in the winter seasons. The spatial distribution of forecast skill in T2m is consistent with that in Z500, indicating that some large-scale pattern is responsible for the forecast skill at the subseasonal time scale. In the next section, we will analyze the model performance related to the influence of the tropical MJO that will likely explain such a forecast skill.
5. Impact of the MJO
The MJO is the dominant mode of variability in the tropics on a subseasonal time scale. The large-scale tropical diabatic heating associated with the MJO induces Rossby waves that propagate poleward and eastward, thus influencing the extratropical atmospheric circulation and weather. As discussed in Jin and Hoskins (1995), the extratropical response pattern as a response to tropical forcing can be established in about two weeks. Indeed, several observational studies have revealed significant lagged associations of the extratropical circulation anomaly with respect to the MJO. For example, as discussed in Lin et al. (2009), about two–three pentads after the occurrence of MJO phase 3, according to the definition of Wheeler and Hendon (2004), which corresponds to dipole of tropical convection anomaly (enhanced convection in the tropical Indian Ocean and reduced convection in the tropical western Pacific), a positive phase of the North Atlantic Oscillation (NAO) tends to occur. The NAO is a dominant circulation pattern in the Northern Hemisphere, influencing weather over North America and Europe (e.g., Hurrell et al. 2003). The reverse phase of the dipole (i.e., MJO phase 7) would then likely be followed by a negative NAO. Such a dipole structure of the tropical convection anomaly is the most effective in exciting extratropical circulation anomalies (e.g., Lin et al. 2010b).
Lagged correlation of the MJO with North American surface air temperature was also observed. For example, Lin and Brunet (2009) found that a large part of eastern Canada wintertime surface temperature is anomalously warm three–five pentads after phase 3 of the MJO. Such a lagged relationship implies predictability of North American temperature anomalies a few weeks in advance given knowledge of the initial state of the MJO. Forecasts using statistical models have demonstrated some useful skill of North American temperature anomalies beyond 20 days, especially for strong MJO cases (e.g., Yao et al. 2011; Rodney et al. 2013; Johnson et al. 2014).
To demonstrate that the MJO contributes to the wintertime temperature forecast skill, forecast cases that start from initial conditions with a strong MJO are selected. Skill is calculated for these strong MJO cases and compared with those forecasts starting from a weak MJO initial condition. The selection of cases is done according to the observed Real-time Multivariate MJO index (RMM) (Wheeler and Hendon 2004). The daily values of RMM index are obtained from the Australian Bureau of Meteorology website (http://www.bom.gov.au/climate/mjo). The strong MJO cases are those forecasts starting from initial conditions with an MJO amplitude greater than 1, whereas the weak MJO cases have an amplitude smaller than 1. Of the 52 forecasts during the winter seasons of 2014 and 2015, half of them (26) are strong MJO cases. The correlation skills of T2m in weeks 3 and 4 for strong and weak MJO cases are compared in Fig. 11. As can be seen, a much better forecast skill is achieved when the initial condition has a strong MJO signal (Figs. 11c and 11d). This indicates that a large part of the skill in the CMC monthly forecasting comes from the tropical MJO.
The correlation skill of weekly averaged T2m: (a) week 3 for weak MJO cases, (b) week 4 for weak MJO cases, (c) week 3 for strong MJO cases, and (d) week 4 for strong MJO cases. Yellow (orange) areas represent those where the correlation is statistically significant at a 0.05 (0.01) level according to the Student’s t test. The contour interval is 0.1. Red contours represent positive correlations and blue contours represent negative correlations.
Citation: Monthly Weather Review 144, 12; 10.1175/MWR-D-16-0138.1
To further assess the influence of the MJO in the CMC monthly forecasting system, hindcasts of the 18-yr period from 1995 to 2012 that correspond to all the real-time forecasts of 2015 are analyzed. In the winter half-year of November–April, we have a total of 468 cases (18 yr × 26 cases per winter), of that 46 cases have an initial condition corresponding to MJO phase 3 according to the observed RMM index, and 44 cases correspond to phase 7. Composites of the four-member ensemble mean forecast pentad-averaged Z500 and T2m anomalies are made for the cases of MJO phase 3 and 7 separately.
To see how the model forecasts the extratropical circulation anomaly after the initial condition of MJO phase 3, Fig. 12 displays the composites of Z500 anomalies from pentads 1 to 3. The observed anomalies (left panels) are generally in agreement with those reported in Lin et al. (2009). Following MJO phase 3 with enhanced convection in the tropical Indian Ocean and reduced convection in the tropical western Pacific, positive Z500 anomalies develop in the North Pacific. A wave train emerges across the North Pacific and North America, and in pentad 2 an anomalous anticyclone is seen over northeastern North America (Fig. 12c). This anticyclonic circulation stays in pentad 3 (Fig. 12c) and pentad 4 (not shown). By pentad 3, the signal is mainly in the North Atlantic sector, which resembles a positive NAO. The CMC monthly forecasting system predicts the Z500 anomaly evolution quite well (right panels of Fig. 12). In pentad 3, the positive NAO structure is in general reproduced, except that the negative center near Greenland is too weak. The positive Z500 anomalies over the North Pacific are too strong in pentad 3, a similar behavior as reported for the ECMWF model (Vitart and Molteni 2010). The composites for MJO phase 7 show similar features as those of MJO phase 3, except with an opposite sign and weaker signals (not shown).
Composites of pentad-averaged Z500 anomalies in pentads 1–3 for the forecasts initialized with an MJO phase 3: (a),(c),(e) observation and (b),(d),(f) forecasts. Yellow (orange) areas represent those where the anomaly is statistically significant at a 0.05 (0.01) level according to the Student’s t test. The contour interval is 10 m. Red contours represent positive correlations and blue contours represent negative correlations.
Citation: Monthly Weather Review 144, 12; 10.1175/MWR-D-16-0138.1
Composites of T2m anomalies in North America for pentads 1–3 following an initial condition of MJO phase 3 are shown in Fig. 13. The evolution of T2m anomaly is in general similar to that reported in Lin and Brunet (2009). Consistent with a delayed response to tropical MJO forcing, no signal is seen in pentad 1. A warm temperature anomaly starts to develop in pentad 2 over central Canada and the eastern United States. In pentad 3 significant warm temperature anomalies cover most of eastern North America, with a maximum of over 2°C (Fig. 13e). This signal then weakens and gradually disappears after pentad 5 (not shown). The model forecasts the T2m anomaly evolution quite well (right panels of Fig. 13). The T2m signal can be explained by the upper-atmospheric circulation pattern. As can be seen from Figs. 12c and 12e, warm advection on the west side of the 500-hPa anticyclonic circulation anomaly would be responsible for the warm temperature anomalies over eastern North America.
Composites of pentad-averaged T2m anomalies in pentads 1–3 for the forecasts initialized with an MJO phase 3: (a),(c),(e) observation and (b),(d),(f) forecasts. Yellow (orange) areas represent those where the anomaly is statistically significant at a 0.05 (0.01) level according to the Student’s t test. The contour interval is 0.5°C. Red contours represent positive correlations and blue contours represent negative correlations.
Citation: Monthly Weather Review 144, 12; 10.1175/MWR-D-16-0138.1
6. Summary and discussion
In this paper, monthly predictions in the Canadian Meteorological Centre are described. The monthly prediction became separated from the seasonal prediction system, CanSIPS, in July 2015, when a new monthly forecasting system based on the operational medium-range global ensemble system (GEPS) went into operation. GEPS is extended to 32 days once a week. Although the new monthly forecasting system is a two-tier system, where an atmospheric model is used with specified SST and sea ice condition during the model integration, it takes advantage of the improved initial condition and higher resolution of the GEPS compared to CanSIPS. Also hindcasts of 18 yr are used to calibrate the monthly forecast (20 yr since December 2015). The hindcasts are produced a few weeks prior to the real-time forecast (i.e., on the fly). Therefore, model improvement can easily take place without worrying about the rerunning of the full hindcast.
Verification of the monthly forecasts is performed with the real-time forecasts of 2014 and 2015. Ensemble forecasts of weekly averages of 500-hPa geopotential height and 2-m air temperature are compared with that of the NCEP–NCAR reanalysis. For the Northern Hemisphere as a whole, the 500-hPa Z500 forecast shows some skill, which is better than persistence and climatology forecast. The skill is better in winter than summer. In winter, significant forecast skill is found in the North Pacific and eastern North America beyond week 2 (days 12–18) for both Z500 and T2m. Over eastern North America, significant forecast skill is observed in winter for all lead times up to days 26–32. In the summer season, some skill of the T2m anomaly forecast is observed in the central and southeastern United States, possibly related to soil moisture initial conditions. These results indicate that a useful subseasonal forecast is possible in these regions of North America.
The ability of the CMC monthly forecasting system in capturing the MJO-related teleconnections is analyzed using the data of 18 yr of hindcasts. It is found that the Rossby wave train following phase 3 of the MJO is well represented in the first four pentads. Three pentads after the MJO phase in the initial condition, a positive phase of the NAO is produced, which is in agreement with the observation. The development of the warm anomaly over eastern North America following phase 3 of the MJO is also well predicted. This indicates that such MJO influences contribute to the forecast skill in eastern North America.
The performance of this two-tier system in producing skillful forecasts indicates that major sources of predictability on the subseasonal time scale are reasonably well represented in the system through initialization and persistence of anomalies in SST and sea ice. For example, an improved initialization helps to capture tropical convection and modes of variability including the MJO. To produce a realistic teleconnection, the initial atmospheric state in the forcing region must be well represented.
Several studies have shown that air–sea interactions are important in simulation and forecast of the MJO (e.g., Vitart et al. 2007). This implies that an air–sea coupled model would be helpful for the subseasonal prediction. When the MJO forecast skill is improved, it can be expected that the forecast skill of T2m in North America would be further improved. Therefore, we are working to develop our next generation of subseasonal forecasting system where the GEM model will be coupled with an ocean–ice model. There are some recent studies (e.g., de Boisseson et al. 2012; Fu et al. 2013; Wang et al. 2015) suggesting that including coupled model forecast daily SST anomalies during an atmosphere-only model forecast period can improve MJO forecast skill comparing to that using persistent SST as in GEPS. This sounds like a practical approach comparing to full-blown coupled GCMs, which are usually troubled with large climate drift and systematic errors. It would be interesting to investigate the impact on subseasonal forecast skill in GEPS by specifying the daily SST anomaly with that from the CanSIPS coupled model predictions.
Acknowledgments
The implementation of GEPS and its extension for monthly forecasting represents a cumulative effort of many people from the modeling and postprocessing sections in research and development divisions, and operational sections of Environment and Climate Change Canada (ECCC) collocated at the Canadian Meteorological Center (CMC) in Dorval, Québec, Canada. In particular, we thank Maria Abrahamowicz, Stéphane Bélair, Gilbert Brunet, Peter Houtekamer, Rochdi Lahlou, Benoit Archambault, and Juan Sebastian Fontecilla, among others, for their help and support.
REFERENCES
Brasnett, B., 2008: The impact of satellite retrievals in a global sea-surface-temperature analysis. Quart. J. Roy. Meteor. Soc., 134, 1745–1760, doi:10.1002/qj.319.
Bretherton, C., M. Widmann, V. P. Dymnikov, J. M. Wallace, and I. Blade, 1999: The effective number of spatial degrees of freedom of a time-varying field. J. Climate, 12, 1990–2009, doi:10.1175/1520-0442(1999)012<1990:TENOSD>2.0.CO;2.
Brunet, G., and Coauthors, 2010: Collaboration of the weather and climate communities to advance subseasonal-to-seasonal prediction. Bull. Amer. Meteor. Soc., 91, 1397–1406, doi:10.1175/2010BAMS3013.1.
Buehner, M., and Coauthors, 2015: Implementation of deterministic weather forecasting systems based on ensemble-variational data assimilation at Environment Canada. Part I: The global system. Mon. Wea. Rev., 143, 2532–2559, doi:10.1175/MWR-D-14-00354.1.
Carrera, M. L., S. Bélair, V. Fortin, B. Bilodeau, D. Charpentier, and I. Doré, 2010: Evaluation of snowpack simulations over the Canadian Rockies with an experimental hydrometeorological modeling system. J. Hydrometeor., 11, 1123–1140, doi:10.1175/2010JHM1274.1.
Charron, M., G. Pellerin, L. Spacek, P. L. Houtekamer, N. Gagnon, H. L. Mitchell, and L. Michelin, 2010: Toward random sampling of model error in the Canadian Ensemble Prediction System. Mon. Wea. Rev., 138, 1877–1901, doi:10.1175/2009MWR3187.1.
Côté, J., S. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 1998: The operational CMC-MRB Global Environmental Multiscale (GEM) model. Part I: Design considerations and formulation. Mon. Wea. Rev., 126, 1373–1395, doi:10.1175/1520-0493(1998)126<1373:TOCMGE>2.0.CO;2.
Dee, D. P., and Coauthors, 2011: The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553–597, doi:10.1002/qj.828.
de Boisseson, E., M. A. Balmaseda, F. Vitart, and K. Mogensen, 2012: Impact of the sea surface temperature forcing on hindcasts of Madden–Julian Oscillation events using the ECMWF model. Ocean Sci., 8, 1071–1084, doi:10.5194/os-8-1071-2012.
Derome, J., G. Brunet, A. Plante, N. Gagnon, G. J. Boer, F. W. Zwiers, S. Lambert, and H. Ritchie, 2001: Seasonal predictions based on two dynamical models. Atmos.–Ocean, 39, 485–501, doi:10.1080/07055900.2001.9649690.
ECMWF, 2015: IFS documentation—Cy41r1, operational implementation 12 May 2015, Part V: Ensemble prediction system. ECMWF, 25 pp. [Available online at http://www.ecmwf.int/sites/default/files/elibrary/2015/9212-part-v-ensemble-prediction-system.pdf.]
Fu, X., J.-Y. Lee, P.-C. Hsu, H. Taniguchi, B. Wang, W. Q. Wang, and S. Weaver, 2013: Multi-model MJO forecasting during DYNAMO/CINDY period. Climate Dyn., 41, 1067–1081, doi:10.1007/s00382-013-1859-9.
Gagnon, N., H. Lin, S. Beauregard, M. Charron, B. Archambault, R. Lahlou, and C. Côté, 2013: Improvements to the Global Ensemble Prediction System (GEPS) from version 3.0.0 to version 3.1.0. Canadian Meteorological Centre Tech. Note, Environment Canada, 38 pp. [Available online at http://collaboration.cmc.ec.gc.ca/cmc/cmoi/product_guide/docs/lib/technote_geps310_20131204_e.pdf.]
Gagnon, N., S. Beauregard, R. Muncaster, M. Abrahamowicz, R. Lahlou, and H. Lin, 2014a: Improvements to the Global Ensemble Prediction System (GEPS) reforecast system from version 3.1.0 to version 4.0.0. Canadian Meteorological Centre Tech. Note, Environment Canada, 35 pp. [Available online at http://collaboration.cmc.ec.gc.ca/cmc/cmoi/product_guide/docs/lib/Tech_Note_GEPS400_reforecast_v1.1_E.pdf.]
Gagnon, N., X.-X. Deng, P. L. Houtekamer, S. Beauregard, A. Erfani, M. Charron, R. Lahlou, and J. Marcoux, 2014b: Improvements to the Global Ensemble Prediction System (GEPS) from version 3.1.0 to version 4.0.0. Canadian Meteorological Centre Tech. Note, Environment Canada, 49 pp. [Available online at http://collaboration.cmc.ec.gc.ca/cmc/cmoi/product_guide/docs/lib/technote_geps-400_20141118_e.pdf.]
Gagnon, N., and Coauthors, 2015: Improvements to the Global Ensemble Prediction System (GEPS) from version 4.0.1 to version 4.1.1. Canadian Meteorological Centre Tech. Note, Environment Canada, 36 pp. [Available online at http://collaboration.cmc.ec.gc.ca/cmc/cmoi/product_guide/docs/lib/technote_geps-411_20151215_e.pdf.]
Gauthier, P., M. Buehner, and L. Fillion, 1999: Background-error statistics modelling in a 3D variational data assimilation scheme: Estimation and impact on the analyses. Proc. ECMWF Workshop on Diagnosis of Data Assimilation Systems, Reading, United Kingdom, ECMWF, 131–145.
Girard, C., and Coauthors, 2014: Staggered vertical discretization of the Canadian Environmental Multiscale (GEM) model using a coordinate of the log-hydrostatic-pressure type. Mon. Wea. Rev., 142, 1183–1196, doi:10.1175/MWR-D-13-00255.1.
Hagedorn, R., T. M. Hamill, and J. S. Whitaker, 2008: Probabilistic forecast calibration using ECMWF and GFS ensemble reforecasts. Part I: Two-meter temperatures. Mon. Wea. Rev., 136, 2608–2619, doi:10.1175/2007MWR2410.1.
Hoskins, B., 2013: The potential for skill across the range of the seamless weather-climate prediction problem: A stimulus for our science. Quart. J. Roy. Meteor. Soc., 139, 573–584, doi:10.1002/qj.1991.
Houtekamer, P. L., H. L. Mitchell, and X. Deng, 2009: Model error representation in an operational ensemble Kalman filter. Mon. Wea. Rev., 137, 2126–2143, doi:10.1175/2008MWR2737.1.
Houtekamer, P. L., X. Deng, H. L. Mitchell, S.-J. Baek, and N. Gagnon, 2014: Higher resolution in an operational ensemble Kalman filter. Mon. Wea. Rev., 142, 1143–1162, doi:10.1175/MWR-D-13-00138.1.
Hurrell, J. W., Y. Kushnir, G. Ottersen, and M. Visbeck, 2003: An overview of the North Atlantic Oscillation. The North Atlantic Oscillation: Climatic Significance and Environmental Impact, Geophys. Monogr., Vol. 134, Amer. Geophys. Union, 1–35, doi:10.1029/134GM01.
Jin, F., and B. J. Hoskins, 1995: The direct response to tropical heating in a baroclinic atmosphere. J. Atmos. Sci., 52, 307–319, doi:10.1175/1520-0469(1995)052<0307:TDRTTH>2.0.CO;2.
Johnson, N. C., D. C. Collins, S. B. Feldstein, M. L. L’Heureux, and E. E. Riddle, 2014: Skillful wintertime North American temperature forecasts out to four weeks based on the state of ENSO and the MJO. Wea. Forecasting, 29, 23–38, doi:10.1175/WAF-D-13-00102.1.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471, doi:10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2.
Kharin, V. V., Q. Teng, F. W. Zwiers, G. J. Boer, J. Derome, and J. S. Fontecilla, 2009: Skill assessment of seasonal hindcasts from the Canadian Historical Forecast Project. Atmos.–Ocean, 47, 204–223, doi:10.3137/AO1101.2009.
Koster, R. D., and Coauthors, 2011: The second phase of the Global Land–Atmosphere Coupling Experiment: Soil moisture contributions to subseasonal forecast skill. J. Hydrometeor., 12, 805–822, doi:10.1175/2011JHM1365.1.
Lin, H., and G. Brunet, 2009: The influence of the Madden–Julian Oscillation on Canadian wintertime surface air temperature. Mon. Wea. Rev., 137, 2250–2262, doi:10.1175/2009MWR2831.1.
Lin, H., G. Brunet, and J. Derome, 2008: Forecast skill of the Madden–Julian Oscillation in two Canadian atmospheric models. Mon. Wea. Rev., 136, 4130–4149, doi:10.1175/2008MWR2459.1.
Lin, H., G. Brunet, and J. Derome, 2009: An observed connection between the North Atlantic Oscillation and the Madden–Julian Oscillation. J. Climate, 22, 364–380, doi:10.1175/2008JCLI2515.1.
Lin, H., G. Brunet, and J. S. Fontecilla, 2010a: Impact of the Madden-Julian Oscillation on the intraseasonal forecast skill of the North Atlantic Oscillation. Geophys. Res. Lett., 37, L19803, doi:10.1029/2010GL044315.
Lin, H., G. Brunet, and R. Mo, 2010b: Impact of the Madden–Julian oscillation on wintertime precipitation in Canada. Mon. Wea. Rev., 138, 3822–3839, doi:10.1175/2010MWR3363.1.
Madden, R. A., and P. R. Julian, 1971: Detection of a 40–50 day oscillation in the zonal wind in the tropical Pacific. J. Atmos. Sci., 28, 702–708, doi:10.1175/1520-0469(1971)028<0702:DOADOI>2.0.CO;2.
Merryfield, W. J., and Coauthors, 2013: The Canadian Seasonal to Interannual Prediction System. Part I: Models and initialization. Mon. Wea. Rev., 141, 2910–2945, doi:10.1175/MWR-D-12-00216.1.
Rodney, M., H. Lin, and J. Derome, 2013: Subseasonal prediction of wintertime North American surface air temperature during strong MJO events. Mon. Wea. Rev., 141, 2897–2909, doi:10.1175/MWR-D-12-00221.1.
Shapiro, M., and Coauthors, 2010: An Earth-system Prediction Initiative for the twenty-first century. Bull. Amer. Meteor. Soc., 91, 1377–1388, doi:10.1175/2010BAMS2944.1.
Vitart, F., 2004: Monthly forecasting at ECMWF. Mon. Wea. Rev., 132, 2761–2779, doi:10.1175/MWR2826.1.
Vitart, F., and F. Molteni, 2010: Simulation of the MJO and its teleconnections in the ECMWF forecast system. Quart. J. Roy. Meteor. Soc., 136, 842–855, doi:10.1002/qj.623.
Vitart, F., S. Woolnough, M. A. Balmaseda, and A. M. Tompkins, 2007: Monthly forecast of the Madden–Julian Oscillation using a coupled GCM. Mon. Wea. Rev., 135, 2700–2715, doi:10.1175/MWR3415.1.
Waliser, D. E., K.-M. Lau, W. Stern, and C. Jones, 2003: Potential predictability of the Madden–Julian Oscillation. Bull. Amer. Meteor. Soc., 84, 33–50, doi:10.1175/BAMS-84-1-33.
Wang, W. Q., A. Kumar, J. X. Fu, and M.-P. Hung, 2015: What is the role of the sea surface temperature uncertainty in the prediction of tropical convection associated with the MJO? Mon. Wea. Rev., 143, 3156–3175, doi:10.1175/MWR-D-14-00385.1.
Wheeler, M., and H. H. Hendon, 2004: An all-season real-time multivariate MJO index: Development of an index for monitoring and prediction. Mon. Wea. Rev., 132, 1917–1932, doi:10.1175/1520-0493(2004)132<1917:AARMMI>2.0.CO;2.
Xie, P., and P. A. Arkin, 1997: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model outputs. Bull. Amer. Meteor. Soc., 78, 2539–2558, doi:10.1175/1520-0477(1997)078<2539:GPAYMA>2.0.CO;2.
Yao, W., H. Lin, and J. Derome, 2011: Submonthly forecasting of winter surface air temperature in North America based on organized tropical convection. Atmos.–Ocean, 49, 51–60, doi:10.1080/07055900.2011.556882.
