1. Introduction
Recent improvements to numerical predictions using global atmospheric general circulation models (AGCMs) have included model physics, better initial states from improved observation, sophisticated data assimilation systems, and improved resolution with fast computational resources (Bauer et al. 2015; Owens and Hewson 2018). However, the initial state of the atmosphere is still rapidly lost within 2 weeks due to its chaotic behavior (Lorenz 1982), making medium-range forecasting after 1 week difficult.
To overcome such AGCM limitations, coupled models have been developed to consider Earth systems such as the atmosphere, ocean, land, and sea ice to improve the predictions in extended ranges. For example, in coupled models, diurnal variations and interactions between the atmosphere and ocean can provide improved local weather forecasting (Fallmann et al. 2017). The ultimate goal of coupled model forecasting is so-called seamless prediction, which covers all ranges from weather to climate (Palmer et al. 2008; Hoskins 2013). Especially, subseasonal-to-seasonal (S2S) range (known as the weather–climate prediction gap) is the most complex period to predict because the atmospheric states are losing initial memory and predictability of oceanic state is not enough (e.g., Meehl 1995; Delworth et al. 2006; Saha et al. 2006; Koster et al. 2010; Dunne et al. 2012; Orsolini et al. 2013; Robertson et al. 2015; Guemas et al. 2016; White et al. 2017).
Regarding the coupled model predictions, many operational systems initialize the Earth system components separately. In this case, existing studies reported that the forecast errors could grow rapidly during the short-range forecasting period (Mulholland et al. 2015), primarily due to the initialization shock derived from differences in analysis and prediction models, data assimilation methods, and the observations referenced differently across the initialization components. Mulholland et al. (2015) suggested the initialization shock comes from the inconsistency of sea surface temperature between the atmosphere and ocean components. However, the physical mechanism of initialization shock is still unclear. Coupled data assimilation (CDA), which can reduce initialization shock, assimilates the ocean and atmosphere simultaneously so that each component interfaces better with the others. A major advantage of CDA is that it can provide balanced initial states (Zhang et al. 2020). Generally, this is classified into strongly and weakly CDA (Penny and Hamill 2017). The former is the most sophisticated method for considering the coupled system as one single system through a cross-domain error covariance matrix. However, this requires building a new data assimilation system with a high computational cost due to a large error covariance matrix. On the other hand, the latter is combined with each independent component but interacts between the components in the coupled system, which has the advantage of using an existing data assimilation system and relatively low cost.
Most operational centers attempt to use weakly coupled data assimilation (WCDA) in their prediction systems. For example, the Met Office developed a WCDA system for fully coupled atmosphere–land–ocean–ice components (Lea et al. 2015; Guiavarc’h et al. 2019). It provided evidence for improving forecasting skills via CDA, although this system did not show significant improvements in short-range weather forecasting. Environment and Climate Change Canada adopted WCDA in their prediction system. They compared it with uncoupled data assimilation (UDA), which improved forecasting skills within 4 days in the tropics and extratropics (Skachko et al. 2019). The National Aeronautics and Space Administration (NASA) also used WCDA for seasonal prediction, including El Niño–Southern Oscillation and Madden–Julian oscillation (MJO; Madden and Julian 1971), but they did not measure the impact of coupled initialization compared to uncoupled initialization (Molod et al. 2020).
The European Centre for Medium-Range Weather Forecasts (ECMWF) developed WCDA with partial coupling due to the possibility of degraded forecasting skills due to the low accuracy of the ocean analysis system. The partial coupling limits to resolve eddies in the ocean model poleward of 25° but uses prescribed SST and sea ice concentration from accurate analysis data. This approach improved the forecasting of low-level atmospheric temperature and humidity (Browne et al. 2019). The ECMWF also examined quasi-strong CDA, which assimilates observations in each component system, but the observations influence cross components during the analysis phase, which shows promising improvement in 10-day forecasting skills, especially near the surface in the tropics (Laloyaux et al. 2016).
Overall, the impact of CDA is very diverse and sensitive to the model used, and it remains unclear how initialization shock degrades forecasting skills and whether CDA can improve forecasting skills in the S2S range or subseasonal phenomena such as MJO. It is therefore necessary to intensively examine model systematic bias and the impact of CDA, which previous studies have not yet achieved.
In this study, we developed a WCDA scheme to be applied to the Global Seasonal Forecasting System, version 5 (GloSea5-GC2.0; MacLachlan et al. 2015; Williams et al. 2015), operated by the Korea Metrological Administration (KMA). This state-of-the-art, fully coupled atmosphere–ocean–land–sea ice global climate model from the Met Office is intended for operational use for seasonal prediction (Vitart et al. 2017). This study suggests improved forecasting skills, including MJO prediction by the developed CDA. Further, it discusses how the initialization shock contributes to systematic biases and degrades the forecast skill in the extended ranges.
2. Methods and data
a. Global Seasonal Forecasting System, version 5
The GloSea5-GC2.0 is a fully coupled seasonal forecasting system developed by the Met Office, which model is based on the Hadley Centre Global Environmental Model, version 3 (HadGEM3). The atmosphere–land component is based on the Met Office Unified Model (UM; Walters et al. 2017) and the Joint UK Land Environment Simulator (JULES; Best et al. 2011) along with the ocean–sea ice component based on the Nucleus for European Modelling of the Ocean (NEMO; Madec 2008) and the Los Alamos Sea Ice Model (CICE; Rae et al. 2015). Each component realizes the interactions with other components through the Ocean Atmosphere Sea Ice Soil (OASIS3) coupler (Valcke 2013). This study follows the KMA operational configuration (GloSea5-GC2.0; Williams et al. 2015). The spatial resolution of the atmospheric model is N216 (0.83° longitude and 0.56° latitude with 85 vertical levels). The top level of the atmosphere is ∼85 km in height. The ocean model uses the ORCA025 tripolar grid (Blockley et al. 2014) with 75 vertical levels.
Initialization of atmosphere and land components of GloSea5-GC2.0 uses the global analysis obtained from the KMA Global Data Assimilation and Prediction System (GDAPS), which is based on the Met Office UM model and a hybrid four-dimensional variational data assimilation (4DVAR) scheme (Clayton et al. 2013). It uses various observations from surface in situ, sonde, aircraft, and satellite sources. KMA produces a 6-hourly atmosphere analysis field at 0000, 0600, 1200, and 1800 UTC. It has a horizontal resolution of N1280 (∼10 km) with 70 vertical levels, so the GDAPS analysis fields are regridded to match the GloSea5 grids.
The KMA GloSea5 system uses ocean initial states from the Met Office Forecast Ocean Assimilation Model (FOAM; Blockley et al. 2014), which is based on the variational data assimilation scheme of NEMO (NEMOVAR; Mogensen et al. 2009). The FOAM also has the ORCA025 tripolar grid.
b. Weakly coupled atmosphere–ocean data assimilation
We develop WCDA as a new initialization method for the seasonal forecasting system. The coupling scheme for WCDA is based on an incremental analysis update (IAU; Bloom et al. 1996), a method for updating prognostic variables with analysis increments defined as the differences between coupled model backgrounds and existing analysis. Therefore, WCDA here refers to the process of replaying the states using an atmosphere–ocean coupled model along with the analysis fields produced from individual data assimilations with uncoupled atmospheric and ocean models. Compared to a simple nudging scheme, this method acts as a low-pass filter, significantly reducing the adverse impacts of analysis increments on the model background. Thus, using IAU minimizes imbalances and spurious waves introduced by the analysis (Bloom et al. 1996; Ham et al. 2016).
Figure 1 illustrates the developed WCDA process. Initialization begins with the existing analysis of independent atmosphere and ocean data assimilations. After initialization, the first guess (step 1 in Fig. 1) is obtained from a 6-h coupled model forecast, which serves as the coupled model background (referred to as the “predictor” steps). The second step (step 2) calculates the increment from the KMA GDAPS atmosphere analysis (from a hybrid 4DVAR scheme) and then rewinds the time step by 3 h (step 3). This is followed by a 6-h integration with incremental forcing based on IAU (step 4, called the “corrector” steps). Instead of applying the entire increment at once, IAU divides the total increment by the number of time steps and applies an equal portion uniformly throughout the 6-h corrector step. Finally, an additional 3-h integration (step 1′) generates a new coupled background. Updated atmospheric prognostic variables include potential temperature, wind, and specific humidity. In this study, the prognostic mass field is not included due to technical issues, but this system well reproduces geopotential height. It presumably comes from the well-corrected temperature and wind fields affecting the mass field. This cycle repeats during the WCDA process (steps 1–4), with the ocean analysis from the NEMOVAR scheme updating oceanic prognostic variables (temperature, salinity, wind, sea surface height) every day at 0000 UTC (2′). Note that ocean increments are applied within a 6-h time window only at 0000 UTC when the ocean analysis is available daily. During IAU, the atmosphere and ocean components are balanced through flux exchanges via the coupler.
Schematic diagram of the CDA system. (1) Initialization and the free coupled model run from the coupled analysis to obtain the coupled background (“predictor” steps), (2) increment development (analysis to background) from atmospheric analysis every 6 h and (2′) from ocean analysis every 24 h, (3) rewinding of the time step by 3 h, and (4) the coupled model run with IAU forcing terms (“corrector” steps) to produce the coupled analysis. Steps 1–4 outline the sequence of the WCDA process. (5) The GloSea5 ensemble forecasts from the coupled initialization can start at 0000, 0600, 1200, and 1800 UTC. See the text for further details.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0205.1
The reanalysis field at the center of the IAU serves as the initial state for forecasting. This WCDA method is relatively simple but has the advantage of utilizing existing data analysis. It is flexible enough to accommodate changes in the analysis field from other data assimilation systems or reanalyses while keeping computational costs relatively low.
c. Experiments and validation
Forecasts start at 0000 UTC every 1st, 9th, 17th, and 25th day of each month to match the hindcast cycle of KMA GloSea5 (Table 1). In this study, the WCDA is tested on the operational setting of KMA, and the analysis period is limited to the operational period of KMA, which starts in July 2018. A full year (July 2018–June 2019), including the whole winter, is used to avoid seasonal sensitivity. Each forecast is integrated for 30 days with four ensemble members generated by a stochastic kinetic energy backscatter (SKEB2) method for the atmosphere (Bowler et al. 2009); the ocean is not perturbed. The number of ensembles is the same as in the KMA operational run. Forecasting runs are conducted for both uncoupled (UFcst) and coupled (CFcst) data assimilation. UFcst is initialized by GDAPS for atmosphere and UKMO FOAM for ocean, while CFcst is initialized by analysis from the developed WCDA method (yellow arrow in Fig. 1).
Experimental setting.
Coupled reanalysis data from GloSea5 with WCDA are used as the reference data. The quality of the GloSea5 reanalysis is validated by gridded Climatic Research Unit (CRU) time series (TS) data, version 4.05 (Harris et al. 2021), and Global Precipitation Climatology Project (GPCP; Adler et al. 2003) for monthly 2-m surface air temperature (SAT) and precipitation, respectively. Further, we compare GloSea5 reanalysis to other widely used reanalyses such as fifth major global reanalysis produced by ECMWF (ERA5; Hersbach et al. 2020) and Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2; Gelaro et al. 2017). GloSea5 reanalysis is used to examine the forecasting skill of GloSea5 in each initialization experiment. The forecasting skill of the GloSea5 is also calculated by using ERA5 and MERRA-2 as reference dataset, which does not significantly differ (not shown). The climatology of the GloSea5 forecasts is obtained from the averaged hindcasts of KMA GloSea5, which covers 1991–2010 with three ensemble members. Because KMA does not provide reanalysis from GDAPS, 1-day forecasting of GDAPS is used to compare coupling impact in GloSea5 reanalysis.
The forecasting skill for the MJO is examined using the real-time multivariate MJO (RMM) index suggested by Wheeler and Hendon (2004). This index is calculated using combined empirical orthogonal functions (EOFs) for outgoing longwave radiation (OLR) anomalies, 850-hPa zonal winds, and 200-hPa zonal winds in the tropics. To assess the MJO forecasting skill, the correlation and RMSE of the bivariate MJO index forecasts are used, following Lin et al. (2008). Statistical significance tests are conducted using the bootstrap method with 10 000 random samples.
3. Results
a. Verification of the coupled reanalysis data
To verify the quality of the coupled reanalysis from GloSea5 with WCDA, the annual mean SAT and precipitation are compared with CRU and GPCP observations, respectively (Fig. 2). The GloSea5 coupled reanalysis shows a warm bias over northeastern Russia (+2.7°C on average with a maximum of +8.5°C) and a cold bias in Greenland and the Himalayan region (−3.5°C on average with a minimum of −13°C). This bias pattern also appears in ERA5 and MERRA-2, as noted in previous studies (Betts et al. 2006; Wang and Zeng 2013; Fig. 1 in the online supplemental material). Due to the absence of data for Antarctica in the CRU dataset, difference values for that region are masked out. For precipitation, GloSea5 exhibits a wet bias over the intertropical convergence zone (ITCZ) and the South Pacific convergence zone (SPCZ), with values reaching up to 17 mm day−1. When compared with state-of-the-art uncoupled reanalyses such as ERA5 and MERRA-2, these biases are similarly present in ERA5 with a comparable magnitude, while MERRA-2 shows a slight dry bias, consistent even when using other observational precipitation datasets (supplemental Fig. 2).
Annual mean of (a) 2-m temperature from CRU and (c) precipitation from GPCP in July 2018–June 2019. (b),(d) A 2-m temperature and precipitation bias of GloSea5 coupled reanalysis.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0205.1
Figure 3 presents the Taylor diagram for the seasonal mean climatology of temperature and precipitation. In the temperature fields, ERA5 demonstrates the best performance, likely due to the assimilation of SAT observations to a limited extent. Nevertheless, all reanalyses show pattern correlation values over 0.99 in both summer and winter, along with realistic global mean temperatures, suggesting a comparably high performance by the GloSea5 coupled reanalysis.
Taylor diagram of (a) 2-m temperature and (b) precipitation for ERA5 (blue), MERRA-2 (green), and GloSea5 (red). Crosses and circles indicate summer (JJA) and winter (DJF), respectively.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0205.1
Unlike temperature, precipitation is not directly assimilated. Therefore, the skill for precipitation is relatively limited due to the dependence on physical schemes and the horizontal resolution of the global models that produce the reanalysis. To highlight the differences in precipitation in reanalysis, a comparison is made for the tropics (20°S–20°N), where the largest mean bias is observed. Among the three reanalyses, MERRA-2 shows the best performance in both spatial patterns and mean precipitation. ERA5 and GloSea5 have comparable performances, with ERA5 showing slightly better. Although the GloSea5 coupled reanalysis has room for further improvement, it still maintains a high pattern correlation of over 0.94. It remains unclear whether the overestimated precipitation in the tropical convergence zones by GloSea5, compared to the other two analyses, is driven by atmosphere–ocean coupling. All reanalyses reproduce precipitation more accurately in winter than in summer, likely due to model deficiencies in representing summer monsoons over the Northern Hemisphere.
In this study, the variability of tropical precipitation from coupled reanalysis is further investigated, where the GloSea5 coupled reanalysis shows overestimated precipitation (Fig. 2d). Figure 4 is the probability density of precipitation over the warm pool (20°S–20°N, 40°E–180°) and the ITCZ (2.5°–10°N, 100°–175.5°W). Following Kim et al. (2014), the probability density is categorized into 51 bins based on precipitation rate. The lowest bin includes the precipitation between 0 and 0.097 97 mm day−1, while the highest ranges from 150 to 1000 mm day−1. In this study, precipitation less than 0.08 mm day−1 is considered as nonprecipitation.
The probability density of precipitation of GPCP (black), GloSea5 (red), MERRA-2 (green), ERA5 (blue), and GDAPS (orange) over (a) warm pool (20°S–20°N, 40°E–180°) and (b) ITCZ (2.5°–10°N, 100°–177.5°W). Note that the precipitation from GDAPS is not produced in data assimilation mode but from atmospheric model forecasts made 1 day after the data assimilation initialization.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0205.1
Compared to the uncoupled reanalysis from ERA5 and MERRA-2, the GloSea5 coupled reanalysis tends to better reproduce precipitation rates above 10 mm day−1. In Kim et al. (2014), outdated versions of uncoupled reanalysis from ERA-Interim and MERRA exhibited difficulty in reproducing heavy precipitation when compared to GPCP, which has now been improved in their updated versions. In this study, ERA5 and MERRA-2 still exhibit more light precipitation and still suffer from similar deficiencies in the tropics, despite their superior performance in seasonal-mean precipitation climatology (Fig. 3b). In Fig. 5, the precipitation of the GDAPS 1-day forecast, which shares the atmospheric model with GloSea5, is also compared. Both GDAPS and GloSea5 exhibit similar behavior in terms of precipitation. However, GDAPS tends to produce more light precipitation, similar to uncoupled reanalyses, indicating that atmosphere–ocean coupling may improve the precipitation distribution.
Lead–lag correlation between SST and precipitation for observation (black), GloSea5 (red), MERRA-2 (green), and ERA5 (blue) over (a) the warm pool and (b) the ITCZ. Observational data are OISSTv2 for SST and GPCP for precipitation.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0205.1
The better distribution of precipitation seems to come from better atmosphere–ocean interaction in the subseasonal time scale. Kumar et al. (2013) and Feng et al. (2018) analyzed atmosphere–ocean coupling in reanalyses with a lead-lag correlation between 10–60-day bandpass-filtered SST and precipitation. In both regions, warming of SSTs leads to increased precipitation, which subsequently cools the SSTs within a 10–20-day period (Fig. 5).
GloSea5 and ERA5 generally reproduce the atmosphere–ocean coupling observed in GPCP for both regions, although ERA5 shows slightly weaker coupling. For MERRA-2 and GDAPS, the atmosphere–ocean coupling appears to be more sensitive to SST in the warm pool region. GloSea5, however, tends to have a shorter time scale for atmosphere–ocean coupling compared to observations, with the peak of the correlation occurring about 2 days earlier. In the ITCZ, none of the reanalyses accurately reproduce the precipitation-to-SST relationship. Notably, the correlation strength over the ITCZ is higher in GloSea5 than in observations. While the uncoupled reanalyses, including GDAPS, struggle to reproduce the SST-to-precipitation relationship, GloSea5 demonstrates better performance, highlighting the advantages of coupling.
The GloSea5 coupled reanalysis reasonably reproduces the subseasonal variability of tropical waves. Figure 6 shows a comparison of the wavenumber–frequency spectra, as suggested by Wheeler and Kiladis (1999), derived from precipitation data over 15°S–15°N using a temporal window of 96 days and an overlap of 65 days. Figure 7 shows the ratio of power in each wave. Notably, GloSea5 tends to overestimate short- and high-frequency waves, presumably due to an overestimated atmosphere–ocean interaction. Overall, all reanalyses demonstrate similar performance in representing tropical waves, and they tend to show weaker power in symmetric MJO and Kelvin waves. Among them, MERRA-2 exhibits the best performance in symmetric MJO, while GloSea5 and GDAPS show the best performance in the Kelvin wave. Note that GloSea5 and GDAPS are nearly indistinguishable in power ratio, suggesting that the representation of tropical waves is less sensitive to the atmosphere–ocean coupling in this comparison. Instead, the results from Figs. 6 and 7 show greater sensitivity to differences in the models’ physics parameterizations and grid-scale dynamics.
Wavenumber–frequency spectra of GPCP precipitation for (a) symmetric component and (f) asymmetric component. Black lines indicate dispersion curves. Differences in GloSea5, ERA5, MERRA-2, and GDAPS are shown from the second to the last columns, respectively.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0205.1
The power ratio of each wave in GloSea5 (red), MERRA-2 (green), ERA5 (blue), and GDAPS (orange) relative to GPCP precipitation.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0205.1
b. Forecasting skill with coupled data assimilation
Figure 8 shows the forecasting skill of the SAT with respect to the forecast lead days averaged for three regions: the Northern Hemisphere (20°–90°N), tropics (20°S–20°N), and the Southern Hemisphere (90°–20°S). In all regions, there is no significant difference in RMSE and the mean bias. Lea et al. (2015) also show no significant improvement in short-range forecasting skills measured by RMSE in the same coupled model system, being consistent with this study. Considering that Laloyaux et al. (2016) provide evidence for improving temperature forecasts, the degree of improvement by CDA can be sensitive to the base forecasting models and the initialization methods.
Time series of (left) root-mean-squared error and (right) mean bias in SAT by UFcst (blue) and CFcst (red) for the (a),(b) Northern Hemisphere, (c),(d) tropics, and (e),(f) Southern Hemisphere. A spread of one standard deviation of ensembles and cases is indicated by shading.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0205.1
This study further examines the forecasting skill of MJO in the subseasonal time scale. Because the reanalysis with WCDA shows a reasonable representation of MJO, one may expect the MJO forecasting skill improvement from the WCDA initialization. Figure 9 compares the bivariate correlation and RMSE of MJO RMM index forecasts between UFcst and CFcst during the boreal cold season (October–March). With a predictable correlation threshold of 0.5 for the MJO, UFcst can predict the MJO up to 11 days, while CFcst extends this to 17 days and remains the correlation of 0.5 up to 25 days. The skill difference becomes indistinguishable after 26 days. Note that this forecasting skill, verified over a single year, appears lower than that of other current S2S models, including the same model, when tested over much longer hindcast periods. Due to the small sample size, the correlation and RMSE do not show a gradual degradation of forecast skills as the forecast lead time increases. Nonetheless, compared to UFcst, using WCDA clearly improves MJO prediction.
(a) Correlation coefficient and (b) RMSE of the RMM index by forecasting time for UFcst (blue) and CFcst (red). Shading indicates the minimum–maximum range of bootstrapping with 10 000 random samplings.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0205.1
The improvement in MJO forecasting appears to result from enhanced eastward propagation of the MJO from the Indian Ocean to the Maritime Continent. Figure 10 presents the lag regression of OLR and 850-hPa zonal wind, averaged over 10°S–10°N, relative to the area-averaged LR in the Indian Ocean (5°S–5°N, 65°–75°E). This illustrates the cases where the MJO convection is initially located in the Indian Ocean. In the GloSea5 coupled reanalysis, clear eastward propagation is observed over 15 days, extending from the Indian Ocean to the western Pacific. In comparison, neither forecast accurately captures the eastward propagation over the Maritime Continent (∼120°E), a phenomenon known as the Maritime Continent barrier effect (Zhang and Ling 2017; Kim et al. 2017). However, CFcst demonstrates a better representation of eastward propagation for up to 10 days with a significant signal (Figs. 10b,e), which aligns with the increase in the forecasting skill of the RMM index (Fig. 9a). In contrast, forecasting skill in UFcst rapidly diminishes after 6 days, coinciding with a weakened signal of eastward propagation (Fig. 10c).
Lag regression of averaged OLR (shading) and 850-hPa zonal wind (contours) from 10°S to 10°N onto the averaged OLR over the Indian Ocean (5°S–5°N, 65°–75°E) for (a) the GloSea5 coupled reanalysis, (c) UFcst, and (e) CFcst. Gray dashed lines indicate the forecast lead time at day 10 (horizontal) and the Maritime Continent centered at 120°E (vertical). Dotted areas indicate the 95% confidence level. (b),(d),(f) Differences (UFcst–CFcst, UFcst–GloSea5, and CFcst–GloSea5, respectively).
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0205.1
While there is no significant difference in temperature forecast skill over the tropics (Fig. 8c), there are large differences in moisture forecast fields between Ufcst and Cfcst, particularly over the Indo-tropical western Pacific (INWP; 20°S–20°N, 60°E–180°), the main active region of MJO. Figure 11 presents the 850-hPa specific humidity, precipitation, and OLR averaged for all forecasting cases in the INWP region as a function of forecast lead time. In specific humidity, both forecasts show a clear decreasing tendency (dry bias) during forecasting; this dry bias of GloSea5 is also reported by Kim et al. (2019).
Averaged time series of (a) 850-hPa specific humidity, (b) precipitation, and (c) OLR along forecasting time in INWP for UFcst (blue) and CFcst (red); black dotted line is the annual mean from GloSea5 reanalysis.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0205.1
The initial conditions of UFcst are warmer across all vertical levels, wetter at low levels around 850 hPa, and more stable than those of CFcst (supplemental Fig. 3). These warm and wet biases in UFcst are attributed to systematic biases in the atmosphere-only GDAPS analysis. As a result, in the early stages of forecasting, CFcst exhibits more precipitation, accompanied by a gradual decrease in low-level moisture. In contrast, precipitation and OLR in UFcst show suppressed moist convection in the tropics at the start of forecasting, followed by a sudden increase after 48 h, which can be attributed to an initialization shock in moisture. CFcst does not show this initial shock but instead shows a gradual decrease in precipitation as it transitions to the coupled model climatology. Additionally, the moisture initialization shock in UFcst seems related to the accumulation of moisture with reduced precipitation during the initial phase of the forecast (Figs. 11a,b).
Most of the tropical precipitation is produced by the convection parameterization in the model (supplemental Fig. 4). This study examines vertical stability to understand why initial suppression of convection exists in UFcst. Using equivalent potential temperature (θe), Fig. 12 shows the instability of the atmosphere represented as
Vertical profile of stability (surface equivalent potential temperature–saturated equivalent potential temperature) for (a) UFcst, (b) CFcst, and (c) difference between UFcst and Cfcst (UFcst–CFcst) within 10 days forecasting in INWP. Black lines indicate the LFC (lower) and equilibrium (upper).
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0205.1
4. Conclusions and summary
We developed a weakly coupled data assimilation system based on the state-of-the-art fully coupled GloSea5-GC2.0 model. It is based on the IAU scheme to update prognostic variables of atmosphere and ocean components with the increments between existing atmosphere and ocean analyses from separately operating data assimilation systems (KMA, GDAPS, and Met Office FOAM, respectively) and the coupled model backgrounds.
First, this study examined the coupled reanalysis from WCDA to compare existing reanalyses ERA5 and MERRA-2. The GloSea5 coupled reanalysis has comparable performance in temperature and precipitation. It reproduces tropical precipitation variability in time and space, presumably due to the enhanced atmosphere–ocean interaction during the coupled data assimilation cycle. It is also well shown in wavenumber–frequency spectra to show a comparable description of MJO variability to other uncoupled reanalyses from ERA5 and MERRA-2.
As shown in Lea et al. (2015), GloSea5-GC2.0 does not show a significant improvement in forecasting skill on SAT. The warm bias is originally induced by the differences in SST referenced by the atmospheric data assimilation and that represented by the oceanic data assimilation. UFcst uses the initial conditions from individual data assimilation but with the same atmospheric and ocean component models. Especially in the uncoupled GDAPS atmospheric data assimilation, SST comes from the FOAM ocean data assimilation. Although this consistency could minimize the initialization shock in temperature in the interface between the atmosphere and ocean, the moisture process has shown to be more sensitive to the initialized states. This suggests that further studies will be needed to fully understand the engaged processes behind the initialization shock. In this study, we identified that initial atmospheric instability differs due to the absence of atmosphere–ocean interaction in GDAPS, leading to inevitable initialization shock in the convective process in UFcst after 1 day. This initialization shock disrupts the convective system in the tropics, making abnormal convection and precipitation within 10 days, which can degrade the eastward propagation of MJO. With WCDA initialization, initialization shock is effectively reduced, and forecasting skills for MJO are also extended to 6 days more than UFcst. However, the systematic bias of GloSea5 is still dominant after 2 weeks, so the forecasting tendency is adjusted to dry and less convection than observation.
This initialization shock in moisture can affect atmosphere–ocean coupled phenomena such as MJO, and WCDA improves the MJO forecasting skill. While the RMM prediction skill is maintained for ∼11 days in UFcst, it is much improved up to 17 days in CFcst. Therefore, this study suggests that improving atmosphere–ocean coupled data assimilation can provide a better opportunity to improve extended-range forecasts.
One limitation of this study is that the verification of prediction performance is limited to just 1 year, as the operational forecasting system has been upgraded to GloSea, version 6 (GloSea6). The coupled data assimilation developed in this study is currently being tested using GloSea6 at the KMA.
Acknowledgments.
This work was funded by the Korea Meteorological Administration Research and Development Program under Grant RS-2024-00403698.
Data availability statement.
The GloSea5-GC2.0 operated by the Korea Meteorological Administration is only available on the computing environment of the National Supercomputing Center under KMA. However, the modified source codes and scripts used in this work are available for the editor and reviewers. Please contact the authors for more information on code availability. ERA5 reanalysis is available at the Copernicus Climate Change Service (https://cds.climate.copernicus.eu/cdsapp#!/home). MERRA-2 reanalysis is provided by the NASA GMAO (https://gmao.gsfc.nasa.gov/reanalysis/MERRA-2/data_access/). CRU data and GPCP data are available at CRU (https://climatedataguide.ucar.edu/climate-data/cru-ts-gridded-precipitation-and-other-meteorological-variables-1901) and GPCP (https://climatedataguide.ucar.edu/climate-data/gpcp-monthly-global-precipitation-climatology-project), respectively.
REFERENCES
Adler, R. F., and Coauthors, 2003: The version-2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present). J. Hydrometeor., 4, 1147–1167, https://doi.org/10.1175/1525-7541(2003)004<1147:TVGPCP>2.0.CO;2.
Bauer, P., A. Thorpe, and G. Brunet, 2015: The quiet revolution of numerical weather prediction. Nature, 525, 47–55, https://doi.org/10.1038/nature14956.
Best, M. J., and Coauthors, 2011: The Joint UK Land Environment Simulator (JULES), model description – Part 1: Energy and water fluxes. Geosci. Model Dev., 4, 677–699, https://doi.org/10.5194/gmd-4-677-2011.
Betts, A. K., M. Zhao, P. A. Dirmeyer, and A. C. M. Beljaars, 2006: Comparison of ERA40 and NCEP/DOE near-surface data sets with other ISLSCP-II data sets. J. Geophys. Res., 111, D22S04, https://doi.org/10.1029/2006JD007174.
Blockley, E. W., and Coauthors, 2014: Recent development of the Met Office operational ocean forecasting system: An overview and assessment of the new Global FOAM forecasts. Geosci. Model Dev., 7, 2613–2638, https://doi.org/10.5194/gmd-7-2613-2014.
Bloom, S. C., L. L. Takacs, A. M. Da Silva, and D. Ledvina, 1996: Data assimilation using incremental analysis updates. Mon. Wea. Rev., 124, 1256–1271, https://doi.org/10.1175/1520-0493(1996)124<1256:DAUIAU>2.0.CO;2.
Bowler, N. E., A. Arribas, S. E. Beare, K. R. Mylne, and G. J. Shutts, 2009: The local ETKF and SKEB: Upgrades to the MOGREPS short-range ensemble prediction system. Quart. J. Roy. Meteor. Soc., 135, 767–776, https://doi.org/10.1002/qj.394.
Browne, P. A., P. de Rosnay, H. Zuo, A. Bennett, and A. Dawson, 2019: Weakly coupled ocean–atmosphere data assimilation in the ECMWF NWP system. Remote Sens., 11, 234, https://doi.org/10.3390/rs11030234.
Clayton, A. M., A. C. Lorenc, and D. M. Barker, 2013: Operational implementation of a hybrid ensemble/4D-Var global data assimilation system at the Met Office. Quart. J. Roy. Meteor. Soc., 139, 1445–1461, https://doi.org/10.1002/qj.2054.
Delworth, T. L., and Coauthors, 2006: GFDL’s CM2 global coupled climate models. Part I: Formulation and simulation characteristics. J. Climate, 19, 643–674, https://doi.org/10.1175/JCLI3629.1.
Dunne, J. P., and Coauthors, 2012: GFDL’s ESM2 global coupled climate–carbon Earth System Models. Part I: Physical formulation and baseline simulation characteristics. J. Climate, 25, 6646–6665, https://doi.org/10.1175/JCLI-D-11-00560.1.
Fallmann, J., H. Lewis, J. M. Castillo, A. Arnold, and S. Ramsdale, 2017: Impact of sea surface temperature on stratiform cloud formation over the North Sea. Geophys. Res. Lett., 44, 4296–4303, https://doi.org/10.1002/2017GL073105.
Feng, X., K. Haines, C. Liu, E. de Boisséson, and I. Polo, 2018: Improved SST-precipitation intraseasonal relationships in the ECMWF coupled climate reanalysis. Geophys. Res. Lett., 45, 3664–3672, https://doi.org/10.1029/2018GL077138.
Gelaro, R., and Coauthors, 2017: The Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2). J. Climate, 30, 5419–5454, https://doi.org/10.1175/JCLI-D-16-0758.1.
Guemas, V., and Coauthors, 2016: A review on Arctic sea-ice predictability and prediction on seasonal to decadal time-scales. Quart. J. Roy. Meteor. Soc., 142, 546–561, https://doi.org/10.1002/qj.2401.
Guiavarc’h, C., J. Roberts-Jones, C. Harris, D. J. Lea, A. Ryan, and I. Ascione, 2019: Assessment of ocean analysis and forecast from an atmosphere–ocean coupled data assimilation operational system. Ocean Sci., 15, 1307–1326, https://doi.org/10.5194/os-15-1307-2019.
Ham, Y.-G., H.-J. Song, J. Jung, and G.-H. Lim, 2016: Development of the nonstationary incremental analysis update algorithm for sequential data assimilation system. Adv. Meteor., 2016, 4305204, https://doi.org/10.1155/2016/4305204.
Harris, I. C., P. D. Jones, and T. Osborn, 2021: CRU TS4.05: Climatic Research Unit (CRU) Time-Series (TS) version 4.05 of high-resolution gridded data of month-by-month variation in climate (Jan. 1901 - Dec. 2020). NERC EDS Centre for Environmental Data Analysis, accessed 22 December 2024, https://catalogue.ceda.ac.uk/uuid/c26a65020a5e4b80b20018f148556681/.
Hersbach, H., and Coauthors, 2020: The ERA5 global reanalysis. Quart. J. Roy. Meteor. Soc., 146, 1999–2049, https://doi.org/10.1002/qj.3803.
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, https://doi.org/10.1002/qj.1991.
Kim, D., M.-I. Lee, D. Kim, S. D. Schubert, D. E. Waliser, and B. Tian, 2014: Representation of tropical subseasonal variability of precipitation in global reanalyses. Climate Dyn., 43, 517–534, https://doi.org/10.1007/s00382-013-1890-x.
Kim, D., H. Kim, and M.-I. Lee, 2017: Why does the MJO detour the Maritime Continent during austral summer? Geophys. Res. Lett., 44, 2579–2587, https://doi.org/10.1002/2017GL072643.
Kim, H., M. A. Janiga, and K. Pegion, 2019: MJO propagation processes and mean biases in the SubX and S2S reforecasts. J. Geophys. Res. Atmos., 124, 9314–9331, https://doi.org/10.1029/2019JD031139.
Koster, R. D., and Coauthors, 2010: Contribution of land surface initialization to subseasonal forecast skill: First results from a multi-model experiment. Geophys. Res. Lett., 37, L02402, https://doi.org/10.1029/2009GL041677.
Kumar, A., L. Zhang, and W. Wang, 2013: Sea surface temperature–precipitation relationship in different reanalyses. Mon. Wea. Rev., 141, 1118–1123, https://doi.org/10.1175/MWR-D-12-00214.1.
Laloyaux, P., M. Balmaseda, D. Dee, K. Mogensen, and P. Janssen, 2016: A coupled data assimilation system for climate reanalysis. Quart. J. Roy. Meteor. Soc., 142, 65–78, https://doi.org/10.1002/qj.2629.
Lea, D. J., I. Mirouze, M. J. Martin, R. R. King, A. Hines, D. Walters, and M. Thurlow, 2015: Assessing a new coupled data assimilation system based on the Met Office coupled atmosphere–land–ocean–sea ice model. Mon. Wea. Rev., 143, 4678–4694, https://doi.org/10.1175/MWR-D-15-0174.1.
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, https://doi.org/10.1175/2008MWR2459.1.
Lorenz, E. N., 1982: Atmospheric predictability experiments with a large numerical model. Tellus, 34, 505–513, https://doi.org/10.3402/tellusa.v34i6.10836.
MacLachlan, C., and Coauthors, 2015: Global Seasonal forecast system version 5 (GloSea5): A high-resolution seasonal forecast system. Quart. J. Roy. Meteor. Soc., 141, 1072–1084, https://doi.org/10.1002/qj.2396.
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, https://doi.org/10.1175/1520-0469(1971)028<0702:DOADOI>2.0.CO;2.
Madec, G, and the NEMO Team, 2008: NEMO ocean engine. Note du Pole de modélisation, Institut Pierre-Simon Laplace (IPSL) No. 27, 401 pp., https://epic.awi.de/id/eprint/39698/1/NEMO_book_v6039.pdf.
Meehl, G. A., 1995: Global coupled general circulation models. Bull. Amer. Meteor. Soc., 76, 951–957, https://doi.org/10.1175/1520-0477-76.6.951.
Mogensen, K. S., M. A. Balmaseda, A. Weaver, M. Martin, and A. Vidard, 2009: NEMOVAR: A variational data assimilation system for the NEMO ocean model. ECMWF Newsletter, No. 120, ECMWF, Reading, United Kingdom, 17–21, https://www.ecmwf.int/sites/default/files/elibrary/2009/17483-nemovar-variational-data-assimilation-system-nemo-ocean-model.pdf.
Molod, A., and Coauthors, 2020: GEOS-S2S version 2: The GMAO high-resolution coupled model and assimilation system for seasonal prediction. J. Geophys. Res., 125, e2019JD031767, https://doi.org/10.1029/2019JD031767.
Mulholland, D. P., P. Laloyaux, K. Haines, and M. A. Balmaseda, 2015: Origin and impact of initialization shocks in coupled atmosphere–ocean forecasts. Mon. Wea. Rev., 143, 4631–4644, https://doi.org/10.1175/MWR-D-15-0076.1.
Orsolini, Y. J., R. Senan, G. Balsamo, F. J. Doblas-Reyes, F. Vitart, A. Weisheimer, A. Carrasco, and R. E. Benestad, 2013: Impact of snow initialization on sub-seasonal forecasts. Climate Dyn., 41, 1969–1982, https://doi.org/10.1007/s00382-013-1782-0.
Owens, R. G., and T. D. Hewson, 2018: ECMWF forecast user guide. ECMWF, Reading, United Kingdom, 10 pp., https://doi.org/10.21957/m1cs7h.
Palmer, T. N., F. J. Doblas-Reyes, A. Weisheimer, and M. J. Rodwell, 2008: Toward seamless prediction: Calibration of climate change projections using seasonal forecasts. Bull. Amer. Meteor. Soc., 89, 459–470, https://doi.org/10.1175/BAMS-89-4-459.
Penny, S. G., and T. M. Hamill, 2017: Coupled data assimilation for integrated earth system analysis and prediction. Bull. Amer. Meteor. Soc., 98, ES169–ES172.
Rae, J. G. L., H. T. Hewitt, A. B. Keen, J. K. Ridley, A. E. West, C. M. Harris, E. C. Hunke, and D. N. Walters, 2015: Development of the Global Sea Ice 6.0 CICE configuration for the Met Office Global Coupled model. Geosci. Model Dev., 8, 2221–2230, https://doi.org/10.5194/gmd-8-2221-2015.
Robertson, A. W., A. Kumar, M. Peña, and F. Vitart, 2015: Improving and promoting subseasonal to seasonal prediction. Bull. Amer. Meteor. Soc., 96, ES49–ES53, https://doi.org/10.1175/BAMS-D-14-00139.1.
Saha, S., and Coauthors, 2006: The NCEP climate forecast system. J. Climate, 19, 3483–3517, https://doi.org/10.1175/JCLI3812.1.
Skachko, S., and Coauthors, 2019: Weakly coupled atmosphere–ocean data assimilation in the Canadian global prediction system (v1). Geosci. Model Dev., 12, 5097–5112, https://doi.org/10.5194/gmd-12-5097-2019.
Valcke, S., 2013: The OASIS3 coupler: A European climate modelling community software. Geosci. Model Dev., 6, 373–388, https://doi.org/10.5194/gmd-6-373-2013.
Vitart, F., and Coauthors, 2017: The Subseasonal to Seasonal (S2S) prediction project database. Bull. Amer. Meteor. Soc., 98, 163–173, https://doi.org/10.1175/BAMS-D-16-0017.1.
Walters, D., and Coauthors, 2017: The Met Office unified model global atmosphere 6.0/6.1 and JULES global land 6.0/6.1 configurations. Geosci. Model Dev., 10, 1487–1520, https://doi.org/10.5194/gmd-10-1487-2017.
Wang, A., and X. Zeng, 2013: Development of global hourly 0.5° land surface air temperature datasets. J. Climate, 26, 7676–7691, https://doi.org/10.1175/JCLI-D-12-00682.1.
Wheeler, M., and G. N. Kiladis, 1999: Convectively coupled equatorial waves: Analysis of clouds and temperature in the wavenumber–frequency domain. J. Atmos. Sci., 56, 374–399, https://doi.org/10.1175/1520-0469(1999)056<0374:CCEWAO>2.0.CO;2.
Wheeler, M. C., 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, https://doi.org/10.1175/1520-0493(2004)132<1917:AARMMI>2.0.CO;2.
White, C. J., and Coauthors, 2017: Potential applications of subseasonal-to-seasonal (S2S) predictions. Meteor. Appl., 24, 315–325, https://doi.org/10.1002/met.1654.
Williams, K. D., and Coauthors, 2015: The Met Office Global Coupled model 2.0 (GC2) configuration. Geosci. Model Dev., 8, 1509–1524, https://doi.org/10.5194/gmd-8-1509-2015.
Zhang, C., and J. Ling, 2017: Barrier effect of the Indo-Pacific Maritime Continent on the MJO: Perspectives from tracking MJO precipitation. J. Climate, 30, 3439–3459, https://doi.org/10.1175/JCLI-D-16-0614.1.
Zhang, S., and Coauthors, 2020: Coupled data assimilation and parameter estimation in coupled ocean–atmosphere models: A review. Climate Dyn., 54, 5127–5144, https://doi.org/10.1007/s00382-020-05275-6.
