1. Introduction
Decadal climate predictions have recently emerged as a new field in climate studies (Hurrell et al. 2009). Situated between short-term forecasts and long-term climate projections, they aim to predict climate conditions for the next 10–30 yr. The prediction horizon thus centers on a time horizon that is considered very relevant for many decisions related to climate mitigation and adaption (infrastructure planning, resources management, and emission mitigation strategies).
Decadal prediction skill can be expected mainly from two sources: on the one hand from anthropogenic (greenhouse gases and aerosols) and natural (volcanic and solar) external forcing (Meehl et al. 2009) and on the other hand from (predictable) modes of internal variability. This is based on the underlying assumption (Hurrell et al. 2009) that all climate system predictions share the same processes and mechanisms, independent of the targeted temporal scale, and consequently benefit from initialization of a global climate model (GCM) with best estimates of the observed climate state (Hazeleger et al. 2013a; Branstator et al. 2012; Hawkins and Sutton 2009). Therefore, a skilful prediction relies on a high quality initialization procedure (Branstator and Teng 2010).
The main drivers for decadal-scale climate variability emerge from oceanic processes in the midlatitudes, such as the Pacific decadal oscillation (Meehl et al. 2010; Mochizuki et al. 2010) or meridional overturning and gyre circulations in the Atlantic (Wouters et al. 2013; Pohlmann et al. 2009; Keenlyside et al. 2008), which are predictable on a decadal time scale. Additionally, these patterns of decadal variability extend into the tropical oceans, and it may be that much of their impact is communicated to the atmosphere through (tropical) SST changes (Goddard et al. 2012).
Although a propagation of oceanic low-frequency variations into the atmosphere and further onto land areas has been proven theoretically by related changes in precipitation, river flow, surface temperature variations, or hurricane activity over land (Enfield et al. 2001; Knight et al. 2006; Zhang and Delworth 2006; Sutton and Hodson 2005; Smith et al. 2010), the skill of decadal predictions over land is generally low (Smith et al. 2007) and drops drastically when long-term temperature trends (which are fairly well predictable from continued greenhouse gas emissions) are removed (van Oldenborgh et al. 2012; Corti et al. 2012).
Yet, oceans and external forcing might not be the only source of predictability at time scales longer than 1 or 2 weeks. Various studies suggest that physical processes over land are capable of providing detectable forecast skill. For example, the relevance of realistic soil moisture initialization for skilful projection of precipitation and temperature up to 2 months ahead has been shown by Koster et al. (2011) and Conil et al. (2007). Also, information on the snow mass state can enhance skill for boreal summer (Douville 2010).
Thus relevant aspects of the land–atmosphere coupling in the model formulation can enhance skill of predictions over land. Since promising results have been achieved by triggering the soil moisture–climate interaction, we extend the exploration of the contribution of land–atmosphere coupling to climate predictability and tap the subsequent control on the atmospheric energy and water budget, which is the type and state of vegetation. The distribution of natural vegetation, its phenology and development state are largely controlled by climate (Köppen 1884), but in return vegetation systematically affects regional climate by modifying the surface energy and water balance (Pielke et al. 2011; Arora 2002).
In an earlier study (Weiss et al. 2012), we illustrated the beneficial impact of a realistic observation-based vegetation state on the potential predictability and skill of a climate model. Here we study whether dynamic vegetation–climate interaction positively affects decadal prediction skill in a model system initialized from observations. In the current study we explore impacts on forecast skill by (i) using a more realistic initialization of vegetation based on long offline spinup runs with a dynamic vegetation model driven by observed climate and (ii) allowing a two-way interaction between the vegetation state and the climate system by coupling the vegetation model closely to the atmospheric module. The experiments to test this hypothesis are carried out with a set of decadal hindcast experiments generated by a fully coupled atmosphere–land–ocean–sea ice model.
2. Experimental setup
a. Methodology
To assess the quality of predictions and the improved skill resulting from dynamic vegetation, a set of hindcast experiments is carried out and evaluated against observed datasets. To this end, we use two versions of the European Consortium Earth System Model (EC-Earth) Hazeleger et al. 2012) following the decadal prediction protocol from phase 5 of the Coupled Model Intercomparison Project (CMIP5) (Meehl et al. 2009; Taylor et al. 2012). The full set of prediction experiments consists of five-member ensembles of 10-yr simulations, initialized on 1 November every 5 yr between 1960 and 1995, resulting in a total of eight start dates.
Similar to the experiments reported by Hazeleger et al. (2013b), full-field initialization is used, with ocean initial conditions based on a multivariate three-dimensional variational data assimilation (3DVAR) method for the Nucleus for European Modelling of the Ocean (NEMO) ocean model (NEMOVAR) (Balmaseda et al. 2012; Weaver et al. 2005), assimilating observed three-dimensional temperature, salinity, and sea surface height. Sea ice conditions are based on NEMO, version 2, and the Louvain-la-Neuve Sea Ice Model, version 2 (LIM2), sea ice model forced by surface fluxes obtained from the Drakkar forcing set, version 4.3 (Brodeau et al. 2010). The atmosphere initial conditions are based on the European Centre for Medium-Range Weather Forecasts (ECMWF) Interim Re-Analysis (ERA-Interim; Dee et al. 2011) and, prior to 1989, on 40-yr ECMWF Re-Analysis (ERA-40) data (Uppala et al. 2005), which are perturbed to create initial states for an ensemble of five members (for details, see Weiss et al. 2012). The external forcing (greenhouse gases, ozone, natural and anthropogenic aerosols, solar activity, and land use) is prescribed based on the CMIP5-recommended historical datasets (Moss et al. 2010).
In the current study, the vegetation state in EC-Earth is characterized solely by a variable value of the leaf area index (LAI) of high and low vegetation, replacing the default configuration using a static prescribed LAI value. These quantities are initialized with fields obtained from a long spinup of the Lund–Potsdam–Jena General Ecosystem Simulator (LPJ-GUESS) dynamic vegetation model (see below for model descriptions) forced by the observational climate dataset of the Climate Research Unit (CRU) time series, version 2.1 (TS2.1) (Mitchell and Jones 2005), whereby on a given start date all five members are initialized with the same vegetation fields. Surface albedo and surface roughness are adjusted to prescribed time-varying land-use characteristics. Exploration of LAI and albedo retrieved from radiances observed by the Moderate Resolution Imaging Spectroradiometer (MODIS) revealed a very low correlation between these variables. This may be related to compensating effects of green leaf area on the visible and near-infrared surface reflectivity (Teuling and Seneviratne 2008). Therefore, no adjustment of surface albedo to updated LAI values was applied. Also, variations of the surface albedo with snow extent do not depend on LAI values in these simulations.
Two parallel sets of experiments were carried out. In the control experiment (labeled V2.3) the hindcast runs are generated with the default EC-Earth, version 2.3 (EC-Earth2.3), configuration with a static description of LAI, whereas the second experiment (labeled V2.4) uses the new EC-Earth, version 2.4 (EC-Earth2.4), including dynamic LAI generated by the fully coupled LPJ-GUESS dynamic vegetation model. The EC-Earth2.4 model configuration is also used to generate an initialized control simulation encompassing the period 1960–2000. This simulation is used for exploring the vegetation drift characteristics in the hindcast runs.
b. Model description
EC-Earth2.4 is the successor of EC-Earth2.3 (Hazeleger et al. 2010, 2012), which is composed of an atmospheric model initially derived from the Integrated Forecast System cycle 31r1 of the ECMWF (http://old.ecmwf.int/research/ifsdocs/CY31r1), NEMO version 2 (Madec 2008), LIM2 (Goosse and Fichefet 1999), and the land surface scheme Hydrology Tiled ECMWF Scheme of Surface Exchanges over Land (HTESSEL) (Balsamo et al. 2009). HTESSEL solves the energy balance for six different land surface tiles including low and high vegetation. Tile fractions are calculated dynamically from a static land-use map (prescribing background values of high and low vegetation and bare ground) and time-varying cover fractions of snow and intercepted water. Total surface fluxes are calculated as a weighted average of tile-specific values, calculated using a resistance approach involving aerodynamic and surface resistances to account for the transfer efficiency of heat and water vapor over a vertical temperature and humidity gradient.
A major update from V2.3 is the coupling with the state-of-the-art dynamic vegetation and ecosystem model LPJ-GUESS (Smith et al. 2001). LPJ-GUESS shares process-based representations of physiology and biogeochemistry with the Lund–Potsdam–Jena Dynamic Global Vegetation Model (LPJ-DGVM; Sitch et al. 2003), but it has more realistic treatments of population dynamics, stand structure, and the competition for resources among the simulated woody plant individuals and their herbaceous understory (Smith et al. 2001). Plant populations of various cohorts of plant functional types (PFTs) are simulated in a number of replicate patches (10 in this study) for each vegetated land surface tile. This is necessary in order to account for the differences arising from disturbance history (including wildfires), establishment, and mortality of simulated individuals, which are all treated as stochastic processes in each modeled patch.
LPJ-GUESS has been evaluated in numerous studies (see, e.g., Ahlström et al. 2012; Hickler et al. 2012; Piao et al. 2013; and references therein). It has also been coupled to the Rossby Centre Regional Atmospheric Climate Model (RCA3; Samuelsson et al. 2006) to yield the coupled model system RCA-GUESS (Wramneby et al. 2010; Smith et al. 2011), which was used to identify hotspots of future vegetation–climate feedbacks over Europe.
To capture the major plant types determining global biomes, the version of LPJ-GUESS used in this study uses 11 PFTs for natural vegetation, where 2 are herbaceous and 9 are woody types. These modeled PFTs are defined using fixed parameters to determine their morphology, phenology, and shade tolerance. They also contain bioclimatic limits to determine windows of climatic conditions in which survival, regeneration, and growth is possible. All PFT parameters used in this study are identical to those given by Ahlström et al. (2012). For each PFT the phenological cycle of the vegetation is calculated. Land-use types related to agricultural crops or pastures can be accommodated as well. The model also includes a specific representation of soil biogeochemistry, and a soil water balance algorithm. The model is set up to run on the same grid as EC-Earth, and similar soil hydraulic characteristics are used.
The coupling allows for the replacement the default land-use characteristics and vegetation parameters in EC-Earth by values calculated in LPJ-GUESS. In return, LPJ-GUESS is forced by shortwave radiation, temperature, and total precipitation from EC-Earth and a prescribed CO2 concentration. It returns LAI values for the high vegetation and the low vegetation tiles, which replace the static lookup table values that are the default in the land surface scheme of EC-Earth. This is accomplished by running LPJ-GUESS twice for each grid cell in the interactive coupling scheme. One run uses all 11 PFTs to generate the LAI of the high vegetation in the cell. The second run restricts the set of modeled PFTs to the two herbaceous types in order to generate the LAI of the low vegetation in each cell. The existing tiling and grid structure of HTESSEL is maintained.
Although HTESSEL and LPJ-GUESS are exposed to the same precipitation and near-surface climate, soil water budgets may diverge because of different treatment of soil water and the related fluxes in the two models. No procedure has been applied to avoid this divergence in this experiment.
c. Bias and measures of skill
Model output of near-surface air temperature is evaluated against the combined ERA-40/ERA-Interim dataset (Uppala et al. 2005; Dee et al. 2011). First we analyze model drift, by comparing hindcast anomalies to reanalysis anomalies. Climatologies are calculated by averaging results from the whole experiment period (1960–2005). In a next step, we examine the monthly and seasonal [June–August (JJA)] mean absolute temperature bias after lead times of 1, 6, and 9 months and 3–10 yr, respectively, averaged over five members and eight start dates.
Following Corti et al. (2012), decadal climate hindcast skill is assessed probabilistically using the estimated relative frequency of a given event [e.g., the 2-m temperature (T2M) exceeding some threshold]. Given a specific forecast predicting an increased likelihood of this event, the probability that the event will actually occur is diagnosed in a reliability diagram (Hartmann et al. 2002). Here we explore the occurrence of T2M above the climatological median for selected regions (defined in Table 1), where vegetation is expected to have a nonnegligible impact. As is common in decadal predictability studies, we analyze lead times of 1, 2–5, and 6–9 yr after initialization (Ho et al. 2013; Goddard et al. 2012).
Definition of analyzed regions.
The reliability diagram groups the forecasts into discrete bins with equal probability. For each forecast bin, the frequency with which the event was observed to occur is then plotted on the vertical axis. For perfect reliability the forecast probability and the frequency of occurrence are equal, leading to a straight diagonal on a reliability plot. The forecast has some reliability as long as it has a positive slope. A slope flatter than the diagonal points at overconfidence of the system, since the forecasted likelihood range of occurrence tends to exceed the observed range. The distribution of forecasts over the likelihood range determines the resolution of the forecast. A low resolution is generated by a forecast population centered around a (climatological) value, so the ability of the forecast to resolve the set of sample events into subsets with characteristically different frequencies is low (Corti et al. 2012).
Finally, we use the ranked probability skill score (RPSS) (Weigel et al. 2007; Müller et al. 2005), which measures the performance of the probabilistic forecast relative to the observed climatology. The RPSS considers both the shape and position of the forecasted probability density function (PDF). The RPSS penalizes forecasts with frequencies distribution deviating greatly from the observed climatology. It ranges from −1 to 1, where 0 indicates no skill in comparison to climatology and 1 indicates a perfect score. Global maps of RPSS are shown for 1, 2–5, and 6–9 yr and Northern and Southern Hemisphere summer. Here we define Southern Hemisphere summer as consisting of November–January (NDJ) in order to include the first month after initialization.
3. Results
a. Model drift and bias
The initialization of a decadal model prediction with observations leads to an inevitable drift due to the discrepancy between the observed state and the preferred model climatology. This drift is visible both in the atmospheric as well as vegetation variables, as shown in Fig. 1 for T2M and Fig. 2 for LAI. Figure 1 compares globally averaged, 12-month running means of ensemble mean T2M anomalies for land cells (excluding Antarctica) to observations [reanalysis and CRU time series, version 3.1 (TS3.1)] for the whole experiment period. Anomalies were computed with respect to the 1960–2000 mean value of all forecasts and observations, leading to a distribution of the ensemble of model results and observations spread around zero in Fig. 1. The cooling drift of the model during approximately the first year becomes particularly evident and is a persistent feature of this model version (Hazeleger et al. 2012). On a global average, ensemble mean anomalies of the two model versions remain relatively close together. In V2.4, the ensemble spread is increasing toward the end of the simulations in one of the eight decades.
Global average of 12-month running mean ensemble mean and std dev of T2M anomalies, relative to the mean observed or simulated values over the 1965–2005 period. For better visibility, we separate (top) even and (bottom) odd start dates for EC-Earth2.3 (blue line), EC-Earth2.4 (red line), ERA-40 (black line), and CRU TS3.1 (green line). Shading represents one std dev of the five ensemble members in the simulations.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
Global average of 12-month running mean ensemble mean and std dev of T2M anomalies, relative to the mean observed or simulated values over the 1965–2005 period. For better visibility, we separate (top) even and (bottom) odd start dates for EC-Earth2.3 (blue line), EC-Earth2.4 (red line), ERA-40 (black line), and CRU TS3.1 (green line). Shading represents one std dev of the five ensemble members in the simulations.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
Global average of 12-month running mean ensemble mean and std dev of T2M anomalies, relative to the mean observed or simulated values over the 1965–2005 period. For better visibility, we separate (top) even and (bottom) odd start dates for EC-Earth2.3 (blue line), EC-Earth2.4 (red line), ERA-40 (black line), and CRU TS3.1 (green line). Shading represents one std dev of the five ensemble members in the simulations.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
Ensemble mean LAI of high vegetation as function of forecast lead time generated with LPJ-GUESS. Each panel shows a different region, defined in Table 1. Green lines in different shades denote coupled EC-Earth2.4 simulations, with each line representing a different start date. Blue lines show values averaged over the eight start dates. Orange and red lines represent LAI values generated by LPJ-GUESS driven by CRU TS2.1 climate data for the 1960–70 and 1995–2005 periods, respectively (taken from Ahlström et al. 2012). Horizontal gray lines are the LAI values in the default HTESSEL configuration in V2.3.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
Ensemble mean LAI of high vegetation as function of forecast lead time generated with LPJ-GUESS. Each panel shows a different region, defined in Table 1. Green lines in different shades denote coupled EC-Earth2.4 simulations, with each line representing a different start date. Blue lines show values averaged over the eight start dates. Orange and red lines represent LAI values generated by LPJ-GUESS driven by CRU TS2.1 climate data for the 1960–70 and 1995–2005 periods, respectively (taken from Ahlström et al. 2012). Horizontal gray lines are the LAI values in the default HTESSEL configuration in V2.3.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
Ensemble mean LAI of high vegetation as function of forecast lead time generated with LPJ-GUESS. Each panel shows a different region, defined in Table 1. Green lines in different shades denote coupled EC-Earth2.4 simulations, with each line representing a different start date. Blue lines show values averaged over the eight start dates. Orange and red lines represent LAI values generated by LPJ-GUESS driven by CRU TS2.1 climate data for the 1960–70 and 1995–2005 periods, respectively (taken from Ahlström et al. 2012). Horizontal gray lines are the LAI values in the default HTESSEL configuration in V2.3.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
Figure 2 shows ensemble mean LAI as function of lead time for each of the eight start dates (and their mean value) generated by V2.4. Also shown is LAI generated by the offline CRU TS2.1–driven LPJ-GUESS spinup realization of the first (1960) and last (1995) decades of the experiment. LAI in the coupled climate model drifts toward higher values than generated by the offline LPJ-GUESS run in all regions (defined in Table 1) except northern Russia. Australia recovers from an overshoot in the beginning of the simulation. The global mean LAI shows a positive drift which saturates after roughly three seasonal cycles (Fig. 2, top left).
Apart from the drift in the surface temperature, shown in Fig. 1, a systematic bias in precipitation and surface radiation is present in the coupled model system. A control simulation, initialized using 1960 start data, was used to generate a 30-yr mean (1971–2000) climatology of temperature, precipitation, and photosynthetically active radiation (PAR). A comparison to the 30-yr mean climatologies derived from the CRU TS2.1 dataset used to drive the offline LPJ-GUESS model is shown in Fig. 3 for JJA [but similar results are obtained for December–February (DJF)]. The comparison reveals a wet bias in most extratropical areas (north of 30°N, south of 30°S) in both DJF and JJA. PAR is biased particularly high in the tropics over the Amazon and Africa.
Differences in the 1971–2000 climatologies of surface temperature, total precipitation, and PAR between a control run with the coupled EC-Earth simulation and the LPJ-GUESS simulation driven by CRU TS2.1 for boreal summer (JJA).
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
Differences in the 1971–2000 climatologies of surface temperature, total precipitation, and PAR between a control run with the coupled EC-Earth simulation and the LPJ-GUESS simulation driven by CRU TS2.1 for boreal summer (JJA).
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
Differences in the 1971–2000 climatologies of surface temperature, total precipitation, and PAR between a control run with the coupled EC-Earth simulation and the LPJ-GUESS simulation driven by CRU TS2.1 for boreal summer (JJA).
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
This drift in the mean climatology leads to different climatological mean LAI values in LPJ-GUESS when driven by the CRU TS2.1 dataset versus the coupled climate model (Fig. 4). We see in both the extratropics and the tropics a tendency for higher LAI values in the coupled run, with the patterns matching the positive bias in precipitation in, for example, North America and eastern Eurasia and the PAR bias in the Amazon and African tropics.
As in Fig. 3, for the (left) high and (right) low LAI.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
As in Fig. 3, for the (left) high and (right) low LAI.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
As in Fig. 3, for the (left) high and (right) low LAI.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
An analysis of the anomaly correlation between LAI and T2M or precipitation (not shown) reveals that during JJA most areas below 50°N display a negative correlation between LAI and T2M, while the correlation between LAI and precipitation is positive at nearly all locations on the globe. The negative precipitation bias of EC-Earth in the Amazon (Fig. 3, middle) does not lead to reduced vegetation growth, as moisture availability is not a limiting factor here. Instead, negative precipitation anomalies are normally associated with enhanced insolation and a higher net radiation (Van der Molen et al. 2011), which may explain the positive LAI drift in these areas. In the extratropical areas (notably the western half of North America) the anomaly correlation analysis displays a dependence of particularly high vegetation LAI on JJA precipitation, which may explain the positive LAI drift at high latitudes. Thus, the LAI drift shown in Fig. 2 in the coupled simulation is mainly a result of different hydrological climatology in EC-Earth and the CRU forcing data, leading to wetter soils and lower soil water stress conditions at high latitudes during JJA and higher PAR in the tropics.
The drifts in temperature and precipitation are further illustrated in Table 2, where temperature and precipitation differences between the decadal predictions with V2.4 and CRU TS2.1 are shown per region. Our findings confirm the results of earlier studies, showing that in midlatitudes EC-Earth version 2 is too cold and too wet compared to observations (Hazeleger et al. 2012).
EC-Earth2.4 bias of annual mean temperature and precipitation relative to the CRU TS2.1 dataset per region. Data are averaged over 6–9-yr lead time for all forecasts.
In spite of the clear LAI drift shown in Fig. 2, the LAI values simulated here still compare well to observed values reported in previous studies. For a global average, Asner et al. (2003) found a value of 4.5 with a standard deviation of 2.5 using a large in situ dataset. For the Amazon, Caldararu et al. (2012) report LAI values of 3.5–6.
The model temperature bias consolidates after a few years. Figure 5 shows this development by snapshots of the bias after month 1 (November), month 6 (April), month 9 (July), and the Northern Hemisphere summer (JJA) averaged for years 3–10 after initialization.
T2M bias after lead times (top)–(bottom) of 1, 6, and 9 months and for Northern Hemisphere summer after 3–10 yr for (left) EC-Earth2.3 and (center) EC-Earth2.4 in comparison to ERA-40 and (right) the difference between V2.4 and V2.3. Wide shadings represent nonsignificant differences at roughly 95% confidence.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
T2M bias after lead times (top)–(bottom) of 1, 6, and 9 months and for Northern Hemisphere summer after 3–10 yr for (left) EC-Earth2.3 and (center) EC-Earth2.4 in comparison to ERA-40 and (right) the difference between V2.4 and V2.3. Wide shadings represent nonsignificant differences at roughly 95% confidence.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
T2M bias after lead times (top)–(bottom) of 1, 6, and 9 months and for Northern Hemisphere summer after 3–10 yr for (left) EC-Earth2.3 and (center) EC-Earth2.4 in comparison to ERA-40 and (right) the difference between V2.4 and V2.3. Wide shadings represent nonsignificant differences at roughly 95% confidence.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
For the month 1 lead times, we find only small differences between the two model versions. A slight reduction in bias is seen over northwestern Russia, the northwestern United States, and Australia, but on the other hand small increases in bias occur over Central America (Fig. 5). In month 6, we find a significant reduction of the (negative) bias over Russia, as well as temperature increases over India and the Sahel. In month 9, a reduction of the (negative) temperature bias is seen in southern Europe, northern Russia, and Canada, as well as largely over the tropics north of the equator, stretching from Africa over to India and Southeast Asia. This tendency is confirmed for the long-term JJA average of years 3–10, when model drift is reduced. The location of the reduction in bias points at a significant role of vegetation in boreal spring and summer conditions, and in transition zones between dry and wet climates, such as the Sahel.
b. Evaluation of the decadal forecasts
With our initial question of whether dynamic vegetation has a positive impact on decadal prediction in mind, we are especially interested in the usefulness of our forecasts: that is, their reliability and difference between the two model versions. Like V2.3, V2.4 drifts toward a lower temperature than the initialization state, but its equilibrium climate is slightly warmer than in V2.3 (see Table 3). The first 14 months illustrate the impact of the two different vegetation initialization strategies. EC-Earth2.4 uses LAI fields based on a CRU climate spinup, while LAI in EC-Earth2.3 is initialized (and prescribed) with a constant lookup table value. Figure 2 shows that this prescribed LAI value is generally higher throughout most of the year (certainly outside the growing season) than the seasonal cycle followed in the coupled simulation.
Globally averaged temperature bias (°C) for EC-Earth2.3 and EC-Earth2.4 as function of forecast lead time.
To illustrate the effect of varying LAI, Fig. 6 shows T2M for the global land area and the five selected regions for the first 14 months of the simulations for one start date (1960). Although the initial LAI values are quite different for the two model versions, the impact of the initialization on T2M during the first months is small as the simulations were started in November (in accordance with the CMIP5 decadal prediction protocol) when vegetation is at a minimum in the Northern Hemisphere. Indeed, we do see an improvement in T2M in the first months of the simulations for Australia, where our initialization falls in spring. Also in North Africa a strong impact is visible.
T2M of the first 14 months after initialization for five ensemble members in 1960 for the global mean land area and the selected regions defined in Table 1 for EC-Earth2.3 (blue lines), EC-Earth2.4 (red lines), and ERA40 (black line).
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
T2M of the first 14 months after initialization for five ensemble members in 1960 for the global mean land area and the selected regions defined in Table 1 for EC-Earth2.3 (blue lines), EC-Earth2.4 (red lines), and ERA40 (black line).
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
T2M of the first 14 months after initialization for five ensemble members in 1960 for the global mean land area and the selected regions defined in Table 1 for EC-Earth2.3 (blue lines), EC-Earth2.4 (red lines), and ERA40 (black line).
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
Apart from the different initializations, the update of LAI information throughout the forecast does generate some noticeable effects on T2M. The global mean land temperature in the peak summer season is approximately 0.2°C warmer in V2.4 than in V2.3. For the tropical regions, particularly Africa, marked differences are shown between the simulations in the warm season.
Figure 7 shows reliability diagrams for the probability that T2M exceeds the climatological median for the regions defined in Table 1, for the summer months in the respective hemispheres. Again, we show the two periods 2–5- and 6–9-yr lead time. Predictions appear to be reliable in all regions except for northern Russia, for both model versions. Although a formal significance test of the difference between the two model versions [e.g., as proposed by Ferro (2014)] was not applied, we find a slight improvement in some of the regions under EC-Earth2.4, by comparing the deviation from the diagonal, but this improvement is fairly small. With increasing lead time, the sample populations tend to become concentrated in the climatological bins around the median (see histograms in the bottom right of the panels in Fig. 7), which hints at a sample size that is insufficiently large. For example, the histogram of North America shifts from an asymmetric shape in the 2–5-yr sample to a dome-shaped distribution in the 6–9-yr lead time sample. The mean bias is reduced at the expense of resolution of the probabilistic forecasts.
Reliability diagrams of the probability of T2M exceeding the climatological median in the respective summer seasons (NDJ for Australia and JJA for all other areas). Regions are as in Table 1. Results are for lead times of (left) 2–5 and (right) 6–9 yr. The histograms in the bottom right of the panels depict the frequency distribution of the EC-Earth forecasts. The blue lines represent EC-Earth2.3 and the red lines represent EC-Earth2.4.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
Reliability diagrams of the probability of T2M exceeding the climatological median in the respective summer seasons (NDJ for Australia and JJA for all other areas). Regions are as in Table 1. Results are for lead times of (left) 2–5 and (right) 6–9 yr. The histograms in the bottom right of the panels depict the frequency distribution of the EC-Earth forecasts. The blue lines represent EC-Earth2.3 and the red lines represent EC-Earth2.4.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
Reliability diagrams of the probability of T2M exceeding the climatological median in the respective summer seasons (NDJ for Australia and JJA for all other areas). Regions are as in Table 1. Results are for lead times of (left) 2–5 and (right) 6–9 yr. The histograms in the bottom right of the panels depict the frequency distribution of the EC-Earth forecasts. The blue lines represent EC-Earth2.3 and the red lines represent EC-Earth2.4.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
Finally, Fig. 8 shows the difference in RPSS between the two model versions for lead times of 1, 2–5, and 6–9 yr, for Northern Hemisphere winter and summer. Particularly in the first year after initialization and more in months 1–3 (winter) than in months 8–10 (summer), RPSS increases in many areas when moving from V2.3 to V2.4. In years 2–5, an increase in RPSS is only visible for Northern Hemisphere summer, while in years 6–9 we find improvements in both seasons. A skill better than a climatological forecast is not found frequently (not shown).
Difference in RPSS, comparing V2.4 vs V2.3, for (top) NDJ and (bottom) JJA for lead times of (left) 1, (center) 2–5, and (right) 6–9 yr. RPSS is calculated with respect to ERA-40 climatology.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
Difference in RPSS, comparing V2.4 vs V2.3, for (top) NDJ and (bottom) JJA for lead times of (left) 1, (center) 2–5, and (right) 6–9 yr. RPSS is calculated with respect to ERA-40 climatology.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
Difference in RPSS, comparing V2.4 vs V2.3, for (top) NDJ and (bottom) JJA for lead times of (left) 1, (center) 2–5, and (right) 6–9 yr. RPSS is calculated with respect to ERA-40 climatology.
Citation: Journal of Climate 27, 22; 10.1175/JCLI-D-13-00684.1
4. Discussion and conclusions
The modest skill over land of state-of-the-art decadal prediction systems prohibits their use in climate adaptation and service communities. In this study, we have taken a first tentative step to increase decadal climate predictability over land by coupling an existing and well-established process- and individual-based dynamic vegetation model to a fully coupled atmosphere–land–ocean–sea ice model. This coupled system has additional degrees of freedom induced by a physically and physiologically realistic response of vegetation to ambient climate conditions, as previously prescribed variables are replaced by an interactive component. In addition, it is able to generate feedbacks of vegetation state anomalies to the atmosphere. The potential ability of the vegetation to memorize climate anomalies at seasonal to multiannual time scales may provide an extra source of predictability over land, when realistic initial states and subsequent vegetation responses are implemented in the decadal prediction system. Apart from the representation of extra memory, the vegetation response to trends like increases in global mean temperature or CO2 concentrations may result in better projections of the climate system.
The development of a new coupled climate–vegetation system, EC-Earth2.4, parallels similar developments carried out earlier by a range of Earth system modeling groups, some of them participating in the CMIP5 protocol. Most of these Earth system models (ESMs) were developed with the main purpose of exploring carbon–climate feedback processes (Friedlingstein et al. 2006) or biogeophysical feedbacks at the global (Levis et al. 2000; Delire et al. 2011) and regional (Wramneby et al. 2010) scale. Here we applied such a coupled model system to explore additional predictability of T2M, following earlier potential predictability studies by, for example, Weiss et al. (2012) and Van den Hurk et al. (2003).
With the new model version, we find a reduced T2M bias in many regions, and forecasts are slightly more reliable. Even though we succeed in increasing the RPSS in a few regions, especially in the first year after initialization, only a very few areas have skill better than climatology. The limited sample size of five ensemble members per experiment and a start date every 5 yr may have limited the potential for significant increases in either reliability or RPSS scores. In addition, the additional degrees of freedom introduced by the interactive vegetation generate additional noise to the model that does not necessarily translate into improved skill.
The regions with the greatest bias improvement are found in the tropical Sahel and India and at high latitudes in boreal Russia and North America. Areas with low bias in the reference version (e.g., continental United States and Europe) are not negatively affected by the new model version. The results found here are indeed in agreement with previous studies by Douville (2010), Koster et al. (2011), and Seneviratne et al. (2010). The model study by Koster et al. (2004) exploring regions where boreal summer precipitation is controlled by soil moisture identified a number of hotspots (North America, Sahel, and northern India). These areas coincide with hotspots for strong observation-based soil moisture–temperature coupling identified by Miralles et al. (2012). Our results show that forecasts are improved in a zone that is exposed to shifts in the ITCZ and monsoon cycles, consistent with the transitional dry to wet climate zones that appear to expose a strong land–atmosphere coupling.
Some practical limitations apply to the setup that was chosen here. Since multiple interactions between vegetation and the atmosphere potentially lead to unstable feedbacks, we chose to narrow the list of exchanged variables in this first coupling approach, accepting certain inconsistencies between models. One consequence of this approach is that generally positive LAI drifts in the coupled system were generated, which have their origin in a systematic temperature, precipitation, and radiation bias in EC-Earth. This leads to generally more favorable vegetation growing conditions in the tropics (where more radiation reaches the surface) and at higher latitudes (where a positive precipitation bias appear to be effective). Additional experiments with the coupled model in which the systematic LAI bias was removed revealed a further improvement of particularly the climatological temperature distribution (not shown). However, application of a bias correction to initialized decadal forecasts, as explored in this study, is not trivial. The bias is a function of the forecast lead time (Fig. 2), and a well-calibrated relaxation scheme must be developed in order to avoid overcompensation or undercompensation of the LAI bias. Preferably, a correction to the model climate output used to drive the offline vegetation model is to be implemented, which would ensure a close correspondence between the offline generated initial vegetation states and the online vegetation evolution. Such a bias correction procedure is currently under development.
To explore the added value of realistic initialization of vegetation, the experimental design must allow the choice of further start dates in addition to November as prescribed by the CMIP5 protocol. For example, it is likely that the initial vegetation state in early NH spring contains more useful and extractable information, certainly for the seasonal time scale. In addition, the findings presented here may be very specific for the current model configuration. A more extensive exploration should be applied involving an ensemble of model systems.
Acknowledgments
Xueli Wang assisted in some of the data processing. This work was funded by the European Commission’s 7th Framework Programme, under Grant 525 Agreement 226520, COMBINE project. PAM, BS, and GS acknowledge support from the Swedish strategic research area “Modelling the Regional and Global Earth System” (MERGE) of the Swedish VR-funded Lund University Centre for the study of Climate and Carbon Cycle (LUCCI).
REFERENCES
Ahlström, A., P. A. Miller, and B. Smith, 2012: Too early to infer a global NPP decline since 2000. Geophys. Res. Lett., 39, L15403, doi:10.1029/2012GL052336.
Arora, V., 2002: Modeling vegetation as a dynamic component in soil-vegetation-atmosphere transfer schemes and hydrological models. Rev. Geophys.,40, 1006, doi:10.1029/2001RG000103.
Asner, G. P., J. M. Scurlock, and J. A. Hicke, 2003: Global synthesis of leaf area index observations: Implications for ecological and remote sensing studies. Global Ecol. Biogeogr., 12, 191–205, doi:10.1046/j.1466-822X.2003.00026.x.
Balmaseda, M. A., K. Mogensen, and A. T. Weaver, 2012: Evaluation of the ECMWF ocean reanalysis system ORAS4. Quart. J. Roy. Meteo. Soc.,139, 1132–1161, doi:10.1002/qj.2063.
Balsamo, G., A. Beljaars, K. Scipal, P. Viterbo, B. van den Hurk, M. Hirschi, and A. K. Betts, 2009: A revised hydrology for the ECMWF model: Verification from field site to terrestrial water storage and impact in the integrated forecast system. J. Hydrometeor., 10, 623–643, doi:10.1175/2008JHM1068.1.
Branstator, G., and H. Teng, 2010: Two limits of initial-value decadal predictability in a CGCM. J. Climate,23, 6292–6311, doi:10.1175/2010JCLI3678.1.
Branstator, G., H. Teng, G. A. Meehl, M. Kimoto, J. R. Knight, M. Latif, and A. Rosati, 2012: Systematic estimates of initial-value decadal predictability for six AOGCMS. J. Climate, 25, 1827–1846, doi:10.1175/JCLI-D-11-00227.1.
Brodeau, L., B. Barnier, A.-M. Treguier, T. Penduff, and S. Gulev, 2010: An ERA40-based atmospheric forcing for global ocean circulation models. Ocean Modell., 31, 88–104, doi:10.1016/j.ocemod.2009.10.005.
Caldararu, S., P. Palmer, and D. Purves, 2012: Inferring Amazon leaf demography from satellite observations of leaf area index. Biogeosciences, 9, 1389–1404, doi:10.5194/bg-9-1389-2012.
Conil, S., H. Douville, and S. Tyteca, 2007: The relative influence of soil moisture and SST in climate predictability explored within ensembles of AMIP type experiments. Climate Dyn., 28, 125–145, doi:10.1007/s00382-006-0172-2.
Corti, S., A. Weisheimer, T. N. Palmer, F. J. Doblas-Reyes, and L. Magnusson, 2012: Reliability of decadal predictions. Geophys. Res. Lett.,39, L21712, doi:10.1029/2012GL053354.
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.
Delire, C., N. de Noblet-Ducoudré, A. Sima, and I. Gouirand, 2011: Vegetation dynamics enhancing long-term climate variability confirmed by two models. J. Climate, 24, 2238–2257, doi:10.1175/2010JCLI3664.1.
Douville, H., 2010: Relative contribution of soil moisture and snow mass to seasonal climate predictability: A pilot study. Climate Dyn.,34, 797–818, doi:10.1007/s00382-008-0508-1.
Enfield, D. B., A. M. Mestas-Nunez, and P. J. Trimble, 2001: The Atlantic multidecadal oscillation and its relation to rainfall and river flows in the continental US. Geophys. Res. Lett., 28, 2077–2080, doi:10.1029/2000GL012745.
Ferro, C. A. T., 2014: Fair scores for ensemble forecasts. Quart. J. Roy. Meteor. Soc., 140, 1917–1923, doi:10.1002/qj.2270.
Friedlingstein, P., and Coauthors, 2006: Climate–carbon cycle feedback analysis: Results from the C4MIP model intercomparison. J. Climate, 19, 3337–3353, doi:10.1175/JCLI3800.1.
Goddard, L., J. W. Hurrell, B. P. Kirtman, J. Murphy, T. Stockdale, and C. Vera, 2012: Two time scales for the price of one (almost). Bull. Amer. Meteor. Soc., 93, 621–629, doi:10.1175/BAMS-D-11-00220.1.
Goosse, H. and T. Fichefet, 1999: Importance of ice-ocean interactions for the global ocean circulation: A model study. J. Geophys. Res.,104, 23 337–23 355, doi:10.1029/1999JC900215.
Hartmann, H. C., T. C. Pagano, S. Sorooshian, and R. Bales, 2002: Confidence builders: Evaluating seasonal climate forecasts from user perspectives. Bull. Amer. Meteor. Soc., 83, 683–698, doi:10.1175/1520-0477(2002)083<0683:CBESCF>2.3.CO;2.
Hawkins, E., and R. Sutton, 2009: The potential to narrow uncertainty in regional climate predictions. Bull. Amer. Meteor. Soc., 90, 1095–1107, doi:10.1175/2009BAMS2607.1.
Hazeleger, W., and Coauthors, 2010: EC-Earth: A seamless Earth-system prediction approach in action. Bull. Amer. Meteor. Soc., 91, 1357–1363, doi:10.1175/2010BAMS2877.1.
Hazeleger, W., and Coauthors, 2012: EC-Earth v2.2: Description and validation of a new seamless Earth system prediction model. Climate Dyn., 39, 2611–2629, doi:10.1007/s00382-011-1228-5.
Hazeleger, W., V. Guemas, B. Wouters, S. Corti, I. Andreu-Burillo, F. J. Doblas-Reyes, K. Wyser, and M. Caian, 2013a: Multiyear climate predictions using two initialization strategies. Geophys. Res. Lett.,40, 1794–1798, doi:10.1002/grl.50355.
Hazeleger, W., and Coauthors, 2013b: Predicting multiyear North Atlantic Ocean variability. J. Geophys. Res. Oceans, 118, 1087–1098, doi:10.1002/jgrc.20117.
Hickler, T., and Coauthors, 2012: Projecting the future distribution of European potential natural vegetation zones with a generalized, tree species-based dynamic vegetation model. Global Ecol. Biogeogr., 21, 50–63, doi:10.1111/j.1466-8238.2010.00613.x.
Ho, C. K., E. Hawkins, L. Sharey, and F. M. Underwood, 2013: Statistical decadal predictions for sea surface temperatures: A benchmark for dynamical GCM predictions. Climate Dyn. ,41,917–935, doi:10.1007/s00382-012-1531-9.
Hurrell, J., G. A. Meehl, D. Bader, T. L. Delworth, B. Kirtman, and B. Wielicki, 2009: A unified modeling approach to climate system prediction. Bull. Amer. Meteor. Soc.,90, 1819–1832, doi:10.1175/2009bams2752.1.
Keenlyside, N., M. Latif, J. Jungclaus, L. Kornblueh, and E. Roeckner, 2008: Advancing decadal-scale climate prediction in the North Atlantic sector. Nature, 453, 84–88, doi:10.1038/nature06921.
Knight, J. R., C. K. Folland, and A. A. Scaife, 2006: Climate impacts of the Atlantic multidecadal oscillation. Geophys. Res. Lett., 33, L17706, doi:10.1029/2006GL026242.
Köppen, W., 1884: Die warmezonen der erde, nach der dauer der heissen, gemassigten und kalten zeit und nach der wirkung der warme auf die organische welt betrachtet (The thermal zones of the earth according to the duration of hot, moderate and cold periods and to the impact of heat on the organic world). Meteor. Z.,1, 215–226.
Koster, R. D., and Coauthors, 2004: Regions of strong coupling between soil moisture and precipitation. Science, 305, 1138–1140, doi:10.1126/science.1100217.
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.
Levis, S., J. A. Foley, and D. Pollard, 2000: Large-scale vegetation feedbacks on a doubled CO2 climate. J. Climate, 13, 1313–1325, doi:10.1175/1520-0442(2000)013<1313:LSVFOA>2.0.CO;2.
Madec, G., 2008: NEMO ocean engine. Note du Pôle de modélisation de l’Institut Pierre-Simon Laplace 27, 217 pp.
Meehl, G. A., and Coauthors, 2009: Decadal prediction: Can it be skillful? Bull. Amer. Meteor. Soc., 90, 1467–1485, doi:10.1175/2009BAMS2778.1.
Meehl, G. A., A. Hu, and C. Tebaldi, 2010: Decadal prediction in the Pacific region. J. Climate, 23, 2959–2973, doi:10.1175/2010JCLI3296.1.
Miralles, D. G., M. J. van den Berg, A. J. Teuling, and R. A. M. de Jeu, 2012: Soil moisture-temperature coupling: A multiscale observational analysis. Geophys. Res. Lett., 39, L21707, doi:10.1029/2012GL053703.
Mitchell, T. D., and P. D. Jones, 2005: An improved method of constructing a database of monthly climate observations and associated high-resolution grids. Int. J. Climatol., 25, 693–712, doi:10.1002/joc.1181.
Mochizuki, T., and Coauthors, 2010: Pacific decadal oscillation hindcasts relevant to near-term climate prediction. Proc. Natl. Acad. Sci. USA, 107, 1833–1837, doi:10.1073/pnas.0906531107.
Moss, R. H., and Coauthors, 2010: The next generation of scenarios for climate change research and assessment. Nature, 463, 747–756, doi:10.1038/nature08823.
Müller, W., C. Appenzeller, F. Doblas-Reyes, and M. Liniger, 2005: A debiased ranked probability skill score to evaluate probabilistic ensemble forecasts with small ensemble sizes. J. Climate, 18, 1513–1523, doi:10.1175/JCLI3361.1.
Piao, S. L., and Coauthors, 2013: Evaluation of terrestrial carbon cycle models for their response to climate variability and to CO2 trends. Global Change Biol., 19, 2117–2132, doi:10.1111/gcb.12187.
Pielke, R. A., Sr., and Coauthors, 2011: Land use/land cover changes and climate: Modeling analysis and observational evidence. Wiley Interdiscip. Rev.: Climate Change,2, 828–850, doi:10.1002/wcc.144.
Pohlmann, H., J. H. Jungclaus, A. Kohl, D. Stammer, and J. Marotzke, 2009: Initializing decadal climate predictions with the GECCO oceanic synthesis: Effects on the North Atlantic. J. Climate, 22, 3926–3938, doi:10.1175/2009JCLI2535.1.
Samuelsson, P., S. Gollvik, and A. Ullerstig, 2006: The land-surface scheme of the Rossby Centre regional atmospheric climate model (RCA3). Swedish Meteorological and Hydrological Institute 122, 38 pp.
Seneviratne, S. I., T. Corti, E. L. Davin, M. Hirschi, E. B. Jaeger, I. Lehner, B. Orlowsky, and A. J. Teuling, 2010: Investigating soil moisture–climate interactions in a changing climate: A review. Earth Sci. Rev., 99, 125–161, doi:10.1016/j.earscirev.2010.02.004.
Sitch, S., and Coauthors, 2003: Evaluation of ecosystem dynamics, plant geography and terrestrial carbon cycling in the LPJ dynamic global vegetation model. Global Change Biol., 9, 161–185, doi:10.1046/j.1365-2486.2003.00569.x.
Smith, B., I. C. Prentice, and M. T. Sykes, 2001: Representation of vegetation dynamics in the modelling of terrestrial ecosystems: Comparing two contrasting approaches within European climate space. Global Ecol. Biogeogr., 10, 621–637, doi:10.1046/j.1466-822X.2001.00256.x.
Smith, B., P. Samuelsson, A. Wramneby, and M. Rummukainen, 2011: A model of the coupled dynamics of climate, vegetation and terrestrial ecosystem biogeochemistry for regional applications. Tellus, 63A, 87–106, doi:10.1111/j.1600-0870.2010.00477.x.
Smith, D. M., S. Cusack, A. W. Colman, C. K. Folland, G. R. Harris, and J. M. Murphy, 2007: Improved surface temperature prediction for the coming decade from a global climate model. Science, 317,796–799, doi:10.1126/science.1139540.
Smith, D. M., R. Eade, N. J. Dunstone, D. Fereday, J. M. Murphy, H. Pohlmann, and A. A. Scaife, 2010: Skilful multi-year predictions of Atlantic hurricane frequency. Nat. Geosci., 3, 846–849, doi:10.1038/ngeo1004.
Sutton, R. T., and D. L. Hodson, 2005: Atlantic Ocean forcing of North American and European summer climate. Science, 309, 115–118, doi:10.1126/science.1109496.
Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485–498, doi:10.1175/BAMS-D-11-00094.1.
Teuling, A. J., and S. I. Seneviratne, 2008: Contrasting spectral changes limit albedo impact on land-atmosphere coupling during the 2003 European heat wave. Geophys. Res. Lett., 35, L03401, doi:10.1029/2007GL032778.
Uppala, S. M., and Coauthors, 2005: The ERA-40 Re-Analysis. Quart. J. Roy. Meteor. Soc.,131, 2961–3012, doi:10.1256/qj.04.176.
Van den Hurk, B. J. J. M., P. Viterbo, and S. Los, 2003: Impact of leaf area index seasonality on annual land surface evaporation in a global circulation model. J. Geophys. Res., 108, 4191–4199, doi:10.1029/2002JD002846.
Van der Molen, M. K., B. J. J. M. van den Hurk, and W. Hazeleger, 2011: A dampened land use change climate response towards the tropics. Climate Dyn.,37, 2035–2043, doi:10.1007/s00382-011-1018-0.
van Oldenborgh, G. J., F. J. Doblas-Reyes, B. Wouters, and W. Hazeleger, 2012: Decadal prediction skill in a multi-model ensemble. Climate Dyn., 38, 1263–1280, doi:10.1007/s00382-012-1313-4.
Weaver, A., C. Deltel, E. Machu, S. Ricci, and N. Daget, 2005: A multivariate balance operator for variational ocean data assimilation. Quart. J. Roy. Meteor. Soc.,131, 3605–3625, doi:10.1256/qj.05.119.
Weigel, A. P., M. A. Liniger, and C. Appenzeller, 2007: The discrete Brier and ranked probability skill scores. Mon. Wea. Rev., 135, 118–124, doi:10.1175/MWR3280.1.
Weiss, M., B. van den Hurk, R. Haarsma, and W. Hazeleger, 2012: Impact of vegetation variability on potential predictability and skill of EC-Earth simulations. Climate Dyn., 39, 2733–2746, doi:10.1007/s00382-012-1572-0.
Wouters, B., W. Hazeleger, S. Drijfhout, G.-J. van Oldenborgh, and V. Guemas, 2013: Decadal predictability of the North Atlantic subpolar gyre. Geophys. Res. Lett., 40, 3080–3084, doi:10.1002/grl.50585.
Wramneby, A., B. Smith, and P. Samuelsson, 2010: Hotspots of vegetation-climate feedbacks under future greenhouse forcing in Europe. J. Geophys. Res., 115, D21119, doi:10.1029/2010JD014307.
Zhang, R., and T. L. Delworth, 2006: Impact of Atlantic multidecadal oscillations on India/Sahel rainfall and Atlantic hurricanes. Geophys. Res. Lett., 33, L17712, doi:10.1029/2006GL026267.
