Abstract

The fact that aerosols are important players in Earth’s radiation balance is well accepted by the scientific community. Several studies have shown the importance of characterizing aerosols in order to constrain surface radiative fluxes and temperature in climate runs. In numerical weather prediction, however, there has not been definite proof that interactive aerosol schemes are needed to improve the forecast. Climatologies are instead used that allow for computational efficiency and reasonable accuracy. At the monthly to subseasonal range, it is still worth investigating whether aerosol variability could afford some predictability, considering that it is likely that persisting aerosol biases might manifest themselves more over time scales of weeks to months and create a nonnegligible forcing. This paper explores this hypothesis using the ECMWF’s Ensemble Prediction System for subseasonal prediction with interactive prognostic aerosols. Four experiments are conducted with the aim of comparing the monthly prediction by the default system, which uses aerosol climatologies, with the prediction using radiatively interactive aerosols. Only the direct aerosol effect is considered. Twelve years of reforecasts with 50 ensemble members are analyzed on the monthly scale. Results indicate that the interactive aerosols have the capability of improving the subseasonal prediction at the monthly scales for the spring/summer season. It is hypothesized that this is due to the aerosol variability connected to the different phases of the Madden–Julian oscillation, particularly that of dust and carbonaceous aerosols. The degree of improvement depends crucially on the aerosol initialization. More work is required to fully assess the potential of interactive aerosols to increase predictability at the subseasonal scales.

1. Introduction

The impact of aerosol particles is widely recognized as an important factor for accurate climate and weather predictions. The role of aerosols in Earth’s radiation balance has been addressed by the climate modeling community since the early 90s (Crutzen and Andreae 1990; Charlson et al. 1992; Hansen et al. 1992, to mention a few). Both natural and anthropogenic aerosols are deemed crucial for a correct representation of present-day and future climate/weather scenarios (Boucher et al. 2013).

The direct effect in cloud-free regions can be substantial and affect variables such as surface temperature, precipitation, and winds. The aerosol direct effect consists of the sum of two phenomena: scattering/absorption of incoming solar radiation and absorption/emission of longwave radiation. The radiative impact of aerosols is very dependent on their vertical distribution, chemical composition, and surrounding environment. For example, the impact of aerosols over high-albedo surfaces at high latitudes could be a warming net effect, whereas the same aerosols over a dark surface might have a cooling impact. This could be particularly true for certain types of aerosols that are absorbing, such as black carbon. In the presence of high humidity, aerosol species such as sulfates have much larger optical depth due to hygroscopic growth. The impact of large aerosol loads generated during extreme events, such as desert dust storms (Rémy et al. 2015) or extensive episodes of biomass burning (Zhang et al. 2016), could also be substantial. Aerosols are also important because they serve as cloud condensation nuclei and affect cloud properties such as the cloud life cycle, the optical properties, and the precipitation activity of clouds. This type of impact on the radiative balance is often referred to as indirect, as it manifests itself through changes in the cloud properties (i.e., Koch and Del Genio 2010; Painemal and Zuidema 2013). A cloud that has been exposed to a large number of aerosol particles is generally made up of smaller particles, given the same liquid water path, and hence is more reflective than one that has formed in pristine conditions. Many uncertainties, however, still exist in terms of the magnitude and sign of these impacts. Aerosol particles are also important due to their impact on human health. It has been acknowledged that they represent a serious public health issue, as shown by recent particulate matter (PM) pollution events in western Europe and China (Wang et al. 2014; Sun et al. 2014).

While it is recognized that incorporating aerosol impacts is essential for accurate numerical weather prediction (NWP), the cost of running a full integrated aerosol system at the high resolution of current NWP systems is considered, at the moment, too high with respect to the perceived benefits. Priority is hence given to defining robust aerosol climatologies that can describe the average effect of aerosols on the radiative balance but that are not computationally expensive. At the European Centre for Medium-Range Weather Forecasts (ECMWF), the climatology described by Tegen et al. (1997; hereafter T97) has been used for several years with satisfying results (Tompkins et al. 2005; Rodwell and Jung 2008).

Recently, updated aerosol climatologies based on the Copernicus Atmosphere Monitoring Service (CAMS) interim reanalysis (hereafter CAMSira; Flemming et al. 2017) have been tested to replace the T97 climatology. A first investigation of the use of the CAMS-based climatology showed it had positive impact on NWP scores in summer in the Northern Hemisphere, particularly in areas with large aerosol loadings such as the Indian Ocean. In the boreal winter, however, the impact was not as positive due to an underestimation of absorbing aerosols. CAMSira, which ran from 2003 to 2015, is known to have problems with aerosol speciation, showing an excess of sulfate aerosols over the ocean and low values of carbonaceous aerosols at the midlatitudes. A full evaluation of CAMSira is given by Flemming et al. (2017). In the aerosol assimilation system, observations of total aerosol optical depth (AOD) from the Moderate Resolution Imaging Spectroradiometer (MODIS), flown on two NASA spacecraft (Terra and Aqua), constrain the total optical mass of the aerosol species in the model, but the redistribution of the analysis increments into the single species is largely model driven. Much effort has gone into representing correctly the aerosol speciation in the most recent version of the CAMSira-based climatology through postprocessing, with the aim of achieving neutral to positive impact in the medium-range forecast scores. The current operational model at ECMWF uses this climatology, which is documented by Bozzo et al. (2017).

In parallel, efforts have been made to study the impact of the fully prognostic and interactive aerosols on relevant NWP variables. One coordinated effort was that of the Working Group on Numerical Experimentation (WGNE), which included several weather forecast centers. The participating centers with aerosol capabilities coupled with the meteorological model were tasked to run selected test cases for which the climatology was not representative of the actual aerosol conditions. One of the cases was a large desert dust storm that swept over most of the Middle East on 18 April 2011. Results from the ECMWF integrations are reported by Rémy et al. (2015). Conclusions from that study showed that the impact of the dust aerosols on surface temperature and winds was noticeable, even though the synoptic situation was not much affected by the radiative forcing of the prognostic aerosols. The impact of wind changes on increasing dust emissions was also highlighted. Similar conclusions were also reached by Colarco et al. (2014), whereas in contrast, Pérez et al. (2006) found that emissions were lowered for a similar intense dust event. The impact of interactive aerosols, particularly dust, on tropical cyclogenesis was investigated by Reale et al. (2011) using the GEOS-5 global forecasting system with interactive aerosols. They found that the effects of dust on static stability increased with time, becoming highly significant after day 5 and producing an environment less favorable to cyclogenesis. Similarly, Reale et al. (2014) investigated the impact of dust on the African easterly jet (AEJ) and found that forecasts in which interactive aerosols were included presented an AEJ at higher altitude and slightly displaced to the north, in agreement with observations and previous results.

Previous studies had also shown an impact of aerosols’ direct and indirect effects on the medium-range NWP. In particular, Mulcahy et al. (2014) showed that accounting for aerosol effects, including the indirect one, improved the forecast of surface radiation and temperature. However, having a full prognostic aerosol scheme was deemed too costly, and climatologies were shown to also be a powerful yet computationally efficient way to include aerosol effects. More recently, Zhang et al. (2016) compared several NWP models for a large biomass burning event that occurred in the United States and found that all models using aerosol climatologies severely overestimated surface temperature, whereas models with prognostic aerosols showed the correct surface cooling. The authors estimated a cooling at the surface of approximately 0.25° to 1.0°C per unit AOD at 550 nm. They also pointed out that the frequency of events with AOD larger than 1 is very low. However, even moderate AODs (below 0.5) could bring a sizable temperature impact in certain regions.

Starting from the idea that aerosol impacts would manifest themselves more at the longer ranges than at the short/medium ranges (see, e.g., Bellucci et al. 2015), an effort has been made to activate the aerosol modules in the coupled ECMWF’s Ensemble Prediction System (EPS; Vitart 2014). The aerosol model is part of the Integrated Forecast System (IFS), which is the model developed at ECMWF for seamless numerical weather prediction from a few days up to seasons. The aerosol component has been developed in the context of precursors’ projects, which were funded by the European Commission [e.g., the Monitoring Atmospheric Composition and Climate (MACC) project]. To the best of our knowledge, this is the first documented attempt at using an interactive aerosol model in an ocean–atmosphere coupled system to assess aerosol impacts on meteorology at the subseasonal-to-seasonal (S2S) scales. The inclusion of interactive aerosols in the EPS opens the doors to new applications in S2S prediction, including an improvement to the meteorological fields at the extended range as well as a novel application, the prediction of aerosols fields at those time scales, which again has not been previously documented.

The paper describes the steps taken toward this implementation and the results obtained from subseasonal reforecasts for the period May–June. The focus is on understanding what type of impact aerosols may have on the subseasonal prediction and what the physics behind this impact may be. Two control runs are obtained by running the EPS in the standard configuration: one uses the T97 climatology to prescribe the aerosol components within the radiation scheme, while the second control run uses the CAMS climatology of Bozzo et al. (2017). In the experiments with the interactive aerosols, not only are the aerosols prognostic variables in the model, but they are also allowed to interact with the radiation via the direct effect. The indirect effect is not considered here. An attempt to include the aerosol indirect effect via the autoconversion scheme was presented by Morcrette et al. (2011), but it has not been possible to duplicate those results in more recent model versions. This should be subject to further research.

The paper is divided into the following sections. Section 2 outlines the experiment setup and includes a brief description of the prognostic aerosol model that is included in the ECMWF’s IFS. Section 3 presents the results of the experiments, while in section 4, possible mechanisms to explain those results are discussed. Finally, section 5 summarizes the main findings and provides an outlook on further developments in this line of research.

2. Technical aspects

a. The operational coupled Ensemble Prediction System

Subseasonal forecasts out to 46 days have been produced routinely at ECMWF since March 2002 and operationally since October 2004 (Vitart 2014). In the current configuration, the monthly forecasts are generated by extending the 15-day ensemble integrations to 46 days twice a week (at 0000 UTC on Mondays and Thursdays). Forecasts are based on the medium range/monthly ensemble forecast (ENS), which is part of the ECMWF’s Integrated Forecast System. ENS includes 51 members run with a horizontal resolution of TCo639 (about 16 km) up to forecast day 15 and TCo319 (about 32 km) thereafter. The atmospheric model is coupled to an ocean model (NEMO) with a 1/4° horizontal resolution.

After a few days of model integrations, the model mean climate begins to differ from the initial conditions. No bias correction is applied to remove or reduce the drift in the model, and no steps are taken to remove or reduce any imbalances in the coupled model initial state. The effect of the drift on the model calculations is estimated a posteriori from integrations of the model in previous years (reforecasts) and removed (calibration). The climatology that is provided by the reforecasts is computed using a suite that includes only 11 members of 46-day integrations with the same configuration as the real-time forecasts, starting on the same day and month as the real-time forecast over the past 20 years. For the model integrations presented in this study, the reforecasts are run with 10 ensemble members over a shorter period (2003–15) due to the limited availability of the aerosol emissions.

Initial perturbations are generated using a combination of singular vectors and perturbations generated using the ECMWF ensemble of data assimilations, and model uncertainties are simulated using two stochastic schemes (Leutbecher et al. 2017). The aerosol fields are not perturbed in the different ensemble members. However, for natural aerosols such as desert dust and sea salt, whose emissions are parameterized based on meteorological variables—most prominently, winds—any perturbations on those will also reflect on perturbations on the aerosol emissions themselves.

b. The aerosol model

In the context of the Copernicus Atmosphere Monitoring Service and precursors’ projects, Global Earth-system Monitoring using Satellite and in-situ data (GEMS) and Monitoring Atmospheric Composition and Climate (MACC), ECMWF has developed a capability to monitor and forecast atmospheric composition, including aerosols, greenhouse gases, and reactive gases, using satellite observations and a combination of global and regional models. The atmospheric composition prediction system is based on the IFS meteorological model, maintained and developed by ECMWF. The version used in this work corresponds to cycle 43R1 of the IFS, for which a detailed description can be found on the ECMWF’s web page.1 Generally, IFS is not run with the full coupled chemistry due to its computational cost. Currently, the operational resolution of IFS with full chemistry is 40 km with 60 vertical levels up to 0.1 hPa, as opposed to the operational NWP without full chemistry, which has a resolution of 9 km and 137 vertical levels up to 0.01 hPa. Aerosols are forecast within the global system by a bulk–bin scheme (Morcrette et al. 2009), based on earlier work by Reddy et al. (2005) and Boucher et al. (2002), that includes five species: dust, sea salt, black carbon, organic carbon, and sulfates. Dust aerosols are represented by three prognostic variables that correspond to three size bins, with bin limits of 0.03, 0.55, 0.9, and 20 μm in radius. Sea salt aerosols are also represented by three size bins with limits of 0.03, 0.5, 5, and 20 μm in radius. Emissions of natural aerosols such as desert dust and sea salt are parameterized based on model variables, with surface winds being the main driver. For all other tropospheric aerosols (carbonaceous aerosols and sulfates), emission sources are defined according to established inventories (Lamarque et al. 2010). Biomass burning emissions contributing to black carbon and organic matter loads are prescribed from the Global Fire Assimilation System (GFAS; Kaiser et al. 2012). Removal processes include sedimentation of all particles, wet and dry deposition, and in-cloud and below-cloud scavenging. For organic matter and black carbon, both the hydrophobic and hydrophilic components are considered. A very simplified representation of the sulfur cycle is also included with only two variables, sulfur dioxide (SO2) and sulfate (SO4); the latter is in the particulate phase. Overall, a total of 12 additional prognostic variables for the mass mixing ratio of the different components (bins or types) of the various aerosols are used in this configuration. Several revisions of the dust emission schemes have been undertaken, as well as developments to include more aerosol species, such as nitrates and secondary organic aerosols (S. Rémy 2016, personal communication).

In the version of the global IFS used in this study, which was operational until July 2017, the direct radiative effect of aerosols is taken into account using the aerosol monthly climatology of T97. A second control experiment was performed using the new CAMS climatology by Bozzo et al. (2017), which became operational after July 2017. In the experimental version of the system, however, the aerosol optical depth, which is then used to calculate the radiative impacts, can be computed directly from the mass mixing ratios of the prognostic aerosols provided by the aerosol module. We make use of this capability to set up experiments with the coupled Ensemble Prediction System, as described in the next section, to investigate the importance of the direct radiative impact of the prognostic aerosols relative to control runs that use the T97 and the CAMS/Bozzo climatologies. We also investigate the sensitivity to initial conditions for the aerosols by choosing different datasets to initialize the simulations.

c. Experiment setup

Four experiments were run to assess the aerosol impacts: one control integration with the T97 climatology in which all settings are similar to the operational setup, but run at lower horizontal resolution (T255 corresponding to 80 km; hereafter CONTROL1); a second control run with the CAMS/Bozzo climatology at the same reduced resolution (CONTROL2); an interactive prognostic aerosol run in which the prognostic aerosols are initialized using the time-varying CAMSira (PROG1); and a second interactive aerosol run in which the prognostic aerosols are initialized using a fixed climatology based on a CAMS experiment without any data assimilation (PROG2; J. Flemming 2016, personal communication). The different choice of initialization allows us to understand the sensitivity of the interactive aerosol runs to the initial conditions. Neither the T97 nor the CAMS/Bozzo climatology is used in the interactive aerosol runs in which aerosols are instead allowed to interact with the radiation. Only the direct aerosol effect is taken in consideration, while indirect aerosol effects on clouds are not modeled.

All simulations were conducted with 91 vertical levels. Prescribed emissions for the anthropogenic species over the years of interest (2003–15) were used. Updating the emissions over the course of the reforecasts is clearly essential, particularly for biomass burning emissions; these cannot be accounted for with persistence over the course of several weeks, as they have a natural life cycle of a few days. It is possible to take into account these emissions using climatologies. However, in this work, we have used emissions estimated from the fire radiative power (FRP) provided by the MODIS instruments on board the Aqua and Terra satellites and processed according to Kaiser et al. (2012) to obtain emission coefficients for biomass burning aerosols. As far as biomass burning emissions are concerned, these simulations represent a “best case scenario” because emissions are based on actual observations of MODIS FRP. It would be possible to prescribe climatologies for the biomass burning to see specifically the impact of the way the biomass burning source is prescribed. Ultimately, and ideally, if one had a prognostic model for biomass burning emissions related to weather parameters, the full impact of prescribing those important emissions versus modeling them could be assessed. This is still subject to much investigation, and it is not covered here.

For computational cost, the size of the ensemble was limited to 10 members plus one unperturbed forecast. To ensure more robust statistics, five initialization dates were considered for each year for the summer integrations (20 and 25 April and 1, 5, and 10 May). The total number of ensemble members was hence increased to 55. All runs were set up to be 6 months long, although only statistics for weeks 1–4 are presented here. The seasonal aspects will be investigated in a companion paper. The meteorological variables were initialized using ERA-Interim (Dee et al. 2011), whereas the initial conditions for soil variables were taken from experiments run in house (G. Balsamo 2016, personal communication).

3. Results

Several diagnostics are applied to assess the relative skill of the experiments. These diagnostics have been developed over a number of years and are now used routinely at ECMWF to assess improvements in the EPS due to changes in various elements (e.g., physical parameterizations or changes in the ocean model, as well as increases in vertical and horizontal resolution).

Bias plots for several meteorological parameters averaged for the weekly period starting at days 26–32 (week 4) are shown in Figs. 13. For a start date of 1 May, this corresponds to the end of May/beginning of June. The bias is estimated by computing the difference between the model weekly climatology as a function of lead time and the weekly mean climatology from ERA-Interim computed over the same years (2003–15). Biases are only shown with respect to CONTROL1.

Fig. 1.

Bias in temperature (K) at 850 hPa for (top) CONTROL1, (middle) PROG1, and (bottom) PROG2 at week 4 (end of May/beginning of June).

Fig. 1.

Bias in temperature (K) at 850 hPa for (top) CONTROL1, (middle) PROG1, and (bottom) PROG2 at week 4 (end of May/beginning of June).

Fig. 2.

As in Fig. 1, but for meridional wind (m s−1) at 850 hPa.

Fig. 2.

As in Fig. 1, but for meridional wind (m s−1) at 850 hPa.

Fig. 3.

As in Fig. 1, but for total precipitation (mm day−1).

Fig. 3.

As in Fig. 1, but for total precipitation (mm day−1).

Aerosols will directly affect lower-troposphere temperature in cloud-free regions due to the radiative cooling in the shortwave. Changes in the diabatic heating profile can also occur as a result of differential warming/cooling induced by absorbing aerosols. Winds are also indirectly affected by aerosols because of the changes in the amplitude and distribution of the diabatic heating (Ramanathan et al. 2005; Lau and Kim 2010).

The bias in temperature at 850 hPa for CONTROL1 is shown in Fig. 1 for 4 weeks into the simulation. The bias for experiments PROG1 and PROG2 with respect to CONTROL1 is shown in the same figure. Blue shades indicate a negative bias, whereas red shades indicate positive bias. Reduction of biases up to 1°–2° is visible in both PROG1 and PROG2, particularly over the Asian dust belt in the northern Pacific Ocean and, to a lesser extent, the North Atlantic dust belt. The bias in the Mediterranean basin also appears reduced. The Arctic also experiences a large reduction in bias; however, few grid points are actually included in that region. Generally, the Northern Hemisphere appears to be cooler in the PROG1 and PROG2 experiments than in the CONTROL1 experiments, with the exception of Russia.

Figure 2 shows reduced biases in meridional wind at 850 hPa, particularly over the Arabian Sea for both the PROG1 and PROG2 experiments with respect to CONTROL1. In this area, meridional winds are usually overestimated by the ECMWF’s model as a result of an overly strong Walker circulation. Wind biases in East Asia are reduced by approximately 1–2 m s−1. Likewise, we observe a small reduction in bias in the Southern Ocean. Precipitation biases are also reduced over several tropical regions, including the Tibetan Plateau and the North Pacific, as shown in Fig. 3. Particularly interesting is the bias reduction in East Asia, which amounts to 0.5–1 mm day−1.

Scorecards for the PROG1 and PROG2 experiments as compared to the CONTROL1 and CONTROL2 runs are shown in Figs. 4 and 5. Scorecards show the difference in ranked probability skill scores (RPSS; see, e.g., Murphy and Winkler 1984) between two experiments for 20 different parameters’ (upper-air and surface fields) weekly means (week 1: days 5–11, week 2: days 12–18, week 3: days 19–25, and week 4: days 26–32) over the northern extratropics (north of 30°N) and the tropics (30°N–30°S band). Yellow and red colors (blue and cyan) indicate that the experiment being scored has lower (higher) RPSS than the control experiment: the higher the RPSS, the more skillful the experiment. A statistical test has been applied to the differences of RPSS scores. It is based on a 10 000 resampling bootstrap procedure. Dark blue and dark red dots indicate that the difference of RPSS is statistically significant within the 5% level of confidence.

Fig. 4.

Scorecards for experiments (left) PROG1 and (right) PROG2, compared to CONTROL1. The abbreviation w1 corresponds to week 1 (days 5–11), w2 corresponds to week 2 (days 12–18), w3 corresponds to week 3 (days 19–25), and w4 corresponds to week 4 (days 26–32) from the start date (1 May). The following variables are verified: total precipitation (tp), 2-m temperature (t2m), surface temperature (stemp), sea surface temperature (sst), mean sea level pressure (mslp), temperature at 50 hPa (t50), horizontal wind at 50 hPa (u50), meridional wind at 50 hPa (v50), streamfunction at 200 hPa (sf200), velocity potential at 200 hPa (vp200), temperature at 200 hPa (t200), horizontal wind at 200 hPa (u200), meridional wind at 200 hPa (v200), geopotential at 500 hPa (z500), temperature at 500 hPa (t500), horizontal wind at 500 hPa (u500), meridional wind at 500 hPa (v500), temperature at 850 hPa (t850), horizontal wind at 850 hPa (u850), and meridional wind at 850 hPa (v850). See text for an explanation of the verification metrics.

Fig. 4.

Scorecards for experiments (left) PROG1 and (right) PROG2, compared to CONTROL1. The abbreviation w1 corresponds to week 1 (days 5–11), w2 corresponds to week 2 (days 12–18), w3 corresponds to week 3 (days 19–25), and w4 corresponds to week 4 (days 26–32) from the start date (1 May). The following variables are verified: total precipitation (tp), 2-m temperature (t2m), surface temperature (stemp), sea surface temperature (sst), mean sea level pressure (mslp), temperature at 50 hPa (t50), horizontal wind at 50 hPa (u50), meridional wind at 50 hPa (v50), streamfunction at 200 hPa (sf200), velocity potential at 200 hPa (vp200), temperature at 200 hPa (t200), horizontal wind at 200 hPa (u200), meridional wind at 200 hPa (v200), geopotential at 500 hPa (z500), temperature at 500 hPa (t500), horizontal wind at 500 hPa (u500), meridional wind at 500 hPa (v500), temperature at 850 hPa (t850), horizontal wind at 850 hPa (u850), and meridional wind at 850 hPa (v850). See text for an explanation of the verification metrics.

Fig. 5.

As in Fig. 4, but for experiments (left) PROG1 and (right) PROG2, compared to CONTROL2.

Fig. 5.

As in Fig. 4, but for experiments (left) PROG1 and (right) PROG2, compared to CONTROL2.

Both the PROG1 and PROG2 experiments have a good performance with respect to the CONTROL1 experiment in terms of a number of variables, particularly in the Northern Hemisphere. Of particular relevance are the significant positive changes in PROG2 for the Northern Hemisphere in meridional and zonal winds at 200 hPa, as well as temperature at the same pressure level. Positive, significant changes are also observed in PROG2 in meridional wind at 500 hPa, temperature at 850 hPa, and surface temperature. Positive, nonsignificant changes occur in most variables at all ranges. The tropics show less of a response to the prognostic aerosols. It is interesting to observe that most of the significant impact occurs at the extended range (week 4), which confirms the initial working hypothesis of a cumulative effect of the aerosol forcing over an extended time period.

With respect to CONTROL2, PROG1 and PROG2 perform in a similar way with overall positive impact even though the number of statistically significant improvements are fewer (see Fig. 5). The performance of PROG2 remains more positive than that of PROG1. In general, the experiments with the radiatively interactive aerosols outperform the control experiments with the climatology.

Differences between the two experiments with prognostic aerosols (PROG1 and PROG2) can be ascribed to the different initializations. PROG1 was initialized with CAMSira, whereas PROG2 was initialized with a climatology derived from a free-running model run with a recent version of the CAMS system. The climatology from the free-running model has more similar aerosol distribution to that of the interactive aerosol runs, given that both are free running and have no observational constraint. On the contrary, the aerosol distribution in the interim reanalysis run is quite different, particularly for the dust aerosols, due to the use of a previous model version and the assimilation of aerosol optical depth from satellite observations. When using the CAMSira data for initialization, the EPS with interactive aerosols receives an initial “shock” being pushed away from its natural state. After a few days, the impact of the initialization is felt less, and both runs relax to a similar aerosol state. This is the possible cause of the better scores in the PROG2 run with respect to PROG1.

Figure 6 shows the impact of using the CAMS/Bozzo climatology with respect to the T97 climatology. Overall, the impact is neutral to mildly positive, indicating that the improvements shown in PROG1 and PROG2 are associated with an improved description of the aerosol variability, rather than the improved average state that the CAMS/Bozzo climatology arguably provides with respect to T97 climatology [see Bozzo et al. (2017) for a full comparison of the two climatologies].

Fig. 6.

As in Fig. 4, but for experiment CONTROL2, compared to CONTROL1.

Fig. 6.

As in Fig. 4, but for experiment CONTROL2, compared to CONTROL1.

To put into perspective the improvement that the interactive prognostic aerosols bring in terms of scores, Fig. 7 shows the scorecard calculated in the same way as for the aerosol experiments here reported, for the current operational version of the ECMWF’s IFS (cycle 43R3) as compared to the previous version (cycle 43R1). In the new model version, many changes were introduced, including the use of a new, more efficient radiation scheme with reduced noise and more accurate longwave radiation transfer calculation; the inclusion of the CAMS/Bozzo climatology; and an adjustment to the convection scheme to increase supercooled liquid water at colder temperatures. Overall, the new model cycle performs better than the previous on many metrics, but the scores for the monthly system are largely neutral.

Fig. 7.

As in Fig. 4, but for an experiment with ECMWF’s model cycle 43R3 against a control experiment with cycle 43R1.

Fig. 7.

As in Fig. 4, but for an experiment with ECMWF’s model cycle 43R3 against a control experiment with cycle 43R1.

4. Aerosols and the Madden–Julian oscillation

Having established that the positive changes in scores produced by the interactive aerosols are significant and of similar magnitude or larger than those caused by other model changes, it is important to address what the physical mechanisms behind this impact might be.

It is accepted that the aerosol climatology provides a good average state but that at the longer scales (e.g., weeks 3 and 4), the aerosol variability may become more important. Of course, this does not preclude the fact that a more accurate climatology may outperform a less accurate one by better representing the average aerosol state. In fact, referring back to Fig. 6, the CONTROL2 run (CAMS/Bozzo climatology) has overall more positive scores with respect to CONTROL1 (T97 climatology). However, regardless of the relative merit of a given aerosol climatology, any run with a static climatology does not allow for a modulation of the aerosol fields, particularly those affected by winds such as desert dust, which could be induced by changes in the circulation happening at the scale of weeks.

Thinking along these lines, one of the most important mechanisms of tropical variability on time scales exceeding 1 week but less than a season is the Madden–Julian oscillation (MJO; Madden and Julian 1971). The MJO has a significant impact on the Indian monsoon (Murakami 1976; Yasunari 1979), which happens in an area of large aerosol loadings, as extensively discussed by Ramanathan et al. (2001) and Collins et al. (2002), to mention a few.

Therefore, it follows naturally to consider the impact that the MJO has on the aerosol fields, which in turn could be responsible for an increase in predictability at the monthly scale. The fact that the MJO plays an important role in aerosol variability was first pointed out by Tian et al. (2008) and further discussed by Tian et al. (2011) and Guo et al. (2013). The authors analyzed several years of AOD data from the MODIS retrievals and concluded that some of the mode of variability in AOD was indeed correlated to the time scales of the MJO. The MJO-related intraseasonal variance accounts for about 25% of the total aerosol optical thickness variance over the tropical Atlantic (Tian et al. 2011), primarily through its influence on the Atlantic low-level zonal winds.

Figure 8 shows the average distribution of aerosols in terms of optical depth from the PROG1 experiment over the period 2003–15 averaged over the first month of integration (May). The top panel shows the total aerosol optical depth, the middle panel shows the optical depth due to dust aerosols, and the bottom panel shows the optical depth due to carbonaceous aerosol. Here, we do not use the term “biomass burning aerosols” because those are a subgroup of the carbonaceous aerosols whose sources also include anthropogenic emissions from industrial or other activities. Of course, in areas where wildfires are predominant, the contribution to the carbonaceous aerosols comes prevalently from biomass burning. The dust distribution appears quite realistic, with maximum loadings over the Sahara and Asian deserts (Gobi and Taklamakan). The same can be said for the patterns of the carbonaceous aerosols where the contributions of the biomass burning areas of central Africa and, to a smaller extent, South America, as well as the eastern Asian pollution, can be clearly seen.

Fig. 8.

Average (a) total, (b) dust, and (c) carbonaceous AOD distribution over the years 2002–15 from the PROG1 experiment for the first month of integration. Units are multiplied by 100.

Fig. 8.

Average (a) total, (b) dust, and (c) carbonaceous AOD distribution over the years 2002–15 from the PROG1 experiment for the first month of integration. Units are multiplied by 100.

To assess the impact of the MJO on aerosols in the model simulations as well as in CAMSira, composites of dust and carbonaceous optical thickness anomalies, relative to the model climatology, have been produced when the active phase of the MJO is located over the Indian Ocean [phases 2 and 3; see Wheeler and Hendon (2004) for the definition of MJO phase], Maritime Continent (phases 4 and 5), western North Pacific (phases 6 and 7), and Western Hemisphere (phases 8 and 1). The left panels of Fig. 9 show the anomaly of the dust field from PROG1 induced by the MJO with respect to the climatological distribution. For comparison, the right panels show the patterns of modulation obtained from CAMSira for the same MJO phases. The similarity of patterns and the fact that opposite phases of the MJO (e.g., phases 2 and 3 or phases 6 and 7) have opposite impacts on the aerosol variability suggest that the MJO modulation is a robust signal, visible both in CAMSira and in the PROG1 forecast, although the amplitude of the impact is smaller in the models’ simulations than in the reanalysis. It is important to note that CAMSira is constrained by observations of aerosol optical depth from MODIS; hence, the variability can be considered equivalent to that found by Tian et al. (2008) using MODIS observations. Figure 10 illustrates the same concept but for carbonaceous aerosols (organic matter and black carbon).

Fig. 9.

Anomaly of dust AOD in the different phases of the MJO for the (left) PROG1 experiment and (right) CAMS interim reanalysis: (a),(b) phases 2 and 3; (c),(d) phases 4 and 5; (e),(f) phases 6 and 7; and (g),(h) phases 8 and 1.

Fig. 9.

Anomaly of dust AOD in the different phases of the MJO for the (left) PROG1 experiment and (right) CAMS interim reanalysis: (a),(b) phases 2 and 3; (c),(d) phases 4 and 5; (e),(f) phases 6 and 7; and (g),(h) phases 8 and 1.

Fig. 10.

As in Fig. 9, but for carbonaceous aerosols.

Fig. 10.

As in Fig. 9, but for carbonaceous aerosols.

There are multiple mechanisms through which the MJO modulation can happen. Precipitation plays a large role in aerosol removal, while winds are responsible for natural emissions and transport. Focusing on dust, which is a natural aerosol whose emissions are driven primarily by surface wind speed, one could imagine that a tight correlation should exist between wind and dust anomalies in the various phases of the MJO. Figure 11 shows the surface wind speed anomalies and the corresponding dust optical depth anomalies in the various phases of the MJO over northern Africa. Points indicate statistical significance of the anomalies at the 90% level as computed from the 11-member ensemble distribution. The wind speed anomalies are perfectly correlated to the dust anomalies. For example, in phases 4 and 5, the wind speed anomalies over the central Sahara are negative, indicating a decreased wind speed with respect to average. Correspondingly, the dust optical depth anomalies are also negative, indicating less dust emitted than average in this phase of the MJO. In phases 6 and 7, however, the wind speed anomalies are positive and so are the corresponding dust optical depth anomalies.

Fig. 11.

Anomaly of (left) surface wind speed and (right) dust optical depth over North Africa in the different phases of the MJO: (a),(b) phases 2 and 3; (c),(d) phases 4 and 5; (e),(f) phases 6 and 7; and (g),(h) phases 8 and 1.

Fig. 11.

Anomaly of (left) surface wind speed and (right) dust optical depth over North Africa in the different phases of the MJO: (a),(b) phases 2 and 3; (c),(d) phases 4 and 5; (e),(f) phases 6 and 7; and (g),(h) phases 8 and 1.

These results suggest that the model is able to simulate a modulation of dust and carbonaceous aerosols by the MJO, which is consistent with the modulation detected in CAMSira and with observed studies by Tian et al. (2008, 2011). The amplitude of this modulation represents about 20% of the total variance and is therefore likely to have a significant impact on the forecasts. Current operational forecasts that use climatological aerosols do not simulate this modulation of the MJO and therefore cannot simulate its possible impact on the climatology. Although these results do not prove that the positive impact on the forecast skill scores when using interactive aerosols comes only from the modulation of the MJO, they provide a plausible mechanism for this improvement.

5. Summary and future outlook

Investigating the question of aerosol impacts in the extended range, the authors conducted three multiyear runs with the ECMWF’s Ensemble Prediction System coupled with the NEMO ocean model at 1/4° resolution. The resolution of the atmospheric model was T255, corresponding to approximately 80 km. Twelve years of reforecasts with 50 ensemble members were analyzed for the period May–June. The two control runs used the T97 climatology and the CAMS/Bozzo climatology implemented in the most recent version of the ECMWF model (Bozzo et al. 2017), whereas the experiments with prognostic aerosols allow the latter to be interactive in the radiation scheme. Only direct effects were included. The two prognostic aerosol integrations were initialized with the time-varying CAMSira and with a fixed climatology based on a free-running version of the CAMS forecast system, respectively.

Results have shown the potential of interactive prognostic aerosols to improve model prediction at the monthly scales. Temperature and wind biases were reduced in both prognostic aerosol runs over several areas in the tropics and the midlatitudes. East Asia and the North Pacific seem to benefit especially from the interactive aerosols, with reduction in temperature biases of up to 0.5°–2.0°. Scorecards also show positive significant impacts of the prognostic aerosols on several meteorological fields, including upper-level winds and lower-tropospheric temperature, particularly over the Northern Hemisphere. The run with the climatological initialization scored generally better, possibly because the initial state of the model with the prognostic aerosols is closer to the initializing climatology than to the CAMSira.

A plausible physical mechanism behind this positive impact was shown to be the aerosol modulation induced by the MJO in its various phases. The patterns of variability in the monthly runs correlate well with the same patterns calculated from CAMSira, indicating 1) a degree of realism in the aerosol fields produced by the monthly run and 2) the importance of quantifying the aerosol variability induced by atmospheric modes, which are active at the subseasonal scales, such as the MJO. It would also be beneficial to see if these mechanisms and their impacts on the prediction are similar in other models with subseasonal prediction capabilities. This could be explored, for example, in the context of the subseasonal-to-seasonal (S2S) prediction project supported by the World Meteorological Organization (WMO).

As a welcome side effect, having prognostic aerosols in the monthly system also means having a monthly prediction of the aerosols themselves. Figure 12 shows the ranked probability skill score (RPSS) for the PROG1 and PROG2 integrations for the tropics. Persistence is also shown for comparison. Both forecast experiments have higher RPSS than persistence for dust aerosols. PROG1 scores the highest, related to the fact that it has been initialized with CAMSira. This could be of interest, for example, for health-related applications. Dust has been linked in northern Africa with outbreaks of meningitis, and having an even moderately skillful model prediction a month ahead could be useful for planning and preparedness by the health authorities. A full assessment of the skill of the aerosol prediction at the monthly scales will be explored in a follow-on paper.

Fig. 12.

RPSS for experiments PROG1 (orange) and PROG2 (green) with respect to a persistence forecast (blue) of dust optical depth for the tropics.

Fig. 12.

RPSS for experiments PROG1 (orange) and PROG2 (green) with respect to a persistence forecast (blue) of dust optical depth for the tropics.

Observed emissions for biomass burning based on fire radiative power were prescribed over the period under investigation. Since those have only been available since 2002, subsequent studies will be performed with biomass burning climatologies such as the Global Fire Emission Database (GFED) inventory (van der Werf et al. 2010), which allows for longer reforecasts. The impact of using fire emission climatologies instead of observed emissions will be assessed in a separate study, as well as the possibility of developing a fire predictive model based on meteorological variables at the subseasonal/seasonal scales.

Acknowledgments

The research leading to these results has received funding from the European Union 7th Framework Programme (FP7/2007-2013) under Grant Agreement 603502 (EU project DACCIWA: Dynamics–aerosol–chemistry–cloud interactions in West Africa). Samuel Rémy is gratefully acknowledged for his expert advice on running the CAMS aerosol model. Johannes Flemming is gratefully thanked for providing the simulation for the climatological initialization. We also acknowledge invaluable help received from Alessio Bozzo, who provided the CAMS climatology and contributed to discussions on the paper. Thanks also to Francesca Di Giuseppe, Rossana Dragani, and Magdalena Alonso Balmaseda for the support around this line of research and useful discussions. Many thanks to Anabel Bowen and Simon Witter for their assistance in improving the figures. The comments of three anonymous reviewers have helped greatly in improving this manuscript and are therefore gratefully acknowledged.

REFERENCES

REFERENCES
Bellucci
,
A.
, and Coauthors
,
2015
:
Advancements in decadal climate predictability: The role of nonoceanic drivers
.
Rev. Geophys.
,
53
,
165
202
, https://doi.org/10.1002/2014RG000473.
Boucher
,
O.
,
M.
Pham
, and
C.
Venkataraman
,
2002
: Simulation of the atmospheric sulfur cycle in the LMD GCM: Model description, model evaluation, and global and European budgets. IPSL Note 23, 26 pp.
Boucher
,
O.
, and Coauthors
,
2013
: Clouds and aerosols. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 571–657.
Bozzo
,
A.
,
S.
Remy
,
A.
Benedetti
,
J.
Flemming
,
P.
Bechtold
,
M.
Rodwell
, and
J.-J.
Morcrette
,
2017
: Implementation of a CAMS-based aerosol climatology in the IFS. ECMWF Tech. Memo. 801, 35 pp., https://www.ecmwf.int/sites/default/files/elibrary/2017/17219-implementation-cams-based-aerosol-climatology-ifs.pdf.
Charlson
,
R. J.
,
S. E.
Schwartz
,
J. M.
Hales
,
R. D.
Cess
,
J. A.
Coakley
,
J.
Hansen
, and
D.
Hofmann
,
1992
:
Climate forcing by anthropogenic aerosols
.
Science
,
255
,
423
430
, https://doi.org/10.1126/science.255.5043.423.
Colarco
,
P. R.
,
E. P.
Nowottnick
,
C. A.
Randles
,
B.
Yi
,
P.
Yang
,
K.-M.
Kim
,
J. A.
Smith
, and
C. G.
Bardeen
,
2014
:
Impact of radiatively interactive dust aerosols in the NASA GEOS-5 climate model: Sensitivity to dust particle shape and refractive index
.
J. Geophys. Res. Atmos.
,
119
,
753
786
, https://doi.org/10.1002/2013JD020046.
Collins
,
W. D.
,
P. J.
Rasch
,
B. E.
Eaton
,
D. W.
Fillmore
,
J. T.
Kiehl
,
C. T.
Beck
, and
C. S.
Zender
,
2002
:
Simulation of aerosol distributions and radiative forcing for INDOEX: Regional climate impacts
.
J. Geophys. Res.
,
107
,
8028
, https://doi.org/10.1029/2000JD000032.
Crutzen
,
P. J.
, and
M. O.
Andreae
,
1990
:
Biomass burning in the tropics: Impact on atmospheric chemistry and biogeochemical cycles
.
Science
,
250
,
1669
1678
, https://doi.org/10.1126/science.250.4988.1669.
Dee
,
D.
, and Coauthors
,
2011
:
The ERA-Interim reanalysis: Configuration and performance of the data assimilation system
.
Quart. J. Roy. Meteor. Soc.
,
137
,
553
597
, https://doi.org/10.1002/qj.828.
Flemming
,
J.
, and Coauthors
,
2017
:
The CAMS interim reanalysis of carbon monoxide, ozone and aerosol for 2003–2015
.
Atmos. Chem. Phys.
,
17
,
1945
1983
, https://doi.org/10.5194/acp-17-1945-2017.
Guo
,
Y.
,
B.
Tian
,
R. A.
Kahn
,
O.
Kalashnikova
,
S.
Wong
, and
D. E.
Waliser
,
2013
:
Tropical Atlantic dust and smoke aerosol variations related to the Madden–Julian oscillation in MODIS and MISR observations
.
J. Geophys. Res. Atmos.
,
118
,
4947
4963
, https://doi.org/10.1002/jgrd.50409.
Hansen
,
J.
,
A.
Lacis
,
R.
Ruedy
, and
M.
Sato
,
1992
:
Potential climate impact of Mount Pinatubo eruption
.
Geophys. Res. Lett.
,
19
,
215
218
, https://doi.org/10.1029/91GL02788.
Kaiser
,
J.
, and Coauthors
,
2012
:
Biomass burning emissions estimated with a global fire assimilation system based on observed fire radiative power
.
Biogeosciences
,
9
,
527
554
, https://doi.org/10.5194/bg-9-527-2012.
Koch
,
D.
, and
A. D.
Del Genio
,
2010
:
Black carbon semi-direct effects on cloud cover: Review and synthesis
.
Atmos. Chem. Phys.
,
10
,
7685
7696
, https://doi.org/10.5194/acp-10-7685-2010.
Lamarque
,
J.-F.
, and Coauthors
,
2010
:
Historical (1850–2000) gridded anthropogenic and biomass burning emissions of reactive gases and aerosols: Methodology and application
.
Atmos. Chem. Phys.
,
10
,
7017
7039
, https://doi.org/10.5194/acp-10-7017-2010.
Lau
,
W. K.
, and
K.-M.
Kim
,
2010
:
Fingerprinting the impacts of aerosols on long-term trends of the Indian summer monsoon regional rainfall
.
Geophys. Res. Lett.
,
37
,
L16705
, https://doi.org/10.1029/2010GL043255.
Leutbecher
,
M.
, and Coauthors
,
2017
:
Stochastic representations of model uncertainties at ECMWF: State of the art and future vision
.
Quart. J. Roy. Meteor. Soc.
,
143
,
2315
2339
, https://doi.org/10.1002/qj.3094.
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.
Morcrette
,
J.-J.
, and Coauthors
,
2009
:
Aerosol analysis and forecast in the European Centre for Medium-Range Weather Forecasts Integrated Forecast System: Forward modeling
.
J. Geophys. Res.
,
114
,
D06206
, https://doi.org/10.1029/2008JD011235.
Morcrette
,
J.-J.
,
A.
Benedetti
,
A.
Ghelli
,
J.
Kaiser
, and
A.
Tompkins
,
2011
: Aerosol-cloud-radiation interactions and their impact on ECMWF/MACC forecasts. ECMWF Tech. Memo. 660, 36 pp., https://www.ecmwf.int/en/elibrary/11283-aerosol-cloud-radiation-interactions-and-their-impact-ecmwf-macc-forecasts.
Mulcahy
,
J. P.
,
D. N.
Walters
,
N.
Bellouin
, and
S. F.
Milton
,
2014
:
Impacts of increasing the aerosol complexity in the Met Office global numerical weather prediction model
.
Atmos. Chem. Phys.
,
14
,
4749
4778
, https://doi.org/10.5194/acp-14-4749-2014.
Murakami
,
M.
,
1976
:
Analysis of summer monsoon fluctuations over India
.
J. Meteor. Soc. Japan
,
54
,
15
31
, https://doi.org/10.2151/jmsj1965.54.1_15.
Murphy
,
A. H.
, and
R. L.
Winkler
,
1984
:
Probability forecasting in meteorology
.
J. Amer. Stat. Assoc.
,
79
,
489
500
.
Painemal
,
D.
, and
P.
Zuidema
,
2013
:
The first aerosol indirect effect quantified through airborne remote sensing during VOCALS-REx
.
Atmos. Chem. Phys.
,
13
,
917
931
, https://doi.org/10.5194/acp-13-917-2013.
Pérez
,
C.
,
S.
Nickovic
,
G.
Pejanovic
,
J. M.
Baldasano
, and
E.
Özsoy
,
2006
:
Interactive dust-radiation modeling: A step to improve weather forecasts
.
J. Geophys. Res.
,
111
,
D16206
, https://doi.org/10.1029/2005JD006717.
Ramanathan
,
V.
, and Coauthors
,
2001
:
Indian Ocean Experiment: An integrated analysis of the climate forcing and effects of the great Indo-Asian haze
.
J. Geophys. Res.
,
106
,
28 371
28 398
, https://doi.org/10.1029/2001JD900133.
Ramanathan
,
V.
, and Coauthors
,
2005
:
Atmospheric brown clouds: Impacts on South Asian climate and hydrological cycle
.
Proc. Natl. Acad. Sci. USA
,
102
,
5326
5333
, https://doi.org/10.1073/pnas.0500656102.
Reale
,
O.
,
K.
Lau
, and
A.
da Silva
,
2011
:
Impact of interactive aerosol on the African easterly jet in the NASA GEOS-5 global forecasting system
.
Wea. Forecasting
,
26
,
504
519
, https://doi.org/10.1175/WAF-D-10-05025.1.
Reale
,
O.
,
K.
Lau
,
A.
da Silva
, and
T.
Matsui
,
2014
:
Impact of assimilated and interactive aerosol on tropical cyclogenesis
.
Geophys. Res. Lett.
,
41
,
3282
3288
, https://doi.org/10.1002/2014GL059918.
Reddy
,
M. S.
,
O.
Boucher
,
N.
Bellouin
,
M.
Schulz
,
Y.
Balkanski
,
J.-L.
Dufresne
, and
M.
Pham
,
2005
:
Estimates of global multicomponent aerosol optical depth and direct radiative perturbation in the Laboratoire de Météorologie Dynamique general circulation model
.
J. Geophys. Res.
,
110
,
D10S16
, https://doi.org/10.1029/2004JD004757.
Rémy
,
S.
, and Coauthors
,
2015
:
Feedbacks of dust and boundary layer meteorology during a dust storm in the eastern Mediterranean
.
Atmos. Phys. Chem.
,
15
,
12 909
12 933
, https://doi.org/10.5194/acp-15-12909-2015.
Rodwell
,
M. J.
, and
T.
Jung
,
2008
:
Understanding the local and global impacts of model physics changes: An aerosol example
.
Quart. J. Roy. Meteor. Soc.
,
134
,
1479
1497
, https://doi.org/10.1002/qj.298.
Sun
,
Y.
,
Q.
Jiang
,
Z.
Wang
,
P.
Fu
,
J.
Li
,
T.
Yang
, and
Y.
Yin
,
2014
:
Investigation of the sources and evolution processes of severe haze pollution in Beijing in January 2013
.
J. Geophys. Res. Atmos.
,
119
,
4380
4398
, https://doi.org/10.1002/2014JD021641.
Tegen
,
I.
,
P.
Hollrig
,
M.
Chin
,
I.
Fung
,
D.
Jacob
, and
J.
Penner
,
1997
:
Contribution of different aerosol species to the global aerosol extinction optical thickness: Estimates from model results
.
J. Geophys. Res.
,
102
,
23 895
23 915
, https://doi.org/10.1029/97JD01864.
Tian
,
B.
, and Coauthors
,
2008
:
Does the Madden-Julian oscillation influence aerosol variability?
J. Geophys. Res.
,
113
,
D12215
, https://doi.org/10.1029/2007JD009372.
Tian
,
B.
,
D. E.
Waliser
,
R. A.
Kahn
, and
S.
Wong
,
2011
:
Modulation of Atlantic aerosols by the Madden-Julian oscillation
.
J. Geophys. Res.
,
116
,
D15108
, https://doi.org/10.1029/2010JD015201.
Tompkins
,
A. M.
,
C.
Cardinali
,
J.-J.
Morcrette
, and
M.
Rodwell
,
2005
:
Influence of aerosol climatology on forecasts of the African easterly jet
.
Geophys. Res. Lett.
,
32
,
L10801
, https://doi.org/10.1029/2004GL022189.
van der Werf
,
G. R.
, and Coauthors
,
2010
:
Global fire emissions and the contribution of deforestation, savanna, forest, agricultural, and peat fires (1997–2009)
.
Atmos. Chem. Phys.
,
10
,
11 707
11 735
, https://doi.org/10.5194/acp-10-11707-2010.
Vitart
,
F.
,
2014
:
Evolution of ECMWF sub-seasonal forecast skill scores
.
Quart. J. Roy. Meteor. Soc.
,
140
,
1889
1899
, https://doi.org/10.1002/qj.2256.
Wang
,
Y.
, and Coauthors
,
2014
:
Mechanism for the formation of the January 2013 heavy haze pollution episode over central and eastern china
.
Sci. China Earth Sci.
,
57
,
14
25
, https://doi.org/10.1007/s11430-013-4773-4.
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.
Yasunari
,
T.
,
1979
:
Cloudiness fluctuations associated with the Northern Hemisphere summer monsoon
.
J. Meteor. Soc. Japan
,
57
,
227
242
, https://doi.org/10.2151/jmsj1965.57.3_227.
Zhang
,
J.
,
J. S.
Reid
,
M.
Christensen
, and
A.
Benedetti
,
2016
:
An evaluation of the impact of aerosol particles on weather forecasts from a biomass burning aerosol event over the Midwestern United States: Observational-based analysis of surface temperature
.
Atmos. Chem. Phys.
,
16
,
6475
6494
, https://doi.org/10.5194/acp-16-6475-2016.

Footnotes

© 2018 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).