Forcing, Precipitation and Cloud Responses to Individual Forcing Agents

by

Tao Tang

Earth and Ocean Sciences Duke University

Date:______Approved:

______Drew Shindell, Supervisor

______Prasad Kasibhatla

______M. Susan Lozier

______Apostolos Voulgarakis

______A. Brad Murray

Dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Department of Earth and Ocean Sciences in the Graduate School of Duke University

2020

i

v

ABSTRACT

Forcing, Precipitation and Cloud Responses to Individual Forcing Agents

by

Tao Tang

Earth and Ocean Sciences Duke University

Date:______Approved:

______Drew Shindell, Supervisor

______Prasad Kasibhatla

______M. Susan Lozier

______Apostolos Voulgarakis

______A. Brad Murray

An abstract of a dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Department of Earth and Ocean Sciences in the Graduate School of Duke University

2020

Copyright by Tao Tang 2020

Abstract

Previously, we usually analyze climate responses to all the climate drivers combined. However, the climate responses to individual climate drivers are far from well-known, as it is nearly impossible to separate the climate responses to individual climate drivers from the pure observational records. In this dissertation, I analyzed the responses of effective radiative forcing (ERF), precipitation and clouds to five individual climate drivers by using the model output from the Precipitation and Driver Response

Model Inter-comparison Project (PDRMIP, consisting of five core experiments: CO2x2,

CH4x3, Solar+2%, BCx10, and SO4x5). Firstly, I compared the ERF values estimated by six different methods and demonstrated that the values estimated using fixed sea- surface temperature and linear regression methods are fairly consistent for most climate drivers. For each individual driver, multi-model mean ERF values vary by 10-50% with different methods, and this difference may reach 70-100% for BC. Then, I analyzed the dynamical responses of precipitation in Mediterranean to well-mixed greenhouse gases

(WMGHGs) and aerosols and found that precipitation in Mediterranean is more sensitive to BC forcing. When scaled to historical forcing level, WMGHG contributed roughly two-thirds to the Mediterranean drying during the past century and BC aerosol contributed the remaining one-third by causing a northward shift of the jet streams and storm tracks. Lastly, I explored the responses of shortwave cloud radiative effect

(SWCRE) to CO2 and the two aerosol species and found that CO2 causes positive

iv

SWCRE changes over most of the Northern Hemisphere during boreal summer, and BC causes similar positive responses over North America, Europe and East China but negative SWCRE over India and tropical Africa. When normalized by global ERF, the change of SWCRE from BC forcing is roughly 3-5 times larger than that from CO2.

SWCRE change is mainly due to cloud cover changes resulting from the changes in relative humidity, and to a lesser extent, changes in circulation and stability. The

SWCRE response to sulfate aerosols, however, is negligible compared to that from CO2 and BC, because the radiation scattered by clouds under all-sky conditions will also be scattered by aerosols under clear-sky conditions. As SW is in effect only during daytime, positive (negative) SWCRE could amplify (dampen) daily maximum temperature

(Tmax). Using a multi-linear regression model, I found that Tmax increases by 0.15 K and 0.13 K given unit increase in local SWCRE under the CO2 and BC experiments, respectively. When domain-averaged, SWCRE changes contributed to summer mean

Tmax changes by 10-30% under CO2 forcing and by 30-50% under BC forcing, varying by regions, which can have important implications extreme climatic events and socio- economic activities.

v

Dedication

This dissertation is dedicated to my family and beloved ones.

vi

Contents

Abstract ...... iv

Dedication ...... vi

Contents ...... vii

List of Tables ...... ix

List of Figures ...... x

Acknowledgements ...... xii

1 Introduction and Overview ...... 1

1.1 Basic Concept ...... 1

1.2 PDRMIP Background ...... 3

1.3 Motivations ...... 5

2. Comparison of Effective Radiative Forcing Calculations using Multiple Methods, Drivers and Models ...... 11

2.1 Introduction ...... 11

2.2 Data and Methods ...... 13

2.2.1 Model data ...... 13

2.2.2 Estimating ERF ...... 16

2.3 Results ...... 20

2.4 Discussion and Summary ...... 33

3. Dynamical Response of Mediterranean Precipitation to Greenhouse Gases and Aerosols ...... 42

3.1 Introduction ...... 42

3.2 Data and Method ...... 44

vii

3.2.1 Data ...... 44

3.2.2 Method ...... 47

3.3 Results ...... 48

3.4 Case Study ...... 53

3.5 Discussions and Conclusion ...... 61

4. Response of Shortwave Cloud Radiative Effect to Greenhouse Gases and Aerosols and its Impact on Daily Maximum Temperature ...... 67

4.1 Introduction ...... 67

4.2 Data and Methods ...... 70

4.2.1 Data ...... 70

4.2.2 Method ...... 71

4.3 Results ...... 73

4.3.1 SWCRE Change ...... 73

4.3.2 Mechanism of the Cloud Changes ...... 75

4.3.3 Fast and Slow Responses ...... 80

4.3.4 SWCRE Response to Sulfate Aerosol ...... 81

4.3.5 Impact on Radiation and Tmax ...... 82

4.4 Discussion and Conclusion ...... 87

5. Conclusion and Summary ...... 92

5.1 Key Findings ...... 92

5.2 Implications, Limitations and Future Work ...... 94

References ...... 99

Biography ...... 115

viii

List of Tables

Table 1: Descriptions of the nine PDRMIP models used in this study...... 15

Table 2: Forcing, feedback and global mean land temperature change (∆T_land) for the CO2×2 experiment...... 22

Table 3: same as Table 2, but for Solar+2% experiment...... 23

Table 4: Same as Table 2, but for SO4×5 experiment...... 24

Table 5: Same as Table 2, but for BC×10 experiment...... 25

Table 6: Same as Table 2, but for CH4×3 experiment...... 26

Table 7: Comparison of multi-model mean (MMM) values of root-mean-squared errors ...... 28

Table 8 Feedbacks parameters for land/ocean in the different time periods for the CO2×2 experiment...... 36

Table 9 Comparison of MMM (mean±1 std) values of ERF derived with different time periods...... 38

Table 10 Aerosol treatments for the PDRMIP models...... 46

Table 11: Domain-averaged Tmax changes from each radiative component estimated from the linear models (unit: K)...... 87

ix

List of Figures

Figure 1: Different definitions of climate forcing ...... 3

Figure 2: The evolution of radiative imbalance (N) with surface temperature change (∆T) in the CO2×2 experiments...... 21

Figure 3: Same as Figure 1, but for Solar+2% experiment...... 28

Figure 4: Same as Figure 1 but for SO4 x 5 experiment...... 29

Figure 5: Same as Figure 1, but for BC×10 experiment...... 30

Figure 6: Same as Figure 1, but for the CH4×3 experiments...... 32

Figure 7: Comparison of different ERF values for each experiment...... 34

Figure 8: Normalized ∆P (change per unit forcing) ...... 49

Figure 9: Domain-averaged ∆P ...... 49

Figure 10: Domain-averaged energy budget change for each forcing and energy component ...... 50

Figure 11: Same as Figure 7, but for sea level pressure (SLP)...... 52

Figure 12: Same as Figure 7, but for zonal wind...... 52

Figure 13: Scaled change for the combination of WMGHGs, BC and SO4 ...... 58

Figure 14: SLP (a & b) and zonal wind change (c & d)...... 59

Figure 15: Precipitation trends of CMIP5 HistoricalGHG simulations ...... 62

Figure 16: Aerosol loadings for the two aerosol experiments in CanESM2 model ...... 71

Figure 17: SWCRE changes (a-c) and cloud cover changes per unit forcing ...... 74

Figure 18: Domain-averaged SWCRE changes for three regions ...... 75

Figure 19: Same as Figure 17, but for humidity at 850 hPa...... 76

Figure 20: Domain-averaged vapor pressure changes per unit forcing ...... 77

x

Figure 21: Same as Fig. 17, but for changes of moisture flux convergence (MFC, a-c) and vertical velocity (omega, d-f) per unit forcing...... 77

Figure 22: Same as Fig. 17, but for MFC changes due to dynamics (a-c) and thermodynamics (d-f)...... 79

Figure 23: Same as Figure 17 (d-f), but for fast (a-c) and slow responses (d-f) of cloud cover changes per unit forcing...... 81

Figure 24: Changes of SW flux per unit negative forcing under all-sky (a), clear-sky (b) conditions and their difference (c) for SO4 experiment...... 82

Figure 25: Changes of Rin and its components ...... 84

Figure 26: Comparison of fitted Tmax from the linear models vs original Tmax values. 85

Figure 27: SWCRE changes for the BC experiment ...... 90

xi

Acknowledgements

I would like to thank Duke University for providing the opportunity for me to pursuit PhD degree and my advisor Prof. Drew Shindell for his mentor, support and guidance. It is really enjoyable to work with him. I would also like to thank my committee members, Prof. M. Susan Lozier (Duke & Georgia Tech), Prof. Prasad

Kasibhatla (Duke), Prof. A. Brad Murray (Duke) and Dr. Apostolos Voulgarakis

(Imperial College London) for their helpful guidance and invaluable advice throughout my PhD and dissertation. I would like to thank my PDRMIP collaborators for their insightful comments in improving my research and paper manuscripts. I would also like to thank my group mates, Yuqiang Zhang, Karl Seltzer and Muye Ru, for their assistance on my research. They also make my PhD life colorful and fun. I would also like to thank

EOS faculty members, staff and other colleagues for the great and colorful environment they provide. This dissertation is impossible without these people.

xii

1 Introduction and Overview

This dissertation is the combined PhD work during my past a few years, aiming to investigate the climate responses to individual climate drivers from three perspectives-forcing, regional precipitation and cloud responses. In this chapter, I will begin with a brief introduction to the basic concept (e.g., the definition of climate driver, climate forcing), and the reasons why it is imperative to study the climate responses to individual climate drivers. This is followed by the introduction of PDRMIP project and the reasons why we need model simulations from this project. Then I will introduce the motivations and the big picture for the three projects. The detailed studies are presented in the following chapters (Chapter 2-4), followed by the concluding remarks in Chapter

5. Brief conclusions and implications for each study will be given. Some interesting questions merit further investigations and possible future work are also discussed.

1.1 Basic Concept

Climate drivers are substances or activities that external to the climate system and could change the radiative flux at the top of the atmosphere (TOA) to warm or cool the climate system. The most well-known climate driver is the greenhouse gases (GHGs) in the atmosphere, such as CO2, CH4 and N2O, which are generally known as well-mixed

GHGs (WMGHGs) as their concentrations are uniform in the atmosphere. Other anthropogenic GHGs includes tropospheric ozone (O3), CFCs and HCFCs. These GHGs can trap outgoing longwave radiation and re-emit back to surface or to outer space with

1

a lower temperature and thus, warming the climate. In addition to GHGs, there are also some other important climate drivers, including but are not limited to aerosols, land use/cover change, solar irradiance. Aerosols are small particles suspended in the air, which can interact with incoming solar radiation either by scattering or absorbing the radiation. They can also interact with clouds, changing the microphysical properties of clouds and rainfall efficiency (Boucher et al., 2013). The anthropogenic aerosols are mainly from transportation, power plants, energy utility. Land us and land cover change activities mainly include agricultural activities, deforestation and urbanization activities, which could impact the climate by modifying the surface albedo, influencing turbulent fluxes and altering the carbon cycle. The solar irradiance influences the incoming solar radiation, as the solar flux is influenced by the sunspots on the solar disk with a 11-year cycle.

When a climate driver is present, the resulting radiative imbalance at TOA is called forcing (Figure 1). If the adjustments of atmospheric temperature, water vapor and clouds are made, but with surface temperature or a portion of surface conditions unchanged, the forcing is called effective radiative forcing (ERF) (Myhre et al., 2013b). In a warming scenario, the net radiative flux (ERF) is positive downward. For instance, the total anthropogenic ERF is 2.3 W/m2 in 2011 relative to 1750 (Myhre et al., 2013b). It has been shown to be a more accurate indicator of the temperature response to forcing agents than the standard stratospherically adjusted radiative forcing, due to the

2

inclusion of tropospheric adjustments (Figure 1). When the energy is imbalanced at the

TOA, the climate will respond to such energy imbalance. In a warming case, the climate will get warmer and emit more longwave radiation to balance such radiative flux until it reaches a new equilibrium state, in which the net radiative flux will become zero.

Figure 1: Different definitions of climate forcing (Myhre et al., 2013b).

1.2 PDRMIP Background

The global mean temperature has risen by nearly 1 K since 1880, and such climate responses have been extensively studied and well documented (Hartmann et al.,

2013). However, this is the climate response to all the climate drivers in combined, and much less attention has been paid to the climate response to individual climate drivers.

The responses to individual drivers are mainly diagnosed by using global climate

3

models (GCMs), as it is nearly impossible to separate the climate responses to individual climate drivers from the observational data. Most previous studies are mainly focused on the doubling or quadrupling the concentration of CO2 (Manabe & Wetherald, 1975;

Washington & Meehl, 1984; Andrews et al., 2012b). There are also some studies exploring the climate responses to aerosols (Allen & Sherwood, 2011; Westervelt et al.,

2017; Xu et al., 2018). These studies provide useful information in understanding the climate responses to individual climate drivers. However, they mostly use only one climate model for CO2 simulations only. For aerosol studies, the aerosols are generally treated as a whole, not by species, and the forced signal is usually small with a large uncertainty range (Bond et al., 2013). In addition, an organized multi-model study is still lacking.

In this regard, a group of international scientists launched the project-

Precipitation and Driver Response Model Inter-comparison Project (PDRMIP) in Oslo,

Norway 2013 (Myhre et al., 2017), which aims to extend the analysis of the impacts of single climate drivers, on short and long time scales. The PDRMIP is designed to perform a thorough investigation of the differences in the effects of anthropogenic and natural drivers on climate. This will be accomplished based on five core simulations, a doubling of CO2 concentration (CO2×2), a tripling of CH4 concentration (CH4×3), a 2% increase in solar irradiance (Solar+2%), a tenfold increase of present-day black carbon concentration/emission (BC×10), and a fivefold increase of present-day SO4

4

concentration/emission (SO4×5). All perturbations were abrupt. This is the first internationally-organized model inter-comparison project that aims to systematically diagnose the climate responses to individual climate drivers. There are 11 GCMs participating PDRMIP project and 9 of them are employed in the this dissertation (For detailed model setup, please see Chapter 2). There are already some studies using

PDRMIP output before I started my PhD work. For example, Samset et al. (2016) analyzed the fast and slow precipitation response. Liu et al. (2018) studied the precipitation response to local and remote aerosol forcings. In order to avoid the overlap with those studies, my analyses in this dissertation are focusing on forcing, regional precipitation and the response of cloud radiative effect. The background and motivations are discussed below.

1.3 Motivations

To better our understanding of climate responses to single climate drivers, I present some results by analyzing observational data, PDRMIP output, as well as output from other climate simulations (e.g., CMIP5) in this dissertation. The first study is focused on the comparison of ERF values calculated by different methods under these five climate drivers (Chapter 2). ERF is important in that it is crucial to examine the climate sensitivity and projecting long-term climate change. For instance, if we can obtain the climate sensitivity with current global ERF and temperature responses and

5

based on this value, we can project how much the temperature will rise based on GHGs emissions in the future. The two most commonly used techniques are linear regression in the coupled simulation (Gregory et al., 2004) or by using fixed-sea surface temperature simulations (Hansen et al., 2002), which has reported that the values from these two methods are consistent in the CO2 experiments (Andrews et al., 2012b).

However, the linear regression method has been challenged by some recent studies arguing that the slopes will become less negative after a few decades (Armour et al.,

2013; Andrews et al., 2015), which may lead to biased lower values. To address this time- varying feedback, two curved regression will be employed for comparison in this study.

Besides, if the consistency of the values from the two methods still hold for other climate drivers is still unclear. Thus, a multi-model and multi-agent comparison will provide useful information on these issues. Another important thing to note is that the ERF is only a metric to describe the forcing. The forcing values will not influence our ongoing climate policy or mitigation measures, such as Paris Agreement. We, in fact, do not know the true value of this forcing and even the definition of ERF is still in debate

(Boucher et al., 2013). I am not intending to find the true value of forcing, but rather to compare the ERF values obtained by six different methods to provide information to the researchers in the community. They can choose the best method based on their own needs, available output and computational resources.

6

When a climate driver is present, it will cause energy imbalance (ERF) and the climate will be responding to this perturbation (e.g., changes in surface temperature, precipitation and circulation). My second study is focused on precipitation in the

Mediterranean region (Chapter 3). The motivation is twofold. Firstly, the precipitation has been decreasing in the Mediterranean region since the early 20th century (Piervitali et al., 1998; Buffoni et al., 1999; Mariotti et al., 2002; Dünkeloh & Jacobeit, 2003; Xoplaki et al., 2004). This drying not only influenced the hydrological cycle and ecosystems

(Piervitali et al., 1998; Mariotti et al., 2002), but also had great socio-economic impacts, causing water shortages and disrupt agricultural and industrial activities, as well as hydro-electric power generations (Fernandez et al., 2003). Secondly, the impacts of individual climate drivers are still uncertain, especially the roles of aerosols. On history, there is a period (1940-1970) when the global mean temperature changes little or slightly decreased, which is often called ‘global dimming’. This ‘dimming’ is mostly related with the large emissions of aerosols from Europe (Wild, 2009). Anthropogenic aerosols have been reported to greatly influence the temperature in the Mediterranean (Wild, 2009;

Nabat et al., 2014), but the effects of aerosols on Mediterranean precipitation have not been carefully examined. Thus, it is crucial to understand the different impacts of the climate drivers that are responsible for the Mediterranean precipitation trend. In

Chapter 3, I will present the results from a dynamical perspective by using PDRMIP model output and observational/reanalysis datasets. The results show that the BC

7

aerosols also played a substantial role in the drying of Mediterranean, despite the dominant role from GHGs.

It is reported that there has been an increasing trend in the frequency and intensity of heatwave events in much of Europe, Asia and Australia (Meehl & Tebaldi,

2004; Della-Marta et al., 2007; Barriopedro et al., 2011; Hartmann et al., 2013), which have had devastating impacts on environment, ecosystems and socio-economy, such as forest fires, power cuts, transport restrictions, crop failure and live loss (De Bono et al.,

2004; Ciais et al., 2005; Robine et al., 2008). These extreme temperature events have been associated with increasing global mean surface temperature and variability due to continued increasing concentrations of greenhouse gases (Schär et al., 2004; Seneviratne et al., 2006; Vautard et al., 2007; Zampieri et al., 2009; Hartmann et al., 2013). The mechanisms of these extreme climatic events have been proposed by many studies and they are generally characterized by two features: (1) quasi-stationary anticyclonic circulation patterns and (2) soil water deficit (Meehl & Tebaldi, 2004; Fischer et al.,

2007a; Vautard et al., 2007; Barriopedro et al., 2011). Anticyclonic conditions enhance subsidence, warm air advection and solar heating (Meehl & Tebaldi, 2004; Miralles et al.,

2014) while moisture deficit could partition more surface net energy into sensible heat instead of latent cooling, exacerbating drought and heatwave events (Fischer et al.,

2007b; Zampieri et al., 2009; Seneviratne et al., 2013). Among these mechanisms, a well- known process is the shortwave cloud radiative effect (SWCRE), in which the cloud

8

cover reduces and more SW radiation reaches the surface, intensifying solar heating and amplifying heatwaves (Rowell & Jones, 2006; Vautard et al., 2007; Zampieri et al., 2009;

Chiriaco et al., 2014; Myers et al., 2018).

In fact, much of global warming projected over the next century comes from feedbacks instead of direct warming from CO2. Among these feedbacks, the most complex and uncertain one is the cloud feedback (Dessler, 2010; Zelinka et al., 2017).

Clouds have a pivotal role in influencing the Earth’s energy budget (Ramanathan et al.,

1989). By enhancing the planetary albedo, clouds exert a global mean SWCRE of about -

50 W m-2 at the top-of-the-atmosphere, and by contributing to the greenhouse effect, exert a mean longwave effect (LWCRE) of approximately +30 W m-2 (Boucher et al.,

2013). On the whole, clouds cause a net cooling of 20 W m-2 relative to a cloud-free Earth, which is approximately five times as large as the heating from a doubling of CO2 concentration. Therefore, a subtle change in cloud properties has a potential to cause significant impact on climate (Boucher et al., 2013; Zelinka et al., 2017). It is suggested that SW cloud feedback is positively correlated with net climate feedback (Zhou et al.,

2016). This is because, when temperature is cooler, the atmosphere is more stable in tropical and subtropical Pacific Ocean, which traps more moisture and permits more low-level clouds. These clouds reflect more SW radiation at TOA, making the energy imbalance more negative (Wood & Bretherton, 2006; Zhou et al., 2016). Thus, the cloud feedback is of great importance to long-term climate change. Moreover, clouds have also

9

been reported to impact daily maximum temperature (Tmax), especially SWCRE, as it is in effect only during daytime, which could enhance or dampen solar heating (Dai et al.,

1999; Tang & Leng, 2012).

All the above studies suggest that the SWCRE plays a crucial role in influencing the surface energy budget and extreme temperature. In addition to the lack of multi- model studis on the cloud response to individual aerosol species, the contributions of

SWCRE to Tmax change from individual forcing agents have not been quantified yet.

Previous studies generally conclude that cloud cover and precipitation could explain the variability of Tmax or changes of diurnal temperature range by a fraction based on the coefficients of determination (R2) derived from a simple correlation analysis (Dai et al.,

1999; Tang & Leng, 2012). In Chapter 4, I examined the changes of SWCRE to CO2, BC and sulfate aerosols in terms of the changes of SWCRE and the mechanisms, and, for the first time, quantified the contributions of SWCRE to Tmax changes based on energy budget analysis.

10

2. Comparison of Effective Radiative Forcing Calculations using Multiple Methods, Drivers and Models

[Tang, T., Shindell, D., Faluvegi, G., Myhre, G., Olivié, D., Voulgarakis, A., ... &

Hodnebrog, Ø . (2019). Comparison of Effective Radiative Forcing Calculations Using

Multiple Methods, Drivers, and Models. Journal of Geophysical Research:

Atmospheres, 124(8), 4382-4394, doi: 10.1029/2018JD030188]

2.1 Introduction

ERF is defined as the net downward radiative flux at the top of the atmosphere

(TOA) after allowing for atmospheric temperature, water vapor and clouds to adjust, but with surface temperature or a portion of surface conditions unchanged (Myhre et al.,

2013b). It has been shown to be a more accurate indicator of the temperature response to forcing agents than the standard stratospherically-adjusted radiative forcing, due to the inclusion of tropospheric adjustments. Better estimation of ERF is crucial to understanding the climate response to different forcings as well as predicting long-term climate change.

Two methods are commonly employed to determine ERF (see methods): one is to simulate the climate response with fsst simulations (Hansen et al., 2002), the other is to linearly regress the net TOA radiative flux against global mean temperature change (∆T) in a transient model simulation (Gregory et al., 2004). It has been reported that the 11

values from these two methods are quantitatively consistent (Andrews et al., 2012b).

However, the ERF values from both methods may be biased. The former method allows land response in the simulations, and although Hansen et al. (2005) showed that accounting for land adjustment substantially increases ERF values, this is typically not included in fsst ERF calculations. The latter assumes a constant feedback parameter of the climate system in the adjustment process, which has been questioned by recent studies (Armour et al., 2013; Andrews et al., 2015; Gregory & Andrews, 2016;

Proistosescu & Huybers, 2017). For example, Armour et al. (2013) suggest that the linear regression technique has fundamental biases because the slope depends on the regions where the surface is warming most rapidly at the time when the regression is performed. Andrews et al. (2015) suggest that the feedback parameters would become less negative after a few decades, making the response concave instead of linear and leading to ERF values that are low biased using linear regression. Thus, a better understanding of ERF estimated with different methods is needed. There is no study so far, to our best knowledge, that compares the ERF values estimated with different approaches, multiple climate drivers and models. This study aims to bridge this gap, comparing the effects of methodological choices on the estimation of ERF values, using a group of global climate models forced with five different climate drivers.

12

2.2 Data and Methods

2.2.1 Model data

This study employs output from models participating in the Precipitation Driver and Response Model Intercomparison Project (PDRMIP), utilizing simulations examining the individual responses to CO2, CH4, solar insolation, BC and sulfate aerosols. The nine models used in this study are CanESM, GISS, HadGEM2, HadGEM3,

MIROC, CAM4, CAM5, NorESM and IPSL (see Table 1). In these simulations, global- scale perturbations were applied to the models: a doubling of CO2 concentration

(CO2×2), a tripling of CH4 concentration (CH4×3), a 2% increase in solar irradiance

(Solar+2%), a tenfold increase of present-day black carbon concentration/emission

(BC×10), and a fivefold increase of present-day SO4 concentration/emission (SO4×5). All perturbations were abrupt. Greenhouse gases and solar perturbations were applied relative to the models’ baseline values. For aerosol perturbations, monthly year 2000 concentrations were derived from the AeroCom Phase II initiative (Myhre et al., 2013a) and multiplied by the stated factors in concentration-driven models. Some models were unable to perform simulations with prescribed concentrations. These models multiplied the emissions by these factors instead, using either the same emissions or in one case another dataset (Table 1). Each perturbation was run in two configurations, a 15-yr fsst simulation and a 100-yr coupled simulation. One model (CESM-CAM4) used a slab ocean setup for the coupled simulation whereas the others used a full dynamic ocean.

13

More detailed descriptions of PDRMIP and its initial findings are given in Samset et al.

(2016), Myhre et al. (2017) and Tang et al. (2018).

14

Table 1: Descriptions of the nine PDRMIP models used in this study.

Model name Version Resolution Ocean setup Aerosol setup References CanESM 2010 2.8×2.8 35 levels Coupled Emission Arora et al. (2011) GISS-E2 E2-R 2×2.5 40 levels Coupled Fixed concentration Schmidt et al. (2014) 1.875×1.25 HadGEM2-ES 6.6.3 Coupled Emissions Collins et al. (2011) 38 levels 1.875×1.25 Bellouin et al. (2011) HadGEM3 GA 4.0 Coupled Fixed concentration 85 levels Walters et al. (2014) Takemura et al. (2005) MIROC- T85 5.9.0 Coupled HTAP2 emissions Takemura et al. (2009) SPRINTARS 40 levels Watanabe et al. (2010)

15 2.5×1.9 Neale et al. (2010) CESM-CAM4 1.0.3 Slab Fixed concentration 26 levels Gent et al. (2011) Hurrell et al. (2013) 2.5×1.9 Kay et al. (2015) CESM-CAM5 1.1.2 Coupled Emissions 30 levels Otto-Bliesner et al. (2016) Bentsen et al. (2013) 2.5×1.9 NorESM 1-M Coupled Fixed concentration Iversen et al. (2013) 26 levels Kirkevåg et al. (2013) IPSL-CM 5A 3.75×1.9 19 levels Coupled Fixed concentration Dufresne et al. (2013) Note: GA = Global Atmosphere. HTAP2 = Hemispheric Transport Air Pollution, Phase 2.

2.2.2 Estimating ERF

For fsst simulations, ERF (ERF_fsst) is typically diagnosed by calculating the change of global mean TOA radiative flux (Hansen et al., 2002):

ERF_fsst = ∆SW + ∆LW (1)

where ∆SW and ∆LW indicate the change of shortwave (SW) and longwave (LW) radiation at the TOA, respectively. For this calculation, I use years 6-15 from the fsst simulations. In ERF_fsst, the sea surface temperature (SST) and sea ice are fixed while the land surface is allowed to adjust, as in practice it is difficult to fix land temperature in the models. This means the global temperature has partially responded to the forcing, which causes the original external forcing to be underestimated. To account for this,

Hansen et al. (2005) proposed a modified definition of ERF based on fsst simulations:

ERF_fsst_∆Tland = ∆SW + ∆LW + ∆T_land/λ (2)

where ∆T_land is the change in land surface air temperature and λ is the climate sensitivity parameter (K per W m-2, in this case evaluated from the PDRMIP CO2×2 experiment). Like all ERFs, ERF_fsst_∆Tland also allows tropospheric and stratospheric conditions to adjust to the presence of the forcing agent.

16

For regression, following Gregory et al. (2004), global mean radiative flux at the

TOA is linearly regressed against ∆T in the coupled simulations to obtain ERF_linr:

N = F – H = F - α×∆T (3)

where N is the net radiation flux (W m-2, positive downward), F is the imposed forcing (W m-2, positive downward), and H is the radiative response caused by the climate change (W m-2, positive upward), which is linearly proportional to global surface temperature change (∆T). All values in Eqn (3) are global averages (W m-2). α is the climate feedback parameter (W m-2 K-1), indicating the strength of the climate system’s net feedback. If F and α are constant, N is a linear function of ∆T with a slope of -α and an intercept of F (N = - α×∆T + F). When ∆T = 0, the intercept F = N, which is thus the

ERF. Like ERFs diagnosed from fsst simulations, rapid adjustments of both the troposphere and stratosphere, including indirect effects of aerosols are included. Note that although the ERF is a short-term concept (e.g.., a few months after the forcing is imposed), the regression results inherently depend on the evolution of the climate system on longer time scales (e.g., a few decades) (Gregory et al., 2004). The first 30-yr of data were primarily used for regression analyses in this study.

17

I also apply two simple curved fits to the evolution of N and ∆T. The first is an exponential fit (ERF_exp):

N = a × exp (b × ∆T) (4)

where a and b are best-fit coefficients. When ∆T = 0, N = a, which thus gives the

ERF.

The second is a quadratic polynomial fit (ERF_poly):

N = p1×∆T2 + p2×∆T + p3 (5)

where p1, p2 and p3 are best-fit coefficients. When ∆T = 0, N = p3, which is hence the ERF.

Another method evaluated in this study involves the technique of radiative kernels (Soden et al., 2008).

ERF_kernel = ERF_fsst - AT_land - AT - Aq - Aa (6)

In this approach, the ERF_kernel value is obtained by subtracting the rapid adjustments associated with the land surface change from the ERF_fsst (obtained by

18

Eqn. 1), while keeping rapid adjustments in the atmosphere that are not as directly associated with the land surface response, such as cloud responses. Ax are the rapid adjustments associated with land surface responses, which include land surface temperature (AT_land) and albedo (Aa). In addition, land surface temperature change also causes changes in the tropospheric temperature (AT) and here I assume a constant lapse rate in the troposphere, therefore the same change in tropospheric temperature as that of the surface. For water vapor, the fraction of the radiative flux change from a constant lapse rate to the full tropospheric temperature change has been used to scale the calculation of total water vapor change in order to account for the portion of water change associated with the surface temperature response (Aq). A rapid adjustment is the product of the direct radiative response to an incremental change in the respective variable and the total climate response of that variable. The former term is the radiative kernel, derived from a single offline radiative transfer model, while the latter is estimated from the response of a given PDRMIP model.More details of the radiative kernel method are given in Zhang and Huang (2014), Chung and Soden (2015), Myhre et al. (2018) and Smith et al. (2018).

Uncertainties are reported for multi-model mean (MMM) values based on the model-to-model variation in results. Additional discussion of uncertainty is presented in the discussion section.

19

2.3 Results

The results from the CO2×2 experiment for each model are shown in Table 2 and

Fig. 1. The multi-model mean (MMM) ERF_fsst is 3.65±0.26 W m-2 and ERF_linr is

3.67±0.55 W m-2. These are quantitatively consistent (Table 2), in line with Gregory et al.

(2004)and Andrews et al. (2012b). All models show fairly strong linear correlations between N and ∆T (Fig. 1). Five models show larger values of ERF_fsst than ERF_linr and vice versa for the remaining models. The inter-model spread in ERF_linr is much larger than that of ERF_fsst, consistent with the results of Forster et al. (2016), which is presumably due to the extra processes involving the ocean in the coupled simulations and faster increase in the signal-to-noise ratio in equilibrium fsst simulations. When it comes to ERF_fsst_∆Tland, the MMM value is 4.78 W m-2, which is roughly 30% larger than either ERF_fsst or ERF_linr, due to the inclusion of land adjustment in the case of comparison with ERF_fsst. The models with greater increases from ERF_fsst to

ERF_fsst_∆Tland generally show stronger feedback (e.g., GISS and MIROC). Examining the two curved regression fits, MMM values are 3.91 W m-2 for ERF_poly and 4.70 W m-2 for ERF_exp, making the latter consistent with ERF_fsst_∆Tland, though large discrepancies occur in some individual models. Three models (CanESM, GISS and

CAM5) show opposite curvature in the polynomial fit relative to the exponential fit. I calculated the root mean squared errors (RMSE) for all three regression methods, which give similar goodness of fit (Table 7). Thus, it is difficult to determine which one is better

20

from a statistical perspective. The MMM of ERF_kernel is 4.13 W m-2, which is roughly

13% larger than ERF_fsst, but smaller then the ERF_fsst_∆Tland. This indicates that the sign of the net adjustments due to albedo, tropospheric temperatures and tropospheric water vapor is opposite to that of the land surface temperature in a warming scenario.

Figure 2: The evolution of radiative imbalance (N) with surface temperature change (∆T) in the CO2×2 experiments, with a linear fit (black dashed line), a quadratic polynomial fit (black solid line) and an exponential fit (blue line). The intercepts of the fits indicate ERF_linr, ERF_poly and ERF_exp, respectively. The y- coordinates of green circles indicate ERF_fsst_∆Tland and red stars indicate ERF_fsst. The x-coordinates of red stars indicate the global mean surface temperature change due to land (∆T_land) in the fsst experiment. Linear correlation coefficients between N and ∆T are shown in the upper-right corner.

21

Table 2: Forcing, feedback and global mean land temperature change (∆T_land) for the CO2×2 experiment.

ERF (W m-2) -α Model ∆T_land (K) fsst linr fsst_∆Tland poly exp kernel (W m-2 K-1) CanESM 3.57 4.38 4.31 4.15 5.15 4.01 -1.37 0.46 GISS 4.06 4.32 6.50 2.85 4.95 4.59 -2.22 0.56 HadGEM2 3.35 3.22 4.08 4.26 3.97 3.92 -0.78 0.49 HadGEM3 3.65 3.52 4.36 3.69 3.98 4.28 -0.72 0.61 MIROC 3.62 4.13 5.22 4.41 5.43 4.02 -1.74 0.45 CAM4 3.62 3.04 4.36 3.64 6.39 4.07 -0.98 0.54

22

CAM5 4.08 4.09 5.55 3.05 4.68 4.63 -1.23 0.67 NorESM 3.50 3.25 4.34 5.54 4.31 3.82 -1.11 0.40 IPSL 3.39 3.09 4.28 3.64 3.47 3.80 -0.76 0.53 MMM±1std 3.65±0.26 3.67±0.55 4.78±0.81 3.91±0.80 4.70±0.89 4.13±0.31 -1.21±0.50 0.52±0.09 Note: N/A indicates the exponential fit gives an error. 1 std indicates one standard deviation across the nine models.

Table 3: same as Table 2, but for Solar+2% experiment.

ERF (W m-2) -α Model ∆T_land (K) fsst linr fsst_∆Tland poly exp kernel (W m-2 K-1) CanESM 4.09 4.20 4.53 4.44 4.95 4.24 -1.10 0.27 GISS 4.48 4.22 5.30 3.85 4.83 4.57 -2.40 0.19 HadGEM2 4.20 3.32 4.58 4.08 3.92 4.38 -0.90 0.26 HadGEM3 4.36 3.56 4.68 4.75 4.14 4.56 -0.62 0.27 MIROC 4.15 3.78 4.85 4.43 4.84 4.31 -1.63 0.20 CAM4 4.11 2.84 4.43 3.60 5.44 4.21 -1.02 0.24 CAM5 4.24 4.01 4.88 3.39 4.85 4.41 -1.41 0.29

23

NorESM 4.01 3.60 4.69 9.40 5.47 4.20 -1.27 0.32 IPSL 4.07 3.89 4.56 4.27 4.47 4.27 -0.96 0.30 MMM±1std 4.19±0.15 3.71±0.44 4.72±0.26 4.70±1.82 4.77±0.52 4.35±0.14 -1.26±0.52 0.26±0.05

Table 4: Same as Table 2, but for SO4×5 experiment.

ERF (W m-2) -α Model ∆T_land (K) fsst linr fsst_∆Tland poly exp kernel (W m-2 K-1) CanESM -3.24 -3.39 -3.71 -2.80 -3.85 -3.50 -1.01 -0.30 GISS -2.79 -3.59 -3.46 -7.70 -10.50 -2.91 -3.41 -0.15 HadGEM2 -4.02 -3.63 -4.44 -5.17 -4.52 -4.26 -0.91 -0.28 HadGEM3 -8.26 -7.46 -8.56 -8.97 -8.20 -8.47 -0.66 -0.26 MIROC -2.77 -2.75 -3.47 -2.27 -3.24 -2.82 -2.05 -0.19 CAM4 -2.04 -2.00 -2.31 -3.05 -5.14 -2.18 -1.33 -0.20 CAM5 -2.10 -2.02 -2.07 -2.90 -2.78 -2.02 -1.43 0.02

24

NorESM -3.79 -3.40 -4.29 -4.06 -4.09 -3.90 -1.49 -0.24 IPSL -2.70 -2.54 -3.09 -2.50 -2.69 -2.87 -0.84 -0.24 MMM±1std -3.52 ±1.90 -3.42 ±1.64 -3.93±1.91 -4.38 ±2.43 -5.00 ±2.65 -3.66 ±1.95 -1.46±0.84 -0.20±0.09

Table 5: Same as Table 2, but for BC×10 experiment.

ERF (W m-2) -α Model ∆T_land (K) fsst linr fsst_∆Tland poly exp kernel (W m-2 K-1) CanESM* 1.55 1.57 1.96 1.59 2.01 1.65 -1.08 0.25 GISS* 1.23 1.06 1.57 1.07 1.91 1.32 -2.15 0.08 HadGEM2* 2.90 1.45 3.18 2.43 2.03 3.05 -0.66 0.19 HadGEM3 0.70 0.31 0.82 -0.02 0.30 0.75 -0.18 0.10 MIROC* 0.63 0.47 0.64 0.48 0.44 0.67 -1.72 0.00 CAM4 0.77 0.08 0.92 0.41 N/A 0.89 -0.14 0.11 CAM5* 0.42 0.22 0.82 0.23 0.14 0.40 -1.13 0.19

25

NorESM 1.41 1.04 1.75 2.94 2.32 1.53 -1.19 0.16 IPSL 0.82 0.69 0.93 0.80 0.76 0.87 -0.68 0.06 MMM±1std 1.35±0.98 0.95±0.59 1.63±1.02 1.16±0.89 1.31±0.93 1.42±1.04 -1.35±0.59 0.14±0.10 Note: models with │r│>0.48 (Fig. 4) are denoted with an asterisk and MMM calculations are based on these models only for all methods.

Table 6: Same as Table 2, but for CH4×3 experiment.

ERF (W m-2) -α Model ∆T_land (K) fsst linr fsst_∆Tland poly exp kernel (W m-2 K-1) CanESM 1.36 0.89 1.65 0.79 0.88 1.46 -1.12 0.18 GISS* 1.34 1.45 2.02 0.89 1.45 1.60 -1.97 0.16 HadGEM2 0.97 0.55 1.05 0.93 0.59 1.03 -0.23 0.05 HadGEM3 1.39 1.15 1.56 1.62 1.43 1.49 -0.87 0.15 MIROC* 0.78 0.85 0.97 0.82 0.99 0.84 -1.60 0.05 CAM4* 1.27 0.80 1.45 1.13 1.86 1.36 -0.73 0.13 CAM5* 0.86 0.96 1.06 0.88 0.93 0.91 -1.77 0.09

26

NorESM 1.25 0.31 1.53 1.37 0.32 1.34 0.43 0.14 IPSL 1.58 1.32 1.88 1.75 1.45 1.75 -0.63 0.18 MMM±1std 1.06±0.28 1.02±0.30 1.38±0.48 0.93±0.14 1.31±0.44 1.18±0.36 -1.52±0.55 0.11±0.05 Note: models with │r│≥0.48 (Fig. 5) are denoted with an asterisk and MMM calculations are based on these models only for all methods.

The results for the Solar+2% and SO4×5 experiments are similar to the CO2×2 experiment (see Fig. 2-3 and Table 3-4). The magnitudes of the forcings are comparable to that of the CO2×2 experiment, with the MMM values of 4.19±0.15 W m-2 and -3.52±1.90

W m-2 for the Solar+2% and SO4×5 experiments (ERF_fsst), respectively. All nine models show significant linear correlations between N and ∆T, and the MMM values of

ERF_linr and ERF_fsst are again consistent. For the Solar+2% experiment, both ERF_poly and ERF_exp MMM values are consistent with ERF_fsst_∆Tland. For SO4×5 experiment, both ERF_poly and ERF_exp methods give larger MMM values than ERF_fsst_∆Tland, but they are within the uncertainty ranges. All three regression methods show similar goodness of fit (Table 7). However, polynomial fits show the opposite curvature in some individual models again. The MMM values using the ERF_kernel method are larger than ERF_fsst, but smaller than ERF_fsst_∆Tland for both experiments. The feedback parameters remain essentially unchanged when compared with the CO2×2 experiment for all the models in both experiments.

27

Table 7: Comparison of multi-model mean (MMM) values of root-mean- squared errors (RMSE, in W m-2), which shows the goodness of the fit. For CH4×3 and BC×10 experiments, only asterisked models (Table 5 and 6) are used for MMM calculations.

Experiment linear polynomial Exponential

CO2×2 0.42±0.11 0.41±0.11 0.43±0.11

CH4×3 0.37±0.15 0.38±0.16 0.40±0.18 Solar+2% 0.38±0.11 0.36±0.10 0.38±0.11 BC×10 0.36±0.10 0.36±0.11 0.37±0.11

SO4×5 0.45±0.12 0.43±0.12 0.44±0.12

Figure 3: Same as Figure 1, but for Solar+2% experiment.

28

Figure 4: Same as Figure 1 but for SO4 x 5 experiment.

29

Figure 5: Same as Figure 1, but for BC×10 experiment.

The behavior is different, however, for the BC×10 experiment (Table 5 and Fig. 4).

Despite the large perturbation (BC×10), the forcing is small in some models, with a

MMM value of 1.35±0.98 W m-2 (ERF_fsst), roughly one-third of that for the CO2×2 experiment. The ERF_linr diverges more substantially from ERF_fsst. In fact, except for the CanESM, GISS and IPSL models, the remaining six models show at least 25% smaller values in ERF_linr than ERF_fsst. There are only five models (CanESM, GISS,

HadGEM2, MIROC and CAM5) that show fairly good correlations (│r│>0.48, this value is chosen because of the large gap between 0.49 and 0.33, see Fig. 4) between N and ∆T,

30

and the largest │r│ is only 0.68, indicating that the relationship between N and ∆T is noisy. When only these five models are considered, ERF_linr is 30% smaller than

ERF_fsst on average. ERF_fsst_∆Tland is 21% larger than ERF_fsst, and 72% larger than

ERF_linr, based on MMM values. Both ERF_poly and ERF_exp are less than

ERF_fsst_∆Tland and the ERF_kernel values are almost the same as ERF_fsst, but 13% smaller than the ERF_fsst_∆Tland values on average. For all mothods, the model-to- model variability is large and so it is difficult to evaluate the robustness of differences between methodologies. For example, ERF_linr and ERF_fsst differs by up to a factor of two in several models, but they are very similar in others. The feedback parameters remain nearly unchanged in these five models relative to the simulations discussed previously.

31

Figure 6: Same as Figure 1, but for the CH4×3 experiments.

The results for the CH4×3 experiment (Table 6 and Fig. 5) are somewhat similar to the BC×10 experiment in that the climate forcing is relatively small (1.06 W m-2 for MMM of ERF_fsst), and the linear regression technique does not work well for some models.

Only four out of the nine models show fairly strong correlations between N and ∆T

(│r│>=0.48), with the largest │r│again being 0.68. When considering these four models only, however, the ERF_fsst (1.06 W m-2) and ERF_linr (1.02 W m-2) are consistent.

ERF_fsst_∆Tland (1.38 W m-2) is 30% larger than ERF_fsst, and ERF_exp (1.31 W m-2) is consistent with ERF_fsst_∆Tland. These results are similar to those found for CO2×2.

32

The ERF_poly (0.93 W m-2), however, is 32% lower than ERF_fsst_∆Tland, and even lower than ERF_linr. The ERF_kernel again stands between ERF_fsst and

ERF_fsst_∆Tland.

2.4 Discussion and Summary

The magnitude of ERF_fsst_∆Tland, on average, is 32% (CO2×2), 30%

(CH4×3), 13% (Solar+2%), 21% (BC×10) and 11% (SO4×5) larger than ERF_fsst for the indicated experiments (Fig. 6), due to the fact that the former accounts for land adjustment (∆T_land/λ). In the greenhouse gas experiments, the adjustment to the forcing is thus roughly one-third of the original ERF_fsst. Interestingly, ∆T_land in the

CO2×2 experiment is more than double that in the Solar+2% and SO4×5 experiments, despite their similar magnitude forcings. When compared with ERF_linr,

ERF_fsst_∆Tland is also substantially larger. The MMM ratios of

ERF_fsst_∆Tland/ERF_linr are 1.30, 1.35, 1.27 and 1.15 for the CO2×2, CH4×3, Solar+2% and SO4×5 experiments, respectively. When it comes to BC, however, the ratio is 2.00±

1.04 (mean±1 std, obtained by calculating individual model ratios and then averaging them), which is much larger than the ratios in other experiments (Fig. 6, blue bar). Note that this value is larger than the 72% reported above (Table 5) due to the different order of averaging; that value was obtained by taking the ratio of MMM values.

33

Figure 7: Comparison of different ERF values for each experiment.The symbols in each bar indicate MMM values of ratios for each experiment while the error bars show one standard deviation across the models. For CH4×3 and BC×10 experiments, only the models with │r│>=0.48 are used.

The consistency between MMM ERF_fsst and ERF_linr seen in prior work also holds in our current study across all experiments, except for BC (Fig. 6, black bars).

When land adjustment is included and values are hence presumably more physically realistic, however, the values derived from fsst simulations (ERF_fsst_∆Tland) are consistently larger than ERF_linr. This can be plausibly attributed to time-varying feedback parameters leading to low biases in linear regression (Armour et al., 2013;

Andrews et al., 2015; Gregory & Andrews, 2016; Proistosescu & Huybers, 2017). To explore this further, I calculated the feedback parameters for land and ocean separately

(by regressing N against global ∆T for each grid box) for all models and all experiments.

The feedback parameters for years 1-30 are -1.14±0.12 W m-2 K-1 over land and -1.33±

34

0.12 W m-2 K-1 over oceans, while for years 31-90 they are -0.49±0.10 W m-2 K-1 over land and -0.68±0.17 W m-2 K-1 over oceans. The feedback parameters thus become significantly less negative for both regions during the later years, but are similar in the two areas, with slightly larger values for the ocean. Thus, it is not clear that land feedbacks differ greatly from ocean feedbacks, although there is a very large spread across models, but there is strong evidence that early feedbacks are larger than later ones. The fsst simulations were short, and hence include only the early, stronger feedbacks. The two curved fits attempt to account for the concavity in feedbacks in the longer coupled runs. Based on MMM, the ERF values derived with a polynomial fit are consistent with the ERF_fsst_∆Tland in the Solar+2% and SO4×5 experiments, but not in the other experiments (Fig. 6, green bars) while the values using an exponential fit largely reconcile the regression results with the ERF_fsst_∆Tland values within the uncertainty range, with especially close correspondence for CO2, CH4, and solar experiments (Fig. 6, red bars). For BC, the MMM ratio of ERF_fsst_∆Tland/ERF_exp remains high at 2.1 (Fig. 6, red bar), though there is a large inter-model spread (mainly due to the CAM5 model). The reason for this discrepancy is not yet clear, though I reiterate that the regression-based results are highly uncertain in many models.

35

Table 8 Feedbacks parameters for land/ocean in the different time periods for the CO2×2 experiment.Note that the values in the text are MMM for all five experiments.

yr 1-30 yr 31-90 yr 1-90 Models land ocean land ocean land ocean CanESM -1.49 -1.24 -1.21 -0.97 -1.30 -1.07 GISS -2.62 -2.19 -1.42 -0.38 -2.40 -1.59 HadGEM2 -0.64 -0.86 -1.11 0.37 -0.81 -0.65 HadGEM3 -0.89 -0.56 -0.38 -0.21 -0.62 -0.26 MIROC -1.20 -1.87 -1.14 -1.78 -1.22 -1.95 CAM4 -0.92 -0.88 0.27 0.25 -0.90 -0.82 CAM5 -0.57 -1.35 -0.38 -2.48 -0.89 -1.43 NorESM -1.35 -0.94 0.18 -0.56 -0.85 -1.06 IPSL -1.50 -0.45 -0.89 -0.31 -1.26 -0.52 MMM±1std -1.24±0.62 -1.15±0.58 -0.68±0.62 -0.67±0.93 -1.14±0.53 -1.04±0.54

To further investigate uncertainties and the sensitivity to the time period used for the regression results, I compared the ERF values for three regressions using results from years 1-30 and years 1-100 of the coupled simulations (Table 9). In other studies, linear regressions have generally been applied to the first 20-30 years of data. Results based on shorter time spans (e.g. 10 years) would be subject to strong internal variability

(Hansen et al., 2005). For ERF_poly and ERF_exp, there are no significant changes when more years are included except for the CO2×2 experiment, which shows a slight increase in both values but they remain within the uncertainty ranges. For ERF_linr, however, there are systematic decreases (6-10%) across all five experiments when a longer time span is used due to time-varying feedback parameters. This analysis indicates that both

36

ERF_poly and ERF_exp are less sensitive to the chosen time period. Moreover, all three regressions show similar goodness of fit.

For comparison, I also calculated the uncertainty ranges for the non-regression

-2 -2 ERF values. For ERF_fsst, the MMM values are ±0.16 W m (CO2×2), ±0.17 W m

-2 -2 -2 (CH4×3 ), ±0.18 W m (Solar+2%), ±0.16 W m (BC×10), and ±0.19 W m (SO4×5), based on the inter-annual variability of radiative fluxes. These uncertainty ranges are generally 0.12 W m-2 larger in magnitude for the ERF_fsst_∆Tland due to the additional variability in climate sensitivity and land response (∆Tland). For ERF_kernel, the uncertainty in the additional kernel-derived adjustments was estimated by analyzing the spread in rapid adjustments calculated using 6 kernels from Smith et al. (2018) for each model. The spread in the rapid adjustments are generally 0.02-0.05 W m-2 across most of the models and experiments. When combined with the uncertainties of ERF_fsst, they produce MMM uncertainties [estimated by: δERF_kernel = (δERF_fsst2 + δrapid

2 1/2 -2 adjustments ) , δ is the uncertainty for each component] of ±0.17 W m (CO2×2), ±

-2 -2 -2 0.17 W m (CH4×3 ), ±0.19 W m (Solar+2%), ±0.18 W m (BC×10), and ±0.19 W

-2 m (SO4×5), similar to those of ERF_fsst and smaller than the regression-based analyses

(Table 8).

37

Table 9 Comparison of MMM (mean±1 std) values of ERF derived with different time periods.

ERF_linr ERF_poly ERF_exp experiment year ERF RMSE ERF RMSE ERF RMSE 1-30 3.67±0.55 0.42±0.11 3.91±0.80 0.41±0.11 4.70±0.89 0.43±0.11 CO2×2 1-100 3.45±0.57 0.39±0.09 4.12±0.66 0.39±0.09 5.30±1.60 0.40±0.10 1-30 1.02±0.30 0.37±0.15 0.93±0.14 0.38±0.16 1.31±0.44 0.40±0.18 CH4×3 1-100 0.83±0.18 0.37±0.12 0.98±0.23 0.36±0.12 1.37±0.79 0.38±0.14 1-30 3.71±0.44 0.38±0.11 4.69±1.82 0.36±0.10 4.77±0.52 0.38±0.11 Solar+2% 1-100 3.45±0.51 0.38±0.08 4.41±1.02 0.38±0.08 5.06±0.88 0.39±0.09 1-30 0.95±0.59 0.36±0.10 1.16±0.89 0.36±0.11 1.31±0.93 0.37±0.11 BC×10 1-100 0.83±0.52 0.38±0.11 1.06±0.73 0.38±0.11 1.26±0.98 0.40±0.12

38 1-30 -3.42±1.64 0.45±0.12 -4.38±2.43 0.43±0.12 -5.00±2.65 0.44±0.12 SO4×5 1-100 -3.06±1.87 0.43±0.10 -4.24±1.91 0.42±0.11 -4.91±2.15 0.42±0.11

Our analyses show that the ERF values may differ by ~10% - 50% with different methods, based on MMM results, and this difference my reach 100% for BC. It is difficult to determine, however, which method for calculating ERF is best. Researchers need to choose the appropriate approach, depending on available model output, computation resources and research needs. With fsst simulations, it is easy and fast to obtain either

ERF_fsst or ERF_fsst_∆Tland values. However, for researchers who only have coupled simulations, regression methods are the only option to obtain the ERF. Linear regression is simple and most widely used (Myhre et al., 2013b), and is also generally consistent with ERF_fsst values. The latter, however, neglects a known process (land temperature adjustment), and so arguably ERF_fsst_∆Tland is a more physically realistic method.

Based on MMM values, an exponential fit is shown to be more consistent with

ERF_fsst_∆Tland values, though exceptions exist in individual models. In addition, it is less sensitive to the chosen time period than linear regression. The polynomial fit is sensitive to natural variability and may give opposite curvature to that inferred from a direct analysis of the feedback’s temporal evolution (e.g., GISS model in Fig. 1). I also investigated higher order polynomial fits, but the curvatures are wrong (e.g., do not fit the evolution of N and ∆T). The radiative kernel method provides another option to obtain ERF values accounting for land-related responses, and leads to only slight increases (4-13%) compared with ERF_fsst (Fig. 6, purple bars). The disadvantage of this

39

method is that it requires more output from the simulations and more computation resources compared with other approaches.

The poor correlations between N and ∆T for the BC×10 experiment suggest that the regression technique may not work well in the BC×10 experiment, at least for some models and for the time scales considered in this study. This may be partly attributable to internal variability. The forcing in the BC×10 experiment is relatively small compared with the CO2×2 experiment, ranging from 0.8-2.0 W m-2 for most models. However, the inter-annual variability of N in the BC×10 experiment ranges from 0.3 to 0.5 W m-2, which is a substantial fraction of the forcing. Such variability blurs the relation of N and

∆T. Further analyses indicate that this variability is mainly associated with SW cloud effects (not shown), suggesting the pivotal role of low-level clouds on the unforced fluctuations of radiation budgets. Due to the relatively larger forcing in CO2×2, Solar+2% and SO4×5 experiments (~ 4 W m-2), the SW cloud effects did not significantly affect the fits in those cases, whereas the regression techniques do not work well for some models in the CH4×3 and BC×10 experiments with small forcing. Therefore, the regression techniques (linear, polynomial and exponential) appear useful in those large-forcing experiments, but should be used with caution in small-forcing simulations.

In this study, I compared six methods for estimating ERF values from nine models participating in the PDRMIP project. The consistency between the values of

ERF_fsst and ERF_linr in prior studies holds for most climate forcings, except for BC.

40

When land response is accounted for in fsst simulations, the ERF_fsst_∆Tland is roughly

10-30% larger than ERF_linr, however, and 70-100% larger for the BC×10 experiments.

Such adjustments can also be accounted for by using radiative kernels, which typically leads to values in between the ERF_fsst and ERF_fsst_∆Tland results. There is a tendency for the values derived from linear regression to be lower than ERF_fsst_∆Tland values, which appears to be explained by the time-varying feedbacks. Such differences can be largely eliminated by using an exponential regression, making them consistent with the fsst values with land adjustment included under most climate forcings based on

MMM results. BC forcing is quite sensitive to the method used, the reasons for which merit further study.

41

3. Dynamical Response of Mediterranean Precipitation to Greenhouse Gases and Aerosols

[Tang, T., Shindell, D., Samset, B. H., Boucher, O., Forster, P. M., Hodnebrog, Ø .,

... & Faluvegi, G. (2018). Dynamical response of Mediterranean precipitation to greenhouse gases and aerosols. Atmospheric Chemistry and Physics, 18(11), 8439-8452, doi:

10.5194/acp-18-8439-2018]

3.1 Introduction

Aerosols, fine particles in the atmosphere produced by both natural processes and anthropogenic activities, impact the Earth’s climate by scattering and absorbing solar radiation (direct effect), or by modifying the properties of clouds (indirect effects) through a variety of mechanisms including atmospheric heating and changes in ice nuclei and cloud condensation nuclei (CCN), including their size, location and concentration. These changes may significantly affect solar radiation and precipitation

(Ramanathan et al., 2001; Kaufman et al., 2002; Shindell et al., 2012; Bond et al., 2013;

Boucher et al., 2013). The effects of aerosols on climate have been widely studied both on global and regional scales (Ramanathan & Carmichael, 2008; Shindell & Faluvegi, 2009).

For example, Menon et al. (2002) reported slight cooling and drying trends in the northern part of China in the 2nd half of the 20th century and attributed such trends to the emissions of BC aerosols based on model simulations. Similarly, Hodnebrog et al.

42

(2016) reported a precipitation decrease in southern Africa due to local biomass burning aerosols based on analyses of model simulations and local energy budget. On the other hand, Koren et al. (2012) argued that aerosols could intensify rainfall events in the lower and mid-latitudes by analyzing satellite observations. However, Stevens and Feingold

(2009) contended that the effects of aerosols on clouds and precipitation are very limited due to the buffering effects of the climate system itself. In addition to their influence on temperature and precipitation, aerosols may also affect large-scale atmospheric circulation. For example, Takahashi and Watanabe (2016) suggested that the Pacific trade winds were accelerated partially by sulfate aerosols during the past two decades.

Jacobson and Kaufman (2006) suggested a surface wind reduction due to aerosol particles in California and China, which may also impact air pollution and wind energy.

Dunstone et al. (2013) also reported that aerosols could modulate Atlantic tropical storm frequency due to aerosol-induced shifts in the Hadley circulation. These differing results suggest that aerosol effects on regional climate may depend on the aerosol types, seasons, and regions of interest.

A decreasing precipitation trend in the Mediterranean area since the 20th century has been reported and its possible causes have been investigated in many studies

(Piervitali et al., 1998; Buffoni et al., 1999; Mariotti et al., 2002; Dünkeloh & Jacobeit, 2003;

Xoplaki et al., 2004). For instance, Quadrelli et al. (2001) observed a strong correlation between winter Mediterranean precipitation and the NAO (Hurrell et al., 2001). Krichak

43

and Alpert (2005) suggested that the East Atlantic-West Russia (EA-WR) pattern may also play an important role in modulating the precipitation in the Mediterranean. Hence, the responses of Mediterranean precipitation to these large-scale variability patterns

(e.g., NAO, EA-WR), and to some extent how these patterns might be responding to external drivers, are fairly well-understood (Black et al., 2010). However, prior studies included all the drivers at once, so cannot discern the relative roles of WMGHGs and other agents. Anthropogenic aerosols have been reported to greatly influence the temperature in the Mediterranean (Nabat et al., 2014), but the effects of aerosols on

Mediterranean precipitation have not been carefully examined. Since precipitation impacts water availability for both ecosystems and human societies, it is crucial to understand the different impacts of the climate drivers that are responsible for the

Mediterranean precipitation trend. To bridge this knowledge gap, here I analyze

Mediterranean precipitation changes based on a group of state-of-the-art GCMs that examined the precipitation response to individual climate drivers, which could help inform management of water resources, regional societal activities such as agriculture, and even emissions mitigation.

3.2 Data and Method

3.2.1 Data

This study employed the same PDRMIP models as that in the last chapter. For detailed model information, please see Table 1. The aerosol treatments of the PDRMIP

44

models are given in the Table 10. In the PDRMIP experiments, all models include dust among the aerosols, but again as the PDRMIP protocol was designed to look at the dynamic climate response to prescribed aerosol and GHG changes, dust was held fixed in the concentration-driven simulations. Further work could usefully explore if changes in dust loading might have contributed to Mediterranean precipitation changes.

45

Table 10 Aerosol treatments for the PDRMIP models.

Aerosol mode/size Cloud drop Model Mixing state Size distribution Evolve bin activation CanESM S, N, BC, dust, SS, OC Internal Log-normal N/A N/A S (1), N (1), OC (1), GISS-E2R Internal & external Log-normal N/A Empirically BC (1), SS (2), dust (4) S (3), BC, OC, BB (3), HadGEM2 External Log-normal Physically Empirically SS (2), dust (6) S (3), BC, OC, BB (3), HadGEM3 External Log-normal N/A Empirically SS (2), dust (6) MIROC- S (1), BC (1), OC (1),

46 Internal & external Log-normal N/A Empirically SPRINTARS dust (6), SS (4)

S, SS (4), dust (4), BC CESM-CAM4 External Log-normal N/A N/A (2), POM (2), SOA S, POM, SOA, SS, BC, CESM-CAM5 Internal Log-normal Physically Physically dust (3) S, OM, BC, SS, dust; NorESM Internal & external Log-normal Physically Physically 13 modes, 33 size bins IPSL-CM5A S, BC, OC, dust, SS External Log-normal N/A Empirically

3.2.2 Method

In addition to direct analysis of meteorological fields (e.g. precipitation, sea-level pressure) in the models, I also analyzed the energy budget associated with the hydrological cycle. Following Hodnebrog et al. (2016) and Muller and O’Gorman (2011), the precipitation change is related to diabatic cooling and the horizontal transport of dry static energy as follows:

Lc ∆P = ∆Q + ∆H (7)

Here Lc is the latent heat of condensation of water vapor, which is 29 W m-2 mm-1 day. ∆P is the precipitation change. ∆H is the column-integrated dry static energy flux divergence and ∆Q is the column-integrated diabatic cooling, which is calculated as:

∆Q = ∆LW + ∆SW - ∆SH (8)

where ∆LW is the change of longwave radiation in the atmospheric column and

∆SW is the change of shortwave radiation in the atmospheric column. ∆SH is the change of upward sensible heat flux.

Since most of the precipitation events occur in the wet season (Oct-Mar) in the

Mediterranean, roughly 70% of total annual precipitation (Mariotti et al., 2002; Kottek et

47

al., 2006), the analysis was restricted to the wet season in the current study unless noted otherwise. All of the data used in this study were re-gridded into 2.5°× 2.5° horizontal resolution for analyses.

3.3 Results

Fig. 7 shows the multi-model mean (MMM) of normalized ∆P for each forcing.

Both CO2 and BC caused a substantial drying over Mediterranean, with a larger magnitude from BC (Fig. 7a-b) whereas SO4 contributed very little in the Mediterranean region (Fig. 7c). Moreover, in stark contrast to the drying of the Mediterranean,

Northern Europe shows increasing precipitation trends for CO2 and BC (Fig. 7a-b), which will be discussed in more details later. To compare the precipitation response quantitatively, the domain-averaged (purple rectangle in Fig. 7, 30°N-45°N, 10°W-40

°E) trends are shown in Fig. 8. For CO2, all of the nine models show drying trends (Fig.

8a). The MMM is -0.03±0.03 mm/day per W/m2, with individual model values ranging from -0.01 to -0.06 mm/day per W/m2. For BC (Fig. 8b), all nine models again show drying trends, with the MMM value -0.12±0.07 mm/day per W/m2, which is four times as large as that of CO2. When it comes to SO4 (Fig. 8c), the model results even differ in the sign of change and the MMM value is small (-0.01±0.04 mm/day per W/m2). These analyses show that the precipitation response is more sensitive to BC forcing than to CO2 and SO4 in this region.

48

Figure 8: Normalized ∆P (change per unit forcing)for (a) CO2, (b) BC and (c) SO4. Black dots indicate the change is significant at 0.95 confidence level. Please note that the sign for SO4 is flipped due to its negative forcing. Thus, the results shown for SO4 is the precipitation change per unit negative forcing.

Figure 9: Domain-averaged ∆P (purple rectangles in Fig. 7) for (a) CO2, (b) BC and (c) SO4. Error bars of multi-model mean (MMM) are 90% inter-model spread.

49

In order to investigate the mechanisms governing the precipitation response, I performed an energy budget analysis for this region (Fig. 9). For CO2, the drying is dominated by horizontal energy transport (gray box in the CO2 panel), albeit some offset by diabatic cooling (pink box in the CO2 panel). For BC, the net radiation change, which is primarily SW (red box in the BC panel), has a larger impact than the horizontal energy change (gray box in the BC panel), but the latter is nonetheless a substantial fraction of the net change. When it comes to SO4, the small precipitation response results from the offsetting of net radiation change (pink box in the SO4 panel) and horizontal energy transport (gray box in the SO4 panel). The energy budget analysis implies that the dynamical responses to CO2 and BC played a crucial role in modulating the precipitation in this region.

Figure 10: Domain-averaged energy budget change for each forcing and energy component as shown in Equation (1) and (2). It holds that Lc ∆P (blue) = ∆Q (pink) + ∆H (gray), where Lc ∆P is the change in total latent heating, ∆Q (pink) = ∆LW (green) + ∆SW (red) - ∆SH (brown) is the change in diabatic cooling of the atmospheric column due to shortwave and longwave radiation, and sensible heat flux, ∆H is the change in column-integrated dry static energy flux divergence. The error bars indicated 90% inter-model spread.

50

I then analyzed the ∆SLP from the model output (Fig. 10). Specifically, it is seen that CO2 induced strong SLP changes. The SLP increased at mid-latitudes, with increases extending from the North Atlantic to Southern Europe and decreased at high latitudes

(Fig. 10a). BC led to a similar pattern of SLP change, but with increased magnitude (Fig.

10b), characterized by two increases centered in Europe and the Western North Atlantic.

Compared with CO2 and BC, SO4 caused an opposite change (Fig. 10c). The CO2 and BC forcings appear to induce a pattern similar to the positive phase of the NAO /AO

(Lorenz, 1951), in which the jet streams and storm tracks are displaced northward, leading to a drier Mediterranean and precipitation increases in Northern Europe (Fig.

7a-b). Such a shift in response to forcings is more clearly seen in the changes of zonal winds (Fig. 11). The CO2 caused a strengthening of zonal winds in the whole upper atmosphere and a strengthening around 60°N from the near-surface to the top of the atmosphere, as well as weakening around 30°N from the near-surface to the mid- troposphere (Fig. 11a), as in prior studies (Shindell et al., 2001). The strengthening around 60°N is more apparent for BC (Fig. 11b). Similar results were seen in response to aerosol forcing in a prior study (Allen & Sherwood, 2011). This shift is possibly due to the enhancement of the tropospheric temperature gradient between mid-latitudes and high-latitudes, as suggested by (Allen et al., 2012).

51

Figure 11: Same as Figure 7, but for sea level pressure (SLP).

Figure 12: Same as Figure 7, but for zonal wind.The green contours represent the 50-yr climatology of the zonal wind in the control simulations. The contours are at the interval of 5m/s, with positive values indicating eastward winds.

Our analyses illustrate that BC aerosols may modulate regional precipitation in part via modifying large-scale circulation patterns. Many previous studies suggest that

BC could impact regional precipitation by changing the local vertical temperature profile, in which BC aerosols absorb solar radiation and heat the atmosphere, thus suppressing convection and cloud formation (Kaufman et al., 2002; Meehl et al., 2008;

Ramanathan & Carmichael, 2008; Hodnebrog et al., 2016). Our results (analyses of the

52

energy budget, SLP and zonal winds) suggest that a portion of the drying is also associated with large-scale circulation responses. In addition, our pattern of jet stream/storm track changes (Fig. 10 & 11) is also in agreement with the projections from the latest IPCC report (Collins et al., 2013) based on a set of CMIP5 models, with increasing storm activities in Northern Europe and decreasing storms in the

Mediterranean. Such a shift of storm tracks may further reduce the precipitation in the

Mediterranean, though reductions in WMGHG or BC emissions may help to mitigate the projected drying.

3.4 Case Study -Historical observations and Scaled Model

Results

The above analyses demonstrated how the precipitation and circulation responded to each forcing both qualitatively and quantitatively. In order to explore their potential relative contributions to the total precipitation change, here I apply linear scaling to the model output. Since PDRMIP utilized large aerosol and greenhouse gas changes in order to achieve strong signals that could be statistically significant with a relatively modest amount of computational time, the precipitation change from those model outputs needs to be scaled in order to compare with observations. Uncertainties related to this approach are discussed further in section 3.5.

In this study, I focus on the period from 1901 to 2010. The scaled precipitation change for each individual forcing is defined as:

53

∆Pscaled = ∆P × (ERF1901-2010 / ERFmodel) (9)

In equation (9), ∆P is the precipitation change over Mediterranean in the model during the last 50 years in the coupled run, since the model has reached near- equilibrium state after 30 years. ERF1901-2010 is the historical global ERF for the period of

1901-2010. The values were obtained from the latest Intergovernmental Panel on Climate

Change (IPCC) assessment report (Myhre et al., 2013b). The ERF1901-2010 value used for

CO2 is 2.33 W/m2, which is larger than the CO2 value from the IPCC report as CO2 was used to represent all WMGHGs in this case study. ERF1901-2010 values for BC and SO4 are

0.28 W/m2 and -0.33 W/m2, respectively. ERFmodel is the global ERF in the PDRMIP models, which was obtained by calculating the energy flux change at the top of the atmosphere from years 6 to 15 of the fixed-SST simulations, since present models largely equilibrate within 5 years of fixed-SST running (Kvalevåg et al., 2013). In addition to the direct effects of the aerosols, the indirect effects of aerosols were also included in most of the models and thus, in the ERFmodel. The value of (ERF1901-2010 / ERFmodel) is the scaling factor applied to model precipitation output to match historical forcing levels. They are

0.64 [0.57, 0.69], 0.33 [0.10, 0.68] and 0.11 [0.04, 0.16] for CO2, BC and SO4, respectively

(where the values indicate the mean [min, max] across the nine models). An important assumption here is that the ∆P changes linearly with ERFmodel.

54

∆Pscaled is calculated for CO2, BC, and SO4 separately. The total ∆Pscaled is the combination of the three, assuming their responses to those forcings can be added linearly. It should be noted that in this analysis, I use the combined responses to

WMGHGs, BC and SO4 to approximate the total historical response over 1901-2010.

Several additional factors may have also played a role, including natural forcing (solar and volcanic activities), land use/land cover change, contrails, ozone (O3) (both tropospheric and stratospheric) and stratospheric water vapor, which have forcings of -

0.03, -0.09, 0.05, 0.26 and 0.06 W/m2, respectively (Myhre et al., 2013b). As all these forcings are fairly small, simulations to isolate their impacts would be extremely computationally expensive and hence were not performed but to first order I expect their exclusion is unlikely to greatly affect our results. Characterization of the influence of these other drivers merits future study, particularly as some operate via different physical processes (e.g. tropospheric ozone is both a greenhouse gas and an absorber of incoming solar radiation). Similar analyses were also performed to obtain scaled SLP change (∆SLPscaled), zonal wind change and energy budget change in the atmospheric column.

Several observational and reanalysis datasets were also employed in this part of our study. For precipitation, Global Precipitation Climatology Center (GPCC) monthly precipitation data (Schneider et al., 2016), provided by NOAA/OAR/ESRL from their website (https://rda.ucar.edu/datasets/ds496.0/), is employed. It is a high-quality gridded

55

dataset that is mainly terrestrial station-derived. For SLP, I use HadSLP2 data (Allan &

Ansell, 2006), which is created by combining marine observations from ICOADS data

(Worley et al., 2005) and land (terrestrial and island) observations (available at https://www.esrl.noaa.gov/psd/data/gridded/data.hadslp2.html). I also use

NCEP/NCAR reanalysis data (Kalnay et al., 1996), downloaded from https://www.esrl.noaa.gov/psd/data/gridded/data.ncep.reanalysis.derived.surface.html, for the comparisons of zonal wind. All these datasets have undergone rigorous quality control and have been widely used in the climate community, including the IPCC 2013 assessment report (Hartmann et al., 2013). The trends of the observed and reanalysis data were estimated by a simple linear regression applied to the same period of the datasets.

The combination of WMGHGs, BC and SO4 exerted a strong drying trend in the

Mediterranean (Fig. 12a). The drying trends shown here are statistically significant and consistent with the observations (Fig. 12b), as well as previous studies (Buffoni et al.,

1999; Mariotti et al., 2002; Dünkeloh & Jacobeit, 2003). When averaged over the whole domain, the scaled drying trends caused by WMGHGs, BC and SO4 are -1.28±1.21 mm/decade, -0.58±0.34 mm/decade and -0.03±0.21 mm/decade, respectively (not shown here). When combined (Fig. 12c), all nine models show decreased precipitation, with MMM value of -1.89±1.39 mm/decade, which is roughly a 5% decrease relative to the climatology of the control simulations. Such a decreasing trend is indistinguishable

56

from the observations (-2.78±1.10 mm/decade, a 10% decrease compared with its 110-yr climatology). In spite of the dominant role of WMGHGs in the drying of the

Mediterranean, BC contributed roughly one-third (31±17%) of the total forced precipitation decrease in this region whereas the contribution of the scattering aerosol-

SO4 is negligible (~1.6%). I also examined the trend of precipitation in the control simulations and found only very weak responses (Fig. 12c), with a mean value of 0.004

±0.03 mm/decade and maximum value of 0.03 mm/decade in any individual model.

Since current GCMs are able to capture the broad spatial and temporal features of internal variability (Flato et al., 2013), and the forced drying signal is almost equal to the total signal (Fig. 12a-c), the consistent drying trend in the models is very unlikely to be attributable to unforced variability and appears realistic. The energy budget change (Fig.

12d) clearly shows that the net precipitation decrease is mainly due to horizontal energy transport (gray box) rather than diabatic cooling (pink box), because the absorption of

SW radiation (red box) and LW radiative cooling (green box) offset one another in total.

57

Figure 13: Scaled change for the combination of WMGHGs, BC and SO4for (a) spatial pattern of precipitation, (c) domain-averaged precipitation change, and (d) energy budget change, along with 1901-2010 (b) GPCC observational data (for which gray indicate missing or incomplete data). The dots in (a) and (b) indicate changes are significant at 0.95 and 0.9 confidence level, respectively. Error bars in (c) and (d) indicate 90% uncertainty ranges.

Fig. 13a shows the overall response of SLP to these forcings, with strong SLP increases at mid-latitudes and strong decreases at higher latitudes. Such patterns of SLP changes are also found in the observed datasets (Fig. 13b), albeit with a larger magnitude. The combined pattern of zonal wind responses shows winds intensified at the northern edge of the jet stream and weakened at the southern edge (Fig. 13c). The

NCEP dataset depicts a similar pattern of changes, with winds intensifying at 60°N and

58

weakening at 30°N, but as with SLP, with a stronger magnitude (Fig. 13d). Some previous studies have pointed out that current GCMs significantly underestimate the tropical expansion and jet stream shift, which could be related to the short observational record, large internal variability or model deficiencies (Johanson & Fu, 2009; Allen et al.,

2012). Despite the underestimations, our analyses clearly demonstrate the shift of the jet stream in response to these forcings that appears qualitatively consistent with observations.

Figure 14: SLP (a & b) and zonal wind change (c & d).(a) and (c) are scaled change for the combination of WMGHGs, BC and SO4, while (b) and (d) are Hadley observational data and NCEP reanalysis data, respectively. Dots indicate the changes are significant at 0.95 confidence level. The green contours in (c) and (d) represent the climatology position of the zonal wind. The contours are at the interval of 5m/s, with positive values indicating eastward winds. 59

Based on the model simulations in the current study, the pattern of climate response to BC forcing over the past ~110 years is similar to the response to WMGHGs over Europe and the North Atlantic, including precipitation, SLP and zonal winds. At the same time, our results suggest that SO4 played a very limited role in modulating

Mediterranean precipitation trends and North Atlantic storm tracks. In other words, the precipitation trends during the past 110 years in the Mediterranean are likely to be only weakly sensitive to scattering aerosols that were not modeled (e.g., organic carbon, nitrate) or the uncertainties in aerosol negative forcing (probably not even for indirect forcing, as they were included in sulfate simulations for most models). The small sensitivity of SO4 is likely due to compensation between local and remote effects (Liu et al., 2018). Combined with its small ERF, the role of SO4 appears to be negligible during this period. However, the simulations examined here were not designed to determine whether the aerosol effects are due to local or remote emissions from the models. Initial analysis from PDRMIP regional experiments (in which BC over Asia only is multiplied by 10, with everything else being held at present-day levels) indicate that BC from Asia contributes as much as 60% to the drying signal in the Mediterranean, and in fact a larger average rainfall change in the Mediterranean than averaged over Asia itself. This suggests that the remote effects of BC may have dominated the Mediterranean precipitation changes. Hence the response to global BC increases may be a reasonable proxy for the 20th century changes, although it would be useful to explore the effects of

60

local reductions from Europe itself in the late 20th century. The relative impacts of local versus remote forcing will be further explored in forthcoming PDRMIP analyses.

3.5 Discussions and Conclusion

Since PDRMIP experiments are equilibrium simulations while the real-world is transient, and I scaled PDRMIP forcing to match historical levels, I examined related experiments to test both these aspects of the methodology used in our comparison with historical observations. Historical GHG-only simulations using the same CMIP5 models

(Taylor et al., 2012) that participated in the PDRMIP project were collected and analyzed

(data available at http://strega.ldeo.columbia.edu:81/CMIP5/.monthly/.byModel/). Six models are available and each model has 1-5 ensemble members. All of the six models show drying trends (Fig. 14), with a MMM value of -1.32±1.65 mm/decade (-1.29 mm/decade when weighted by ensemble size) which is quite close to the WMGHGs result of our scaled equilibrium PDRMIP output (-1.28±1.21 mm/decade). In fact, the overlap of their probability density functions is 0.85, assuming a normal distribution.

This comparison indicates that our methodology does not appear to be a large source of uncertainty in the current study, though response to other agents may not be as linear as those to WMGHGs (unfortunately, simulations are not currently available to evaluate other forcers, and, given the enormous expense of running enough ensemble members to isolate the relatively small signals for individual aerosols, are unlike to be anytime soon). Similar analyses were also performed for SLP and zonal winds, and again there is

61

no appreciable difference between the historical transients and the scaled equilibrium responses. The consistent results suggest that the methodology works surprisingly well.

Figure 15: Precipitation trends of CMIP5 HistoricalGHG simulations during 1901-2005, (a) domain-averaged (purple rectangles in Figure 7) trends for each model. Error bars of each model indicate 90% inter-ensemble spread and error bars of MMM indicate the 90% inter-model spread. The numbers in parentheses indicate the ensembles collected for each model. (b) spatial pattern of MMM value for the trends. Dots indicate that the change is significant at 0.95 confidence level.

In addition to the wet season, precipitation during the dry season (Apr-Sep) for the PDRMIP model was also analyzed. The modelled dry season precipitation trends, however, do not match the observations well (not shown). The modelled results also show a statistically strong drying trend while the observations do not show significant 62

changes. Two possible reasons may be responsible for the apparent discrepancies. One is that only 30% of the total precipitation occurs during the dry season (boreal summer months) and it is difficult to simulate the uneven distribution of infrequent rainfall events. The other is that there are large uncertainties in the observational data itself.

Unlike the wet season, in which nearly half of the grid boxes show statistically significant trends (Fig. 12b), almost none of the grid boxes show statistically significant trends in the dry season, undermining the robustness of the observational results.

The drying influence of WMGHGs will be more prominent in the future due to their projected continued growth. In contrast, many studies suggest that aerosol concentrations may decrease rapidly in the future due to air quality and climate policies along with their relatively short lifetime compared with WMGHGs (Andreae et al., 2005;

Myhre et al., 2013b; Shindell et al., 2013). Reductions of BC could, to some extent, slow down the drying trend in the Mediterranean. Overall, a drier Mediterranean region is expected owing to increasing WMGHGs, but the pace of change in global BC emissions may substantially modify the drying rate in the near term.

Some limitations and uncertainties still exist in our current study. First, it is important to keep in mind that the case study in Section 3.4 is not a formal attribution analysis, despite the estimation of BC contribution. Our aim is to give a first grasp of the effects of aerosol on regional precipitation in the Mediterranean. Second, although our comparison of scaled equilibrium and unscaled transient simulations indicates that our

63

methodology works well at least for WMGHGs, there is no systematic study so far exploring the linearity (or non-linearity) of the precipitation responses to BC or the linearity of responses to multiple versus individual forcings on regional scales. Third is that the ERF1901-2010 of BC represents direct effects only (Myhre et al., 2013b). Semidirect and indirect effects, however, are included in many of our PDRMIP models, and thus in

ERFmodel. I did not include these effects in the scaling in this study for two reasons: first, the indirect effects of BC in the PDRMIP models do not include ice particles, as well as internal cloud absorption (Jacobson, 2012) and are difficult to evaluate as BC concentrations were prescribed in several of the models so that they cannot interact fully with clouds, indicating that they are not fully resolved, and second, the net ERF1901-2010 of semidirect plus indirect effects is likely small (-0.1 to +0.2 W/m2) with a very large overall uncertainty range (-0.4 to +0.9 W/m2) (Bond et al., 2013). If the semi-direct and indirect effects of BC (-0.1 to +0.2 W/m2) are considered in the scaling, the ∆Pscaled of BC aerosol would be -0.44 to -0.87 mm/decade and still contribute a substantial part (25 to

40 %) to the drying. The situation is similar for sulfate aerosol, for which indirect effects are included in ERFmodel, but not in ERF1901-2010. I did not include indirect effects in our scaling as these were not attributed to individual aerosol species in the IPCC AR5

(Boucher et al., 2013). If the indirect effects are considered, the negative ERF1901-2010 could increase roughly by a factor of two (assuming indirect effects are largely associated with sulfate). However, the ∆Pscaled of sulfate aerosol would still be very small compared with

64

WMGHGs or BC, which would not impact our conclusions. As noted previously, the use of prescribed concentrations will also limit the ability of models to capture aerosol-cloud interactions realistically, affecting precipitation responses as well as ERF estimates. Since the responses do not obviously vary systematically between concentration-driven and emissions-driven models, such effects may be relatively small but merit future study.

The final issue is related to the design of the model simulations. The perturbations are 5

× or 10× present-day aerosol concentrations, which are time-invariant. The aerosols, however, have significant spatial and temporal variations. For instance, aerosol concentrations have been increasing in Asia continuously since 1950, but decreasing in

Europe since the 1970s (Allen et al., 2013). As noted, further work is needed to determine how much of the Mediterranean trends result from local relative to remote forcing. To the extent that the trends are driven by remote forcing the potential influence of such spatio-temporal variations will be small. This will be explored in future PDRMIP simulations.

Our analyses show that both WMGHGs and BC influence wet season

Mediterranean rainfall by causing an enhanced positive NAO/AO-like SLP pattern as well as by some local heating due to SW absorption. The SLP pattern is characterized by higher SLP in the North Atlantic and Mediterranean and lower SLP in the Northern part of Europe, which diverts the jet stream and storm tracks further northward, reducing the precipitation in the Mediterranean and increasing precipitation in Northern Europe. In

65

contrast, global perturbations of the scattering aerosol SO4 have a negligible impact. The results from this study may have important implications to the management of regional water resources, agricultural practice, ecosystems, environment, and economics as well as social development and behavior in a warming climate. They also stress the importance of accounting for the aerosols (and generally short-lived forcers) for short- term (e.g., decadal) regional climate prediction.

66

4. Response of Shortwave Cloud Radiative Effect to Greenhouse Gases and Aerosols and its Impact on Daily Maximum Temperature

4.1 Introduction

Clouds have a pivotal role in influencing the Earth’s energy budget (Ramanathan et al., 1989). By enhancing the planetary albedo, clouds exert a global mean shortwave cloud radiative effect (SWCRE) of about -50 W m-2 at the top-of-the-atmosphere, and by contributing to the greenhouse effect, exert a mean longwave effect (LWCRE) of approximately +30 W m-2 (Boucher et al., 2013). On the whole, clouds cause a net cooling of 20 W m-2 relative to a cloud-free Earth, which is approximately five times as large as the radiative forcing from a doubling of CO2 concentration. Therefore, a subtle change in cloud properties has a potential to cause significant impact on climate (Boucher et al.,

2013; Zelinka et al., 2017). Recent studies contended that the cloud feedback, especially the SW cloud feedback is very likely to be positive (Clement et al., 2009; Dessler, 2010;

Zelinka et al., 2017). As the SW cloud feedback is positively correlated with the net climate feedback parameter (Andrews et al., 2012b; Andrews et al., 2015; Zhou et al.,

2016), a stronger positive SW cloud feedback will lead to higher climate sensitivity and may lead to a future warming towards the high end of current projections (Zhai et al.,

2015; Andrews et al., 2018).

On seasonal scales, SWCRE is strongest in the summer months when the solar heating is strongest (Harrison et al., 1990). Because SWCRE is in effect only during

67

daytime, it can substantially modify Tmax. For instance, Dai et al. (1999) found that increased cloud cover can reduce daily maximum temperature (Tmax), thereby decreasing diurnal temperature range. Tang and Leng (2012) reported that the damped

Tmax over Eurasia could be partially explained by the cloud cover increase during 1982-

2009. As a positive feedback, SWCRE has also been reported to play a role in heatwave and drought events over Europe by enhancing solar heating (Rowell & Jones, 2006;

Vautard et al., 2007; Zampieri et al., 2009; Chiriaco et al., 2014; Myers et al., 2018). This has influenced the environment, ecosystems and socio-economy through affecting the frequency and intensity of forest fires, power cuts, transport restrictions, crop failure and loss of life (De Bono et al., 2004; Ciais et al., 2005; Robine et al., 2008). For example,

Wetherald and Manabe (1995) reported that in the summer of mid-latitude continents, higher temperature enhances evaporation in the spring and then evaporation decreases in the summer due to depleted soil moisture. Combined with higher temperature, this summertime evaporation reduction leads to lower relative humidity (RH), which reduces cloud cover and thereby invigorates solar heating. Cheruy et al. (2014) revealed that the inter-model spread of summer temperature projections in Northern mid- latitudes in CMIP5 (Climate Model Inter-comparison Project Phase 5) models is greatly influenced by SWCRE.

All the above studies suggest that the SWCRE plays an important role in influencing the surface energy budget and extreme temperature. Well-mixed

68

greenhouse gases (WMGHGs) and aerosols are currently the two largest anthropogenic forcings (Myhre et al., 2013b). A better understanding on the climate response to these individual forcing agents is increasingly needed, considering their different trends across the globe and opposite impacts on climate (Shindell & Faluvegi, 2009). Due to the difficulty of separating the forced climate signal of a single agent within observational records, these studies are generally based on model simulations, such as the widely used increased CO2 experiment (Andrews et al., 2012a). Many attempts have also been made to explore the aerosol impact on the cloud properties and Earth’s energy balance

(Lohmann & Feichter, 2005; Chung & Soden, 2017), mean temperature (Ruckstuhl et al.,

2008; Philipona et al., 2009), as well as extreme temperature (Sillmann et al., 2013; Xu et al., 2018). However, all these studies treated aerosols as a whole and the individual impacts from absorbing and scattering aerosols are still less understood. Though some studies investigated the impact from individual aerosol species (Williams et al., 2001;

Chuang et al., 2002; Koch & Del Genio, 2010), they generally used only a single model, and the results may be subject to model biases (Flato et al., 2013). Moreover, due to the continuing increase in the likelihood of hot temperature extremes (Seneviratne et al.,

2014), as well as their serious consequences (De Bono et al., 2004), it is imperative to have a better understanding on the role of SWCRE from individual forcing agents in hot extremes. However, a multi-model study on the cloud response to individual aerosol species and the impact of that response in Tmax is still lacking. Given these knowledge

69

gaps, here we investigated the changes of SWCRE to CO2, BC and sulfate aerosols individually and explored its potential impact on Tmax by using a set of state-of-the-art global climate models. CO2 is the most dominant WMGHG while the latter two represent absorbing and scattering aerosols respectively. This chapter will proceed as follows: Data and methods are described in Section 4.2. Results are presented in Section

4.3, discussions and summary are given in Section 4.4.

4.2 Data and Methods

4.2.1 Data

This study employs the same PDRMIP models as in the last two chapters (Table

1). The aerosol loadings in the CanESM2 model for the two aerosol perturbations are shown in Fig. 16 for illustrative purpose, the spatial patterns for other models are similar. In the BC experiment, the concentration is highest in East China (E. China hereafter), followed by India and tropical Africa. For the SO4 simulations, the aerosols are mainly restricted to the Northern Hemisphere (NH), with the highest loading observed in E. China, followed by India and Europe. Eastern US also has moderately high concentrations. More detailed descriptions of PDRMIP and its initial findings are given in Samset et al. (2016), Myhre et al. (2017), Liu et al. (2018) and Tang et al. (2018).

70

Figure 16: Aerosol loadings for the two aerosol experiments in CanESM2 model (as in illustrative example).

4.2.2 Method

In this study, we focus on the SWCRE at the surface in the lower and mid- latitudes during boreal summer months (June-July-August, JJA hereafter), which is calculated as the difference in the SW radiative flux at the surface between all-sky and clear-sky conditions (Ramanathan et al., 1989). Changes in SWCRE are obtained by subtracting the control simulations from the perturbations using the data of the last 20 years in each coupled simulation. The changes are then normalized by the effective radiative forcing (ERF) in the corresponding experiments to obtain the changes per unit forcing. The ERF values for each model are obtained from Tang et al. (2019), which is diagnosed from the 6-15 year data of the fsst simulations of each perturbation by calculating the radiative flux changes at the top-of-the-atmosphere (Hansen et al., 2002).

The multi-model mean (MMM) ERF values are 3.65±0.09 W m-2 (CO2×2), 1.16±0.25 W m-2 71

(BC×10), and -3.52±0.63 W m-2 (SO4×5) for indicated experiments, respectively (MMM±1 standard error). Then the MMM changes are estimated by averaging all the nine models’ results, giving the same weighting factor to each model. A two-sided student t-test is used to examine whether the MMM results are significantly different from zero. The same process was also repeated to other variables analyzed (i.e., temperature and humidity).

In order to investigate the impact of circulation changes on specific humidity, the horizontal moisture flux convergence (MFC) is calculated (Banacos & Schultz, 2005) as:

(1)

In Equation (1), q is the specific humidity in g kg-1, and V is horizontal wind.

Equation (1) could be further written as:

(2)

In which and are zonal and meridional wind components in m s-1.

72

4.3 Results

4.3.1 SWCRE Change

Figure 17a-c show the SWCRE changes in response to abrupt changes in CO2, BC and SO4. CO2 causes positive SWCRE over most areas in the NH, indicating that more

SW radiation reaches the surface. BC causes similar changes, but with enhanced (ERF- normalized) magnitude, especially in North America (N. America), Europe and East

Asia (E. Asia). In some source regions of BC aerosols (tropical Africa and India), however, the SWCRE changes are negative, which means more SW was reflected. These changes are all statistically significant and are unlikely to be caused by natural variability. Besides, these patterns are also robust and consistent across at least eight of the nine models analyzed under all three experiments (figure not shown). For SO4, the

SWCRE changes are relatively small compared with the other two forcings and few significant changes were found over low-to-mid latitude regions. When domain averaged (green boxes in Fig. 17), the MMM SWCRE from CO2 forcing is, 1.7 W m-2 (N.

America), 2.0 W m-2 (Europe) and 1.5 W m-2 (E. China) respectively for the indicated regions. The SWCRE of BC forcing is 7.0 W m-2 (N. America), 9.0 W m-2 (Europe) and 9.4

W m-2 (E. China) respectively, which is roughly 3 to 5 times larger than that from CO2 forcing whereas sulfate aerosols induced 1.2 W m-2 over E. China and near-zero impact in N. America and Europe, with even the sign of change being uncertain (Fig. 18). Such

SWCRE changes could be largely explained by the changes of cloud cover (Fig. 17d-f).

73

Low-level cloud cover decreased significantly in regions where SWCRE is positive for

CO2 and BC forcing, with a stronger decrease from the latter, indicating that the cloud response is more sensitive to BC forcing than to WMGHGs. The sulfate aerosols caused increased cloud cover over mid-to-high latitudes (fig. 17f). The cloud cover in other levels (e.g., 500 hPa and 300 hPa) show similar pattern of change (not shown). In order to better understand these cloud responses, we will explore a set of potential mechanisms driving such changes.

Figure 17: SWCRE changes (a-c) and cloud cover changes per unit forcing at 850 hPa (d-f) in JJA, results for SO4 are changes per negative forcing. Grey dots indicate changes are significant at 0.05 level. Positive anomalies in a-c indicate more radiation reaching the surface.

74

Figure 18: Domain-averaged SWCRE changes for three regions (green boxes in Fig. 17). Bars represent MMM results and errorbars indicate one standard error across the models.

4.3.2 Mechanism of the Cloud Changes

Clouds form when air rises and cools to saturation, and are thus closely linked to changes in RH. (Fig. 19a-c). The general pattern of RH changes corresponds well with cloud cover changes (Fig. 17d-f). That is, the cloud cover decreases in regions where the

RH drops and vice versa for most areas. A larger RH reduction due to BC compared with CO2 also aligns with larger cloud cover decrease under BC forcing, especially in N.

America and Europe. This spatial pattern is not surprising as it is easier for air masses to reach saturation in conditions with higher RH. By definition, RH depends on both specific humidity and saturation vapor pressure (which, in turn, depends on temperature). In order to probe which factor determines the RH changes, we further analyzed specific humidity changes (Fig. 19d-f). Specific humidity increased

75

ubiquitously under both CO2 and BC scenarios, as a result of increased evaporation in a warmer climate. Thus, the main driver of the RH drop is the atmospheric temperature that drives a faster increase of saturation vapor pressure. Figure 20 shows the changes of vapor pressure as the function of temperature change over Europe at 850 hPa. For example, the temperature increased by ~1.1 K under CO2 forcing, accompanied by ~0.02 kPa vapor pressure increase. Such vapor pressure increase, however, cannot keep pace with the rising saturation vapor pressure, which is about 0.1 kPa. Consequently, the RH decreased in Europe and this is also the case for most other land areas. BC caused stronger temperature increases (and hence larger RH drop) in Europe and N. America, explaining the larger cloud cover reductions compared with CO2. In the source regions of BC, such as India and tropical Africa, the RH increases because of strong increases of specific humidity, combined with weak or no temperature changes (figure not shown).

Figure 19: Same as Figure 17, but for humidity at 850 hPa.

76

Figure 20: Domain-averaged vapor pressure changes per unit forcing as a function of temperature at 850 hPa for Europe. Errorbars indicate one standard error across the models. The thick black line represents saturation vapor pressure.

Figure 21: Same as Fig. 17, but for changes of moisture flux convergence (MFC, a-c) and vertical velocity (omega, d-f) per unit forcing. For vertical velocity, positive anomalies indicate the air is less convective.

Changes in moisture flux and stability may also play a role in altering specific humidity and cloud formation. Here we analyze the changes of MFC and vertical 77

velocity (omega) and find significant changes under the BC experiment (Fig. 21). It is seen that more moisture is transported to tropical Africa and India (Fig. 21b), which could explain the abovementioned increases of specific humidity in these regions despite their lack of warming. A similar response was noted by Liu et al. (2018), which suggested that more moisture could be brought into monsoon regions due to BC forcing.

Koch and Del Genio (2010) noted that BC particles could promote cloud cover in convergent regions as they enhance deep convection and low-level convergence when drawing in moisture from ocean to land regions. This is also observed in our analyses, for example over Africa, North India and Pakistan and part of North China (Fig. 21b and e). However, these impacts may be further compounded by cloud type, circulation, local humidity, and the altitude of BC particles relative to the clouds (Koch & Del Genio, 2010;

Samset & Myhre, 2015). Following Seager et al. (2007), the MFC changes could be further decomposed into changes due to mean circulations (dynamics) and changes due to mean moisture content (thermodynamics). On the whole, the MFC changes under BC forcing is largely explained by dynamic changes (Fig. 21b & 22b), while in tropical

Africa, Middle East and India, thermodynamics may also play an important role, either enhance or dampen the changes from dynamics (Fig. 22e). Taking India as an example, the MFC change due to dynamics is 0.05 g kg-1 s-1 and MFC change due to thermodynamics is 0.29 g kg-1 s-1, indicating a larger role due to increase in mean humidity in this region. The changes in moisture flux and stability in the CO2

78

experiment are relatively weaker compared with those from BC, and most of the changes are only observed in low-latitude regions (Fig. 21a & d). The sulfate aerosols, on the other hand, generally show opposite changes to those from CO2 and BC (Fig. 19c and f), owing to sulfate’s cooling effect. The above analyses illustrate that the cloud cover changes could be primarily explained by RH changes and, to a lesser extent, circulation and stability changes. It should be noted that our current analyses aim to give a first- order grasp on the responses of cloud instead of quantifying the contributions from each individual mechanism as these mechanisms are all arise from temperature change and are inherently superimposed each other, which is nearly impossible to separate their contributions individually.

Figure 22: Same as Fig. 17, but for MFC changes due to dynamics (a-c) and thermodynamics (d-f).

79

4.3.3 Fast and Slow Responses

The above responses shown are total responses, which could be further split into fast responses (also called rapid adjustments) and slow responses (Andrews et al., 2010;

Boucher et al., 2013). The fast responses generally occur within weeks to a few months with the global mean temperature unchanged with the expectation of a small change over land, which could be obtained by fsst simulations. The slow response is mainly depending on global mean temperature change, which could be estimated by the difference between coupled simulations and fsst simulations, assuming the total response is a linear combination of fast response and slow response (Samset et al., 2016;

Stjern et al., 2017). For the CO2 experiment, fast responses dominated in E. US and

Europe while both fast and slow responses influence Asia (Fig. 23). When it comes to

BC, both fast and slow responses are important in these regions, and in some regions the fast and slow response even show opposite changes (e.g., N. Europe). This is consistent with the findings of Stjern et al. (2017) that the response of cloud amount under BC forcing typically consists of opposite rapid adjustments. Regarding sulfate aerosols, the results are similar to CO2 induced changes, with fast responses dominating in E. US and

Europe while both fast and slow responses influence Asia. As discussed in Section 4.3.2, the slow responses in Asia is likely to be associated with MFC changes (Fig. 22), which could be, but are not limited to, shifts in the monsoons or ITCZ, higher mean humidity and tropical expansion, as both greenhouse gases and aerosols have been reported to

80

impact these circulations (Menon et al., 2002; Wang, 2007; Meehl et al., 2008; Seidel et al.,

2008; Allen et al., 2012; Turner & Annamalai, 2012).

Figure 23: Same as Figure 17 (d-f), but for fast (a-c) and slow responses (d-f) of cloud cover changes per unit forcing.

4.3.4 SWCRE Response to Sulfate Aerosol

Another interesting phenomenon worth noting is the relatively small changes in

SWCRE induced by sulfate aerosols compared with CO2 and BC. SWCRE is obtained as the difference of SW fluxes between all-sky and clear-sky conditions (Fig. 24). However, both clouds and aerosol particles scatter solar radiation, so that at least part of the radiation scattered by clouds under all-sky conditions will also be scattered by aerosols under clear-sky conditions (no clouds). That means the SW radiation change at the surface due to scattering may not be sensitive to cloud fraction changes, which leads to very small change in their difference (SWCRE), at least in the source regions (Fig. 24).

The SWCRE under sulfate aerosols will not be further discussed due to its small impact. 81

Figure 24: Changes of SW flux per unit negative forcing under all-sky (a), clear-sky (b) conditions and their difference (c) for SO4 experiment.

4.3.5 Impact on Radiation and Tmax

From the energy perspective, the net incoming radiation (Rin) at the surface is the combination of downward SW radiation and downward longwave (LW) radiation minus the reflected SW radiation (Rin = ↓SW - ↑SW + ↓LW). Rin represents the total energy available to maintain the surface temperature and to sustain the turbulent fluxes

(Philipona et al., 2009). The surface responds to the imposed Rin by redistributing the altered energy content among the outgoing LW radiation and nonradiative fluxes

(ground heat flux and turbulent flux) (Wild et al., 2004). Because SW radiation is in effect only during daytime while LW radiation works both day and night, Rin is directly related to Tmax. In a perturbed climate, both SW and LW radiation will change, thereby changing Rin and Tmax. The net SW radiation change is further linearly decomposed into SW changes under clear-sky conditions and SWCRE changes. The changes of Rin and its individual components, as well as Tmax are shown in Fig. 25. For the CO2×2 experiment, the SW under clear-sky conditions shows slight decreases over most of land

82

surfaces, mainly due to the absorption of SW radiation by enhanced water vapor, except for some high-latitude regions where albedo effect is important (Fig. 25a). Combined with the changes of SWCRE and ↓LW radiation, Rin shows significant increases over all land surfaces and thus, increasing Tmax (Fig. 25g and i). The BC×10 experiment shows similar responses, with significantly negative SW radiation under clear-sky conditions due to SW absorption by BC particles (Fig. 25b) and enhanced ↓LW radiation resulted from atmospheric heating (Fig. 25f). The resulting Rin changes largely explained Tmax changes on the first order, with cooling observed in source regions (India and tropical

Africa) and warming elsewhere (Fig. 25h and j). Nonetheless, some exceptions occurred

(i.e., E. China), with decreased Rin but increased Tmax, possibly due to the atmospheric heat transport (Menon et al., 2002) and reduced turbulent fluxes (Wild et al., 2004).

83

Figure 25: Changes of Rin and its components (a-h) as well as changes of Tmax (i-j) for the CO2×2 (left) and BC×10 (right) experiments (original output, no normalization applied).

In order to further determine the contributions in Tmax changes from each individual radiative component, a multilinear regression model is applied by regressing

Tmax changes to SW clear-sky, SWCRE and ↓LW radiation changes with zero intercept, obtaining the following models:

84

CO2×2:

Tmax = 0.08×SWclear-sky + 0.15×SWCRE + 0.14×↓LW (R2 = 0.73, p < 0.001)

BC×10:

Tmax = 0.05×SWclear-sky + 0.13×SWCRE + 0.15×↓LW (R2 = 0.80, p < 0.001)

Figure 26: Comparison of fitted Tmax from the linear models vs original Tmax values. Blue triangles are values for all grid boxes over NH and black solid line represents one-one line.

All values in the linear models are MMM changes in each experiment. The models could explain 73% and 80% of the Tmax change in CO2×2 and BC×10 experiment respectively. The coefficients represent the Tmax change under unit radiative flux change, in which the Tmax increases by 0.15 K (0.13 K) per unit increase in local SWCRE

85

under the CO2 (BC) experiment respectively. A comparison of the original Tmax values and the fitted values from the linear models is shown in Figure 26. The linear models predict the Tmax changes fairly well, with the values scattering along the one-to-one line. The contributions from each radiative component to Tmax changes were estimated with the linear models and the domain-averaged changes for N. America, Europe, E.

China and India (purple boxes in Fig. 25a) are listed in Table 11. Physically, Tmax increases in these regions are mainly due to the increased flux from SWCRE and ↓LW, and partially offset by the reduced flux from SWclear-sky (Table 11 & Fig. 25). Taking N.

America under CO2×2 experiment as an example, the warming in Tmax from SWCRE and ↓LW are 0.95 K and 3.24 K respectively, in which SWCRE contributed roughly by

23% to the total warming and the remaining 77% is from the ↓LW radiation change.

Such warming is offset by the 0.27 K cooling from SW changes under clear-sky conditions, leading to a net increase of 3.92 K in Tmax. The contributions of SWCRE in

Tmax increases are 29% (Europe), 20% (E. China) and 9% (India) for the indicated regions under the CO2×2 experiment. For the BC×10 experiment, the contributions from

SWCRE are larger than those in the CO2 experiment, i.e. 34% (N. America), 47%

(Europe) and 34% (E. China) for each region. The response over India under the BC experiment is opposite, in which both SW components cause cooling in Tmax due to reduced fluxes and such cooling is slightly offset by the warming from increased ↓LW radiation. In this case, the negative SWCRE change contributed 54% to the reduction in

86

Tmax. It is noted that the radiation change might not explain all Tmax changes, as other factors may come into play such as soil moisture, horizontal heat transport and precipitation (Dai et al., 1999). For instance, the temperature response would be different for a dry surface compared to a wet surface, given the same radiative fluxes. This is because more net radiation is realized as sensible heat instead of latent heat under dry conditions, which has been suggested to play an important role in recent European heatwaves (Seneviratne et al., 2006; Fischer et al., 2007b).

Table 11: Domain-averaged Tmax changes from each radiative component estimated from the linear models (unit: K).

CO2×2 Region SWclear-sky SWCRE ↓LW Total N. America -0.27 ± 0.01 0.95 ± 0.02 3.24 ± 0.03 3.92 ± 0.06 Europe -0.24 ± 0.01 1.14 ± 0.03 2.79 ± 0.02 3.69 ± 0.06 E. China -0.23 ± 0.01 0.71 ± 0.02 2.82 ± 0.02 3.30 ± 0.05 India -0.29 ± 0.01 0.26 ± 0.01 2.59 ± 0.02 2.56 ± 0.04 BC×10 Region SWclear-sky SWCRE ↓LW Total N. America -0.56 ± 0.03 1.00 ± 0.02 1.94 ± 0.04 2.38 ± 0.10 Europe -0.73 ± 0.04 1.15 ± 0.03 1.32 ± 0.03 1.74 ± 0.10 E. China -1.40 ± 0.08 0.98 ± 0.02 1.92 ± 0.04 1.50 ± 0.15 India -0.89 ± 0.05 -1.05 ± 0.02 1.10 ± 0.02 -0.84 ± 0.05

4.4 Discussion and Conclusion

Our study shows that the cloud cover in the summer is reduced in a warming climate over most mid-latitude land regions. The reduction of clouds, at the same time, may also reduce the warming effect by reducing ↓LW radiation (LWCRE). Specifically, 87

the LWCRE changes under unit CO2 forcing, in MMM, are -1.1 W m-2 (N. America), -0.8

W m-2 (Europe) and -1.0 W m-2 (E. China) respectively, resulting in net CRE

(SWCRE+LWCRE) changes of 0.6 W m-2 (N. America), 1.2 W m-2 (Europe) and 0.5 W m-2

(E. China) at the surface. The LWCRE changes per unit BC forcing are -1.7 W m-2 (N.

America), -2.1 W m-2 (Europe) and -1.5 W m-2 (E. China) respectively, leading to net CRE changes of 5.3 W m-2 (N. America), 6.9 W m-2 (Europe) and 7.9 W m-2 (E. China). The net

CRE changes are positive under both forcings and work as a positive feedback in these areas. As SWCRE is only active during daytime, the CRE changes have an even more pronounced amplifying effect on summer extreme temperature in these populated regions.

Recent European heatwave events have been linked to the shift of mean temperature (Schär et al., 2004; Barriopedro et al., 2011). Thus, the enhanced increase in summer mean Tmax may significantly increase the number of hot days and the probability of heatwave events. Our model simulations show that both N. America and

Europe show faster increases in Tmax than in Tmin (daily minimum temperature) under both CO2 and BC experiments (figure not shown), indicating an increase in diurnal temperature range, which has also been reported by Wang and Dillon (2014). These changes can have substantial socio-economic impacts (De Bono et al., 2004; Ciais et al.,

2005), influencing human health (Robine et al., 2008), labor productivity (Kjellstrom et

88

al., 2018), disease transmission (Paaijmans et al., 2010), as well as environmental and other ecological functions (Vasseur David et al., 2014; Wang & Dillon, 2014).

Some limitations also exist in the current study. Firstly, aerosol-cloud interactions cannot be realistically represented, as more than half of the PDRMIP simulations were run with fixed concentrations, where changes in cloud lifetime cannot affect aerosols. For the BC simulations, three models include aerosol indirect effects

(MIROC, NorESM and IPSL) while the remaining ones have only aerosol-radiation interactions included (instantaneous and rapid adjustments). The responses of SWCRE for the two categories are shown in Figure 27. For the regions of interest in the current study, the positive SWCRE over N. America, Europe and E. China and negative SWCRE over India are still observed in the models including indirect effects, but with slightly reduced magnitude. Thus, our main conclusions still hold in both sets of models, since the responses do not qualitatively vary between those with indirect effects and models without those effects, though the quantification of the response to BC is model- dependent. Such effects are not likely to be a large source of uncertainty but merit future study. Secondly, the aerosol perturbations are 10× and 5× present-day aerosol concentrations, which are time-invariant and therefore idealized. Such simulations have provided valuable physical insights into the effects of different forcings on a variety of aspects of the climate system. The aerosol concentrations, however, changed remarkably during the historical period and in recent decades, both spatially and temporally. For

89

example, aerosol concentrations have been decreasing in Europe and N. America since the 1980s and have been increasing in Asia since the 1950s (Smith et al., 2011). Future simulations may use aerosol forcing with realistic spatio-temporal changes.

Figure 27: SWCRE changes for the BC experiment, (a) for models without aerosol indirect effects and (b) for models with indirect effects.

In conclusion, our study shows that both CO2 and BC could cause positive

SWCRE changes over most regions in the NH, with a stronger response caused by BC, except over some key source regions of BC aerosols (e.g., India, tropical Africa) which show opposite changes. The SWCRE changes under sulfate aerosol forcing are, however, relatively small compared with the other two forcers. The SWCRE changes are mainly a consequence of RH changes and, to a lesser extent, circulation and stability changes. The

SWCRE may have contributed by 10~50% of summer mean Tmax increases, depending on forcing agent and region, and contributed substantially to Tmax decreases in the

90

source regions of India and Africa, which has great implications for extreme climatic events and socio-economic activities.

91

5. Conclusion and Summary

5.1 Key Findings

In this dissertation, I present three studies on the climate responses to individual climate drivers from three perspectives.

In the first study (Chapter 2), I compared the climate forcings estimated with six different methods under five drivers, and found that previously reported apparent consistency between the ERF values derived from fixed sea surface temperature simulations and linear regression holds for most climate forcings, excluding black carbon (BC). When land adjustment is accounted for, however, the fsst ERF values are generally 10-30% larger than ERFs derived using linear regression across all forcing agents, with a much larger (~70-100%) discrepancy for BC. Except for BC, this difference can be largely reduced by either using radiative kernel techniques or by exponential regression. Responses of clouds and their effects on shortwave radiation show the strongest variability in all experiments, limiting the application of regression-based ERF in small forcing simulations.

In Chapter 3, I compared the modelled dynamical response of Mediterranean precipitation to individual forcing agents in a set of global climate models. Our analyses show that both greenhouse gases and aerosols can cause drying in the Mediterranean, and that precipitation is more sensitive to black carbon forcing than to WMGHGs or sulfate aerosol. In addition to local heating, BC appears to reduce precipitation by

92

causing an enhanced positive NAO/AO-like sea level pressure (SLP) pattern, characterized by higher SLP at mid-latitudes and lower SLP at high-latitudes. WMGHGs cause a similar SLP change, and both are associated with a northward diversion of the jet stream and storm tracks, reducing precipitation in the Mediterranean while increasing precipitation in Northern Europe. Though the applied forcings were much larger, if forcings are scaled to those of the historical period of 1901-2010, roughly one- third (31±17%) of the precipitation decrease would be attributable to global BC forcing with the remainder largely attributable to WMGHGs whereas global scattering sulfate aerosols have negligible impacts. Aerosol-cloud interactions appear to have minimal impacts on Mediterranean precipitation in these models, at least in part as many simulations did not fully include such processes; these merit further study. The findings from this study suggest that future BC and WMGHG emissions may significantly affect regional water resources, agricultural practices, ecosystems, and the economy in the

Mediterranean region.

In Chapter 4, my analyses show that CO2 causes positive SWCRE changes over most of the NH, and BC causes similar positive responses over North America, Europe and East China but negative SWCRE over India and tropical Africa. When normalized by effective radiative forcing, the SWCRE from BC is roughly 3-5 times larger than that from CO2. SWCRE change is mainly due to cloud cover changes resulting from the changes in RH and, to a lesser extent, changes in circulation and stability. The SWCRE

93

response to sulfate aerosols, however, is negligible compared to that for CO2 and BC.

Using a multilinear regression model, it is found that mean daily maximum temperature

(Tmax) increases by 0.15 K and 0.13 K per W m-2 increase in local SWCRE under the CO2 and BC experiment, respectively. When domain-averaged, the SWCRE change contribution to summer mean Tmax changes was 10-30% under CO2 forcing and 30-50% under BC forcing, varying by region, which can have important implications for extreme climatic events and socio-economic activities.

5.2 Implications, Limitations and Future Work

This dissertation presents results on the climate responses to individual forcing agents, mainly GHGs and aerosols, from three perspectives-forcing, precipitation and cloud, aiming at providing useful information to the climate community and policy- makers. For example, as the concentration of GHGs continues unabated, combined with the dominance of GHGs in the future forcing, the responses of climate to GHGs are quite vital in projecting and understanding future climate change. The analyses in Chapter 3 and 4 provide such information. However, this is far from enough, as our understanding is still limited. As noted in Chapter 1, it is impossible to separate the climate responses to individual climate drivers from the observational data. Thus, the research mainly relies on climate modeling. PDRMIP project is designed for this purpose. However, there are many other climate drivers that are missing, such as land use and land cover

94

change, tropospheric ozone, and some other aerosol particles. Similar to the PDRMIP core experiments, large perturbations are needed to perform these simulations in order to obtain a strong climate signal. And this leads to another issue regarding the design of

PDRMIP. The results from PDRMIP can better our understanding on the mechanisms and physical processes of those forced responses, but could not provide accurate information on to what extent, they influenced our climate or contributed to the historical changes as the forcing is unrealistic large. Consequently, it is imperative to include more simulations with realistic forcing scenarios and more modeling groups in the future PDRMIP-like projects. Despite these shortcomings, I would still argue that

PDRMIP simulations are useful in climate research for the following reasons; (i) strong forced signals could be obtained and the processes could be thoroughly probed. (ii) most of my results shown in this dissertation are in changes per unit forcing. The readers could make a rough estimation about the historical contributions of these forcing agents.

And (iii) a better understanding of the climate system could be obtained only with large perturbations in some cases, though not related in my dissertation. For example, with

CO2×2 or ×4 experiment, we can know when will the ice sheets over Greenland melt or if the Atlantic Meridional Overturning Circulation (AMOC) will slow down and what will happen after that, though we will probably never reach those high CO2 concentrations.

(iv), the large aerosol perturbations could also provide information on how our climate system will respond to volcanic eruptions and geoengineering. Another thing to note is

95

that the impacts of climate drivers are not through temperature or precipitation only, but may also directly via physiological impact. For instance, the aerosols could impact human health both by changing temperature and by getting into our lungs to cause tissue damage, which is a focus health community. Future work of climate impact would be more interdisciplinary and require more collaborations and communications from researchers, funding agencies and policy-makers.

In chapter 2, we find that the consistency between the ERF values derived from fixed sea surface temperature simulations and linear regression holds for most climate forcings, except for BC. Future work may test if this finding still holds under those missing drivers. Besides, the unique behavior of BC found in this study also merits further investigations. Why there is such significant discrepancy between the ERF values obtained from those methods. Is it resulted from the mechanism of BC heating, or due to model deficiency or model errors? More work is needed to answer these questions. It is noted that ERF is only a metric to describe the forcing. The forcing values will not influence our ongoing climate policy or mitigation measures, such as Paris Agreement.

We, in fact, do not know the true value of this forcing and even the definition of ERF is still in debate (Boucher et al., 2013). I am not intending to find the true value of forcing, but rather to compare the ERF values obtained by six different methods to provide information to the researchers in this community. They can choose the best method based on their own needs, available output and computational resources.

96

In Chapter 3, the analysis is mainly focused on the dynamical responses as aerosol-cloud interactions are not included in some of those PDRMIP models. Thus, the results only represent the impacts from aerosol direct effects (aerosol-radiation interactions). On top of that, although our comparison of scaled equilibrium and unscaled transient simulations indicates that our methodology works well at least for

WMGHGs, there is no systematic study so far exploring the linearity (or non-linearity) of the precipitation responses to BC or the linearity of responses to multiple versus individual forcing on regional scales. Lastly, the aerosols have significant spatial and temporal variations. For instance, aerosol concentrations have been increasing in Asia continuously since 1950, but decreasing in Europe since the 1970s (Allen et al., 2013).

However, the concentrations in the PDRMIP simulations are fixed at present-day level, with higher concentrations in the Asia. As noted, further work is needed to determine how much of the Mediterranean trends result from local relative to remote forcing. To the extent that the trends are driven by remote forcing, the potential influence of such spatio-temporal variations will be small. This will be explored in future PDRMIP simulations. The results from this study shed light on how GHGs and air pollutants influence regional hydrological cycles and could better inform policy-makers in taking measures to mitigate the consequences of rainfall shortage.

In Chapter 4, on top of the aerosol indirect effects mentioned earlier, we urge more research should be focused on BC source regions about energy budget and

97

temperature. For example, E. China shows increased Tmax with a decreased Rin while

India shows decreased Tmax with a decreased Rin. Previous studies reported that the temperature response in BC sources regions are different, due to the different heating mechanism of BC, with a slight cooling in source regions and warming elsewhere

(Krishnan & Ramanathan, 2002; Menon et al., 2002). Our analyses also show that the local energy budget analysis may not be enough to predict the temperature in source regions. On top of that, the stronger impact of BC on circulations and moisture fluxes

(Fig. 21) than WMGHGs would be another interesting topic to probe, as such response is observed over the whole NH, not only in previously reported low-latitudes, monsoon regions, but also in part of N. America, Europe and the Atlantic storm tracks, which is much less-understood.

98

References

Allan, R., & Ansell, T. (2006). A new globally complete monthly historical gridded mean sea level pressure dataset (hadslp2): 1850–2004. Journal of Climate, 19(22), 5816- 5842. doi:10.1175/JCLI3937.1

Allen, R. J., Norris, J. R., & Wild, M. (2013). Evaluation of multidecadal variability in cmip5 surface solar radiation and inferred underestimation of aerosol direct effects over europe, china, japan, and india. Journal of Geophysical Research- Atmospheres, 118(12), 6311-6336. doi:10.1002/jgrd.50426

Allen, R. J., & Sherwood, S. C. (2011). The impact of natural versus anthropogenic aerosols on atmospheric circulation in the community atmosphere model. Climate Dynamics, 36(9-10), 1959-1978. doi:10.1007/s00382-010-0898-8

Allen, R. J., Sherwood, S. C., Norris, J. R., & Zender, C. S. (2012). Recent northern hemisphere tropical expansion primarily driven by black carbon and tropospheric ozone. Nature, 485(7398), 350-354. doi:10.1038/nature11097

Andreae, M. O., Jones, C. D., & Cox, P. M. (2005). Strong present-day aerosol cooling implies a hot future. Nature, 435(7046), 1187-1190. doi:10.1038/nature03671

Andrews, T., Forster, P. M., Boucher, O., Bellouin, N., & Jones, A. (2010). Precipitation, radiative forcing and global temperature change. Geophysical Research Letters, 37. doi:10.1029/2010gl043991

Andrews, T., Gregory, J. M., Paynter, D., Silvers, L. G., Zhou, C., Mauritsen, T., et al. (2018). Accounting for changing temperature patterns increases historical estimates of climate sensitivity. Geophysical Research Letters, 45(16), 8490-8499. doi:10.1029/2018gl078887

Andrews, T., Gregory, J. M., & Webb, M. J. (2015). The dependence of radiative forcing and feedback on evolving patterns of surface temperature change in climate models. Journal of Climate, 28(4), 1630-1648. doi:10.1175/JCLI-D-14-00545.1

Andrews, T., Gregory, J. M., Webb, M. J., & Taylor, K. E. (2012a). Forcing, feedbacks and climate sensitivity in cmip5 coupled atmosphere-ocean climate models. Geophysical Research Letters, 39. doi:10.1029/2012gl051607

99

Andrews, T., Gregory, J. M., Webb, M. J., & Taylor, K. E. (2012b). Forcing, feedbacks and climate sensitivity in cmip5 coupled atmosphere‐ocean climate models. Geophysical Research Letters, 39(9). doi:10.1029/2012GL051607

Armour, K. C., Bitz, C. M., & Roe, G. H. (2013). Time-varying climate sensitivity from regional feedbacks. Journal of Climate, 26(13), 4518-4534. doi:10.1175/JCLI-D-12- 00544.1

Arora, V., Scinocca, J., Boer, G., Christian, J., Denman, K., Flato, G., et al. (2011). Carbon emission limits required to satisfy future representative concentration pathways of greenhouse gases. Geophysical Research Letters, 38(5). doi:10.1029/2010GL046270

Banacos, P. C., & Schultz, D. M. (2005). The use of moisture flux convergence in forecasting convective initiation: Historical and operational perspectives. Weather and Forecasting, 20(3), 351-366. doi:10.1175/WAF858.1

Barriopedro, D., Fischer, E. M., Luterbacher, J., Trigo, R. M., & García-Herrera, R. (2011). The hot summer of 2010: Redrawing the temperature record map of europe. Science, 332(6026), 220. doi:10.1126/science.1201224

Bellouin, N., Rae, J., Jones, A., Johnson, C., Haywood, J., & Boucher, O. (2011). Aerosol forcing in the climate model intercomparison project (cmip5) simulations by hadgem2‐es and the role of ammonium nitrate. Journal of Geophysical Research: Atmospheres, 116(D20). doi:10.1029/2011JD016074

Bentsen, M., Bethke, I., Debernard, J., Iversen, T., Kirkevåg, A., Seland, Ø ., et al. (2013). The norwegian earth system model, noresm1-m—part 1: Description and basic evaluation of the physical climate. Geosci. Model Dev, 6(3), 687-720. doi:10.5194/gmd-6-687-2013

Black, E., Brayshaw David, J., & Rambeau Claire, M. C. (2010). Past, present and future precipitation in the middle east: Insights from models and observations. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 368(1931), 5173-5184. doi:10.1098/rsta.2010.0199

Bond, T. C., Doherty, S. J., Fahey, D. W., Forster, P. M., Berntsen, T., DeAngelo, B. J., et al. (2013). Bounding the role of black carbon in the climate system: A scientific assessment. Journal of Geophysical Research-Atmospheres, 118(11), 5380-5552. doi:10.1002/jgrd.50171

100

Boucher, O., Randall, D., Artaxo, P., Bretherton, C., Feingold, G., Forster, P., et al. (2013). Clouds and aerosols. In T. F. Stoker, D. Qin, G.-K. Plattner, M. Tignor, S. K. Allen, J. Boschung, A. Nauels, Y. Xia, V. Bex, & P. M. Midgley (Eds.), Climate change 2013: The physical science basis. Contribution of working group i to the fifth assessment report of the intergovernmental panel on climate change (pp. 571-657). Cambridge, UK and New York, USA: Cambridge University Press.

Buffoni, L., Maugeri, M., & Nanni, T. (1999). Precipitation in italy from 1833 to 1996. Theoretical and Applied Climatology, 63(1), 33-40. doi:10.1007/s007040050089

Cheruy, F., Dufresne, J. L., Hourdin, F., & Ducharne, A. (2014). Role of clouds and land- atmosphere coupling in midlatitude continental summer warm biases and climate change amplification in cmip5 simulations. Geophysical Research Letters, 41(18), 6493-6500. doi:10.1002/2014GL061145

Chiriaco, M., Bastin, S., Yiou, P., Haeffelin, M., Dupont, J.-C., & Stéfanon, M. (2014). European heatwave in july 2006: Observations and modeling showing how local processes amplify conducive large-scale conditions. Geophysical Research Letters, 41(15), 5644-5652. doi:10.1002/2014GL060205

Chuang, C. C., Penner, J. E., Prospero, J. M., Grant, K. E., Rau, G. H., & Kawamoto, K. (2002). Cloud susceptibility and the first aerosol indirect forcing: Sensitivity to black carbon and aerosol concentrations. Journal of Geophysical Research: Atmospheres, 107(D21), AAC 10-11-AAC 10-23. doi:10.1029/2000JD000215

Chung, E.-S., & Soden, B. (2015). An assessment of direct radiative forcing, radiative adjustments, and radiative feedbacks in coupled ocean–atmosphere models. Journal of Climate, 28(10), 4152-4170. doi:10.1175/JCLI-D-14-00436.1

Chung, E.-S., & Soden, B. J. (2017). Hemispheric climate shifts driven by anthropogenic aerosol–cloud interactions. Nature Geoscience, 10, 566. doi:10.1038/ngeo2988

Ciais, P., Reichstein, M., Viovy, N., Granier, A., Ogée, J., Allard, V., et al. (2005). Europe- wide reduction in primary productivity caused by the heat and drought in 2003. Nature, 437(7058), 529-533. doi:10.1038/nature03972

Clement, A. C., Burgman, R., & Norris, J. R. (2009). Observational and model evidence for positive low-level cloud feedback. Science, 325(5939), 460. doi:10.1126/science.1171255

101

Collins, M., Knutti, R., Arblaster, J., Dufresne, J.-L., Fichefet, T., Friedlingstein, P., et al. (2013). Long-term climate change: Projections, commitments and irreversibility. In T. F. Stoker, D. Qin, G.-K. Plattner, M. Tignor, S. K. Allen, J. Boschung, A. Nauels, Y. Xia, V. Bex, & P. M. Midgley (Eds.), Climate change 2013: The physical science basis. Contribution of working group i to the fifth assessment report of the intergovernmental panel on climate change (pp. 1029-1136). Cambridge, UK and New York, USA: Cambridge University Press.

Collins, W., Bellouin, N., Doutriaux-Boucher, M., Gedney, N., Halloran, P., Hinton, T., et al. (2011). Development and evaluation of an earth-system model–hadgem2. Geoscientific Model Development, 4(4), 1051-1075. doi:10.5194/gmd-4-1051-2011

Dai, A., Trenberth, K. E., & Karl, T. R. (1999). Effects of clouds, soil moisture, precipitation, and water vapor on diurnal temperature range. Journal of Climate, 12(8), 2451-2473. doi:10.1175/1520-0442(1999)012<2451:EOCSMP>2.0.CO;2

De Bono, A., Peduzzi, P., Kluser, S., & Giuliani, G. (2004). Impacts of summer 2003 heat wave in europe. Retrieved from https://www.unisdr.org/files/1145_ewheatwave.en.pdf

Della-Marta, P. M., Haylock, M. R., Luterbacher, J., & Wanner, H. (2007). Doubled length of western european summer heat waves since 1880. Journal of Geophysical Research: Atmospheres, 112(D15). doi:10.1029/2007JD008510

Dessler, A. E. (2010). A determination of the cloud feedback from climate variations over the past decade. Science, 330(6010), 1523-1527. doi:10.1126/science.1192546

Dufresne, J.-L., Foujols, M.-A., Denvil, S., Caubel, A., Marti, O., Aumont, O., et al. (2013). Climate change projections using the ipsl-cm5 earth system model: From cmip3 to cmip5. Climate Dynamics, 40(9-10), 2123-2165. doi:10.1007/s00382-012-1636-1

Dünkeloh, A., & Jacobeit, J. (2003). Circulation dynamics of mediterranean precipitation variability 1948–98. International Journal of Climatology, 23(15), 1843-1866. doi:10.1002/joc.973

Dunstone, N. J., Smith, D. M., Booth, B. B. B., Hermanson, L., & Eade, R. (2013). Anthropogenic aerosol forcing of atlantic tropical storms. Nature Geoscience, 6(7), 534-539. doi:10.1038/Ngeo1854

102

Fernandez, J., Saenz, J., & Zorita, E. (2003). Analysis of wintertime atmospheric moisture transport and its variability over southern europe in the ncep reanalyses. Climate Research, 23(3), 195-215. doi:10.3354/cr023195

Fischer, E. M., Seneviratne, S. I., Lüthi, D., & Schär, C. (2007a). Contribution of land- atmosphere coupling to recent european summer heat waves. Geophysical Research Letters, 34(6). doi:10.1029/2006GL029068

Fischer, E. M., Seneviratne, S. I., Vidale, P. L., Lüthi, D., & Schär, C. (2007b). Soil moisture–atmosphere interactions during the 2003 european summer heat wave. Journal of Climate, 20(20), 5081-5099. doi:10.1175/JCLI4288.1

Flato, G., Marotzke, J., Abiodun, B., Braconnot, P., Chou, S.-C., Collins, W., et al. (2013). Evaluation of climate models. In Climate change 2013 – the physical science basis: Working group i contribution to the fifth assessment report of the intergovernmental panel on climate change (pp. 741-866). Cambridge, UK and New York, USA: Cambridge University Press.

Forster, P. M., Richardson, T., Maycock, A. C., Smith, C. J., Samset, B. H., Myhre, G., et al. (2016). Recommendations for diagnosing effective radiative forcing from climate models for cmip6. Journal of Geophysical Research: Atmospheres, 121(20). doi:10.1002/2016JD025320

Gent, P. R., Danabasoglu, G., Donner, L. J., Holland, M. M., Hunke, E. C., Jayne, S. R., et al. (2011). The community climate system model version 4. Journal of Climate, 24(19), 4973-4991. doi:10.1175/2011JCLI4083.1

Gregory, J., & Andrews, T. (2016). Variation in climate sensitivity and feedback parameters during the historical period. Geophysical Research Letters, 43(8), 3911- 3920. doi:10.1002/2016GL068406

Gregory, J., Ingram, W., Palmer, M., Jones, G., Stott, P., Thorpe, R., et al. (2004). A new method for diagnosing radiative forcing and climate sensitivity. Geophysical Research Letters, 31(3). doi:10.1029/2003GL018747

Hansen, J., Sato, M., Nazarenko, L., Ruedy, R., Lacis, A., Koch, D., et al. (2002). Climate forcings in goddard institute for space studies si2000 simulations. Journal of Geophysical Research: Atmospheres, 107(D18). doi:10.1029/2001JD001143

103

Hansen, J., Sato, M., Ruedy, R., Nazarenko, L., Lacis, A., Schmidt, G., et al. (2005). Efficacy of climate forcings. Journal of Geophysical Research: Atmospheres, 110(D18). doi:10.1029/2005JD005776

Harrison, E. F., Minnis, P., Barkstrom, B. R., Ramanathan, V., Cess, R. D., & Gibson, G. G. (1990). Seasonal-variation of cloud radiative forcing derived from the earth radiation budget experiment. Journal of Geophysical Research-Atmospheres, 95(D11), 18687-18703. doi:DOI 10.1029/JD095iD11p18687

Hartmann, D., Klein Tank, A., Rusticucci, M., Alexander, L., Brönnimann, S., Charabi, Y., et al. (2013). Observations: Atmosphere and surface. In T. F. Stoker, D. Qin, G.-K. Plattner, M. Tignor, S. K. Allen, J. Boschung, A. Nauels, Y. Xia, V. Bex, & P. M. Midgley (Eds.), Climate change 2013: The physical science basis. Contribution of working group i to the fifth assessment report of the intergovernmental panel on climate change (pp. 159-254). Cambridge, UK and New York, USA: Cambridge University Press.

Hodnebrog, O., Myhre, G., Forster, P. M., Sillmann, J., & Samset, B. H. (2016). Local biomass burning is a dominant cause of the observed precipitation reduction in southern africa. Nat Commun, 7, 11236. doi:10.1038/ncomms11236

Hurrell, J. W., Holland, M. M., Gent, P. R., Ghan, S., Kay, J. E., Kushner, P. J., et al. (2013). The community earth system model: A framework for collaborative research. Bulletin of the American Meteorological Society, 94(9), 1339-1360. doi:10.1175/BAMS- D-12-00121.1

Hurrell, J. W., Kushnir, Y., & Visbeck, M. (2001). The north atlantic oscillation. 291(5504), 603-605. doi:10.1126/science.1058761

Iversen, T., Bentsen, M., Bethke, I., Debernard, J., Kirkevåg, A., Seland, Ø ., et al. (2013). The norwegian earth system model, noresm1-m-part 2: Climate response and scenario projections. Geoscientific Model Development, 6(2), 389. doi:10.5194/gmd-6- 389-2013

Jacobson, M. Z. (2012). Investigating cloud absorption effects: Global absorption properties of black carbon, tar balls, and soil dust in clouds and aerosols. Journal of Geophysical Research-Atmospheres, 117. doi:10.1029/2011jd017218

Jacobson, M. Z., & Kaufman, Y. J. (2006). Wind reduction by aerosol particles. Geophysical Research Letters, 33(24). doi:10.1029/2006gl027838

104

Johanson, C. M., & Fu, Q. (2009). Hadley cell widening: Model simulations versus observations. Journal of Climate, 22(10), 2713-2725. doi:10.1175/2008JCLI2620.1

Kalnay, E., Kanamitsu, M., Kistler, R., Collins, W., Deaven, D., Gandin, L., et al. (1996). The ncep/ncar 40-year reanalysis project. Bulletin of the American Meteorological Society, 77(3), 437-472. doi:10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2

Kaufman, Y. J., Tanre, D., & Boucher, O. (2002). A satellite view of aerosols in the climate system. Nature, 419(6903), 215-223. doi:10.1038/nature01091

Kay, J., Deser, C., Phillips, A., Mai, A., Hannay, C., Strand, G., et al. (2015). The community earth system model (cesm) large ensemble project: A community resource for studying climate change in the presence of internal climate variability. Bulletin of the American Meteorological Society, 96(8), 1333-1349. doi:10.1175/BAMS-D-13-00255.1

Kirkevåg, A., Iversen, T., Seland, Ø ., Hoose, C., Kristjánsson, J., Struthers, H., et al. (2013). Aerosol–climate interactions in the norwegian earth system model– noresm1-m. Geoscientific Model Development, 6(1), 207-244. doi:10.5194/gmd-6-207- 2013

Kjellstrom, T., Freyberg, C., Lemke, B., Otto, M., & Briggs, D. (2018). Estimating population heat exposure and impacts on working people in conjunction with climate change. International Journal of Biometeorology, 62(3), 291-306. doi:10.1007/s00484-017-1407-0

Koch, D., & Del Genio, A. D. (2010). Black carbon semi-direct effects on cloud cover: Review and synthesis. Atmos. Chem. Phys., 10(16), 7685-7696. doi:10.5194/acp-10- 7685-2010

Koren, I., Altaratz, O., Remer, L. A., Feingold, G., Martins, J. V., & Heiblum, R. H. (2012). Aerosol-induced intensification of rain from the tropics to the mid-latitudes. Nature Geoscience, 5(2), 118-122. doi:10.1038/Ngeo1364

Kottek, M., Grieser, J., Beck, C., Rudolf, B., & Rubel, F. (2006). World map of the köppen- geiger climate classification updated. Meteorologische Zeitschrift, 15(3), 259-263. doi:10.1127/0941-2948/2006/0130

105

Krichak, S. O., & Alpert, P. (2005). Decadal trends in the east atlantic–west russia pattern and mediterranean precipitation. International Journal of Climatology, 25(2), 183- 192. doi:10.1002/joc.1124

Krishnan, R., & Ramanathan, V. (2002). Evidence of surface cooling from absorbing aerosols. Geophysical Research Letters, 29(9). doi:10.1029/2002GL014687

Kvalevåg, M. M., Samset, B. H., & Myhre, G. (2013). Hydrological sensitivity to greenhouse gases and aerosols in a global climate model. Geophysical Research Letters, 40(7). doi:10.1002/grl.50318

Liu, L., Shawki, D., Voulgarakis, A., Kasoar, M., Samset, B. H., Myhre, G., et al. (2018). A pdrmip multimodel study on the impacts of regional aerosol forcings on global and regional precipitation. Journal of Climate, 31(11), 4429-4447. doi:10.1175/jcli-d- 17-0439.1

Lohmann, U., & Feichter, J. (2005). Global indirect aerosol effects: A review. Atmospheric Chemistry and Physics, 5, 715-737. doi:DOI 10.5194/acp-5-715-2005

Lorenz, E. N. (1951). Seasonal and irregular variations of the northern hemisphere sea- level pressure profile. Journal of Meteorology, 8(1), 52-59. doi:10.1175/1520- 0469(1951)008<0052:SAIVOT>2.0.CO;2

Manabe, S., & Wetherald, R. T. (1975). The effects of doubling the co2 concentration on the climate of a general circulation model. Journal of the Atmospheric Sciences, 32(1), 3-15. doi:10.1175/1520-0469(1975)032<0003:TEODTC>2.0.CO;2

Mariotti, A., Struglia, M. V., Zeng, N., & Lau, K. M. (2002). The hydrological cycle in the mediterranean region and implications for the water budget of the mediterranean sea. Journal of Climate, 15(13), 1674-1690. doi:10.1175/1520- 0442(2002)015<1674:THCITM>2.0.CO;2

Meehl, G. A., Arblaster, J. M., & Collins, W. D. (2008). Effects of black carbon aerosols on the indian monsoon. Journal of Climate, 21(12), 2869-2882. doi:10.1175/2007jcli1777.1

Meehl, G. A., & Tebaldi, C. (2004). More intense, more frequent, and longer lasting heat waves in the 21st century. Science, 305(5686), 994. doi:10.1126/science.1098704

106

Menon, S., Hansen, J., Nazarenko, L., & Luo, Y. (2002). Climate effects of black carbon aerosols in china and india. Science, 297(5590), 2250-2253. doi:10.1126/science.1075159

Miralles, D. G., Teuling, A. J., van Heerwaarden, C. C., & Vilà-Guerau de Arellano, J. (2014). Mega-heatwave temperatures due to combined soil desiccation and atmospheric heat accumulation. Nature Geoscience, 7, 345. doi:10.1038/ngeo2141

Muller, C. J., & O’Gorman, P. A. (2011). An energetic perspective on the regional response of precipitation to climate change. Nature Climate Change, 1, 266-271. doi:10.1038/nclimate1169

Myers, T. A., Mechoso, C. R., Cesana, G. V., DeFlorio, M. J., & Waliser, D. E. (2018). Cloud feedback key to marine heatwave off baja california. Geophysical Research Letters, 45(9), 4345-4352. doi:10.1029/2018GL078242

Myhre, G., Forster, P., Samset, B., Hodnebrog, Ø ., Sillmann, J., Aalbergsjø, S., et al. (2017). Pdrmip: A precipitation driver and response model intercomparison project, protocol and preliminary results. Bulletin of the American Meteorological Society(2016). doi:10.1175/BAMS-D-16-0019.1

Myhre, G., Kramer, R., Smith, C., Hodnebrog, Ø ., Forster, P., Soden, B., et al. (2018). Quantifying the importance of rapid adjustments for global precipitation changes. Geophysical Research Letters. doi:10.1029/2018GL079474

Myhre, G., Samset, B., Schulz, M., Balkanski, Y., Bauer, S., Berntsen, T., et al. (2013a). Radiative forcing of the direct aerosol effect from aerocom phase ii simulations. Atmospheric Chemistry and Physics, 13(4), 1853. doi:10.5194/acp-13-1853-2013

Myhre, G., Shindell, D., Bréon, F.-M., Collins, W., Fuglestvedt, J., Huang, J., et al. (2013b). Anthropogenic and natural radiative forcing. In T. F. Stoker, D. Qin, G.-K. Plattner, M. Tignor, S. K. Allen, J. Boschung, A. Nauels, Y. Xia, V. Bex, & P. M. Midgley (Eds.), Climate change 2013: The physical science basis. Contribution of working group i to the fifth assessment report of the intergovernmental panel on climate change (pp. 659-740). Cambridge, UK and New York, USA: Cambridge University Press.

Nabat, P., Somot, S., Mallet, M., Sanchez-Lorenzo, A., & Wild, M. (2014). Contribution of anthropogenic sulfate aerosols to the changing euro-mediterranean climate since 1980. Geophysical Research Letters, 41(15), 5605-5611. doi:10.1002/2014GL060798

107

Neale, R. B., Chen, C.-C., Gettelman, A., Lauritzen, P. H., Park, S., Williamson, D. L., et al. (2010). Description of the ncar community atmosphere model (cam 4.0). Retrieved from Boulder, CO, USA: https://www.ccsm.ucar.edu/models/ccsm4.0/cam/docs/description/cam4_desc.pd f

Otto-Bliesner, B. L., Brady, E. C., Fasullo, J., Jahn, A., Landrum, L., Stevenson, S., et al. (2016). Climate variability and change since 850 ce: An ensemble approach with the community earth system model. Bulletin of the American Meteorological Society, 97(5), 735-754. doi:10.1175/BAMS-D-14-00233.1

Paaijmans, K. P., Blanford, S., Bell, A. S., Blanford, J. I., Read, A. F., & Thomas, M. B. (2010). Influence of climate on malaria transmission depends on daily temperature variation. Proceedings of the National Academy of Sciences, 107(34), 15135. doi:10.1073/pnas.1006422107

Philipona, R., Behrens, K., & Ruckstuhl, C. (2009). How declining aerosols and rising greenhouse gases forced rapid warming in europe since the 1980s. Geophysical Research Letters, 36(2). doi:10.1029/2008GL036350

Piervitali, E., Colacino, M., & Conte, M. (1998). Rainfall over the central-western mediterranean basin in the period 1951-1995. Part i: Precipitation trends. Nuovo Cimento, 21(3), 331-344.

Proistosescu, C., & Huybers, P. J. (2017). Slow climate mode reconciles historical and model-based estimates of climate sensitivity. Science advances, 3(7), e1602821. doi:10.1126/sciadv.1602821

Quadrelli, R., Pavan, V., & Molteni, F. (2001). Wintertime variability of mediterranean precipitation and its links with large-scale circulation anomalies. Climate Dynamics, 17(5), 457-466. doi:10.1007/s003820000121

Ramanathan, V., & Carmichael, G. (2008). Global and regional climate changes due to black carbon. Nature Geoscience, 1, 221. doi:10.1038/ngeo156

Ramanathan, V., Cess, R. D., Harrison, E. F., Minnis, P., Barkstrom, B. R., Ahmad, E., et al. (1989). Cloud-radiative forcing and climate: Results from the earth radiation budget experiment. Science, 243(4887), 57-63. doi:10.1126/science.243.4887.57

108

Ramanathan, V., Crutzen, P. J., Kiehl, J. T., & Rosenfeld, D. (2001). Aerosols, climate, and the hydrological cycle. Science, 294(5549), 2119-2124. doi:10.1126/science.1064034

Robine, J.-M., Cheung, S. L. K., Le Roy, S., Van Oyen, H., Griffiths, C., Michel, J.-P., et al. (2008). Death toll exceeded 70,000 in europe during the summer of 2003. Comptes Rendus Biologies, 331(2), 171-178. doi:10.1016/j.crvi.2007.12.001

Rowell, D. P., & Jones, R. G. (2006). Causes and uncertainty of future summer drying over europe. Climate Dynamics, 27(2), 281-299. doi:10.1007/s00382-006-0125-9

Ruckstuhl, C., Philipona, R., Behrens, K., Collaud Coen, M., Dürr, B., Heimo, A., et al. (2008). Aerosol and cloud effects on solar brightening and the recent rapid warming. Geophysical Research Letters, 35(12). doi:10.1029/2008GL034228

Samset, B., Myhre, G., Forster, P., Hodnebrog, Ø ., Andrews, T., Faluvegi, G., et al. (2016). Fast and slow precipitation responses to individual climate forcers: A pdrmip multimodel study. Geophysical Research Letters, 43(6), 2782-2791. doi:10.1002/2016GL068064

Samset, B. H., & Myhre, G. (2015). Climate response to externally mixed black carbon as a function of altitude. Journal of Geophysical Research: Atmospheres, 120(7), 2913- 2927. doi:10.1002/2014JD022849

Schär, C., Vidale, P. L., Lüthi, D., Frei, C., Häberli, C., Liniger, M. A., et al. (2004). The role of increasing temperature variability in european summer heatwaves. Nature, 427(6972), 332-336. doi:10.1038/nature02300

Schmidt, G. A., Kelley, M., Nazarenko, L., Ruedy, R., Russell, G. L., Aleinov, I., et al. (2014). Configuration and assessment of the giss modele2 contributions to the cmip5 archive. Journal of Advances in Modeling Earth Systems, 6(1), 141-184. doi:10.1002/2013MS000265

Schneider, U., Becker, A., Finger, P., Meyer-Christoffer, A., Rudolf, B., & Ziese, M. (2016). Gpcc full data reanalysis version 7.0: Monthly land-surface precipitation from rain gauges built on gts based and historic data. Retrieved from: https://doi.org/10.5065/D6000072

Seager, R., Ting, M., Held, I., Kushnir, Y., Lu, J., Vecchi, G., et al. (2007). Model projections of an imminent transition to a more arid climate in southwestern north america. Science, 316(5828), 1181-1184. doi:10.1126/science.1139601

109

Seidel, D. J., Fu, Q., Randel, W. J., & Reichler, T. J. (2008). Widening of the tropical belt in a changing climate. Nature Geoscience, 1(1), 21-24. doi:10.1038/ngeo.2007.38

Seneviratne, S. I., Donat, M. G., Mueller, B., & Alexander, L. V. (2014). No pause in the increase of hot temperature extremes. Nature Climate Change, 4, 161. doi:10.1038/nclimate2145

Seneviratne, S. I., Luthi, D., Litschi, M., & Schar, C. (2006). Land-atmosphere coupling and climate change in europe. Nature, 443(7108), 205-209. doi:10.1038/nature05095

Seneviratne, S. I., Wilhelm, M., Stanelle, T., van den Hurk, B., Hagemann, S., Berg, A., et al. (2013). Impact of soil moisture-climate feedbacks on cmip5 projections: First results from the glace-cmip5 experiment. Geophysical Research Letters, 40(19), 5212-5217. doi:10.1002/grl.50956

Shindell, D., & Faluvegi, G. (2009). Climate response to regional radiative forcing during the twentieth century. Nature Geoscience, 2(4), 294-300. doi:10.1038/Ngeo473

Shindell, D. T., Lamarque, J. F., Schulz, M., Flanner, M., Jiao, C., Chin, M., et al. (2013). Radiative forcing in the accmip historical and future climate simulations. Atmospheric Chemistry and Physics, 13(6), 2939-2974. doi:10.5194/acp-13-2939-2013

Shindell, D. T., Schmidt, G. A., Miller, R. L., & Rind, D. (2001). Northern hemisphere winter climate response to greenhouse gas, ozone, solar, and volcanic forcing. Journal of Geophysical Research: Atmospheres, 106(D7), 7193-7210. doi:10.1029/2000JD900547

Shindell, D. T., Voulgarakis, A., Faluvegi, G., & Milly, G. (2012). Precipitation response to regional radiative forcing. Atmospheric Chemistry and Physics, 12(15), 6969-6982. doi:10.5194/acp-12-6969-2012

Sillmann, J., Pozzoli, L., Vignati, E., Kloster, S., & Feichter, J. (2013). Aerosol effect on climate extremes in europe under different future scenarios. Geophysical Research Letters, 40(10), 2290-2295. doi:10.1002/grl.50459

Smith, C., Kramer, R., Myhre, G., Forster, P., Soden, B., Andrews, T., et al. (2018). Understanding rapid adjustments to diverse forcing agents. Geophysical Research Letters, 45. doi:10.1029/2018GL079826

110

Smith, S. J., van Aardenne, J., Klimont, Z., Andres, R. J., Volke, A., & Delgado Arias, S. (2011). Anthropogenic sulfur dioxide emissions: 1850–2005. Atmos. Chem. Phys., 11(3), 1101-1116. doi:10.5194/acp-11-1101-2011

Soden, B. J., Held, I. M., Colman, R., Shell, K. M., Kiehl, J. T., & Shields, C. A. (2008). Quantifying climate feedbacks using radiative kernels. Journal of Climate, 21(14), 3504-3520. doi:10.1175/2007JCLI2110.1

Stevens, B., & Feingold, G. (2009). Untangling aerosol effects on clouds and precipitation in a buffered system. Nature, 461(7264), 607-613. doi:10.1038/nature08281

Stjern, C. W., Samset, B. H., Myhre, G., Forster, P. M., Hodnebrog, O., Andrews, T., et al. (2017). Rapid adjustments cause weak surface temperature response to increased black carbon concentrations. Journal of Geophysical Research-Atmospheres, 122(21), 11462-11481. doi:10.1002/2017jd027326

Takahashi, C., & Watanabe, M. (2016). Pacific trade winds accelerated by aerosol forcing over the past two decades. Nature Climate Change, 6(8), 768-+. doi:10.1038/Nclimate2996

Takemura, T., Egashira, M., Matsuzawa, K., Ichijo, H., O'ishi, R., & Abe-Ouchi, A. (2009). A simulation of the global distribution and radiative forcing of soil dust aerosols at the last glacial maximum. Atmospheric Chemistry & Physics, 9(9). doi:10.5194/acp-9-3061-2009

Takemura, T., Nozawa, T., Emori, S., Nakajima, T. Y., & Nakajima, T. (2005). Simulation of climate response to aerosol direct and indirect effects with aerosol transport‐radiation model. Journal of Geophysical Research: Atmospheres, 110(D2). doi:10.1029/2004JD005029

Tang, Q., & Leng, G. (2012). Damped summer warming accompanied with cloud cover increase over eurasia from 1982 to 2009. Environmental Research Letters, 7(1), 014004. doi:10.1088/1748-9326/7/1/014004

Tang, T., Shindell, D., Faluvegi, G., Myhre, G., Olivié, D., Voulgarakis, A., et al. (2019). Comparison of effective radiative forcing calculations using multiple methods, drivers, and models. Journal of Geophysical Research: Atmospheres, 124(8), 4382- 4394. doi:10.1029/2018JD030188

111

Tang, T., Shindell, D., Samset, B. H., Boucher, O., Forster, P. M., Hodnebrog, Ø ., et al. (2018). Dynamical response of mediterranean precipitation to greenhouse gases and aerosols. Atmospheric Chemistry and Physics, 18(11), 8439-8452. doi:10.5194/acp-18-8439-2018

Taylor, K. E., Stouffer, R. J., & Meehl, G. A. (2012). An overview of cmip5 and the experiment design. 93(4), 485-498. doi:10.1175/bams-d-11-00094.1

Turner, A. G., & Annamalai, H. (2012). Climate change and the south asian summer monsoon. Nature Climate Change, 2(8), 587-595. doi:10.1038/nclimate1495

Vasseur David, A., DeLong John, P., Gilbert, B., Greig Hamish, S., Harley Christopher, D. G., McCann Kevin, S., et al. (2014). Increased temperature variation poses a greater risk to species than climate warming. Proceedings of the Royal Society B: Biological Sciences, 281(1779), 20132612. doi:10.1098/rspb.2013.2612

Vautard, R., Yiou, P., D'Andrea, F., de Noblet, N., Viovy, N., Cassou, C., et al. (2007). Summertime european heat and drought waves induced by wintertime mediterranean rainfall deficit. Geophysical Research Letters, 34(7). doi:10.1029/2006GL028001

Walters, D., Williams, K., Boutle, I., Bushell, A., Edwards, J., Field, P., et al. (2014). The met office unified model global atmosphere 4.0 and jules global land 4.0 configurations. Geoscientific Model Development, 7(1), 361-386. doi:10.5194/gmd-7- 361-2014

Wang, C. (2007). Impact of direct radiative forcing of black carbon aerosols on tropical convective precipitation. Geophysical Research Letters, 34(5). doi:10.1029/2006GL028416

Wang, G., & Dillon, M. E. (2014). Recent geographic convergence in diurnal and annual temperature cycling flattens global thermal profiles. Nature Climate Change, 4(11), 988-992. doi:10.1038/Nclimate2378

Washington, W. M., & Meehl, G. A. (1984). Seasonal cycle experiment on the climate sensitivity due to a doubling of co2 with an atmospheric general circulation model coupled to a simple mixed-layer ocean model. Journal of Geophysical Research: Atmospheres, 89(D6), 9475-9503. doi:10.1029/JD089iD06p09475

112

Watanabe, M., Suzuki, T., O’ishi, R., Komuro, Y., Watanabe, S., Emori, S., et al. (2010). Improved climate simulation by miroc5: Mean states, variability, and climate sensitivity. Journal of Climate, 23(23), 6312-6335. doi:10.1175/2010JCLI3679.1

Westervelt, D. M., Conley, A. J., Fiore, A. M., Lamarque, J.-F., Shindell, D., Previdi, M., et al. (2017). Multimodel precipitation responses to removal of u.S. Sulfur dioxide emissions. Journal of Geophysical Research-Atmospheres, 122(9), 5024-5038. doi:doi:10.1002/2017JD026756

Wetherald, R. T., & Manabe, S. (1995). The mechanisms of summer dryness induced by greenhouse warming. Journal of Climate, 8(12), 3096-3108. doi:10.1175/1520- 0442(1995)008<3096:TMOSDI>2.0.CO;2

Wild, M. (2009). Global dimming and brightening: A review. Journal of Geophysical Research: Atmospheres, 114(D10). doi:10.1029/2008JD011470

Wild, M., Ohmura, A., Gilgen, H., & Rosenfeld, D. (2004). On the consistency of trends in radiation and temperature records and implications for the global hydrological cycle. Geophysical Research Letters, 31(11). doi:10.1029/2003GL019188

Williams, K. D., Jones, A., Roberts, D. L., Senior, C. A., & Woodage, M. J. (2001). The response of the climate system to the indirect effects of anthropogenic sulfate aerosol. Climate Dynamics, 17(11), 845-856. doi:10.1007/s003820100150

Wood, R., & Bretherton, C. S. (2006). On the relationship between stratiform low cloud cover and lower-tropospheric stability. Journal of Climate, 19(24), 6425-6432. doi:10.1175/JCLI3988.1

Worley, S. J., Woodruff, S. D., Reynolds, R. W., Lubker, S. J., & Lott, N. (2005). Icoads release 2.1 data and products. International Journal of Climatology, 25(7), 823-842. doi:10.1002/joc.1166

Xoplaki, E., González-Rouco, J. F., Luterbacher, J., & Wanner, H. (2004). Wet season mediterranean precipitation variability: Influence of large-scale dynamics and trends. Climate Dynamics, 23(1), 63-78. doi:10.1007/s00382-004-0422-0

Xu, Y. Y., Lamarque, J. F., & Sanderson, B. M. (2018). The importance of aerosol scenarios in projections of future heat extremes. Climatic Change, 146(3-4), 393- 406. doi:10.1007/s10584-015-1565-1

113

Zampieri, M., D’Andrea, F., Vautard, R., Ciais, P., de Noblet-Ducoudré, N., & Yiou, P. (2009). Hot european summers and the role of soil moisture in the propagation of mediterranean drought. Journal of Climate, 22(18), 4747-4758. doi:10.1175/2009JCLI2568.1

Zelinka, M. D., Randall, D. A., Webb, M. J., & Klein, S. A. (2017). Clearing clouds of uncertainty. Nature Climate Change, 7(10), 674-678. doi:10.1038/nclimate3402

Zhai, C. X., Jiang, J. H., & Su, H. (2015). Long-term cloud change imprinted in seasonal cloud variation: More evidence of high climate sensitivity. Geophysical Research Letters, 42(20), 8729-8737. doi:10.1002/2015gl065911

Zhang, M., & Huang, Y. (2014). Radiative forcing of quadrupling co2. Journal of Climate, 27(7), 2496-2508. doi:10.1175/JCLI-D-13-00535.1

Zhou, C., Zelinka, M. D., & Klein, S. A. (2016). Impact of decadal cloud variations on the earth’s energy budget. Nature Geoscience, 9, 871. doi:10.1038/ngeo2828

114

Biography

Tao Tang completed his bachelor study in environmental science at Beijing

Forestry University on July/2012 and master study in environmental engineering at the

University of Hong Kong on Dec/2013. Then he joined Duke University as a PhD student in the division of Earth and Ocean Sciences in the fall of 2014, working with

Prof. Drew Shindell. In June/2017, he became a PhD candidate and since then he has been working on the PDRMIP project (Precipitation and Driver Response Model Inter- comparison Project), exploring the climate responses to individual climate forcing agents via using observations and GCMs output. Now he has published two peer-reviewed articles with PDRMIP data and has been serving as a teaching assistant during the past five years for multiple courses in the EOS department at Duke University, such as

Global warming, Climate and society, Climate system, as well as Dynamic Oceans.

115