Demystifying Climate Models

Front matter

Earth Systems Data and Models

Andrew Gettelman Richard B. Rood

Demystifying Climate Models

Part I Basic Principles and the Problem of Climate Forecasts

LI Whatls Clhmate?;. ios CAPES Se Gs ERR hee 12: “What Is a Model? 2.2. ioc RR Res 13. Uncertainty «endet eI Va SEE Sees A ed L4. Summary eos Sa eee eas Noe ae Gosia eg RU ERG edat UR 2:3. .Sunmimaty yeo 222.22. gw E mA RR SHE 3.3. Climate History ot scs ee Se et Sk ela eS 3.59" Summary c.g RR eee ERREUR REPE 4.2.1. “Finite: Pieces i.c le RÀ ed eee eee 42:2. .Processeske c kx ee Eee gee REA a ee oe viii

Contents

Part II Model Mechanics

5.3 General Circulation -cossa a llle 69 JAL- (Cloüds-—— ecce mRERUERE a5 eee a 74

5.8. Sümmary i.e eR eR Redon sb e REND 85

63.2 Deep Oceans. coe o ghee tee DIR TRA 93

6.5 The Ocean Carbon Cycle s siise a rria o a a 104 6.6. “Challenges ineca ogg eee MER ee A ERE 104

Contents

6.7 6.8

Summary ota ne wl E n WE UAPCERQEPe. EAE BS

7.1

7.2

7.3 7.4 7.5 7.6 7.7 7.8

7.9

7.10

8.1

8.2 8.3

8.4

Contents

X 843 ELNO ines ke AERE Ven 152 85. Challenges: oss n vp meus EPA PERS uin 154 8^ Summary els te PAY awe De REIS Ri REM tats 157 Part III Using Models 91:2 “Observations: os sos pees ieee oR ee PS tee 164 9.2.3 Using Model Evaluation to Guide Further Observations: svg vw a ee GE dae REA e RU 172 9.3.TL. FOrecasts 34-23 si Aw eu E a aa ees 173 0°32.» "Projections, 2-. ie ec exeo opt REOR RA 173 9.4 Applications of Climate Model Evaluation: Ozone Assessment... . vss e eS ES ee 174 9.5 Summary ..2 22. dp ew WE omn RE Vu HE d 175 10 Predictability: 2.4532 660.4 ke see sey esas E 177 10.1.2 Variability’: ii be 180 10.2.1 Predicting the Near Term: Initial Condition Uncertainty 4. iturin oe UAE B SESS 182 10.2.2 Predicting the Next 30-50 Years: Scenario Uncertàinty.. e iu b ER redes 183 10.2.3 Predicting the Long Term: Model Uncertainty 10.5; Summary o caneca 24. fb dak eee ERR bI EIUS 196

Contents xi

13:6.3: ‘Challenges. cce xe Ae eS ate Rees LE 250

13:7 -Final Thoughts... vL ESE Ae ee ee 251 Index... oen c bow xe wes died E HUP 271

About the Authors

Andrew Gettelman is a Scientist in the Climate and Global Dynamics and Atmospheric Chemistry and Modeling Laboratories at the National Center for Atmospheric Research (NCAR). He is actively involved in developing atmosphere and chemistry components for global climate models at NCAR. Dr. Gettelman specializes in understanding and simulating cloud processes and their impact on climate, especially ice clouds. He has numerous publications on cloud physics representations in global models, as well as research on climate forcing and feedbacks. He has participated in several international assessments of climate models, particularly for assessing atmospheric chemistry. Gettelman holds a doctorate in Atmospheric Science from the University of Washington, Seattle. He is a recent recipient of the American Geophysical Union Ascent Award, and is a Thompson-Reuters Highly Cited Researcher.

Richard B. Rood is a Professor in the Department of Climate and Space Sciences and Engineering (CLaSP) at the University of Michigan. He is also appointed in the School of Natural Resources and Environment. Prior to joining the University of Michigan, he worked in modeling and high performance computing at the National Aeronautics and Space Administration (NASA). His recent research is focused on the usability of climate knowledge and data in management planning and practice. He has started classes in climate-change problem solving, climate change uncertainty in decision making, climate-change informatics (with Paul Edwards). In addition to publications on numerical models, his recent publications include software engineering, informatics, political science, social science, forestry and public health. Rood's professional degree is in Meteorology from Florida State University. He recently served on the National Academy of Sciences Committee on A National Strategy for Advancing Climate Modeling. He writes expert blogs on climate change science and problem solving for the Weather Underground Richard Rood is a Fellow of American Meteorological Society and a winner of the World Meteorological Organization's Norbert Gerbier Award.

xiii

Introduction

Human-caused climate change is perhaps the defining environmental issue of the early twenty-first century. We observe the earth's climate in the present, but observations of future climate are not available yet. So in order to predict the future, we rely on simulation models to predict future climate.

This book is designed to be a guide to climate simulation and prediction for the non-specialist and an entry point for understanding uncertainties in climate models. The goal is not to be simply a popular guide to climate modeling and prediction, but to help those using climate models to understand the results. This book provides background on the earth's climate system and how it might change, a detailed qualitative analysis of how climate models are constructed, and a discussion of model results and the uncertainty inherent in those results. Throughout the text, terms in bold will be referenced in the glossary. References are provided as footnotes in each chapter.

Who uses climate models? Climate model users are practitioners in many fields who desire to incorporate information about climate and climate change into planning and management decisions. Users may be scientists and engineers in fields such as ecosystems or water resources. These scientists are familiar with models and the roles of models in natural science. In other cases, the practitioners are engineers, urban planners, epidemiologists, or architects. Though not necessarily familiar with models of natural science, experts in these fields use quantitative information for decision-making. These experts are potential users of climate models. We hope in the end that by understanding climate models and their uncertainties, the reader will understand how climate models are constructed to represent the earth's climate system. The book is intended to help the reader become a more competent interpreter or translator of climate model output.

Climate is best thought of as the distribution of weather states, or the probability of finding a particular weather state (usually described by temperature and precipitation) at any place and time. Climate science seeks to be able to describe this distribution. In contrast, the goal of predicting the weather is to figure out exactly which weather state will occur for a specific place and time (e.g., what the high temperature and total precipitation will be on Tuesday for a given city). Even in

XV xvi Introduction modern societies, we are still more dependent on the weather than we like to admit. Think of a winter storm snarling traffic and closing schools. Windstorms and hailstorms can cause significant damage. Or think of the impact of severe tropical cyclones (also called hurricanes or typhoons, depending on their location), personified and immortalized with names like Sandy, Andrew, or Katrina. Persistence (or absence) of weather events is also important. Too little rain (leading to drought and its resulting effects on agriculture and even contributing to wildfires) and too much rain (leading to flooding) are both damaging.

Although we are tempted to speak of a single "climate," there are many climates. Every place has its own. We build our societies to be comfortable during the expected weather events (the climate) in each place. Naturally, different climates mean different expected weather events, and our societies adapt. Buildings in Minneapolis are built to standards different from buildings in Miami or San Francisco. City planning is also different in different climates. Minneapolis has connected buildings so that people do not need to walk outside in winter, for example. Singapore has connected buildings so that people do not need to walk outside in heat and humidity. Not just the built environment, but the fabric of society may be different with local climates. In warmer climates, social life takes place outdoors, for example, or the flow of a day includes a rest period (siesta in Spanish) during the hottest period of the day.

We construct our lives for possible and sometimes rare weather events: putting on snow tires for winter even though snow may not be around for over half a winter. We build into our lives the ability to deal with variations in the weather. A closet contains coats, gloves, hats, rain jackets, umbrellas, sun hats, and sunglasses: We are ready for a range and for a distribution of possible weather states. Some events are rare: Snow occasionally has fallen in Los Angeles, for example. But it is usually the rare or extreme weather events that are damaging. These outliers of the distribution are typically damaging because they are unexpected and therefore we do not adequately prepare for them. Or rather, the expectation (probability) is so rare that it is not cost-effective for society to prepare for them. This applies to the individual as well: if you live in Miami your closet contains more warm weather gear, and less cold weather gear. It is unlikely you would be able to dress for temperatures well below freezing. The impacts of extreme weather are dependent on the climate of a place as well. For example, a few inches (centimeters) of snow is typical for Denver, Minneapolis, or Oslo, but it will shut down Rome or Atlanta. One inch (25 millimeters) of annual rainfall is typical for Cairo, but a disaster in most other places.

Where the most damaging weather events occur are at the extremes of the climate distribution. One problem is that we often do not know the distribution very well. Every time we hit a record (e.g., a high temperature, rainfall in 24 hours, days without rain), we expand the range of observed events a little, and we learn more about what might happen in a particular place. Because extreme events are rare, we do not really know the true chance of their occurrence. Think about your knowledge of the climate where you live or in a place you have visited several times. At first, you might not have a good grasp of what the seasons are like. After a few years,

Introduction xvii you think you know how the seasons evolve. But there are always events that will surprise you. The record events are those that surprise everyone. What is the probability of a hurricane flooding Houston as New Orleans was flooded by Katrina? It has not happened since the city of Houston has been there, so we may not know. The extremes of the distribution of possible weather states are not well known.

This creates even more of a problem when these extremes change. Changing the distribution of weather is what we mean by climate change. The cause of those changes might be natural or they might be human caused (anthropogenic).

So how do we predict the future of weather and the distribution of weather that represents climate in a location? To understand and predict the future, we need a way to represent the system. In other words, we need a model. This book is about how we attempt to use models to represent the complex climate system and predict the future. Our goal is to explain and provide a better understanding of the models we use to describe the past, present, and future of the earth system. These are commonly known as climate models. Scientists often refer to these models formally as earth system models, but we use the term climate model.

The purpose of this book is to demystify the models we use to simulate present and future climate. We explain how the models are constructed, why they are uncertain, and what level of confidence we should place in those models. Uncertainty is not a weakness. Understanding uncertainty is a strength, and a key part of using any model, including climate models. One key message is that the level of confidence depends on the questions we ask. What are we certain about in the future and why? What are we less certain of and why? For policy-makers, this is a critical issue. Understanding how climate models work and how we get there is an important step in making intelligent decisions using (or not using) these climate models. Climate models are being used not just to understand the earth system but also to provide input for policy decisions to address human-caused climate change. The direction of our environment and economy is dependent on policy options chosen based on results of these models.

The chapters in Part I serve as a basic primer on climate and climate change. We hope to give readers an appreciation for the complexity and even beauty of the complex earth system so that they can better understand how we simulate it. In Part II, we discuss the mechanics of models of the earth system: How they are built and what they are trying to represent. Models are built to simulate each region of the climate system (e.g., atmosphere, ocean, and land), critical processes within each region, as well as critical interactions between regions and processes. Finally, in Part IIL, we focus on uncertainties and probabilities in prediction, with a focus on understanding what is known, what is unknown, and the degree of certainty. We also discuss how climate models are evaluated. In the concluding chapter, we discuss what we know, what we may learn in the future, and why we should (or should not) use models.

Part I Basic Principles and the Problem of Climate Forecasts

Chapter 1 Key Concepts in Climate Modeling

In order to describe climate modeling and the climate system, it is necessary to have a common conception of exactly what we are trying to simulate, and what a model actually is. What is climate? What is a model? How do we measure the uncertainty in a model? This chapter introduces some key terms and concepts. We start with some basic definitions of climate and weather. Everyone will come to this book with a preconceived definition of what climate and weather are, but separating these concepts is important for understanding how modeling of climate and weather are similar and why they are different. It also makes sense to discuss what a model is. Even if we do not realize it, we use models all the time. So we describe a few different conceptual types of models and put climate models in context. Finally we introduce the concept of uncertainty. As we discuss later in the book, models may have errors and still be useful, but this requires understanding the errors (the uncertainties) and understanding where they come from. Most of these concepts are common to many types of modeling, and we provide examples throughout the text.

1.4 What Is Climate?

Climate is perhaps easiest to explain as the distribution of possible weather states. On any given day and in any given place, the history of weather events can be compiled into a distribution with probabilities of what the weather might be (see Fig. 1.1). This figure is called a probability distribution function, representing a probability distribution.’ The horizontal axis represents a value (e.g., temperature), and the vertical axis represents the probability (or frequency of occurrence) of that event's (i.e., a given temperature) occurring or having occurred. If based on observations, then the frequency can be the number of times a given temperature occurs. The higher the line, the more probable the event. The most frequent

"There is quite a bit of statistics in climate, by definition. For a technical background, a good reference is Devore, J. L. (2011). Probability and Statistics for Engineering and the Sciences, 8th ed. Duxbury, MA: Duxbury Press. Any specific aspect of statistics (e.g., standard deviation, probability distribution function) can be looked up on Wikipedia (www.wikipedia.com).

© The Author(s) 2016 3 A. Gettelman and R.B. Rood, Demystifying Climate Models, Earth Systems Data and Models 2, DOI 10.1007/978-3-662-48959-8 1

Frequency of Occurrence

Cold Mean T, Hot Value (e.g., temperature)

Fig. 1.1 A probability distribution function with the value on the horizontal axis, and the frequency of occurrence on the vertical axis occurrence is the highest probability (the mode). The total area under the line is the probability. If the total area is given a value of 1, then the area under each part of the curve is the fractional chance that an event exceeding some threshold will occur. In Fig. 1.1, the area to the right of point T, is the probability that the temperature will be greater than T,, which might be about 20 % of the curve. The mean, or expected value, is the weighted average of the points. It need not be the point with the highest frequency. The mean value is the point at which half the probability (50 96) is on one side of the mean, and half on the other. The median is the value at which half the points are on one side and half on the other.

Here is an important and obvious question: How can we predict the climate (for next season, next year, or 50 years from now) if we cannot predict the weather (in 5 or 10 days)? The answer is, we use probability: The climate is the distribution of probable weather. The weather is a particular location in that distribution, and it is conditional on the current state of the system. The chance of a hurricane hitting Miami next week depends mostly on whether one has formed or is forming, and if one has formed, whether it is heading in the direction of Miami. As another example, the chance of having a rainy day in Seattle in January is high. But, given a particular day in January, with a weather state that might be pushing storms well to the north or south, the probability of rain the next day might be very low. In 50 Januaries, though, the probability of rain would be high. So climate is the distribution of weather (sometimes unknown). Weather is a given state in that distribution (often uncertain).

In a probability distribution of climate, the probabilities and the curve change over time: In the middle latitudes, the chance of snow is higher in winter than in summer. The curves will look different from place to place: Some climates have narrow distributions (see Fig. 1.2a), which means the weather is often very close to the average. Think about Hawaii, where the average of the daily highs and lows do not change much over the course of the year or Alaska, where the daily highs and lows may be the same as in Hawaii in summer, but not in winter. For Alaska the annual distribution of temperature is a very broad distribution (more like Fig. 1.2b).

1.1 What Is Climate? 5 Fig. 1.2 Different probability (a) | Narrow distribution functions: a narrow, b wide and € skewed distributions (b) Wide

(c) Uneven (Skewed)

Of course, even in Hawaii, extreme events occur. For a distribution like precipitation, which is bounded at one end by zero (no precipitation) the distribution might be “skewed” (Fig. 1.2c) with a low frequency of high events marking the *extreme'. As events are more extreme (think about hurricanes like Katrina or Sandy), there are fewer such events in the historical record. There may even be possible extreme events that have not occurred. So our description of climate is incomplete or uncertain. This is particularly true for rare (low-probability) events. These events are also the events that cause the most damage.

One aspect of shifting distributions is that extremes can change a lot more than the mean value (see Fig. 1.3). The mean is the value at which the area is equal on each side of the distribution. The mean is the same as the median if the distribution is symmetric. Simply moving the distribution to the right or left causes the area (meaning, the probability) beyond some fixed threshold to increase (or decrease). If the curve represents temperature, then shifting it to warmer temperatures (Fig. 1.3a) decreases the chance of cold events and really increases the chance of warm events. But note that some cold events still occur. Also, you can change the distribution without changing the mean by making the distribution wider (or broader). The mathematical term for the width of a distribution is variance, a statistical term for variability. This situation is illustrated in Fig. 1.3b. The mean is unchanged in

Lots more 'hot' (some off scale: records) Less cold

More hot and cold: both off scale Same mean

Lots more hot, off scale Less cold

Fig. 1.3 Shifting probability distribution functions are illustrated in different ways going from the blue to red distribution. The thick lines are the distribution, the thin dashed lines are the mean of the distributions and the dotted lines are fixed points to illustrate probability. Shown is a increase in mean, b increase in variance (width), c increase in mean and variance

Fig. 1.3b, but the chance of exceeding a given threshold for warm or cold temperatures changes. In other words, the climate (particularly a climate extreme) changes dramatically, even if the mean stays the same. The change need not be symmetric: Hot may change more than cold (or vice versa). Figure 1.3c is not a symmetric distribution. The key is to see climate as the distribution, not as a fixed number (often the mean).

This brings us to the fundamental difference between weather and climate forecasting. In weather forecasting,” we need to know the current state of the system and have a model for projecting it forward. Often the model can be simple. One “model” we all use is called persistence: What is the weather now? It may be like that tomorrow. In many places (e.g., Hawaii), such a model is not bad, but sometimes it is horribly wrong (e.g., when a hurricane hits Hawaii). So we try to use more sophisticated models, now typically numerical ones. These models go by the name numerical weather prediction (NWP) models, and they are used to "forecast" the evolution of the earth system from its current state.

?Bor an overview of the history of weather forecasting, see Edwards, P. N. (2013). A Vast Machine: Computer Models, Climate Data, and the Politics of Global Warming. Cambridge, MA: MIT Press.

1.1 What Is Climate? 7

Climate forecasting uses essentially the same type of model. But the goal of climate forecasting is to characterize the distribution. This may mean running the model for a long time to describe the distribution of all possible weather states correctly. If you start up a weather model from two different states (two different days), you hope to get a different answer each time. But for climate, you want to get the same distribution (the same climate), regardless of when the model started. We return to these examples again later.

1.2 What Is a Model?

A model, in essence, is a representation of a system. A model can be physical (building blocks) or abstract (an image on paper like a plan, or in your head). Abstract models can also be mathematical (monetary or physical totals in a spreadsheet). Ordinary physics that describes how cars go (or, more importantly, how they stop suddenly) is a model for how the physical world behaves. Numbers themselves are abstract models. A financial statement is a model of the money and resource flows of a household, corporation or country. Models are all around us, and we use them to abstract, make tractable and understand our human and natural environment.

As a concrete example, think about different *models" of a building. There can be many types of models of a building. A physical model of the building would usually be at a smaller scale that you can hold in your hand. There are several different abstract models of a building, and they are used for different purposes. Architects and engineers produce building plans: two-dimensional representations of the building, used to construct and document the building. Some of these are highly detailed drawings of specific parts of the building, such as the exterior, or the electrical and plumbing systems. The engineer may have built not just a physical model of the building, but perhaps even a more detailed structural model designed to understand how the building will react to wind or ground motion (earthquakes). Increasingly, these models are “virtual”: The structure is simulated on a computer.

We are familiar with all these sorts of model, but there are other more abstract models that deal with flows and budgets of materials or money. The owner and builder also probably have a spreadsheet model of the costs of construction of the building. This financial model is not certain, because it is really an estimate (or forecast, see below) of all the different costs of construction. And, finally, the owner likely has another model of the financial operation of the building: the money borrowed to finance the building, any income from a commercial building, and the costs of maintenance and operations of the building. The operating plan is really a projection into the future: It depends on a lot of uncertainties, like the cost of electricity or the value of the income from a building. The projection depends on these inputs, which the spreadsheet does not try to predict. The prediction is conditional on the inputs.

Models All Around Us

Models are everywhere in our world. Many models are familiar and physical, such as a small-scale model of a building or a bridge, or a mockup of a satellite or an airplane. Some models we use every day are made up of numbers. Many people use a model with numbers to manage income and expenses, savings and debts; that is, a budget. When the model of a bridge is placed in a computer-assisted design program, or a financial budget is put into a spreadsheet on a personal computer, then one has a "numerical" model. These models have a set of mathematical equations that behave with a specific set of rules, principles or laws.

Climate models are numerical models that calculate budgets of mass, momentum (velocity) and energy based on the physical laws of conservation. For example, energy is conserved (neither created or destroyed) and can, therefore, be counted. The physical laws on which climate models are based are discussed in more detail in Chap. 4. Weather and climate are dynamical systems; that is, they evolve over time. We rely on models of dynamical systems for many aspects of modern life. Here we illustrate a few examples of models that affect our everyday lives.

Climate models are closely related to weather forecast models; both simulate how fluids (air or water) move and interact, and how they exchange heat. An obvious example of a model that affects daily life is the weather forecast model, which is used in planning by individuals, governments, corporations and finance. The exact same physical principles of fluid flow are used to simulate a process in a chemical plant that takes different substances as liquids or gasses, reacts them together under controlled temperature and pressure, and produces new substances. Water and sewage treatment plants share similar principles and models. Internal combustion engines used for cars, trucks, ships and power plants are developed using models to understand how fuel enters the engine and produces heat, and that heat produces motion. Airplanes are also developed using modeling of the airflow around an aircraft. In this case, computational modeling has largely replaced design using wind tunnels. All of these models involve fluid flow and share physical principles with climate models. The details of the problem, for example, flow in pipes as contrasted to flow in the free atmosphere, define the specific requirements for the model construction.

So do you trust a model? Intuitively, we trust models all the time. You are using the results of a model every time you start your car, flush your toilet, turn on a light switch or get in an airplane. You count on models when you drive over a bridge. When NASA sends a satellite or rover to another planet, the path and behavior of the space vehicle relies on a model of simple physics describing complex systems.

Models do not just describe physical objects. Models of infectious diseases played an important role in management of the 2014 outbreak of Ebola. One function of models is to provide plausible representations of events to come,

12 What Is a Model?

and then to place people into those plausible futures. It is a way to anticipate and manage complexity. Though most times these models do not give an exact story of the future, the planning and decision making that comes from these modeling exercises improves our ability to anticipate the unexpected and to manage risk. Think of modeling as a virtual, computational world in which to exercise the practice of trial and error, and therefore, a method for reducing the "error" in trial and error. Reducing these errors saves lives and property. Models reduce the chance of errors: Airplanes do not regularly fall out of the sky, bridges do not normally collapse and chemical plants do not typically leak.

Ultimately, trust of models is anchored in evaluation of models compared to observations and experiences. Weather models are evaluated every day with billions of observations as well as billions of individuals" experiences. Trust is often highly personal. By many objective measures, weather models have remarkable accuracy, for example, letting a city know more than five days in advance that a major tropical cyclone is likely to make landfall near that city. Of course, if the tropical cyclone makes landfall just 60 miles (100 km) away from the city, many people might conclude the models cannot be trusted.

Objectively, however, a model that simulated the tropical cyclone and represented its evolution with an error of 60 miles (100 km) on a globe that spans many thousands of miles can, also, be construed as being quite accurate. This represents the fact that models provide plausible futures that inform decision making.

Like weather models, climate models are evaluated with billions of Observations and investigations of past events. The results of models have been scrutinized by thousands of scientists and practitioners. With virtual certainty, we know the Earth will warm, sea level will rise, ice will melt and weather will change. They provide plausible futures, not prescribed futures. There are uncertainties, and there will always be uncertainties. However, our growing experiences and vigilant efforts to evaluate and improve will help us to understand, manage and, sometimes, reduce uncertainty. As the models improve, trust and usability increase. There remains some uncertainty in most physical models, but that can be accounted for, and we discuss uncertainty, and its value in modeling, at length.