Bibliography - Klaus Keller

1. Baehr, J., D. McInerney, Klaus Keller, and J. Marotzke, 2008: Optimization of an observing system design for the North Atlantic meridional overturning circulation. Journal of Atmospheric and Oceanic Technology, 25(4), doi:10.1175/2007JTECHO535.1 625-634
   [ Abstract ]

   Three methods are analyzed for the design of ocean observing systems to monitor the meridional over-turning circulation (MOC) in the North Atlantic. Specifically, a continuous monitoring array to monitor the MOC at 1000 m at different latitudes is "deployed" into a numerical model. The authors compare array design methods guided by (i) physical intuition (heuristic array design), (ii) sequential optimization, and (iii) global optimization. The global optimization technique can recover the true global solution for the analyzed array design, while gradient-based optimization would be prone to misconverge. Both global optimization and heuristic array design yield considerably improved results over sequential array design. Global optimization always outperforms the heuristic array design in terms of minimizing the root-mean-square error. However, whether the results are physically meaningful is not guaranteed; the apparent success might merely represent a solution in which misfits compensate for each other accidentally. Testing the solution gained from global optimization in an independent dataset can provide crucial information about the solution's robustness.

2. Brennan, C. E., R. J. Matear, and Klaus Keller, 2008: Measuring oxygen concentrations improves the detection capabilities of an ocean circulation observation array. Journal of Geophysical Research – Oceans, 113(C02019), doi:10.1029/2007JC004113
   [ Abstract ]

   The North Atlantic meridional overturning circulation (MOC) may weaken or even collapse in response to anthropogenic climate forcing, with potentially nontrivial socioeconomic impacts. One currently implemented MOC observation system uses temperature and salinity (as well as other) observations along a zonal transect in the North Atlantic. The resulting MOC estimate has, however, a relatively low signal-to-noise ratio due to large internal variability and observation errors. Observations of hydrographic tracers that are mechanistically linked to MOC changes may increase the signal-to-noise ratio. A MOC slowdown is associated in model simulations with a shoaling of the boundary between North Atlantic Deep Water and Antarctic Bottom Water. This shoaling results in detectable trends in water mass tracers. Here we deploy a virtual observation array into a numerical model starting in model year 2006 to test whether observing the apparent oxygen utilization (AOU) in addition to the MOC estimate improves detection capabilities. Our detection method accounts for observation errors, autocorrelated variability, and uncertainty about the initial conditions. Neglecting the effects of observation errors and the uncertainty about the initial conditions results in artificially early detection times. The MOC signal alone enables reliable detection in roughly five decades. Adding AOU observations reduces this detection time by approximately 40%.

3. Keller, Klaus, and D. McInerney, 2008: The dynamics of learning about a climate threshold. Climate Dynamics, 30(2-3), doi:10.1007/s00382-007-0290-5 321-332
   [ Abstract ]

   Anthropogenic greenhouse gas emissions may trigger threshold responses of the climate system. One relevant example of such a potential threshold response is a shutdown of the North Atlantic meridional overturning circulation (MOC). Numerous studies have analyzed the problem of early MOC change detection (i.e., detection before the forcing has committed the system to a threshold response). Here we analyze the early MOC prediction problem. To this end, we virtually deploy an MOC observation system into a simple model that mimics potential future MOC responses and analyze the timing of confident detection and prediction. Our analysis suggests that a confident prediction of a potential threshold response can require century time scales, considerably longer that the time required for confident detection. The signal enabling early prediction of an approaching MOC threshold in our model study is associated with the rate at which the MOC intensity decreases for a given forcing. A faster MOC weakening implies a higher MOC sensitivity to forcing. An MOC sensitivity exceeding a critical level results in a threshold response. Determining whether an observed MOC trend in our model differs in a statistically significant way from an unforced scenario (the detection problem) imposes lower requirements on an observation system than the determination whether the MOC will shut down in the future (the prediction problem). As a result, the virtual observation systems designed in our model for early detection of MOC changes might well fail at the task of early and confident prediction. Transferring this conclusion to the real world requires a considerably refined MOC model, as well as a more complete consideration of relevant observational constraints.

4. Keller, Klaus, G. Yohe, and M. Schlesinger, 2008: Managing the Risks of Climate Thresholds: Uncertainties and Information Needs. Climatic Change, 91(1-2), doi:10.1007/s10584-006-9114-6 5-10
   [ Abstract ]

   Human activities are driving atmospheric greenhouse-gas concentrations beyond levels experienced by previous civilizations. The uncertainty surrounding our understanding of the resulting climate change poses nontrivial challenges for the design and implementation of strategies to manage the associated risks. One challenge stems from the fact that the climate system can react abruptly and with only subtle warning signs before climate thresholds have been crossed (Stocker 1999; Alley et al. 2003). Model predictions suggest that anthropogenic greenhouse-gas emissions increase the likelihood of crossing these thresholds (Cubasch and Meehl 2001; Yohe et al. 2006). Coping with deep uncertainty in our understanding of the mechanisms, locations, and impacts of climate thresholds presents another challenge. Deep uncertainty presents itself when the relevant range of systems models and the associated probability density functions for their parameterizations are unknown and/or when decision-makers strongly disagree on their formulations (Lempert 2002). Furthermore, the requirements for creating feasible observation and modeling systems that could deliver confident and timely prediction of impending threshold crossings are mostly unknown. These challenges put a new emphasis on the analysis, design, and implementation of Earth observation systems and strategies to manage the risks of potential climate threshold responses.

5. Keller, Klaus, D. McInerney, and David F. Bradford, 2008: Carbon dioxide sequestration: When and how much? Climatic Change, 88(3-4), doi:10.1007/s10584-008-9417-x 267-291
   [ Abstract ]

   Carbon dioxide (CO₂) sequestration has been proposed as a key component in technological portfolios for managing anthropogenic climate change, since it may provide a faster and cheaper route to significant reductions in atmospheric CO₂ concentrations than abating CO₂ production. However, CO₂ sequestration is not a perfect substitute for CO₂ abatement because CO₂ may leak back into the atmosphere (thus imposing future climate change impacts) and because CO₂ sequestration requires energy (thus producing more CO₂ and depleting fossil fuel resources earlier). Here we use analytical and numerical models to assess the economic efficiency of CO₂ sequestration and analyze the optimal timing and extent of CO₂ sequestration. The economic efficiency factor of CO₂ sequestration can be expressed as the ratio of the marginal net benefits of sequestering CO₂ and avoiding CO₂ emissions. We derive an analytical solution for this efficiency factor for a simplified case in which we account for CO₂ leakage, discounting, the additional fossil fuel requirement of CO₂ sequestration, and the growth rate of carbon taxes. In this analytical model, the economic efficiency of CO₂ sequestration decreases as the CO₂ tax growth rate, leakage rates and energy requirements for CO₂ sequestration increase. Increasing discount rates increases the economic efficiency factor. In this simple model, short-term sequestration methods, such as afforestation, can even have negative economic efficiencies. We use a more realistic integrated-assessment model to additionally account for potentially important effects such as learning-by-doing and socio-economic inertia on optimal strategies. We measure the economic efficiency of CO₂ sequestration by the ratio of the marginal costs of CO₂ sequestration and CO₂ abatement along optimal trajectories. We show that the positive impacts of investments in CO₂ sequestration through the reduction of future marginal CO₂ sequestration costs and the alleviation of future inertia constraints can initially exceed the marginal sequestration costs. As a result, the economic efficiencies of CO₂ sequestration can exceed 100% and an optimal strategy will subsidize CO₂ sequestration that is initially more expensive than abatement. The potential economic value of a feasible and acceptable CO₂ sequestration technology is equivalent – in the adopted utilitarian model – to a one-time investment of several percent of present gross world product. It is optimal in the chosen economic framework to sequester substantial CO₂ quantities into reservoirs with small or zero leakage, given published estimates of marginal costs and climate change impacts. The optimal CO₂ trajectories in the case of sequestration from air can approach the pre-industrial level, constituting geoengineering. Our analysis is silent on important questions (e.g., the effects of model and parametric uncertainty, the potential learning about these uncertainties, or ethical dimension of such geoengineering strategies), which need to be addressed before our findings can be translated into policy-relevant recommendations.

6. McInerney, D., and Klaus Keller, 2008: Economically optimal risk reduction strategies in the face of uncertain climate thresholds. Climatic Change, 91(1-2), doi:10.1007/s10584-006-9137-z 29-41
   [ Abstract ]

   Anthropogenic greenhouse gas emissions may trigger climate threshold responses, such as a collapse of the North Atlantic meridional overturning circulation (MOC). Climate threshold responses have been interpreted as an example of “dangerous anthropogenic interference with the climate system” in the sense of the United Nations Framework Convention on Climate Change (UNFCCC). One UNFCCC objective is to “prevent” such dangerous anthropogenic interference. The current uncertainty about important parameters of the coupled natural–human system implies, however, that this UNFCCC objective can only be achieved in a probabilistic sense. In other words, climate management can only reduce – but not entirely eliminate – the risk of crossing climate thresholds. Here we use an integrated assessment model of climate change to derive economically optimal risk-reduction strategies. We implement a stochastic version of the DICE model and account for uncertainty about four parameters that have been previously identified as dominant drivers of the uncertain system response. The resulting model is, of course, just a crude approximation as it neglects, for example, some structural uncertainty, and focuses on a single threshold, out of many potential climate responses. Subject to this caveat, our analysis suggests five main conclusions. First, reducing the numerical artifacts due to sub-sampling the parameter probability density functions to reasonable levels requires sample sizes exceeding 10³. Conclusions of previous studies that are based on much smaller sample sizes may hence need to be revisited. Second, following a business-as-usual (BAU) scenario results in odds for an MOC collapse in the next 150 years exceeding 1 in 3 in this model. Third, an economically “optimal” strategy (that maximizes the expected utility of the decision-maker) reduces carbon dioxide (CO₂) emissions by approximately 25% at the end of this century, compared with BAU emissions. Perhaps surprisingly, this strategy leaves the odds of an MOC collapse virtually unchanged compared to a BAU strategy. Fourth, reducing the odds for an MOC collapse to 1 in 10 would require an almost complete decarbonization of the economy within a few decades. Finally, further risk reductions (e.g., to 1 in 100) are possible in the framework of the simple model, but would require faster and more expensive reductions in CO₂ emissions.

7. Keller, Klaus, L. I. Miltich, A. Robinson, and R.S.J. Tol, 2007: How Overconfident are Current Projections of Anthropogenic Carbon Dioxide Emissions? Working Papers, Paper 39, http://www.bepress.com/cgi/viewcontent.cgi?article=1095&context=feem,
   [ Abstract ]

   Analyzing the risks of anthropogenic climate change requires sound probabilistic projections of CO₂ emissions. Previous projections have broken important new ground, but many rely on out-of-range projections, are limited to the 21^st century, or provide only implicit probabilistic information. Here we take a step towards resolving these problems by assimilating globally aggregated observations of population size, economic output, and CO₂ emissions over the last three centuries into a simple economic model. We use this model to derive probabilistic projections of business-as-usual CO₂ emissions to the year 2150. We demonstrate how the common practice to limit the calibration timescale to decades can result in biased and overconfident projections. The range of several CO₂ emission scenarios (e.g., from the Special Report on Emission Scenarios) misses potentially important tails of our projected probability density function. Studies that have interpreted the range of CO₂ emission scenarios as an approximation for the full forcing uncertainty may well be biased towards overconfident climate change projections.

8. Keller, Klaus, C. Deutsch, M. G. Hall, and David F. Bradford, 2007: Early detection of changes in the North Atlantic meridional overturning circulation: Implications for the design of ocean observation systems. Journal of Climate, 20, doi:10.1175/JCLI3993.1 145-157
   [ Abstract ]

   Many climate models predict that anthropogenic greenhouse gas emissions may cause a threshold response of the North Atlantic meridional overturning circulation (MOC). These model predictions are, however, uncertain. Reducing this uncertainty can have an economic value, because it would allow for the design of more efficient risk management strategies. Early information about the MOC sensitivity to anthropogenic forcing (i.e., information that arrives before the system is committed to a threshold response) could be especially valuable. Here the focus is on one particular kind of information: the detection of anthropogenic MOC changes. It is shown that an MOC observation system based on infrequent (decadal scale) hydrographic observations may well fail in the task of early MOC change detection. This is because this system observes too infrequently and the observation errors are too large. More frequent observations and reduced observation errors would result in earlier detection. It is also shown that the economic value of information associated with a confident and early prediction of an MOC threshold response could exceed the costs of typically implemented ocean observation systems by orders of magnitude. One open challenge is to identify a feasible observation system that would enable such a confident and early MOC prediction across the range of possible MOC responses.

9. Keller, Klaus, S.-R. Kim, J. Baehr, David F. Bradford, and Michael Oppenheimer, 2007: What is the economic value of information about climate thresholds? Human-Induced Climate Change: An Interdisciplinary Assessment, Cambridge University Press, Cambridge, MA, http://www.geosc.psu.edu/~kkeller/Keller_snowmass_pp_07.pdf, (Chapter 28), 343-354
   [ Abstract ]

   The field of integrated assessment of climate change is undergoing a paradigm shift towards the analysis of potentially abrupt and irreversible climate changes (Alley et al., 2003; Keller et al., 2006b). Early integrated studies broke important new ground in exploring the relationship between the costs and benefits of reducing carbon dioxide (CO₂) emissions (e.g., Nordhaus, 1991; Manne and Richels, 1991; or Tol, 1997). These studies project the climate response to anthropogenic CO₂ emissions to be relatively smooth and trypically conclude that the projected benefits of reducing CO₂ emissions would justify only small reductions in CO₂ emissions in a cost-benefit framework. The validity of the often-assumed smooth climate response is, however, questionable, given how that climate system has responded to forcing in the geological past. Before the Anthropocene, the geological time period where humans have started to influence the global biogeochemical cycles considerably (Crutzen, 2002) , the predominant responses of the climate system were forced by small changes in solar insolation occurring on timescales of thousands of years (Berger and Loutre, 1991). Yet this slow and smooth forcing apparently triggered abrupt climate changes - a threshold response where the climate system moved between different basins of attraction (Berger, 1990; Clement et al., 2001). Anthropogenic forcing may trigger climate threshold responses in the future (Alley et al., 2003; Keller et al., 2006b). Examples of such threshold responses include (i) a collapse of the North Atlantic meridional overturning circulation (Rahmstorf, 2000; Stommel, 1961), (ii) a disintegration of the West Antarctic Ice Sheet (Mercer, 1978; Oppenheimer, 1998), (iii) abrupt vegetation changes (Claussen et al., 1999; Scheffer et al., 2001), or (iv) changes in properties of the El Nino Southern Oscillation, ENSO (Fedorov and Philander, 2000; Timmermann, 201). Here we focus on the first two examples: a possible collapse of the North Atlantic meridional overturning circulation and a possible disintegration of the West Antarctic Ice Sheet. Our analysis address three main questions. (i) What underlying mechanisms define a climate threshold? (ii) What are the key scientific uncertainties in predicting whether these thresholds may be crossed in the future? (iii) What might be reasonable order-of-magnitude estimates of the expected economic value of reducing these uncertainties? Our analysis suggests that climate strategies designed to reduce the risk of crossing climate thresholds have to be designed in the face of large uncertainties. Some of the key uncertainties (e.g., the sensitivity of the meridional overturning circulation to anthropogenic greenhouse gas emissions) can be reduced by observation systems. We conclude by identifying future research needs.

10. Keller, Klaus, A. Robinson, David F. Bradford, and Michael Oppenheimer, 2007: The regrets of procrastination in climate policy. Environmental Research Letters, (Lett 2, No. 2), doi:10.1088/1748-9326/2/2/024004
    [ Abstract ]

    Anthropogenic carbon dioxide (CO₂) emissions are projected to impose economic costs due to the associated climate change impacts. Climate change impacts can be reduced by abating CO₂ emissions. What would be an economically optimal investment in abating CO₂ emissions? Economic models typically suggest that reducing CO₂ emissions by roughly ten to twenty per cent relative to business-as-usual would be an economically optimal strategy. The currently implemented CO₂ abatement of a few per cent falls short of this benchmark. Hence, the global community may be procrastinating in implementing an economically optimal strategy. Here we use a simple economic model to estimate the regrets of this procrastination—the economic costs due to the suboptimal strategy choice. The regrets of procrastination can range from billions to trillions of US dollars. The regrets increase with increasing procrastination period and with decreasing limits on global mean temperature increase. Extended procrastination may close the window of opportunity to avoid crossing temperature limits interpreted by some as ‘dangerous anthropogenic interference with the climate system’ in the sense of Article 2 of the United Nations Framework Convention on Global Climate Change.

11. Budescu, D. V., R. Lempert, S. Broomell, and Klaus Keller, 2006: Aided and unaided decision making with imprecise probabilities. , http://www.geosc.psu.edu/~kkeller/Budescu_jep_09.pdf,
    [ Abstract ]

    We report results of a series of experiments on decision-making in the presence of irreducibly imprecise probabilities. The choices faced by the subjects resemble loosely the policy choices faced by policy makers in the presence of the threat of abrupt climate change. Consistent with the vagueness avoidance hypothesis, subjects displayed systematic preferences for riskless actions even at a high premium. This tendency increased as a function of the level of vagueness, characterized by the width of an interval of plausible probabilities. The vagueness avoidance tendency was reduced significantly when the subjects had access to one of two different decisions aids with distinct approaches to imprecision. The “Summary” aid provides quantitative comparisons of expected values of alternative actions while the “Display” aid graphically compares the performance of actions over the entire range of plausible probabilities. Although the decisions made in the presence of the two decisions aids were very similar in most respects, we found evidence for an interaction between the presentation format (operationalized by the type of decision aid used) and the subsequent decisions. Exposure to the graphical displays caused more subjects to favor the action that was perceived as superior for larger portions of the probability range, compared to subjects who had access to an expected value calculator. These findings suggest some initial implications for the debate over how to best characterize imprecise probabilistic information for policy-makers involved with climate change.

12. O'Neill, B., P. Crutzen, A. Grübler, M. H. Duong, Klaus Keller, C. Kolstad, A. Lange, M. Obersteiner, Michael Oppenheimer, W. Pepper, W. Sanderson, and M. Schlesinger, et al., 2006: Learning and Climate Change. Climate Policy, http://www.geosc.psu.edu/~kkeller/O%27Neill_cp_07.pdf, 6,
    [ Abstract ]

    Learning – i.e. the acquisition of new information that leads to changes in our assessment of uncertainty – plays a prominent role in the international climate policy debate. For example, the view that we should postpone actions until we know more continues to be influential. The latest work on learning and climate change includes new theoretical models, better informed simulations of how learning affects the optimal timing of emissions reductions, analyses of how new information could affect the prospects for reaching and maintaining political agreements and for adapting to climate change, and explorations of how learning could lead us astray rather than closer to the truth. Despite the diversity of this new work, a clear consensus on a central point is that the prospect of learning does not support the postponement of emissions reductions today.

13. Keller, Klaus, M. G. Hall, S.-R. Kim, David F. Bradford, and Michael Oppenheimer, 2005: Avoiding dangerous anthropogenic interference with the climate system. Climatic Change, 73(3), doi:10.1007/s10584-005-0426-8 227-238
    [ Abstract ]

    The UN Framework Convention on Climate Change calls for the avoidance of “dangerous anthropogenic interference with the climate system”. Among the many plausible choices, dangerous interference with the climate system may be interpreted as anthropogenic radiative forcing causing distinct and widespread climate change impacts such as a widespread demise of coral reefs or a disintegration of the West Antarctic ice sheet. The geological record and numerical models suggest that limiting global warming below critical temperature thresholds significantly reduces the likelihood of these eventualities. Here we analyze economically optimal policies that may ensure this risk-reduction. Reducing the risk of a widespread coral reef demise implies drastic reductions in greenhouse gas emissions within decades. Virtually unchecked greenhouse gas emissions to date (combined with the inertia of the coupled natural and human systems) may have already committed future societies to a widespread demise of coral reefs. Policies to reduce the risk of a West Antarctic ice sheet disintegration allow for a smoother decarbonization of the economy within a century and may well increase consumption in the long run.

14. Min, D.-H., and Klaus Keller, 2005: Errors in estimated temporal tracer trends due to changes in the historical observation network: A case study of oxygen trends in the Southern Ocean. Ocean and Polar Research, http://www.geosc.psu.edu/~kkeller/Min_and_Keller_opr_05.pdf, 27(2), 189-195
    [ Abstract ]

    Several models predict large and potentially abrupt ocean circulation changes due to anthropogenic greenhouse-gas emissions. These circulation changes drive-in the models-considerable oceanic oxygen trend. A sound estimate of the observed oxygen trends can hence be a powerful tool to constrain predictions of future changes in oceanic deepwater formation, heat and carbon dioxide uptake. Estimating decadal scale oxygen trends is, however, a nontrivial task and previous studies have come to contradicting conclusions. One key potential problem is that changes in the historical observation network might introduce considerable errors. Here we estimate the likely magnitude of these errors for a subset of the available observations in the Southern Ocean. We test three common data analysis methods south of Australia and focus on the decadal-scale trends between the 1970’s and the 1990’s. Specifically, we estimate errors due to sparsely sampled observations using a known signal (the time invariant, temporally averaged, World Ocean Atlas 2001) as a negative control. The crossover analysis and the objective analysis methods are far less prone to spatial sampling location biases than the area averaging method. Subject to numerous caveats, we find that errors due to sparse sampling for the area averaging method are on the order of several micromoles kg⁻¹. For the crossover and the objective analysis method, these errors are much smaller. For the analyzed example, the biases due to changes in the spatial design of the historical observation network are relatively small compared to the trends predicted by many model simulations. This raises the possibility to use historic oxygen trends to constrain model simulations, even in sparsely sampled ocean basins.

15. Keller, Klaus, B. M. Bolker, and David F. Bradford, 2004: Uncertain climate thresholds and optimal economic growth. Journal of Environmental Economics and Management, 48(1), doi:10.1016/j.jeem.2003.10.003 723-741
    [ Abstract ]

    We explore the combined effects of a climate threshold (a potential ocean thermohaline circulation collapse), parameter uncertainty, and learning in an optimal economic growth model. Our analysis shows that significantly reducing carbon dioxide (CO₂) emissions may be justified to avoid or delay even small (and arguably realistic) damages from an uncertain and irreversible climate change—even when future learning about the system is considered. Parameter uncertainty about the threshold specific damages and the CO₂ level triggering a threshold can act to decrease near-term CO₂ abatements that maximize expected utility.

16. Kim, S.-R., Klaus Keller, and David F. Bradford, July 2004: Optimal Technological Portfolios for Climate-Change Policy under Uncertainty: A Computable General Equilibrium Approach. Computing in Economics and Finance 2004, Society for Computational Economics, http://ideas.repec.org/p/sce/scecf4/140.html,
    [ Abstract ]

    When exploring solutions to long-term environmental problems such as climate change, it is crucial to understand how the rates and directions of technological change may interact with environmental policies in the presence of uncertainty. This paper analyzes optimal technological portfolios for global carbon emissions reductions in an integrated assessment model of the coupled social-natural system. The model used here is a probabilistic, two-technology extension of Nordhaus’ earlier model (Nordhaus and Boyer, 2000) by incorporating endogenous technological choice between conventional and carbon-free technologies. Taking into account the possible competitions among the technological options, we address the issues of optimal timing, costs and burden-sharing of optimal carbon mitigation strategies in the inherently uncertain world. We perform various analyses related to the major uncertainties about natural, socioeconomic and technological parameters, and investigate the effects of uncertainties resolution, risks and alternative political preferences. The results show that analyses ignoring uncertainty could lead to inefficient and biased technology-policy recommendations for the future.

17. Min, D.-H., and Klaus Keller, 2004: How robust are estimated decadal-scale trends in dissolved oxygen concentrations in the Southern Ocean with respect to methodological assumptions? EOS Trans. AGU, 84(52),
    [ Abstract ]

    The Southern Ocean (SO) plays a central role in deep water formation as well as oceanic heat and carbon dioxide uptake. Several modeling studies suggest considerable circulation changes in the SO over the past few decades in response to anthropogenic carbon dioxide emissions. These circulation changes are accompanied - in the models - by a substantial decrease of dissolved oxygen concentrations. Detailed analysis of temporal and spatial changes of dissolved oxygen in the SO may hence be a promising method to detect the ocean's response to climate change. The few observational studies estimating decadal scale oxygen trends in the SO come, however, to different conclusions. Here, we analyze different statistical methods to estimate decadal scale [O₂] trends in the SO. In a companion study (Keller et al., 2004), we use the most robust method to estimate SO oxygen trends.

18. Moles, C. G., J. R. Banga, and Klaus Keller, 2004: Solving nonconvex climate control problems: Pitfalls and algorithm performances. Applied Soft Computing, 5(1), doi:10.1016/j.asoc.2004.03.011 35-44
    [ Abstract ]

    Global optimization can be used as the main component for reliable decision support systems. In this contribution, we explore numerical solution techniques for nonconvex and nondifferentiable economic optimal growth models. As an illustrative example, we consider the optimal control problem of choosing the optimal greenhouse gas emissions abatement to avoid or delay abrupt and irreversible climate damages. We analyze a number of selected global optimization methods, including adaptive stochastic methods, evolutionary computation methods and deterministic/hybrid techniques. Differential evolution (DE) and one type of evolution strategy (SRES) arrived to the best results in terms of objective function, with SRES showing the best convergence speed. Other simple adaptive stochastic techniques were faster than those methods in obtaining a local optimum close to the global solution, but mis-converged ultimately.

19. Keller, Klaus, D.-H. Min, and R. M. Key, 2003: Detecting decadal-scale oxygen trends in the Southern Ocean. EOS Trans. AGU, 84(52),
    [ Abstract ]

    Several modeling studies suggest a significant oxygen loss of the Southern Ocean over the last few decades. The main cause is a decrease in Southern Ocean ventilation due to climate change. The model predictions of generally decreasing Southern Ocean oxygen concentrations are consistent with the few available observational studies in a small fraction of the Southern Ocean. The observational studies suggest, however, different mechanisms to explain the decreased oxygen concentrations (e.g. changes in water mass fractions or changes in the water density structure). A mechanistic understanding of the Southern Ocean oxygen inventory is critical to predictions of the oceanic response to global climate change. For example, a decreased Southern Ocean ventilation in response to anthropogenic carbon dioxide emissions would cause a decreased oceanic heat and carbon dioxide uptake - a positive climate feedback. Further, an oceanic oxygen loss would bias estimates of the oceanic carbon dioxide sink derived from atmospheric oxygen observations. Here, we estimate decadal-scale dissolved oxygen trends in the Southern Ocean and their statistical relevance using the WOD 2001 hydrographic database. We explore several hypotheses regarding the driving mechanisms for the observed concentration trends and estimate the changes in the total oxygen inventory of the Southern Ocean.

20. Kraepiel, A. M., Klaus Keller, H. B. Chin, E. G. Malcolm, and Francois Morel, 2003: Sources and Variations of Mercury in Tuna. Environmental Science and Technology, 37, doi:10.1021/es0340679 5551-5558
    [ Abstract ]

    While the bulk of human exposure to mercury is through the consumption of marine fish, most of what we know about mercury methylation and bioaccumulation is from studies of freshwaters. We know little of where and how mercury is methylated in the open oceans, and there is currently a debate whether methylmercury concentrations in marine fish have increased along with global anthropogenic mercury emissions. Measurements of mercury concentrations in Yellowfin tuna caught off Hawaii in 1998 show no increase compared to measurements of the same species caught in the same area in 1971. On the basis of the known increase in the global emissions of mercury over the past century and of a simple model of mercury biogeochemistry in the Equatorial and Subtropical Pacific ocean, we calculate that the methylmercury concentration in these surface waters should have increased between 9 and 26% over this 27 years span if methylation occurred in the mixed layer or in the thermocline. Such an increase is statistically inconsistent with the constant mercury concentrations measured in tuna. We conclude tentatively that mercury methylation in the oceans occurs in deep waters or in sediments.

21. Moles, C. G., A. S. Lieber, J. R. Banga, and Klaus Keller, 2003: Global optimization of climate control problems using evolutionary and stochastic algorithms. Advances in Soft Computing: Engineering Design and Manufacturing, Springer-Verlag, Heidelberg, http://www.geosc.psu.edu/~kkeller/Moles_asc_02.pdf, 331-342
    [ Abstract ]

    Global optimization can be used as the main component for reliable decision support systems. In this contribution, we explore numerical solution techniques for nonconvex and nondifferentiable economic optimal growth models. As an illustrative example, we consider the optimal control problem of choosing the optimal greenhouse gas emissions abatement to avoid or delay abrupt and irreversible climate damages. We analyze a number of selected global optimization methods, including adaptive stochastic methods, evolutionary computation methods and deterministic/ hybrid techniques. Differential evolution (DE) and one type of evolution strategy (SRES) arrived to the best results in terms of objective function, with SRES showing the best convergence speed. Other simple adaptive stochastic techniques were faster than those methods in obtaining a local optimum close to the global solution, but misconverged ultimately.

22. Keller, Klaus, R. D. Slater, Michael Bender, and R. M. Key, 2001: Possible biological or physical explanations for decadal scale trends in North Pacific nutrient concentrations and oxygen utilization. Deep Sea Research II, 49(103), doi:10.1016/S0967-0645(01)00106-0 345-362
    [ Abstract ]

    We analyze North Pacific GEOSECS (1970s) and WOCE (1990s) observations to examine potential decadal trends of the marine biological carbon pump. Nitrate concentrations {[NO₃]} and apparent oxygen utilization (AOU) decreased significantly in intermediate waters (by - 0.6 and - 2.9 μmol kg^-1; respectively, at σ_θ = 27.4 kg m^-3; corresponding to ≈ 1050 m). In shallow waters (above roughly 750 m) [NO₃] and AOU increased, though the changes were not statistically significant. A sensitivity study with an ocean general circulation model indicates that reasonable perturbations of the biological carbon pump due to changes in export production or remineralization efficiency are insufficient to account for the intermediate water tracer trends. However, changes in water ventilation rates could explain the intermediate water tracer trends and would be consistent with trends of water age derived from radiocarbon. Trends in AOU and [NO₃] provide relatively poor constraints on decadal scale trends in the marine biological carbon pump for two reasons. First, most of the expected changes due to decadal scale perturbations of the marine biota occur in shallow waters, where the available data are typically too sparse to account for the strong spatial and temporal variability. Second, alternative explanations for the observed tracer trends (e.g., changes in the water ventilation rates) cannot be firmly rejected. Our data analysis does not disprove the null hypothesis of an unchanged biological carbon pump in the North Pacific.

23. Baehr, J., Klaus Keller, and J. Marotzke, 0000: Detecting potential changes in the meridional overturning circulation at 26°N in the Atlantic. Climatic Change, doi:10.1007/s10584-006-9153-z
    [ Abstract ]

    We analyze the ability of an oceanic monitoring array to detect potential changes in the North Atlantic meridional overturning circulation (MOC). The observing array is ‘deployed’ into a numerical model (ECHAM5/MPI-OM), and simulates the measurements of density and wind stress at 26°N in the Atlantic. The simulated array mimics the continuous monitoring system deployed in the framework of the UK Rapid Climate Change program. We analyze a set of three realizations of a climate change scenario (IPCC A1B), in which – within the considered time-horizon of 200 years – the MOC weakens, but does not collapse. For the detection analysis, we assume that the natural variability of the MOC is known from an independent source, the control run. Our detection approach accounts for the effects of observation errors, infrequent observations, autocorrelated internal variability, and uncertainty in the initial conditions. Continuous observation with the simulated array for approximately 60 years yields a statistically significant (p < 0.05) detection with 95 percent reliability assuming a random observation error of 1 Sv (1 Sv = 10⁶ m³s⁻¹). Observing continuously with an observation error of 3 Sv yields a detection time of about 90 years (with 95 percent reliability). Repeated hydrographic transects every 5 years/ 20 years result in a detection time of about 90 years/120 years, with 95 percent reliability and an assumed observation error of 3 Sv. An observation error of 3 Sv (one standard deviation) is a plausible estimate of the observation error associated with the RAPID UK 26°N array.

Direct link to page: http://cmi.princeton.edu/bibliography/results.php?author=3741