Publications
64 results found
Watson-Parris D, Rao Y, Olivie D, et al., 2022, ClimateBench v1.0: A Benchmark for Data-Driven Climate Projections, JOURNAL OF ADVANCES IN MODELING EARTH SYSTEMS, Vol: 14
- Author Web Link
- Cite
- Citations: 1
Weng X, Forster GL, Nowack P, 2022, A machine learning approach to quantify meteorological drivers of ozone pollution in China from 2015 to 2019, ATMOSPHERIC CHEMISTRY AND PHYSICS, Vol: 22, Pages: 8385-8402, ISSN: 1680-7316
- Author Web Link
- Cite
- Citations: 2
Watson-Parris D, Rao Y, Olivié D, et al., 2022, ClimateBench: A benchmark for data-driven climate projections
<jats:p>&lt;p&gt;Exploration of future emissions scenarios mostly relies on one-dimensional impulse response models, or simple pattern scaling approaches to approximate the physical climate response to a given scenario. Such approaches are unable to reliably predict climate variables which respond non-linearly to emissions or forcing (such as precipitation) and must rely on heavily simplified representations of e.g., aerosol, neglecting important spatial dependencies.&lt;/p&gt;&lt;p&gt;Here we present ClimateBench - a benchmark dataset based on a suite of CMIP, AerChemMIP and DAMIP simulations performed by NorESM2, and a set of baseline machine learning models that emulate its response to a variety of forcers. These surrogate models can skilfully predict annual mean global distributions of temperature, diurnal temperature range and precipitation (including extreme precipitation) given a wide range of emissions and concentrations of carbon dioxide, methane and spatially resolved aerosol. We discuss the accuracy and interpretability of these emulators and consider their robustness to physical constraints such as total energy conservation. Future opportunities incorporating such physical constraints directly in the machine learning models and using the emulators for detection and attribution studies are also discussed. This opens a wide range of opportunities to improve prediction, consistency and mathematical tractability.&lt;/p&gt;&lt;p&gt;We hope that by defining a clear baseline with appropriate metrics and providing a variety of baseline models we can bring the power of modern machine learning techniques to bear on the important problem of efficiently and robustly sampling future climates.&lt;/p&gt;</jats:p>
Tuncer D, Babacan O, Guiazon R, et al., 2022, Engineering data-driven solutions for future mobility: perspectives and challenges
The automotive industry is currently undergoing major changes. These includea general shift towards decarbonised mode of transportation, the implementationof mobility as an end-to-end service, and the transition to vehicles thatincreasingly rely on software and digital tools to function. Digitalisation isexpected to play a key role in shaping the future of mobility ecosystems byfostering the integration of traditionally independent system domains in theenergy, transportation and information sectors. This report discussesopportunities and challenges for engineering data-driven solutions that supportthe requirements of future digitalised mobility systems based on three usecases for electric vehicle public charging infrastructures, services andsecurity.
Weng X, Forster G, Nowack P, 2022, A machine learning approach to quantify meteorological drivers of recent ozone pollution in China
<jats:p>Abstract. Surface ozone concentrations have been increasing in many regions of China for the past few years, in contrast to policy-driven declines in other key air pollutants such as particulate matter. While the central role of meteorology in modulating ozone pollution is widely acknowledged, its quantitative contribution remains highly uncertain. Here, we use a data-driven machine learning approach to assess the impacts of meteorology on surface ozone variations in China for the years 2015 to 2019, considering the months of highest ozone pollution from April to October. To quantify the importance of various meteorological driver variables, we apply non-linear random forest regression (RFR) and linear ridge regression (RR) to learn relationships between meteorological variability and surface ozone in China, and contrast the results to those obtained with the widely used multiple linear regression (MLR) and stepwise MLR. We show that RFR outperforms the three linear methods when predicting ozone using only local meteorological predictor variables. This implies the importance of non-linear relationships between local meteorological factors and ozone, which are not captured by linear regression algorithms. In addition, we find that including non-local meteorological predictors can further improve the modelling skill of RR, particularly for Southern China, highlighting the importance of large-scale meteorological phenomena for ozone pollution in that region. Overall, RFR and RR are in close agreement concerning the leading meteorological drivers behind regional ozone pollution. For example, we find that temperature variations are the dominant meteorological driver for ozone pollution in Northern China (e.g., Beijing Tianjin Hebei region), whereas variations in relative humidity are the most important factor in Southern China (e.g., Pearl River Delta). Variability in surface solar radiation modulates photochemistry but was not considered as such in previou
Weng X, Forster G, Nowack P, 2022, Supplementary material to &quot;A machine learning approach to quantify meteorological drivers of recent ozone pollution in China&quot;
Watson-Parris D, Rao Y, Olivié D, et al., 2022, ClimateBench: A benchmark dataset for data-driven climate projections
Watson-Parris D, Rao Y, Olivié D, et al., 2021, ClimateBench: A benchmark dataset for data-driven climate projections
Nowack P, Konstantinovskiy L, Gardiner H, et al., 2021, Machine learning calibration of low-cost NO2 and PM10 sensors: non-linear algorithms and their impact on site transferability, Atmospheric Measurement Techniques, Vol: 14, Pages: 5637-5655, ISSN: 1867-1381
Low-cost air pollution sensors often fail to attain sufficient performance compared with state-of-the-art measurement stations, and they typically require expensive laboratory-based calibration procedures. A repeatedly proposed strategy to overcome these limitations is calibration through co-location with public measurement stations. Here we test the idea of using machine learning algorithms for such calibration tasks using hourly-averaged co-location data for nitrogen dioxide (NO2) and particulate matter of particle sizes smaller than 10 µm (PM10) at three different locations in the urban area of London, UK. We compare the performance of ridge regression, a linear statistical learning algorithm, to two non-linear algorithms in the form of random forest regression (RFR) and Gaussian process regression (GPR). We further benchmark the performance of all three machine learning methods relative to the more common multiple linear regression (MLR). We obtain very good out-of-sample R2 scores (coefficient of determination) >0.7, frequently exceeding 0.8, for the machine learning calibrated low-cost sensors. In contrast, the performance of MLR is more dependent on random variations in the sensor hardware and co-located signals, and it is also more sensitive to the length of the co-location period. We find that, subject to certain conditions, GPR is typically the best-performing method in our calibration setting, followed by ridge regression and RFR. We also highlight several key limitations of the machine learning methods, which will be crucial to consider in any co-location calibration. In particular, all methods are fundamentally limited in how well they can reproduce pollution levels that lie outside those encountered at training stage. We find, however, that the linear ridge regression outperforms the non-linear methods in extrapolation settings. GPR can allow for a small degree of extrapolation, whereas RFR can only predict values within the training ra
Ceppi P, Nowack P, 2021, Observational evidence that cloud feedback amplifies global warming, Proceedings of the National Academy of Sciences, Vol: 118, ISSN: 0027-8424
Global warming drives changes in Earth’s cloud cover, which, in turn, may amplify or dampen climate change. This “cloud feedback” is the single most important cause of uncertainty in Equilibrium Climate Sensitivity (ECS)—the equilibrium global warming following a doubling of atmospheric carbon dioxide. Using data from Earth observations and climate model simulations, we here develop a statistical learning analysis of how clouds respond to changes in the environment. We show that global cloud feedback is dominated by the sensitivity of clouds to surface temperature and tropospheric stability. Considering changes in just these two factors, we are able to constrain global cloud feedback to 0.43 ± 0.35 W⋅m<jats:sup>−2</jats:sup>⋅K<jats:sup>−1</jats:sup> (90% confidence), implying a robustly amplifying effect of clouds on global warming and only a 0.5% chance of ECS below 2 K. We thus anticipate that our approach will enable tighter constraints on climate change projections, including its manifold socioeconomic and ecological impacts.
Thomas C, Voulgarakis A, Lim G, et al., 2021, An unsupervised learning approach to identifying blocking events:the case of European summer, Weather and Climate Dynamics, Vol: 2, ISSN: 2698-4016
Atmospheric blocking events are mid-latitudeweather patterns, which obstruct the usual path of the polar jet streams. They are often associated with heat wavesin summer and cold snaps in winter. Despite being centralfeatures of mid-latitude synoptic-scale weather, there is nowell-defined historical dataset of blocking events. Variousblocking indices (BIs) have thus been suggested for automatically identifying blocking events in observational and inclimate model data. However, BIs show significant regionaland seasonal differences so that several indices are typicallyapplied in combination to ensure scientific robustness. Here,we introduce a new BI using self-organizing maps (SOMs),an unsupervised machine learning approach, and compare itsdetection skill to some of the most widely applied BIs. Toenable this intercomparison, we first create a new groundtruth time series classification of European blocking basedon expert judgement. We then demonstrate that our method(SOM-BI) has several key advantages over previous BIs because it exploits all of the spatial information provided in theinput data and reduces the dependence on arbitrary thresholds. Using ERA5 reanalysis data (1979–2019), we find thatthe SOM-BI identifies blocking events with a higher precision and recall than other BIs. In particular, SOM-BI alreadyperforms well using only around 20 years of training data sothat observational records are long enough to train our newmethod. We present case studies of the 2003 and 2019 European heat waves and highlight that well-defined groups ofSOM nodes can be an effective tool to diagnose such weatherevents, although the domain-based approach can still lead toerrors in the identification of certain events in a fashion similar to the other BIs. We further test the red blocking detectionskill of SOM-BI depending on the meteorological variableused to study blocking, including geopotential height, sealevel pressure and four variables related to potential vorticity,and t
Thomas C, Voulgarakis A, Lim G, et al., 2021, An unsupervised learning approach to identifying blocking events: the case of European summer, Weather and Climate Dynamics, Vol: 2, Pages: 581-608, ISSN: 2698-4016
Atmospheric blocking events are mid-latitude weather patterns, which obstruct the usual path of the polar jet streams. They are often associated with heat waves in summer and cold snaps in winter. Despite being central features of mid-latitude synoptic-scale weather, there is no well-defined historical dataset of blocking events. Various blocking indices (BIs) have thus been suggested for automatically identifying blocking events in observational and in climate model data. However, BIs show significant regional and seasonal differences so that several indices are typically applied in combination to ensure scientific robustness. Here, we introduce a new BI using self-organizing maps (SOMs), an unsupervised machine learning approach, and compare its detection skill to some of the most widely applied BIs. To enable this intercomparison, we first create a new ground truth time series classification of European blocking based on expert judgement. We then demonstrate that our method (SOM-BI) has several key advantages over previous BIs because it exploits all of the spatial information provided in the input data and reduces the dependence on arbitrary thresholds. Using ERA5 reanalysis data (1979–2019), we find that the SOM-BI identifies blocking events with a higher precision and recall than other BIs. In particular, SOM-BI already performs well using only around 20 years of training data so that observational records are long enough to train our new method. We present case studies of the 2003 and 2019 European heat waves and highlight that well-defined groups of SOM nodes can be an effective tool to diagnose such weather events, although the domain-based approach can still lead to errors in the identification of certain events in a fashion similar to the other BIs. We further test the red blocking detection skill of SOM-BI depending on the meteorological variable used to study blocking, including geopotential height, sea level pressure and four variables related to
Kuhn- Regnier A, Voulgarakis A, Nowack P, et al., 2021, The importance of antecedent vegetation and drought conditions as global drivers of burnt areas, Biogeosciences, Vol: 18, Pages: 3861-3879, ISSN: 1726-4170
The seasonal and longer-term dynamics of fuel accumulation affect fire seasonality and the occurrence of extreme wildfires. Failure to account for their influence may help to explain why state-of-the-art fire models do not simulate the length and timing of the fire season or interannual variability in burnt area well. We investigated the impact of accounting for different timescales of fuel production and accumulation on burnt area using a suite of random forest regression models that included the immediate impact of climate, vegetation, and human influences in a given month and tested the impact of various combinations of antecedent conditions in four productivity-related vegetation indices and in antecedent moisture conditions. Analyses were conducted for the period from 2010 to 2015 inclusive. Inclusion of antecedent vegetation conditions representing fuel build-up led to an improvement of the global, climatological out-of-sample R2 from 0.579 to 0.701, but the inclusion of antecedent vegetation conditions on timescales ≥ 1 year had no impact on simulated burnt area. Current moisture levels were the dominant influence on fuel drying. Additionally, antecedent moisture levels were important for fuel build-up. The models also enabled the visualisation of interactions between variables, such as the importance of antecedent productivity coupled with instantaneous drying. The length of the period which needs to be considered varies across biomes; fuel-limited regions are sensitive to antecedent conditions that determine fuel build-up over longer time periods (∼ 4 months), while moisture-limited regions are more sensitive to current conditions that regulate fuel drying.
Kuhn-Régnier A, Voulgarakis A, Nowack P, et al., 2021, Quantifying the Importance of antecedent fuel-related vegetationproperties for burnt area using random forests, Biogeosciences, Vol: 8, ISSN: 1726-4170
The seasonal and longer-term dynamics of fuel accumulation affect fire seasonality and the occurrence of extreme wildfires. Failure to account for their influence mayhelp to explain why state-of-the-art fire models do not simulate the length and timing of the fire season or interannual variability in burnt area well. We investigated the impact of accounting for different timescales of fuel production and accumulation on burnt area using a suite of random forest regression models that included the immediateimpact of climate, vegetation, and human influences in agiven month and tested the impact of various combinationsof antecedent conditions in four productivity-related vegetation indices and in antecedent moisture conditions. Analyses were conducted for the period from 2010 to 2015 inclusive. Inclusion of antecedent vegetation conditions representing fuel build-up led to an improvement of the global,climatological out-of-sample R2from 0.579 to 0.701, but theinclusion of antecedent vegetation conditions on timescales≥ 1 year had no impact on simulated burnt area. Currentmoisture levels were the dominant influence on fuel drying. Additionally, antecedent moisture levels were importantfor fuel build-up. The models also enabled the visualisationof interactions between variables, such as the importanceof antecedent productivity coupled with instantaneous drying. The length of the period which needs to be consideredvaries across biomes; fuel-limited regions are sensitive to antecedent conditions that determine fuel build-up over longertime periods (∼ 4 months), while moisture-limited regionsare more sensitive to current conditions that regulate fuel drying.
Kuhn-Régnier A, Voulgarakis A, Nowack P, et al., 2021, Supplementary material to "Quantifying the Importance of Antecedent Fuel-Related VegetationProperties for Burnt Area using Random Forests", Biogeosciences, ISSN: 1726-4170
Keeble J, Hassler B, Banerjee A, et al., 2021, Evaluating stratospheric ozone and water vapour changes in CMIP6 models from 1850 to 2100, Atmospheric Chemistry and Physics, Vol: 21, Pages: 5015-5061, ISSN: 1680-7316
Stratospheric ozone and water vapour are key components of the Earth system, and past and future changes to both have important impacts on global and regional climate. Here, we evaluate long-term changes in these species from the pre-industrial period (1850) to the end of the 21st century in Coupled Model Intercomparison Project phase 6 (CMIP6) models under a range of future emissions scenarios. There is good agreement between the CMIP multi-model mean and observations for total column ozone (TCO), although there is substantial variation between the individual CMIP6 models. For the CMIP6 multi-model mean, global mean TCO has increased from ∼ 300 DU in 1850 to ∼ 305 DU in 1960, before rapidly declining in the 1970s and 1980s following the use and emission of halogenated ozone-depleting substances (ODSs). TCO is projected to return to 1960s values by the middle of the 21st century under the SSP2-4.5, SSP3-7.0, SSP4-3.4, SSP4-6.0, and SSP5-8.5 scenarios, and under the SSP3-7.0 and SSP5-8.5 scenarios TCO values are projected to be ∼ 10 DU higher than the 1960s values by 2100. However, under the SSP1-1.9 and SSP1-1.6 scenarios, TCO is not projected to return to the 1960s values despite reductions in halogenated ODSs due to decreases in tropospheric ozone mixing ratios. This global pattern is similar to regional patterns, except in the tropics where TCO under most scenarios is not projected to return to 1960s values, either through reductions in tropospheric ozone under SSP1-1.9 and SSP1-2.6, or through reductions in lower stratospheric ozone resulting from an acceleration of the Brewer–Dobson circulation under other Shared Socioeconomic Pathways (SSPs). In contrast to TCO, there is poorer agreement between the CMIP6 multi-model mean and observed lower stratospheric water vapour mixing ratios, with the CMIP6 multi-model mean underestimating observed water vapour mixing ratios by ∼ 0.5 ppmv at 70
Debeire K, Eyring V, Nowack P, et al., 2021, Constraining uncertainty in projected precipitation over land with Causal Discovery
<jats:p>&lt;p&gt;The models participating in the Coupled Model Intercomparison Project Phase 6 (CMIP6) deliver insights on the evolution of the Earth's &amp;#160;climate. The global precipitation changes follow the magnitude of the warming according to a recent study (Tebaldi, Debeire, Eyring et al., 2020) of the CMIP6 ensemble-mean. However, Earth systems models exhibit a large range in simulated precipitation projection over land. In this study, we present a potential approach to constrain the precipitation changes over land globally and regionally. This approach performs a process-oriented model evaluation similar to Nowack et al. study. We evaluate the ability of models to represent atmospheric dynamical interactions by applying Causal Discovery algorithm. We find a relationship between the ability to represent dynamical interactions close to the observations and the projected precipitation changes over land of the model. We show how this relationship can be used to constrain projection of precipitation over land.&lt;/p&gt;&lt;p&gt;&amp;#160;&lt;/p&gt;&lt;p&gt;References:&lt;/p&gt;&lt;p&gt;Nowack, P., Runge, J., Eyring, V. et al. Causal networks for climate model evaluation and constrained projections. Nat Commun 11, 1415. https://doi.org/10.1038/s41467-020-15195-y, 2020.&lt;/p&gt;&lt;p&gt;Runge, J., P. Nowack, M. Kretschmer, S. Flaxman, D. Sejdinovic, Detecting and quantifying causal associations in large nonlinear time series datasets. Sci. Adv. 5, eaau4996, 2019.&lt;/p&gt;&lt;p&gt;Runge, J, Causal Network Reconstruction from Time Series: From Theoretical Assumptions to Practical Estimation. Chaos: An Interdisciplinary Journal of Nonlinear Science 28 (7): 075310. https://aip.scitation.org/doi/10.1063/1.5025050, 2018.&lt;/p&gt;&lt;p&gt
Keeble J, Hassler B, Banerjee A, et al., 2021, Evaluating stratospheric ozone and water vapor changes in CMIP6&#160;models from 1850-2100&#160;
<jats:p>&lt;p&gt;Stratospheric ozone and water vapor are key components of the Earth system, and past and future changes to both have important impacts on global and regional climate. Here we evaluate long-term changes in these species from the pre-industrial (1850) to the end of the 21&lt;sup&gt;st&lt;/sup&gt; century in CMIP6 models under a range of future emissions scenarios. There is good agreement between the CMIP multi-model mean and observations for total column ozone (TCO), although there is substantial variation between the individual CMIP6 models. For the CMIP6 multi-model mean, global mean TCO has increased from ~300 DU in 1850 to ~305 DU in 1960, before rapidly declining in the 1970s and 1980s following the use and emission of halogenated ozone depleting substances (ODSs). TCO is projected to return to 1960&amp;#8217;s values by the middle of the 21&lt;sup&gt;st&lt;/sup&gt; century under the SSP2-4.5, SSP3-7.0, SSP4-3.4, SSP4-6.0 and SSP5-8.5 scenarios, and under the SSP3-7.0 and SSP5-8.5 scenarios TCO values are projected to be ~10 DU higher than the 1960&amp;#8217;s values by 2100. However, under the SSP1-1.9 and SSP1-1.6 scenarios, TCO is not projected to return to the 1960&amp;#8217;s values despite reductions in halogenated ODSs due to decreases in tropospheric ozone mixing ratios. This global pattern is similar to regional patterns, except in the tropics where TCO under most scenarios is not projected to return to 1960&amp;#8217;s values, either through reductions in tropospheric ozone under SSP1-1.9 and SSP1-2.6, or through reductions in lower stratospheric ozone resulting from an acceleration of the Brewer-Dobson Circulation under other SSPs. In contrast to TCO, there is poorer agreement between the CMIP6 multi-model mean and observed lower stratospheric water vapour mixing ratios, with the CMIP6 multi-model mean underestimating o
Chiodo G, Ball WT, Nowack P, et al., 2021, The response of the ozone layer under abrupt 4xCO2 in CMIP6
<jats:p>&lt;p&gt;Previous studies indicate a possible role of stratospheric ozone chemistry feedbacks in the climate response to 4xCO2, either via a reduction in equilibrium climate sensitivity (ECS) or via changes in the tropospheric circulation (Nowack et al., 2015; Chiodo and Polvani, 2017). However, these effects are subject to uncertainty. Part of the uncertainty may stem from the dependency of the feedback on the pattern of the ozone response, as the radiative efficiency of ozone largely depends on its vertical distribution (Lacis et al., 1990). Here, an analysis is presented of the ozone layer response to 4xCO2 in chemistry&amp;#8211;climate models (CCMs) which participated to CMIP inter-comparisons. In a previous study using CMIP5 models, it has been shown that under 4xCO2, ozone decreases in the tropical lower stratosphere, and increases over the high latitudes and throughout the upper stratosphere (Chiodo et al., 2018). It was also found that a substantial portion of the spread in the tropical column ozone is tied to inter-model spread in tropical upwelling, which is in turn tied to ECS. Here, we revisit this connection using 4xCO2 data from CMIP6, thereby exploiting the larger number of CCMs available than in CMIP5. In addition, we explore the linearity of the ozone response, by complementing the analysis with simulations using lower CO2 forcing levels (2xCO2). We show that the pattern of the ozone response is similar to CMIP5. In some models (e.g. WACCM), we find larger ozone responses in CMIP6 than in CMIP5, partly because of the larger ECS and thus larger upwelling response in the tropical pipe. In this presentation, we will discuss the relationship between radiative forcing, transport and ozone, as well as further implications for CMIP6 models.&lt;/p&gt;</jats:p>
Thomas C, Voulgarakis A, Lim G, et al., 2021, An unsupervised learning approach to identifying blocking events: the case of European summer
<jats:p>&lt;p&gt;Atmospheric blocking events are mid-latitude weather patterns, which obstruct the usual path of the polar jet stream. Several blocking indices (BIs) have been developed to study blocking patterns and their associated trends, but these show significant seasonal and regional differences. Despite being central features of mid-latitude synoptic-scale weather, there is no well-defined historical dataset of blocking events. Here, we introduce a new blocking index using self-organizing maps (SOMs), an unsupervised machine learning approach, and compare its detection skill to some of the most widely applied BIs. To enable this intercomparison, we first create a new ground truth time series classification of European blocking based on expert judgement. We then demonstrate that our method (SOM-BI) has several key advantages over previous BIs because it exploits all the spatial information provided in the input data and avoids the need for arbitrary thresholds. Using ERA5 reanalysis data (1979-2019), we find that the SOM-BI identifies blocking events with a higher precision and recall than other BIs. We present a case study of the 2003 European heat wave and highlight that well-defined groups of SOM nodes can be an effective tool to reliably and accurately diagnose such weather events. This contrasts with the way SOMs are commonly used, where an individual SOM node can be wrongly assumed to represent a weather pattern. We also evaluate the SOM-BI performance on about 100 years of climate model data from a preindustrial simulation with the new UK Earth System Model (UK-ESM1). For the model data, all blocking detection methods have lower skill than for the ERA5 reanalysis, but SOM-BI performs significantly better than the conventional indices. This shows that our method can be effectively applied to climate models to develop our understanding of how climate change will affect regional blocking characteristics. Overall, our results demonstra
Mansfield L, Nowack P, Voulgarakis A, 2021, Predicting climate model response to changing emissions
<jats:p>&lt;p&gt;In order to make predictions on how the climate would respond to changes in global and regional emissions, we typically run simulations on Global Climate Models (GCMs) with perturbed emissions or concentration fields. These simulations are highly expensive and often require the availability of high-performance computers. Machine Learning (ML) can provide an alternative approach to estimating climate response to various emissions quickly and cheaply.&amp;#160;&lt;/p&gt;&lt;p&gt;We will present a Gaussian process emulator capable of predicting the global map of temperature response to different types of emissions (both greenhouse gases and aerosol pollutants), trained on a carefully designed set of simulations from a GCM. This particular work involves making short-term predictions on 5 year timescales but can be linked to an emulator from previous work that predicts on decadal timescales. We can also examine uncertainties associated with predictions to find out where where the method could benefit from increased training data. This is a particularly useful asset when constructing emulators for complex models, such as GCMs, where obtaining training runs is costly.&amp;#160;&lt;/p&gt;</jats:p>
Nowack P, Konstantinovskiy L, Gardiner H, et al., 2020, Towards low-cost and high-performance air pollution measurementsusing machine learning calibration techniques
<jats:p>Abstract. Air pollution is a key public health issue in urban areas worldwide. The development of low-cost air pollution sensors is consequently a major research priority. However, low-cost sensors often fail to attain sufficient measurement performance compared to state-of-the-art measurement stations, and typically require calibration procedures in expensive laboratory settings. As a result, there has been much debate about calibration techniques that could make their performance more reliable, while also developing calibration procedures that can be carried out without access to advanced laboratories. One repeatedly proposed strategy is low-cost sensor calibration through co-location with public measurement stations. The idea is that, using a regression function, the low-cost sensor signals can be calibrated against the station reference signal, to be then deployed separately with performances similar to the original stations. Here we test the idea of using machine learning algorithms for such regression tasks using hourly-averaged co-location data for nitrogen dioxide (NO2) and particulate matter of particle sizes smaller than 10 μm (PM10) at three different locations in the urban area of London, UK. Specifically, we compare the performance of Ridge regression, a linear statistical learning algorithm, to two non-linear algorithms in the form of Random Forest (RF) regression and Gaussian Process regression (GPR). We further benchmark the performance of all three machine learning methods to the more common Multiple Linear Regression (MLR). We obtain very good out-of-sample R2-scores (coefficient of determination) > 0.7, frequently exceeding 0.8, for the machine learning calibrated low-cost sensors. In contrast, the performance of MLR is more dependent on random variations in the sensor hardware and co-located signals, and is also more sensitive to the length of the co-location period. We find that, subject to certain condition
Malik A, Nowack PJ, Haigh JD, et al., 2020, Tropical Pacific climate variability under solar geoengineering: impacts on ENSO extremes, ATMOSPHERIC CHEMISTRY AND PHYSICS, Vol: 20, Pages: 15461-15485, ISSN: 1680-7316
Mansfield L, Nowack P, Kasoar M, et al., 2020, Predicting global patterns of long-term climate change from short-term simulations using machine learning, npj Climate and Atmospheric Science, Vol: 3, ISSN: 2397-3722
Understanding and estimating regional climate change under different anthropogenic emission scenarios is pivotal for informing societal adaptation and mitigation measures. However, the high computational complexity of state-of-the-art climate models remains a central bottleneck in this endeavour. Here we introduce a machine learning approach, which utilises a unique dataset of existing climate model simulations to learn relationships between short-te¬rm and long-term temperature responses to different climate forcing scenarios. This approach not only has the potential to accelerate climate change projections by reducing the costs of scenario computations, but also helps uncover early indicators of modelled long-term climate responses, which is of relevance to climate change detection, predictability and attribution. Our results highlight challenges and opportunities for data-driven climate modelling, especially concerning the incorporation of even larger model datasets in the future. We therefore encourage extensive data sharing among research institutes to build ever more powerful climate response emulators, and thus to enable faster climate change projections.
Mansfield L, Nowack P, Kasoar M, et al., 2020, Can we predict global patterns of long-term climate change from short-term simulations?
<jats:p> &lt;p&gt;&lt;span&gt;Furthering our understanding of regional climate change responses to different greenhouse gas and aerosol emission scenarios is pivotal to inform societal adaptation and mitigation measures. However, complex General Circulation Models (GCMs) used for decadal to centennial climate change projections are computationally expensive. Here we have utilised a unique dataset of existing global climate model simulations to show that a novel machine learning approach can learn relationships between short-term and long-term temperature responses to different climate forcings, which in turn can accelerate climate change projections. This approach could reduce the costs of additional scenario computations and uncover consistent early indicators of long-term climate responses. &lt;/span&gt;&lt;/p&gt;&lt;p&gt;&lt;span&gt;We have explored several statistical techniques for this supervised learning task and here we present predictions made with Ridge regression and Gaussian process regression. We have compared the results to pattern scaling as a standard simplified approach for estimating regional surface temperature responses under varying climate forcing scenarios. In this research, we highlight key challenges and opportunities for data-driven climate model emulation, especially with regards to the use of even larger model datasets and different climate variables. We demonstrate the potential to apply our method for gaining new insights into how and where ongoing climate change can be best detected and extrapolated; proposing this as a blueprint for future studies and encouraging data collaborations among research institutes in order to build ever more accurate climate response emulators.&lt;/span&gt;&lt;/p&gt; </jats:p>
Debeire K, Eyring V, Nowack P, et al., 2020, Causal Discovery as a novel approach for CMIP6 climate model evaluation
<jats:p> &lt;p&gt;Causal discovery algorithms are machine learning methods that estimate the dependencies between different variables. One of these algorithms, the recently developed PCMCI algorithm (Runge et al., 2019) estimates the time-lagged causal dependency structures from multiple time series and is adapted to common properties of Earth System time series data. The PCMCI algorithm has already been successfully applied in climate science to reveal known interaction pathways between Earth regions commonly referred to as teleconnections, and to explore new teleconnections (Kretschmer et al., 2017). One recent study used this method to evaluate models participating in the Coupled Model Intercomparison Project Phase 5&amp;#160; (CMIP5) (Nowack et al., 2019).&lt;/p&gt;&lt;p&gt;Here, we build on the Nowack et al. study and use PCMCI on dimension-reduced meteorological reanalysis data and the CMIP6 ensembles data. The resulting causal networks represent teleconnections (causal links) in each of the CMIP6 climate models. The models&amp;#8217; performance in representing realistic teleconnections is then assessed by comparing the causal networks of the individual CMIP6 models to the one obtained from meteorological reanalysis. We show that causal discovery is a promising and novel approach that complements existing model evaluation approaches.&lt;/p&gt;&lt;p&gt;&amp;#160;&lt;/p&gt;&lt;p&gt;References:&lt;/p&gt;&lt;p&gt;Runge, J., P. Nowack, M. Kretschmer, S. Flaxman, D. Sejdinovic, Detecting and quantifying causal associations in large nonlinear time series datasets. Sci. Adv. 5, eaau4996, 2019.&lt;/p&gt;&lt;p&gt;Kretschmer, M., J. Runge, and D. Coumou, Early prediction of extreme stratospheric polar vortex states based on causal precursors, Geophysical Resear
Nowack P, Runge J, Eyring V, et al., 2020, Causal networks for climate model evaluation and constrained projections, Nature Communications, Vol: 11, ISSN: 2041-1723
Global climate models are central tools for understanding past and future climate change. The assessment of model skill, in turn, can benefit from modern data science approaches. Here we apply causal discovery algorithms to sea level pressure data from a large set of climate model simulations and, as a proxy for observations, meteorological reanalyses. We demonstrate how the resulting causal networks (fingerprints) offer an objective pathway for process-oriented model evaluation. Models with fingerprints closer to observations better reproduce important precipitation patterns over highly populated areas such as the Indian subcontinent, Africa, East Asia, Europe and North America. We further identify expected model interdependencies due to shared development backgrounds. Finally, our network metrics provide stronger relationships for constraining precipitation projections under climate change as compared to traditional evaluation metrics for storm tracks or precipitation itself. Such emergent relationships highlight the potential of causal networks to constrain longstanding uncertainties in climate change projections.
Keeble J, Hassler B, Banerjee A, et al., 2020, Evaluating stratospheric ozone and water vapor changes in CMIP6models from 1850–2100
<jats:p>Abstract. Stratospheric ozone and water vapour are key components of the Earth system, and past and future changes to both have important impacts on global and regional climate. Here we evaluate long-term changes in these species from the pre- industrial (1850) to the end of the 21st century in CMIP6 models under a range of future emissions scenarios. There is good agreement between the CMIP multi-model mean and observations, although there is substantial variation between the individual CMIP6 models. For the CMIP6 multi-model mean, global total column ozone (TCO) has increased from ∼300 DU in 1850 to ∼305 DU in 1960, before rapidly declining in the 1970s and 1980s following the use and emission of halogenated ozone depleting substances (ODSs). TCO is projected to return to 1960s values by the middle of the 21st century under the SSP2-4.5, SSP3-7.0, SSP4-3.4, SSP4-6.0 and SSP5-8.5 scenarios, and under the SSP3-7.0 and SSP5-8.5 scenarios TCO values are projected to be ∼10 DU higher than the 1960s values by 2100. However, under the SSP1-1.9 and SSP1-1.6 scenarios, TCO is not projected to return to the 1960s values despite reductions in halogenated ODSs due to decreases in tropospheric ozone mixing ratios. This global pattern is similar to regional patterns, except in the tropics where TCO under most scenarios is not projected to return to 1960s values, either through reductions in tropospheric ozone under SSP1-1.9 and SSP1-2.6, or through reductions in lower stratospheric ozone resulting from an acceleration of the Brewer-Dobson Circulation under other SSPs. CMIP6 multi-model mean stratospheric water vapour mixing ratios in the tropical lower stratosphere have increased by ∼0.5 ppmv from the pre-industrial to the present day and are projected to increase further by the end of the 21st century. The largest increases (∼2 ppmv) are simulated under the future scenarios with the highest assumed forcing pathway (e.g. SSP5-8.5)
Nowack P, Ong QYE, Braesicke P, et al., 2019, Machine learning parameterizations for ozone: climate model transferability, https://sites.google.com/view/climateinformatics2019/proceedings, 9th International Workshop on Climate Informatics, Publisher: UCAR, Pages: 263-268
Many climate modeling studies have demon-strated the importance of two-way interactions betweenozone and atmospheric dynamics. However, atmosphericchemistry models needed for calculating changes in ozoneare computationally expensive. Nowack et al. [1] high-lighted the potential of machine learning-based ozoneparameterizations in constant climate forcing simulations,with ozone being predicted as a function of the atmo-spheric temperature state. Here we investigate the roleof additional time-lagged temperature information underpreindustrial forcing conditions. In particular, we testif the use of Long Short-Term Memory (LSTM) neuralnetworks can significantly improve the predictive skill ofthe parameterization. We then introduce a novel workflowto transfer the regression model to the new UK EarthSystem Model (UKESM). For this, we show for the firsttime how machine learning parameterizations could betransferred between climate models, a pivotal step tomaking any such parameterization widely applicable inclimate science. Our results imply that ozone parame-terizations could have much-extended scope as they arenot bound to individual climate models but, once trained,could be used in a number of different models. We hope tostimulate similar transferability tests regarding machinelearning parameterizations developed for other Earthsystem model components such as ocean eddy modeling,convection, clouds, or carbon cycle schemes.
This data is extracted from the Web of Science and reproduced under a licence from Thomson Reuters. You may not copy or re-distribute this data in whole or in part without the written consent of the Science business of Thomson Reuters.