Does Model Calibration Reduce Uncertainty in Climate Projections?

Uncertainty in climate projections is large as shown by the likely uncertainty ranges in Equilibrium Climate Sensitivity (ECS) of 2.5-4K and in the Transient Climate Response (TCR) of 1.4-2.2K. Uncertainty in model projections could arise from the way in which unresolved processes are represented, the parameter values used, or the targets for model calibration. We show that, in two climate model ensembles which were objectively calibrated to minimise differences from observed large scale atmospheric climatology, uncertainties in ECS and TCR are about two to six times smaller than in the CMIP5 or CMIP6 multi-model ensemble. We also find that projected uncertainties in surface temperature, precipitation and annual extremes are relatively small. Residual uncertainty largely arises from unconstrained sea-ice feedbacks. The 20+ year old HadAM3 standard model configuration simulates observed hemispheric scale observations and pre-industrial surface temperatures about as well as the median CMIP5 and CMIP6 ensembles while the optimised configurations simulates these better than almost all the CMIP5 and CMIP6 models. Hemispheric scale observations and pre-industrial temperatures are not systematically better simulated in CMIP6 than in CMIP5 though the CMIP6 ensemble seems to better simulate patterns of large-scale observations than the CMIP5 ensemble and the optimised HadAM3 configurations. Our results suggest that most CMIP models could be improved in their simulation of large scale observations by systematic calibration. However, the uncertainty in climate projections (for a given scenario) likely largely arises from the choice of parametrisation schemes for unresolved processes (“structural uncertainty”), with different tuning targets another possible contributor.

Probabilistic Forecast of the Extended Range Heatwave Over Eastern China

Several probabilistic forecast methods for heatwave (HW) in extended-range scales over China are constructed using four models (ECMWF, CMA, UKMO, and NCEP) from the Subseasonal-to-Seasonal (S2S) database. The methods include four single-model ensembles (SME; ECMWF, CMA, UKMO, and NCEP), multi-model ensemble (MME), and Bayesian model averaging (BMA). The construction and verification of reforecasts are implemented by a defined heat wave index (HWI) which is not only able to reflect the actual occurrence of heatwaves, but also to facilitate forecast and verification. The performance is measured by traditional verification method at each grid point of the 105°E to 132°E; 20°N to 45°N domain for the July, August, and September (JAS) of 1999–2010. For deterministic evaluations of HWI forecast, BMA shows a better pattern correlation coefficient than SME and MME and comparable equitable threat score (ETS) with ECMWF and MME. The good performance of ECMWF and MME take advantage of setting the percentile thresholds for forecasting HW. For the probabilistic forecast, the Brier score of BMA is comparable (superior) to that of MME and ECMWF at short (long) lead-time. BMA also demonstrates an improvement on the reliability of probabilistic forecast, indicating that BMA method is a useful tool for an extended-range forecast of HW. Meanwhile, in the real-time extended-range probabilistic forecast, the beginning date, end date, and probability of HW event can be predicted by the HWI probabilistic forecast of BMA.

On the Southern Hemisphere stratospheric response to ENSO and its impacts on tropospheric circulation

As the leading mode of Pacific variability, the El Niño-Southern Oscillation (ENSO) causes vast and wide-spread climatic impacts, including in the stratosphere. Following discovery of a stratospheric pathway of ENSO to the Northern Hemisphere surface, here we aim to investigate if there is a substantial Southern Hemisphere (SH) stratospheric pathway in relation to austral winter ENSO events. Large stratospheric anomalies connected to ENSO occur on average at high SH latitudes as early as August, peaking at around 10 hPa. An overall colder austral spring Antarctic stratosphere is generally associated with the warm phase of the ENSO cycle, and vice versa. This behavior is robust among reanalysis and six separate model ensembles encompassing two different model frameworks. A stratospheric pathway is identified by separating ENSO events that exhibit a stratospheric anomaly from those that don’t and comparing to stratospheric extremes that occur during neutral-ENSO years. The tropospheric eddy-driven jet response to the stratospheric ENSO pathway is the most robust in the spring following a La Niña, but extends into summer, and is more zonally-symmetric compared to the tropospheric ENSO teleconnection. The magnitude of the stratospheric pathway is weaker compared to the tropospheric pathway and therefore when it is present, has a secondary role. For context, the magnitude is approximately half that of the eddy-driven jet modulation due to austral spring ozone depletion in the model simulations. This work establishes that the stratospheric circulation acts as an intermediary in coupling ENSO variability to variations in the austral spring and summer tropospheric circulation.

Monsoonal precipitation over Peninsular Malaysia in the CMIP6 HighResMIP experiments: the role of model resolution

This study investigates the ability of 20 model simulations which contributed to the CMIP6 HighResMIP to simulate precipitation in different monsoon seasons and extreme precipitation events over Peninsular Malaysia. The model experiments utilize common forcing but are run with different horizontal and vertical resolutions. The impact of resolution on the models’ abilities to simulate precipitation and associated environmental fields is assessed by comparing multi-model ensembles at different resolutions with three observed precipitation datasets and four climate reanalyses. Model simulations with relatively high horizontal and vertical resolution exhibit better performance in simulating the annual cycle of precipitation and extreme precipitation over Peninsular Malaysia and the coastal regions. Improvements associated with the increase in horizontal and vertical resolutions are also found in the statistical relationship between precipitation and monsoon intensity in different seasons. However, the increase in vertical resolution can lead to a reduction of annual mean precipitation compared to that from the models with low vertical resolutions, associated with an overestimation of moisture divergence and underestimation of lower-tropospheric vertical ascent in the different monsoon seasons. This limits any improvement in the simulation of precipitation in the high vertical resolution experiments, particularly for the Southwest monsoon season.

Malaria Parasite Detection using Efficient Neural Ensembles

Caused by the bite of the Anopheles mosquito infected with the parasite of genus Plasmodium, malaria has remained a major burden towards healthcare for years with an approximate 400,000 deaths reported globally every year. The traditional diagnosis process for malaria involves an examination of the blood smear slide under the microscope. This process is not only time consuming but also requires pathologists to be highly skilled in their work. Timely diagnosis and availability of robust diagnostic facilities and skilled laboratory technicians are very much vital to reduce the mortality rate. This study aims to build a robust system by applying deep learning techniques such as transfer learning and snapshot ensembling to automate the detection of the parasite in the thin blood smear images. All the models were evaluated against the following metrics - F1 score, Accuracy, Precision, Recall, Mathews Correlation Coefficient (MCC), Area Under the Receiver Operating Characteristics (AUC-ROC) and the Area under the Precision Recall curve (AUC-PR). The snapshot ensembling model created by combining the snapshots of the EfficientNet-B0 pre-trained model outperformed every other model achieving a f1 score - 99.37%, precision - 99.52% and recall - 99.23%. The results show the potential of model ensembles which combine the predictive power of multiple weal models to create a single efficient model that is better equipped to handle the real world data. The GradCAM experiment displayed the gradient activation maps of the last convolution layer to visually explicate where and what a model sees in an image to classify them into a particular class. The models in this study correctly activate the stained parasitic region of interest in the thin blood smear images. Such visuals make the model more transparent, explainable, and trustworthy which are very much essential for deploying AI based models in the healthcare network.

Malaria Parasite Detection using Efficient Neural Ensembles

Caused by the bite of the Anopheles mosquito infected with the parasite of genus Plasmodium, malaria has remained a major burden towards healthcare for years with an approximate 400,000 deaths reported globally every year. The traditional diagnosis process for malaria involves an examination of the blood smear slide under the microscope. This process is not only time consuming but also requires pathologists to be highly skilled in their work. Timely diagnosis and availability of robust diagnostic facilities and skilled laboratory technicians are very much vital to reduce the mortality rate. This study aims to build a robust system by applying deep learning techniques such as transfer learning and snapshot ensembling to automate the detection of the parasite in the thin blood smear images. All the models were evaluated against the following metrics - F1 score, Accuracy, Precision, Recall, Mathews Correlation Coefficient (MCC), Area Under the Receiver Operating Characteristics (AUC-ROC) and the Area under the Precision Recall curve (AUC-PR). The snapshot ensembling model created by combining the snapshots of the EfficientNet-B0 pre-trained model outperformed every other model achieving a f1 score - 99.37%, precision - 99.52% and recall - 99.23%. The results show the potential of model ensembles which combine the predictive power of multiple weal models to create a single efficient model that is better equipped to handle the real world data. The GradCAM experiment displayed the gradient activation maps of the last convolution layer to visually explicate where and what a model sees in an image to classify them into a particular class. The models in this study correctly activate the stained parasitic region of interest in the thin blood smear images. Such visuals make the model more transparent, explainable, and trustworthy which are very much essential for deploying AI based models in the healthcare network.

Natural variability vs forced signal in the 2015–2019 Central American drought

The recent multi-year 2015–2019 drought after a multi-decadal drying trend over Central America raises the question of whether anthropogenic climate change (ACC) played a role in exacerbating these events. While the occurrence of the 2015–2019 drought in Central America has been asserted to be associated with ACC, we lack an assessment of natural vs anthropogenic contributions. Here, we use five different large ensembles—including high-resolution ensembles (i.e., 0.5∘ horizontally)—to estimate the contribution of ACC to the probability of occurrence of the 2015–2019 event and the recent multi-decadal trend. The comparison of ensembles forced with natural and natural plus anthropogenic forcing suggests that the recent 40-year trend is likely associated with internal climate variability. However, the 2015–2019 rainfall deficit has been made more likely by ACC. The synthesis of the results from model ensembles supports the notion of a significant increase, by a factor of four, over the last century for the 2015–2019 meteorological drought to occur because of ACC. All the model results further suggest that, under intermediate and high emission scenarios, the likelihood of similar drought events will continue to increase substantially over the next decades.

An Assessment of Climate Feedbacks in Observations and Climate Models Using Different Energy Balance Frameworks

This study evaluates the performance of Coupled Model Intercomparison Project (CMIP) phase 5 and phase 6 models by comparing feedbacks in models to those inferred from observations. Overall, we find no systematic disagreements between the feedbacks in the model ensembles and feedbacks inferred from observations, although there is a wide range in the ability of individual models to reproduce the observations. In particular, 40 of 52 models have best estimates that fall within the uncertainty of the observed total feedback. We quantify two sources of uncertainty in the model ensembles: (1) the structural difference, due to the differences in model parameterizations, and (2) the unforced pattern effect, due to unforced variability, and find that both are important when comparing to an 18-year observational data set. We perform the comparison using two energy balance frameworks: the traditional energy balance framework, in which it is assumed that changes in energy balance are controlled by changes in global average surface temperatures, and an alternative framework that assumes the changes in energy balance are controlled by tropical atmospheric temperatures. We find that the alternative framework provides a more robust way of comparing the models to observations, with both smaller structural differences and smaller unforced pattern effect. However, when considering the relation of feedbacks in response to interannual variability and long-term warming, the traditional framework has advantages. There are no great differences between the CMIP5 and CMIP6 ensembles’ ability to reproduce the observed feedbacks.