If you weren’t convinced by Lüning’s and Vahrenholt’s essay on the failures of climate models posted here two days ago, then here’s more!
By Sebastian Lüning and Fritz Vahrenholt
(Translated/edited by P Gosselin)
In March 2013 journalist Joachim Müller-Jung put it aptly in an article in the FAZ:
Whoever simulates the world, has not understood the truth
How much reality is in the models? Fact is: The complexities increase, but so do the uncertainties. How can science remain credible here?”
Read more at the faz.net.
Today once again we wish to take a cruise through the world of modeling. The field currently finds itself deep in crisis. Earlier they had energetically produced a number of models but the numerous misses are now taking a heavy toll. A first wave of self-criticism is sweeping across the field. by far not everything is as rosy as what the state funders once claimed.
On February 21, 2013 the University Göteborg issued a press release with the title: “Climate models are not good enough“. Within the framework of a promotion project, it was discovered that climate models have been unable to reproduce the observed changes in extreme rainfalls in China over the last 50 years:
Climate models are not good enough
Only a few climate models were able to reproduce the observed changes in extreme precipitation in China over the last 50 years. This is the finding of a doctoral thesis from the University of Gothenburg, Sweden. Climate models are the only means to predict future changes in climate and weather. “It is therefore extremely important that we investigate global climate models’ own performances in simulating extremes with respect to observations, in order to improve our opportunities to predict future weather changes,” says Tinghai Ou from the University of Gothenburg’s Department of Earth Sciences. Tinghai has analysed the model simulated extreme precipitation in China over the last 50 years. “The results show that climate models give a poor reflection of the actual changes in extreme precipitation events that took place in China between 1961 and 2000,” he says. “Only half of the 21 analysed climate models analysed were able to reproduce the changes in some regions of China. Few models can well reproduce the nationwide change.”
Problems with the rainfall modeling are found at every location. Also in the USA models have failed to reproduce the historical precipitation development, as Mishra et al. (2012) and Knappenberger and Michaels (2013) were able to show. Stratton & Stirling (2012) and Ramirez-Villegas et al. (2013) found the same at the global level. Here’s an excerpt of the latter:
Climatological means of seasonal mean temperatures depict mean errors between 1 and 18 ° C (2–130% with respect to mean), whereas seasonal precipitation and wet-day frequency depict larger errors, often offsetting observed means and variability beyond 100%. Simulated interannual climate variability in GCMs warrants particular attention, given that no single GCM matches observations in more than 30% of the areas for monthly precipitation and wet-day frequency, 50% for diurnal range and 70% for mean temperatures. We report improvements in mean climate skill of 5–15% for climatological mean temperatures, 3–5% for diurnal range and 1–2% in precipitation. At these improvement rates, we estimate that at least 5–30 years of CMIP work is required to improve regional temperature simulations and at least 30–50 years for precipitation simulations, for these to be directly input into impact models. We conclude with some recommendations for the use of CMIP5 in agricultural impact studies.”
Soncini & Bocchiola (2011) examined snowfall in the Italian Alps. Also here they found the same: the real, measured development cannot be reproduced by models. What’s even worse: the future projections of various models deviate widely from each other. Here’s an excerpt from the noteworthy paper:
General Circulation Models GCMs are widely adopted tools to achieve future climate projections. However, one needs to assess their accuracy, which is only possible by comparison of GCMs’ control runs against past observed data. Here, we investigate the accuracy of two GCMs models delivering snowfall that are included within the IPCC panel’s inventory (HadCM3, CCSM3), by comparison against a comprehensive ground data base (ca. 400 daily snow gauging stations) located in the Italian Alps, during 1990–2009. The GCMs simulations are objectively compared to snowfall volume by regionally evaluated statistical indicators. The CCSM3 model provides slightly better results than the HadCM3, possibly in view of its finer computational grid, but yet the performance of both models is rather poor. We evaluate the bias between models and observations, and we use it as a bulk correction for the GCMs’ snowfall simulations for the purpose of future snowfall projection. We carry out stationarity analysis via linear regression and Mann Kendall tests upon the observed and simulated snowfall volumes for the control run period, providing contrasting results. We then use the bias adjusted GCMs output for future snowfall projections from the IPCC-A2 scenario. The two analyzed models provide contrasting results about projected snowfall during the 21st century (until 2099). Our approach provides a first order assessment of the expected accuracy of GCM models in depicting past and future snowfall upon the (Italian) Alps. Overall, given the poor depiction of snowfall by the GCMs here tested, we suggest that care should be taken when using their outputs for predictive purposes.”
Out of thin air?
In June 2013 Axel Lauer and Kevin Hamilton in the Journal of Climate looked at cloud models. Also here nothing is different: Every model does its own thing and the real development simply refuses to play along. Next is the abstract of the paper:
Clouds are a key component of the climate system affecting radiative balances as well as the hydrological cycle. Previous studies from the Coupled Model Intercomparison Project Phase 3 (CMIP3) showed quite large biases in the simulated cloud climatology affecting all GCMs [global climate models] as well as a remarkable degree of variation among the models, which represented the state-of-the-art circa 2005. Here we measure the progress that has been made in recent years by comparing mean cloud properties, interannual variability, and the climatological seasonal cycle from the CMIP5 models with satellite observations and with results from comparable CMIP3 experiments. We focus on three climate-relevant cloud parameters: cloud amount, liquid water path, and cloud radiative forcing. We show that intermodel differences are still large in the CMIP5 simulations. We find some small improvements of particular cloud properties in some regions in the CMIP5 ensemble over CMIP3. In CMIP5 there is an improved agreement of the modeled interannual variability of liquid water path as well as of the modeled longwave cloud forcing over mid and high latitude oceans with observations. However, the differences in the simulated cloud climatology from CMIP3 and CMIP5 are generally small and there is very little to no improvement apparent in the tropical and subtropical regions in CMIP5. Comparisons of the results from the coupled CMIP5 models with their atmosphere-only versions run with observed SSTs show remarkably similar biases in the simulated cloud climatologies. This suggests the treatments of subgrid-scale cloud and boundary layer processes are directly implicated in the poor performance of current GCMs [global climate models or general circulation models] in simulating realistic cloud fields.”
No matter which modeling parameter one looks at, the result is off. Another example is ground moisture ground moisture, which according to an analysis by Tim Ball is not correctly given in the IPCC models. There are also problems with thunderstorms, as Anthony Watts at WUWT showed. Or take a look at atmospheric pressure. According to Przybylak et al. 2012, it has not changed much since the beginning of the 19th century. The models, on the other hand, foresaw a significant trend, which scientists sold as an “anthropogenic fingerprint”. That is now turning out to be completely wrong.
Also when looking back further into the geological past, climate models do not make a good impression at all. During the last interglacial, i.e. the warm period of 120,000 years ago, it was warmer than today. Yet, for whatever reason, climate models are unable to reproduce this, as a paper appearing in the 29 August 2014 Climate of the Past journal criticized:
We find that for annual temperatures, the overestimation is small, strongly model-dependent (global mean 0.4 ± 0.3 °C) and cannot explain the recently published 0.67 °C difference between simulated and reconstructed annual mean temperatures during the LIG thermal maximum. However, if one takes into consideration that temperature proxies are possibly biased towards summer, the overestimation of the LIG thermal maximum based on warmest month temperatures is non-negligible with a global mean of 1.1 ± 0.4 °C.”
Another paper that appeared just days before by Dolan et al. looked at the Pilocene 3 million years ago. The task for the 9 modeling groups was to calculate the ice cover over Greenland for the conditions of the Pilocene warm period. Back then it was considerably warmer than today and the sea level was higher, i.e. about the same conditions that are predicted by today’s climate models for the end of the 21st century. The study ended with a big surprise: Everything between “ice-free” and “ice somewhat like today” were given by the models. The reason for the divergence: Every model simulated the local albedo properties in Greenland differently. Yet this is decisive for the ice cover extent of the island. In summary future modeling resembles playing the lottery.
Finally we go back yet another step further, to the mid-Milocene 14 million years ago. A paper by Goldner et al. from March 2014 found that the climate models were off by a full 4°C. Among other things, the authors suspect that certain climate factors are missing in the models. An interesting thought…
What follows is an excerpt from the abstract of that paper:
The mid-Miocene climatic optimum (MMCO) is an intriguing climatic period due to its above-modern temperatures in mid-to-high latitudes in the presence of close-to-modern CO2 concentrations. We use the recently released Community Earth System Model (CESM1.0) with a slab ocean to simulate this warm period, incorporating recent Miocene CO2 reconstructions of 400 ppm (parts per million). We simulate a global mean annual temperature (MAT) of 18 °C, ~4 °C above the preindustrial value, but 4 °C colder than the global Miocene MAT we calculate from climate proxies. […] Our results illustrate that MMCO warmth is not reproducible using the CESM1.0 forced with CO2 concentrations reconstructed for the Miocene or including various proposed Earth system feedbacks; the remaining discrepancy in the MAT is comparable to that introduced by a CO2 doubling. The model’s tendency to underestimate proxy derived global MAT and overestimate the Equator to pole temperature gradient suggests a major climate problem in the MMCO akin to those in the Eocene. Our results imply that this latest model, as with previous generations of climate models, is either not sensitive enough or additional forcings remain missing that explain half of the anomalous warmth and pronounced polar amplification of the MMCO.”