A new paper has been published in Geophysical Research Letters that shows oce again, that climate models and reality significantly vary. It confirms what Dr. John Christy has been saying (see figure below). The paper also references Dr. Judith Curry and her work.

Pronounced differences between observed and CMIP5-simulated multidecadal climate variability in the twentieth century

Plain Language Summary

Global and regional warming trends over the course of the twentieth century have been nonuniform, with decadal and longer periods of faster or slower warming, or even cooling. Here we show that state-of-the-art global models used to predict climate fail to adequately reproduce such multidecadal climate variations. In particular, the models underestimate the magnitude of the observed variability and misrepresent its spatial pattern. Therefore, our ability to interpret the observed climate change using these models is limited.

Abstract

Identification and dynamical attribution of multidecadal climate undulations to either variations in external forcings or to internal sources is one of the most important topics of modern climate science, especially in conjunction with the issue of human-induced global warming. Here we utilize ensembles of twentieth century climate simulations to isolate the forced signal and residual internal variability in a network of observed and modeled climate indices. The observed internal variability so estimated exhibits a pronounced multidecadal mode with a distinctive spatiotemporal signature, which is altogether absent in model simulations. This single mode explains a major fraction of model-data differences over the entire climate index network considered; it may reflect either biases in the models’ forced response or models’ lack
of requisite internal dynamics, or a combination of both.

Some key quotes:

Here we show that state-of-the-art global models used to predict climate fail to adequately reproduce such multidecadal climate variations. In particular, the models underestimate the magnitude of variability in the twentieth century.

…

Our study documents pronounced differences between the observed and CMIP5-simulated climate variability in the twentieth century. These differences are dominated by a coherent multidecadal hemispheric-scale signal present in the observed SST and SLP fields but completely missing in any of the CMIP5 simulations.

…

Our results are also broadly consistent with recent analyses of Cheung et al. [2017], who documented substantial mismatches between their estimated internal components of the observed and CMIP5-simulated AMO, PMO, and NMO variability. However, these authors used subtraction of the scaled CMIP5 MMEM signal to deduce the internal variability in historical simulations of individual CMIP5 models. Kravtsov et al. [2015] and Kravtsov and Callicutt [2017] showed that the residual variability so defined misrepresents the true internal variability in CMIP5 simulations and is, in fact, dominated by model error, that is, the differences between the true forced response of individual models and the MMEM response. The magnitude of the CMIP5 “internal” variability estimated by this method is, hence, much larger than that of the true simulated internal variability, and the spectral characteristics of the true and estimated internal variability are entirely different.

…

Despite our explicit decomposition of the climate variability into the forced and internally generated components, dynamical attribution of the multidecadal model-data differences still remains uncertain. On one hand, if our derived CMIP5-based forced signals are realistic, these differences must arise from internal climate system dynamics presumably misrepresented in CMIP5 models, such as sea ice dynamics [Wyatt and Curry, 2014], oceanic mesoscale eddies [Siqueira and Kirtman, 2016], positive cloud and dust feedbacks [Evan et al., 2013; Martin et al., 2014; Brown et al., 2016; Yuan et al., 2016], or SST-forced NAO response [Kushnir et al., 2002; Eade et al., 2014; Stockdale et al., 2015; Siegert et al., 2016]. On the other hand, however, it is possible that CMIP5 models underestimate multidecadal variations in the true response of the climate system to external forcing or misrepresent the forcing itself [Booth et al., 2012; Murphy et al., 2017]; if this is true, the model-data differences reflect the mismatch between the actual and CMIP5-simulated forced signals, whereas the real world’s internal climate variability may be consistent with that simulated by the models. In either case, we strongly believe that model development activities should strive to alleviate the present large discrepancies between the observed and simulated multidecadal climate variability, as these discrepancies hinder our fundamental understanding of the observed climate change.

The paper is here:  http://ift.tt/2toJIrB

The SI is here: http://ift.tt/2tXOk50

One of the figures from the SI shows the differences between the models forcing response and observed natural variability seen in AMO, NAO, etc cycles:

Figure S4: Raw observed indices (thin lines) and their estimated forced components — ensemble mean (thick lines) and uncertainty (error bars) — with the forced-signal estimates based on the Community Earth System Model (CESM) Large Ensemble Project (LENS) simulations (Kay et al., 2015). Forced signals were estimated using Kravtsov and Callicutt (2017) methodology, as (a) the rescaled (unfiltered) ensemble mean over the 40 historical LENS simulations (left panels), or (b) — as the rescaled 5-yr low-pass filtered ensemble means for 20 synthetic sub-ensembles of 5 simulations, each randomly drawn from the parent 40-member LENS ensemble. The index abbreviations are given in panel captions. Comment: The forced signals based on the entire LENS ensemble and its 5-member sub-ensembles are consistent.

But here is the real smoking gun:

Figure 1. Standard deviations (STDs) of the estimated observed (blue) and CMIP5-simulated historical (red) and control-run (black) internal variability for the five indices considered; top-to-bottom rows correspond to the results for the AMO, PMO, NMO, NAO, and ALPI indices, respectively. Also included are the estimates of the observed internal variability based on the one-, two-, and three-factor scaling methods of Frankcombe et al. [2015]; see legend. The STDs were computed for raw and boxcar running mean low-pass filtered time series using different window sizes of 2 ×K + 1 yr, K = 0 , 1, … , 30 (shown on the horizontal axis); K = 0 corresponds to raw annual data, K = 1—to 3 year low-pass filtered data, and so on. Error bars show the 70% spread of the STDs, between 15th and 85th percentiles of the available estimates of internal variability (see text for details). Shading indicates the range in which the observed internal variability is statistically larger than its historical (light shading only) or control-run counterparts (dark shading and light shading regions combined), at the 5% level; here KC2017 methodology was used to estimate the observed and simulated internal variability over the historical period. The NAO plot also includes the results (heavy black curve) based on an alternative, station-based observed NAO index (http://ift.tt/1r0R5cp /climate-data/ hurrell-north- atlantic-oscillation-nao-index-station-based). (left column) The results based on the full annual data; (right column) the results based on the anomalies with respect to the leading M-SSA pair of the corresponding observed or simulated realization of internal variability (see text for details); the M-SSA embedding dimension M = 20. Comments: (i) The simulated multidecadal variability is much weaker than observed (Figure 1, left column). (ii) Much of this model-data difference is rationalized by the leading M-SSA pair (Figure 1, right column).

h/t to Dr. Leif Svalgaard