1. Discussions  Observationally, the two leading principal modes of ISCCP high clouds are highly correlated with MEI and EMI (Fig 3) and the spatial patterns show characteristics of ENSO in the western (canonical; Fig 1) and central (Modoki; Fig 2) Pacific.  The ratio of explained variance by the second mode to that of the first mode (s 2 /s 1 ) is 8%/24% = 0.33, which is the same as that of SST. Therefore, the high cloud-SST relationship is linear in ISCCP.  In AMIP5 simulations, the spatial patterns of the two leading modes also show good correlations with the ISCCP patterns. But the temporal variations differ greatly among models; only the multi-model mean captures the temporal pattern of EMI.  The s 2 /s 1 ratios in AMIP5 vary greatly from 0.25 to 0.7 while that of ISCCP is 24%, implying a nonlinear high cloud-SST relationship in models (Fig 4). Figure 3. Temporal evolutions of the first (top) and second (bottom) principal modes. Numbers in parentheses are correlation coefficients of the time series with the Multivariate ENSO index (MEI) and ENSO Modoki index (EMI). An Analysis of High Cloud Variability: Imprints from the El Niño–Southern Oscillation * King-Fai Li 1,$, Sze-Ning Mak 2, Hui Su 3, Tiffany M. Chang 4, Jonathan H. Jiang 3, Joel R. Norris 5, and Yuk L. Yung 6 1 University of Washington, Seattle, WA 2 Chinese University of Hong Kong, Hong Kong 3 Jet Propulsion Laboratory, Pasadena, CA 4 Brown University, Providence, RI 5 Scripps Institution of Oceanography, La Jolla, CA 6 California Institute of Technology, Pasadena, CA Summary  Principal modes of near-global (60 o N–60 o S) ISCCP high cloud fraction are identified using the Principal Component Analysis.  The two types of El Niño-Southern Oscillation (ENSO) — the canonical ENSO and the ENSO Modoki — are leading sources of variability in cloudiness, accounting for 24% and 8% of the total high cloud variance, respectively.  14 models from the Atmospheric Model Intercomparison Project Phase 5 (AMIP5) show the spatial pattern and the temporal variations of high cloud fraction associated with the canonical ENSO very well but the magnitudes of the canonical ENSO vary among the models.  The multi-model mean appears to capture the temporal behavior of high cloud fraction associated with the ENSO Modoki but individual AMIP5 models show large discrepancies in capturing the ENSO Modoki temporal variations.  A new metric for the high cloud–SST sensitivity, defined by the explained variances of the first two principal components, suggests that most of the AMIP5 models overestimate the ENSO Modoki effects in high clouds.  These interannual signatures in high cloud are useful for diagnosing high cloud–SST interactions in the models. Data and Methods  ISCCP high cloud is defined as integrated column of cloud fraction with cloud top pressure lower than 440 hPa. Satellite artifacts in ISCCP data have been corrected [Norris and Evan, 2015].  Cloud simulations in 14 AMIP5 models are used: CCCma CanAM4, CNRM-CM5, CSIRO-MK3.6.0, GISS-E2-R, GFDL CM3, INM-CM4, IPSL-CM5A, MIROC-ESM, MIROC5, MPI-ESM-LR, MRI- CGCM3, NCAR CAM5, NCC NorESM1-M, and UKMO HadGEM2-A.  AMIP high cloud fraction in each model layer below 44 hPa are calculated using a random overlapping assumption [Stephens, 1984] before integration.  All data have been deseasonalized before a 3-month running average is applied.  The common principal component analysis (PCA), by means of the singular-value decomposition [Press et al., 2007], is applied to the covariance matrix of the resulting high cloud fraction datasets, which have been weighted by the square root of the cosine of latitudes. Figure 1. The spatial patterns of the first principal modes Figure 2. The spatial patterns of the second principal modes Figure 4. The ratios of the explained variances (s 1 and s 2 ) of the first and second principal modes. The observed value obtained from ISCCP and SST is shown as a dotted line. The simulated values in the 14 AMIP models are shown as filled bars. The models have been arranged in descending order of correlation coefficients between EOF2 and EMI shown in Fig 3. References. Press et al. (2007), Numerical Recipes; Stephens (1984), Mon. Weather Rev., 112, 826–867; Norris and Evan (2015), J. Atmos. Ocean Tech., 32, 691–702. $ Correspondence author: kfli@uw.edu * This work has been submitted to Climate Dynamics