MODES OF NATURAL AND FORCED CLIMATE VARIABILITY IN 6 YEARS OF AIRS AND AMSU DATA Alexander Ruzmaikin, Hartmut H. Aumann Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Drive, Pasadena, CA 91109, USA; emails: Alexander.Ruzmaikin@jpl.nasa.gov; Hartmut H. Aumann@jpl.nasa.gov ABSTRACT We use the Atmospheric Infrared Sounder (AIRS) and Advance Microwave Sounding Unit (AMSU) data obtained on Aqua spacecraft to study mid-tropospheric temperature variability in 2002-2008. The AIRS and AMSU deliver accurate, simultaneous measurements of the state of the at- mosphere twice per day. We investigate the temperature variability at the surface and in a broad layer centered on 400 hPa in a zonally averaged N region over the ocean. Taking into account the nonlinear and non-stationary behav- ior of the temperature we use the data analysis adaptive to the data, the Empirical Mode Decomposition, to separate the atmosphere response to the CO increase from the modes of natural climate variability. Our tentative conclusions are as follows: (1) the AIRS record shows a phase shift relative to CO and a trend in accord with the increase of CO ; (2) the simultaneous AMSU record is in agreement with the cooling of the tropical ocean that may be caused in part by the decline in solar activity. Index Terms— IR Sounder, Empirical Mode Decomposi- tion 1. INTRODUCTION The current notion of the global warming, reflected in the IPCC report (2007), is mainly based on the global temper- ature trends of about 0.1-0.2 K/decade predicted by the long- term model simulations forced by increasing concentration of greenhouse gases. Although the ground-based thermometer measurements are in general agreement with model simula- tions the satellite-based observations are not well reconciled with the predicted trends [Santer et al, 2000]. Detection of weak long-term trends depends on the time span of the data, on the magnitude of variability and the noise in the data. One of the main problems impeding the identification of trends on a strong background of natural variability and noise is the rel- atively short uniform records produced by satellites. Another problem could be the use of linear data analysis techniques. A commonly used method of trend identification consists of a representation of the measurement of a climate variable (say temperature or water vapor) as a sum of a climatological con- stant, a seasonal component (represented by a sinusoid or a few sinusoids), a linear trend and a (correlated or uncorre- lated) noise (c.f. Weatherhead et al., 1998), or as a sum of a linear trend and a term that includes noise and natural vari- ability (Leroy et al., 2009). The follow-up application of the least square estimators results in conclusions that one needs several decades of observations to identify the trend predicted by the GCMs at a reasonable level of statistical significance (95% or 90%). Here we investigate a non-linear, adaptive data analysis method called the Empirical Mode Decomposition [Huang and Wu, 2008] to separate the noise, natural variabil- ity and trend. We apply it method to the Sep. 1, 2002-Aug. 30, 2008 temperature record taken on board of AQUA satel- lite by Airs and AMSU instruments [Aumann et al, 2004]. 2. THE METHOD The EMD method is designed to deal with non-stationary, nonlinear time series such as climate records. The EMD represents the data as a sum of a small number of quasi- orthogonal empirical modes that have time-variable ampli- tudes and frequencies data = Re[ where Re means the real part and the is a non-oscillating residual term [Huang and Wu, 2008]. The number of modes depends on the number of the data points as . Each mode is equivalent to an adaptively filtered signal in an em- pirically determined (not imposed!) frequency band. A mode has an envelope defined by local maxima and minima so that its mean amplitude is zero everywhere. A mean period of a mode can be determined by the number of its maxima. Since data contain noise, which can be subjected to the same decomposition, it is important to know whether each mode represents a true signal or a component of noise. EMD modes of white or colored noise have progressively double III - 81 978-1-4244-3395-7/09/$25.00 ©2009 IEEE IGARSS 2009