Convective signals from surface measurements at AR

Authors: Yi Wang Charles N Long James H Mather Xiaodong Liu

Publish Date: 2010/01/19

Volume: 36, Issue: 3-4, Pages: 431-449

PDF Link

Abstract

Madden–Julian Oscillation MJO signals have been detected using highly sampled observations from the US DOE ARM Climate Research Facility located at the Tropical Western Pacific Manus site Using downwelling shortwave radiative fluxes and derived shortwave fractional sky cover and the statistical tools of wavelet cross wavelet and Fourier spectrum power we report finding major convective signals and their phase change from surface observations spanning from 1996 to 2006 Our findings are confirmed with the satellitegauge combined values of precipitation from the NASA Global Precipitation Climatology Project and the NOAA interpolated outgoing longwave radiation for the same location We find that the Manus MJO signal is weakest during the strongest 1997–1998 El Niño Southern Oscillation ENSO year A significant 3–5month lead in boreal winter is identified further between Manus MJO and NOAA NINO34 sea surface temperature former leads latter A striking inverse relationship is found also between the instantaneous synoptic and intraseasonal phenomena over Manus To further study the interaction between intraseasonal and diurnal scale variability we composite the diurnal cycle of cloudiness for 21MJO events that have passed over Manus Our diurnal composite analysis of shortwave and longwave fractional sky covers indicates that during the MJO peak strong convection the diurnal amplitude of cloudiness is reduced substantially while the diurnal mean cloudiness reaches the highest value and there are no significant phase changes We argue that the increasing diurnal mean and decreasing diurnal amplitude are caused by the systematic convective cloud formation that is associated with the wet phase of the MJO while the diurnal phase is still regulated by the welldefined solar forcing This confirms our previous finding of the antiphase relationship between the synoptic and intraseasonal phenomena The detection of the MJO over the Manus site provides further opportunities in using other groundbased remote sensing instruments to investigate the vertical distributions of clouds and radiative heatings of the MJO that currently is impossible from satellite observationsThis work has been supported primarily by the Climate Change Research Division of the US Department of Energy as part of the Atmospheric Radiation Measurement ARM Climate Research Facility Dr Y Wang is supported also by K C Wong Education Foundation of Chinese Academy of Sciences Dr X Liu is supported by the NSFC National Excellent Young Scientists Fund No 40825008 The authors thank the NOAA Earth System Research Laboratory Physical Sciences Division for interpolated OLR and the NASA Goddard Space Flight Center Laboratory for Atmospheres for GPCP precipitation datasets The cross wavelet and coherence analyses are conducted using a MatLab package see Appendix 3 The NCAR command language NCL is used to process data and plot other graphs see Appendix 3 During the course of our research we greatly benefited from discussions with Drs M Wheeler C Zhang A Grinsted C Torrence and B Tian We also gratefully acknowledge three reviewers for their constructive comments that improved our studyBecause the Fourier transform assumes the data are cyclic errors will occur at the beginning and end of the wavelet power spectrum when dealing with finitelength time series Before calculating the wavelet power the time series is padded with sufficient zeroes to bring the total length up to the closest power of two hence limiting the edge effects and speeding up the Fourier transform Padding with zeros introduces discontinuities at the endpoints The edge effect is the region of wavelet spectrum where edge effects become important and is defined as the efolding time for autocorrelation of wavelet power at each scale The null hypothesis is defined for the wavelet analysis because it assumes that the time series has a background power spectrum If a peak power is significantly above this spectrum then it can be considered as a true feature with a certain percent confidence In this study we assume the background power comes from the theoretical Markov process with a corresponding lag1 correlation coefficient Torrence and Compo 1998We then calculate the Morelet wavelet power spectra and the corresponding 95 significance confidence level against the theoretical Markov process with a corresponding lag1 correlation coefficient Weng and Lau 1994 Torrence and Compo 1998 Torrence and Webster 1999 The wavelet pattern provides the seasonal and yeartoyear evolution of the extracted MJO signal Due to finite sampling the edge effect Weng and Lau 1994 Torrence and Compo 1998 is considered To calculate the Fourier power spectrum we taped 10 zeroes at the beginning and end of the time series to minimize the spectral leakage The bestfit red noise is based on the autoregressive process with a corresponding lag1 autocorrelation coefficient of the original time series The test of significance of the correlation coefficients is based on the assumption that the distribution of the time series follows the binormal distribution around its mean value null hypothesis We also apply the Fourier power spectrum analysis with a 95 confidence envelope over the bestfit red noise Torrence and Compo 1998As a backup check we derive the global wavelet power which is supposed to be consistent with the Fourier power spectrum The global wavelet power is estimated by integrating the total wavelet power in time at each frequency bands in the calculation In addition we also derive the scaleaveraged wavelet power at certain frequency bands This facilitates a comparison of variability among those frequency bands that have been averaged eg synopticscale intraseasonalscale interannualscale The scaleaveraged wavelet power is derived by integrating the wavelet power over certain frequency bandsHowever one should be very cautious about interpreting the mean phase differences into real timelags This cannot always be done and when it can it should be done with care A 90° lead might as well be a 90° lag to the antiphase Therefore there is a nonuniqueness to the solution when doing the conversion In addition a phase angle can be converted only to a time lag for a specific wavelength when two time series are near phaselocked

Keywords: