Estimation of fractional cloud cover from all-sky digital images at sea  Motivation and background  MORE cruises: all-sky digital images of cloud  Testing existing algorithm for estimation  Comprising existing algorithm with visual: - for low clouds condition - for high and mid clouds condition  Conclusions Alexey Sinitsyn Sergey Gulev Krinitsky Mikhail

Motivation:  Accurate estimation of cloud fraction is critical for the quantitative computation of surface short wave and long wave solar radiation at sea.  Today science advanced to a point when we can engage information about cloud types in the analysis of solar radiation considerably extending the abilities of routine observations of cloud fraction taken visually.  This is particularly important under overcast conditions frequently associated with the mixed cloud types and, therefore, optical thickness of the cloudy atmosphere

MORE cruises: (for all-sky image)  October 2007 R/V “Academician Ioffe” Santa Cruz (Spain) – Montevideo (Uruguay)  October – November 2009 R/V “Academician Ioffe” Kaliningrad (Russia) – Cape Town (RSA)  September 2010 R/V “Academician Ioffe” cape Farewell (Greenland) – Szczecin (Poland) Days

All-sky image package  Nikon D80 with 10,2 effective pixels  Sigma 8mm F3,5 Circular-Fisheye  Angle of view for APS-C sensor 180 ° х 120 °  Projection of fisheye lens is equidistant projection  All shoots were taken with hands Data collected: :  62 days of all-sky digital images with 1-hour resolution  Simulteneous hourly observations of standard meteorological variables (clouds, air temperature, humidity, SST) Example of all-sky digital image Sigma 8mm F3,5 EX DG Circular-Fisheye

Existing algorithm for cloud recognizing (ii) Sky Index for Kalisch J., Macke A. (IFM-GEOMAR), 2008: Sky Index (IFM SI) = (DNRed) / (DNBlue ), DNBlue = digital number of blue channel DNRed = digital number of red channel Sky Index (SI) = (DNBlue – DNRed) / (DNBlue + DNRed), DNBlue = digital number of blue channel DNRed = digital number of red channel Empirical threshold 0,1. If SI <= 0,1 then consider pixel as cloudy (i) Sky Index for Yamashita M. and etc, 2004: Empirical threshold 0,8. If SI > 0,8 then consider pixel as cloudy

Comparisons of the algorithm cloud cover estimation for batch processing Applied to batch processing both algorithm for cloud estimation shown same result. Corr = 0.99 Stddev = 0.19

Comparisons of the SI algorithm cloud cover estimation with visual data 2007 Corr = 0.79 Stddev = Corr = 0.78 Stddev = Corr = 0.84 Stddev = 1.89

2007 Corr = 0.92 Stddev = Corr = 0.77 Stddev = Corr = 0.92 Stddev = 1.97 Comparison of calculations with observations in the case of low clouds

Comparison of calculations with observations in the case of high and mid clouds 2007 Corr = 0.69 Stddev = Corr = 0.84 Stddev = 2.10

Potential causes of errors 20% of FOV = 2 cloud amount Rigid threshold

 Using these parameters retrieved from the sky images we improve the accuracy of computation of cloud dependent radiative fluxes at sea surface.  Use only one threshold value - incorrectly. Within each series of photos needed correction threshold.  For further work with the cloud amount and the type of cloud we need to use Circular-Fisheye for crop matrix. Conclusions: