ALBERT

Notes: Summary A method for predicting the movement of cloud bands within the Southwest Monsoon based on infrared satellite imagery is described. The prediction scheme utilizes empirical orthogonal functions of large order time-space fields derived from daily sequences of TIROS-N AVHRR/GAC temperature data. Forward prediction stems from the dominant eigen-modes corresponding to the low-order fluctuations of monsoon cloudiness. The scheme is tested over the 1979 Southwest Monsoon domain throughout the active May–August monsoon season. Pixel averaging is employed to construct manageable time sequences of equivalent black-body temperature arrays which are then transformed into orthonormal representation. The right hand singular vectors correspond to eigenvectors in time; the left-hand singular vectors correspond to principal components in space. Physical interpretation of the initial eigenvectors considers the impact of both fixed geographic land/ocean configuration and monsoon oscillations. The effect of eigenvector forcing arising from the orthogonality constraint is discussed. A hierarchy of eigenvectors selection rules is developed in order to insure that there is some statistical and physical significance associated with the selected eigenvectors used in actual forecasts. Two prediction models are incorporated; the first is a straightforward Fourier extension of the eigenvector set which passes the selection rule criteria. The second is a modified Fourier extension procedure which considers dominant harmonics not necessarily associated with the discrete Fourier frequencies. The scheme is designed for predicting surges and breaks in cloudiness for the mid-and late-season monsoon, after it has become established. The foundation of the prediction scheme is based on the quasi-periodic intraseasonal modes associated with the Southwest Monsoon. Two objective verification schemes are used to assess the accuracy of the forecasts. Measures of skill are compared to persistence forecasts for a point of reference. A subjective approach which consists of visual inspection of video imagery is also incorporated. There is forecast skill above persistence although the overall objective skill scores tend to be lower than subjective evaluation would indicate. This arises from geography influences and eigenvector forcing which impacts contrast and clarity degradation to the predicted cloud fields. It is then shown how the predictions can be improved by first removing the time invariant properties of the infrared scenes. This procedure effectively reduces the influence of geography in the highest order eigenvectors, and thus improves forecast skill scores.