Predicting wind field in the Bay of Bengal from scatterometer observations using genetic algorithm

Abstract

A technique based on genetic algorithm (GA) is applied for predicting wind field in the Bay of Bengal (BOB) using satellite scatterometer observations. Empirical orthogonal function (EOF) analysis is used for compressing the spatial variability into a set of eigenmodes. The time series of each principal component (PC) is subjected to singular spectrum analysis (SSA) and GA is applied to the resulting filtered time series. The forecast PCs are weighted by the spatial eigenmodes for computing forecast wind fields. Predictions made up to 5 days in advance are found to be superior to forecast by persistence method.

Forecasting of scatterometer‐derived wind fields in the north Indian Ocean with a data‐adaptive approach

“…[6] The scope of the present article is to investigate the performance of GA for forecasting wind field simultaneously in both the basins of the north Indian Ocean (AS and BOB). This is the major difference between this study and the previous studies by Sharma et al [2007aSharma et al [ , 2007b. Also, Sharma et al [2007a] did not use singular spectrum analysis (SSA) for filtering out the random noise in the principal component (PC).…”

“…However, even for third-day forecast, the RMSE is mostly of the order of 1.5-2 m/s all over the AS. Earlier, Sharma et al [2007aSharma et al [ , 2007b carried out similar studies in the AS and BOB. It is interesting to compare the RMSEs of the present study in these two regions with the RMSEs of these two previous studies for the purpose of ascertaining the improvement, if any.…”

“…Hence the PC time series should be subjected to some kind of filtering to isolate the deterministic signal in the data. Such an adequate approach is the data‐adaptive approach of SSA The approach is described in detail by Penland et al [1991] and Casdagli et al [1992] and has been used by Alvarez [2003] and Sharma et al [2007b]. It has to be, however, borne in mind, that SSA can induce at least some degree of degeneracy induced by autocorrelated noise, affecting forecast performance.…”

“…In other regions the RMSEs are comparable. As regards the BOB, it can be seen from the previous study Sharma et al, 2007b] that the RMSE of speed forecast in the northwestern part is more than 3.5 m/s. It is quite clear from Figure 4 that the RMSEs for zonal as well as meridional components are much lower, signifying improvement in speed forecast.…”

“…It worked, possibly because the deterministic part was predominant in the time series. Later, Sharma et al [2007b] used SSA in the case of BOB to improve upon the result. It is, however, instructive to use the SSA for filtering out noise in the PCs of the entire north Indian Ocean basin consisting of AS and BOB.…”