CRW Experimental Bleaching Outlook Background

1. Sea Surface Temperature Prediction

A sea surface temperature forecast model was developed by NOAA/ESRL Physical Sciences Division using the Linear Inverse Modeling (LIM). The SST forecast is the first step of the bleaching outlook system.

Weekly SSTs from the Reynolds and Smith OI data sets are consolidated onto 2x2 degree grids. The monthly mean annual cycle, averaged from 1982-2005, was interpolated to small temporal resolution and removed from the weekly data. The resulting anomalies were cast in terms of Empirical Orthogonal Functions (EOFs). That is, we compressed most of the variability and almost all of the predictable variability into a basis having a much smaller dimensionality than the original data set. Thirty of the leading EOFs that contain 75% of the data are retained in the model. Spatial scales too fine to be accurately predicted, given the amount of data available to train the forecast model, are thus eliminated. Using this compressed description of the data, we applied Linear Inverse Modeling (LIM) to estimate linear operators which, when applied to a vector of SST initial values, give the best (in the least squares sense) forecast of SST anomalies at some future time.

Both initial conditions and forecasts have smaller amplitude than the original data set as we expected because only 75% of the variance was retained. This variance is an average over the entire geographical region, but retained variance does vary with geographical location. In order to account for this reduced variance, a predicted SST anomaly at a particular geographical location is inflated by a factor reflecting how much variance at that location was retained (Fig 1). For example, SST anomalies in the main El Niño signal region, where retained variance is close to unity, are affected very little by this inflation factor. SST anomalies in a location where, say, only half the variance is retained, are inflated by a factor of the square root of 2. Fig 2 shows a sample of predicted SST anomalies. At this point, the climatology is added to the adjusted prediction of the SST anomaly to give an estimate of the total SST forecast. The predicted SST is then fed to the second part of the model (see the next subsection) to produce predicted bleaching thermal stress.