AOSN-II August 2003

ESSE Uncertainty Forecasts

ESSE uncertainty initialization and forecast procedure

The ESSE forecast for August 14, 0000 GMT was initialized from an error nowcast for August 12, 0000 GMT. The background ocean field on August 12, 0000 GMT is a HOPS forecast simulation which assimilates all calibrated data up to 1300 GMT on August 11.

The most recent observations were collected during August 11 and made available on August 12, 0300 GMT. The impact of these recent data on the the August 12 0000 GMT error nowcast is illustrated by the non-dimensional data impact field for August 11 (see Table below). Of course, the ESSE error forecast depends on such data impacts. Nowcast errors are reduced at the data locations and uncertainties are then evolved and redistributed via advection, internal growth and external forcings. All of these effects are estimated in the ESSE error forecast for August 14 0000 GMT (see Table below).

Error Subspace Statistical Estimation (ESSE) Results - 60 members
Full Domain
Temperature Salinity U velocity component V velocity component
Monterey Bay Zoom
Temperature Salinity U velocity component V velocity component
Barotropic Streamfunction (including Monterey Bay) Impact of data collected August 11


ESSE Results - 272 members
Full Domain
Temperature Salinity U velocity component V velocity component
Monterey Bay Zoom
Temperature Salinity U velocity component V velocity component
Barotropic Streamfunction (including Monterey Bay)

U and V velocities are the two components of the internal velocity (i.e. total velocity - vertically averaged velocity) along the domain coordinate system (zonal is 29.4 degrees northeast from a latitude, meridional is 29.4 degrees northwest from a longitude)

Analyses of ESSE Results

A comparison of the error standard deviation estimates computed with 272 members with these computed with 60 members illustrates that: - adding more ensemble members leads to more uniform and less spotty error estimates. It is mainly because each perturbed simulations contributes to the total error fields at different locations and with different amplitudes. This can also be explained mathematically. - an ensemble size of about 50 members already provides a good estimates of error variances (note that for error covariances, one usually needs more members than for the variances).

In both the Monterey Bay and full domains, it is interesting to look at the data impact figure and to compare the forecast error fields of yesterday (for August 13) with the forecast error fields of today (in tables above). For example, in the full domain, let's compare the forecast surface T errors of the August 11 product (does not contain the August 11 data) with the same field but for today's product (contains the August 11 data). One sees that the regional data impacts have been advected by the flow and have reduced errors accordingly. One also sees that the calibrated data for August 11 influences much more errors in and around the Bay than in the full domain which is relatively un-sampled. Similar remarks can be made for the other fields. In Monterey Bay, looking at the data impact zoom, and comparing the surface T errors of the August 11 and August 12 products again shows the reduction of errors due to data and advection. It also shows that certain areas of the Bay (e.g. near 36' 48'' and near Pt Sur) were not yet much affected by the August 11 sampling.

Focusing on the full domain, the salinity error field is again clearly influenced by the meandering of fronts offshore. In the southern portion of the domain, the strong wind forcing, strong internal dynamical responses, and open-boundary conditions lead to a surface error amplitude which is getting close to the variability that occurred in the past 2 weeks. In this southern surface region, without new data, we will soon reach the predictive capability limit of our current data assimilative system.

In the Monterey Bay zoom, velocity uncertainties are largest just along the coast, near Pt Sur. Locally, uncertainties are forecast to increase substantially due to the strong southward winds and uncertain position of the corresponding upwelling front and strong currents. Lack of data in this area (too strong currents) also limits the control of uncertainty growth.

Adaptive sampling recommendations

  • The WHOI adaptive gliders should rapidly move to areas upstream of the strong currents and high error spots in the Bay, possibly crossing the corresponding fronts if possible.

  • The SIO gliders and ships should try to collected data between 36.25N-36.15N, just upstream of the large southern error spots, so as to regain predictive capability in this area.



    Return to the Harvard AOSN-II home page