March 25, 2010 Meeting Summary
Mark DeMaria presented some preliminary results on the comparison of ocean heat content from HYCOM with satellite retrievals. Mark introduced his talk by showing the equation he used to calculate the ocean heat content (OHC), which he defined as the amount of energy in a column of the ocean per unit area. In the equation presented, density (rho w) and the specific heat (c w) were multiplied by the integration of temperature in excess of 26 degrees Celsius over the depth. Mark used 1030 kg/m3 for the density and 3850 J/kg-degrees Celsius for the specific heat value. An example calculation of OHC was presented, and Mark mentioned that the magnitude of OHC for this work was expected to be near 100 kJ/cm2. Mark then gave a description of OHC saying that it provided a bulk measure of the upper ocean structure and showed more variability than SST. A positive correlation can also be seen in statistical intensity models between OHC and intensity changes in storms. Mark also mentioned that he used HYCOM in this work because satellite-altimetry based retrievals were available for comparison with its ocean model forecasts. These retrievals were used as a ground "truth" for this work.
Next Mark presented an overview of the datasets used in this work. The OHC data from HWRF-HYCOM runs was provided by Hyun-Sook Kim for Hurricane Ike and was initialized at 00 UTC from the 2-13 of September 2008. Forecast fields were provided every 24 hours through 120h forecasts with a variable resolution from 6-16km. The corresponding OHC data from the satellite retrievals didn't match HYCOM data in resolution or timescales. Satellite data from the previous 10-20 days were used along with Reynolds weekly SST values at a resolution of 50km. Mark mentioned that the satellite retrieval data was useful to capture the pre- and post-storm environment. To further illustrate the difference between the HYCOM and satellite OHC values, Mark presented a HYCOM grid showing every 5th point compared to a grid of every 5th point for satellite retrievals. The HYCOM grid points were much denser than those for the satellite retrieval, especially towards the U.S., Central, and South American coasts. To be able to fairly compare HYCOM to the satellite retrievals, Mark interpolated and smoothed the HYCOM grid to a 0.5 degree retrieval grid using Barnes analysis where neighboring grid points were averaged together.
Then Mark presented plots comparing OHC values from HYCOM and satellite retrievals for the 20080902 00 UTC run of Ike. The first plot, showed the original HYCOM OHC data (top plot of slide 6) featuring many small scale features. The bottom plot on slide 6 showed the smoothed HYCOM OHC field. Some small-scale features are still visible, but not as many as in the original. The top plot of slide 7 again showed the smoothed HYCOM OHC field shown in slide 6 compared to the satellite OHC in the bottom plot. Mark noted that the satellite OHC had a similar structure to the HYCOM data, but the magnitude was off by 2.5. Next, Mark showed a comparison of HYCOM smoothed OHC (top plot in slide 8) to the scaled satellite OHC (bottom plot in slide 8). The HYCOM OHC showed a narrow Gulf Stream while the satellite OHC showed a "washed out" version of the same feature. The satellite OHC was, however, similar in magnitude to HYCOM OHC and had the same basic features.
Next, Mark discussed some preliminary results from his HYCOM/satellite OHC comparisons for the first Ike forecast which was initialized at 00 UTC on September 2, 2008. Scatter plots were shown comparing the HYCOM OHC (x-axis) to the satellite OHC (y-axis) over a 0.5 degree domain, making for approximately 8200 comparison points per validation time basin-wide. While comparisons were made every 24 hours from 0-120h, only results from 0h and 120h forecasts were shown here. The scatter plot for t=0h shows an obvious offset between HYCOM and satellite OHC data with much higher OHC values in HYCOM. However, the correlation between the two was near 0.75. Mark reasoned that the lower satellite OHC values compared to HYCOM could be due to landmass interference. The 120h forecast scatter plot showed more spread than for 0h and a little lower variance. Next, Mark showed a plot of variance explained or regression slope (on the y-axis) versus time (x-axis) for one HYCOM forecast. Here, the variance is pretty constant (in blue), while the slope of the regression increases slightly with time. Mark noted that for these values, there was no bias in this plot except that due to the offset between the HYCOM and satellite OHC data.
Mark concluded by summarizing his preliminary conclusions and future steps for this work. Overall, the basin-wide OHC from HYCOM and satellite retrievals appeared to be highly correlated, however, more work on determining the source of the offset between the data must be done. The correlation also decreases slowly with time between the two, which should not be surprising when working with an oceanic time scale. For future work, Mark planned to perform statistics on the complete sample of Ike runs, and also perform statistics on a more storm-centric domain. Mark also wanted to address the issue of there being a large amount of satellite data over time, so perhaps only using pre- or post-storm retrievals would be useful for this work. Mark also planned to test HYCOM's OHC values in the SHIPS statistical intensity model instead of the satellite OHC. He presented a plot showing intensity error for HWRF (in dark blue), GFDL (light blue), SHIPS (yellow) and LGEM (red) over 2007-2009. Mark speculated that the use of OHC data with the SHIPS model would further reduce the intensity error for that model.