Thanks for sending those. I can't say for sure what the issue is, and unfortunately sometimes model simulations just don't get the forecast accurately, but there are a few things you can try.
1) Anytime you're running for more than 5 days, you really should be using additional (higher resolution SST input, with the sst_update option)
2) Your domain is 10km, which can be somewhat of a "gray zone" for cumulus parameterization (read more about that
here) - meaning you may or may not need to turn on a cumulus scheme. You could consider having a nested domain inside the 10km domain (for e.g., you could use a 5:1 parent_grid_ratio and have a 15km outer domain where you keep the cumulus scheme turned on, and then a 3km nest, where it's turned off - set to 0).
4) You may also need to play around with various physics options. You are currently using the CONUS options, which typically work well for large continental areas, but I believe your domain is in the middle of the ocean, which means you would probably get better results using the "tropical"
physics suite.
Another note: you are using the "lambert" map projection during WPS, but your domain is closer to the Equator, meaning you may want to consider using a mercator projection.
