The data were colleted at 15 s intervals throughout the duration of each cruise.

Wind data collected on board were post-processed to true winds using a software algorithms developed by Shawn R. Smith and Mark A. Bourassa for the WOCEMET software analysis package (wocemet@coaps.fsu.edu). The algorithms was implemented in a matlab function (truewind1.m; written by Hal Batchelder, hbatchelder@oce.orst.edu) that takes 1) direction the bow is pointing, 2) course over which the vessel is moving (may be different from bow direction), 3) speed of vessel over the ground, 4) wind direction referenced to the ship, zero line reference (e.g., angle between the bow and the zero line n the anemometer), and a convention for reporting the output (conv = 0 is meteorological; conv = 1 is oceanographic). The function returns 1) true wind direction, referenced to the fixed earth, 2) true wind speed, referenced to the fixed earth, and 3) apparent wind direction.

Relative fluorescence and wind data were significantly noisier than most other data types. To reduce the high-f equency noise, east wind, north wind and relative fluorescence were filtered (averaged) over a 3 min sampling window (12 observations), although the data are still reported here at 15 s intervals.