7 The Son of NABE - Two Simple Models
Chris Garside
Bigelow Laboratory for Ocean Sciences
W. Boothbay Harbor, ME 04575
At the first planning workshop for NABE activities in the late 1990's we concluded that the objective of the exercise should be to measure, or at least reduce the uncertainties in the carbon sink at the basin scale. The means was to be through an improved understanding of the relevant biogeochemical and physical processes. A critical step in transferring an improved understanding generated from field observation to the basin scale is the development of appropriate tools for the purpose. We have developed two simple syntheses of data from NABE (the elder) and EqPac that have promise as models that allow spatial and temporal extrapolation to larger, and possibly basin and annual scales.
At the conclusion of the occupation of the NABE site (47°N 20°W) by the Atlantis II in early June 1988 the vessel proceeded north on 20°W to Iceland. During this transit we measured surface nutrients hourly. Nitrate increased from the limit of detection at 47°N to about 10 µM at 60°N with considerable variance at the eddy scale. We used the historical data set to construct the NO3 transect on 20°W and used Levitus' and Robinson's atlas data for the depth of the winter mixed layer to estimate winter surface NO3. The difference between these data and a smoothed fit to our transect data represent NO3 drawdown for the year to June, which in conjunction with estimates of mixed layer depth allowed us to estimate seasonal new production.
14C primary production measurements taken during NABE were regressed on PAR with a standard error of the estimate of about 15%. We were then able to use a temporal and latitudinal integration of daily PAR to estimate seasonal total production up to the start of June. Our original choice was to integrate from the onset of stratification at 47°N in late April, since water column stability and a reduced mixing depth are thought to be necessary conditions for bloom development. We obtained the unsatisfactory result that new production exceeded total production over the entire transect by about a factor of two (Figure 7.1). We obtained two other conditions from the model: 1) if the f-ratio was not to exceed 1 anywhere in the transect, primary production must commence no later than late March in advance of continuous stratification, and 2) the least value for the f-ratio is obtained when production starts in late February. This results in the highest value of the f-ratio in the transect being about 0.6 - 0.7, and earlier start dates make little difference because light levels and potential daily production are very low.
There are so few field observations of either NO3 or primary production in these months that verification of the model results is impossible, but significantly measured NO3 at the NABE site in early April was some 3 µM (25%) less than the predicted winter mixed layer value. We concluded that a significant part of seasonal primary production before June (up to 50%) probably occurs in the one to two months preceding stratification, and that the fate of this material is unknown.
The logistics of sampling at this time of year are almost as daunting as the logistics of even quasi synoptic measurements at the basin scale. The strength of satellite remote sensing lies in its ability to make measurements at all seasons and a wide range of scales. Its utility to ocean sciences depends on finding algorithms that allow remotely sensed data to stand as proxies for sea surface or upper water column properties. In this respect there have been a few published
Figure 7.1: Model predictions of total production from PAR on 20°W between 45°N and 60°N assuming production starts at the onset of stratification (Year Day 118). Measured new production from seasonal nitrate draw-down, and the resultant "f" ratio are also plotted. New production exceeds total production and the "f" ratio is greater than one, suggesting that production must start before the onset of stratification.
attempts to relate water column NO3 concentrations to temperature, but these have worked well only at timescales of days and space scales of a few tens of kilometers, and have not been otherwise satisfactory.
We began to investigate the use of T/S data to develop predictions of NO3 concentrations with the hypothesis that winter or newly upwelled water would undergo predictable (not necessarily quantifiable) temporal trends of all three properties. Temperature should increase because of solar heating, NO3 would decline as a result of photosynthesis, and salinity would either increase or decrease depending on the net of precipitation and evaporation. All of these properties vary as continua through time and are somewhat coupled to the same time scales through solar radiation (Figure 7.2.) Consequently, it should be possible to describe NO3 by a polynomial function in T and S.
Our approach was to fit progressively higher order polynomials in T and S including cross product terms to our data sets. Starting with a first order fit of NO3 to T we obtained a standard error of the prediction, then added terms stepwise, only retaining them if their addition reduced the standard error of the prediction. The general form of the equation is:
NO3 = a + bT + cS + dT2 + eS2 + fTS + gT3 ......lS4
Figure 7.2: A hypothetical depiction of the evolution of seawater properties following upwelling or seasonal stratification in temperate waters. Since nitrate, temperature and salinity all vary continuously in time, we hypothesize that nitrate can be expressed as some function of T and S.
In most data sets we found that less than 10% of the data points contributed 40% or more of the variance contributing to the standard error of the prediction, and we usually rejected these data. Little improvement in the fit was obtained beyond fourth order terms. For both EqPac and NABE we obtained algorithms that predict upper (125m) water column NO3 with a precision of ± 0.5 - 1.0 µM (Table 7.1).
These algorithms are surprisingly precise and relevant at length scales of 2300 km (EqPac), areal scales of 2 x 106 km2 and all seasons for 20 years (NABE). They can be used for a variety of purposes including validation of coupled bio-physical numerical model output, modeling new production, validation and QA/QC of data sets, interpolation and retrofitting of long term records such as moorings and historical T/S data. We also wished to see how well they could be adapted to remote sensing, for which only temperature can be adequately measured.
We took a subset of the NABE regional data set (see Table 7.1 for a description) consisting only of data from the upper 20m, and fit it with a third order polynomial in T alone. The standard error of the prediction was ± 1.05 µM. Thus in combination with historical AVHRR data (Miami AVHRR Monthly MCSST and CZCS pigment concentration data are available on CD-ROM) it will be possible to construct regional or even basin scale monthly maps of surface NO3. Monthly changes in NO3 would then be interpretable as NO3 drawdown. Depth integration would provide a monthly estimate of new production and this could be integrated through time and for the basin to obtain seasonal new production.
Large and complex (expensive) numerical models are often employed to produce descriptive information that is useful for planning purposes when major field research efforts are being designed. However, there is a diversity of less complicated (relatively inexpensive) modeling of available data that can also have a valuable role for project planning purposes. Exploitation of this diversity should be encouraged to provide a more robust basis for future planning of Son of NABE.
Table 7.1
Data Set n standard error of the estimate
TT007 901 0.82TT007* 450 0.84
TT011 899 0.68
NABEX 799 0.45
NABE 1433 1.23
NABET 449 1.05
* Fit using a randomly selected half of the TT007 data set. The standard error of the estimate was computed using the other half of the data set
TT007 is the JGOFS EQPAC spring survey cruise data set obtained during January through March 1994 data set.
TT011 is the JGOFS EQPAC fall survey cruise during August and September 1994.
NABEX is a data set collected during the JGOFS North Atlantic Bloom Experiment in April and May of 1989 in the vicinity of 47°N and 20°W. The data set consists of 54 standard hydrocasts and 13 pump stations to a maximum depth of 120m with 4m resolution, with a total of 848 observations.
NABE is a composite data set of NODC and BOFS data described in Garside and Garside (1993). Data are from many cruises and all seasons between July 1970 and June 1990 but not including NABEX, in an area between 15 - 25°W and 35 - 60°N to a maximum depth of 125m.
NABET is a subset of NABE with data from only the upper 20m fit with a polynomial in temperature alone. The purpose of this exercise was to examine the possibility of using remotely sensed AVHRR data to predict mixed layer NO3 concentrations. 449 observations were available, and in this case none was rejected.