GLOBAL BIOGEOCHEMICAL CYCLES, VOL. 10, NO. 1, PAGES 57-69, MARCH 1996 Oceanic primary production 2. Estimation at global scale from satellite (coastal zone color scanner) chlorophyll David Antoine, Jean-Michel Andrt, ��� and Andr6 Morel Laboratoire de Physique et Chimie Marines, Universit6 Pierre et Marie Curie et CNRS, Villefranche sur Mer, France Abstract. A fast method has been proposed [Antoine and Morel, this issue] to compute the oceanic primary production from the upper ocean chlorophyll-like pigment concentration, as it can be routinely detected by a spaceborne ocean color sensor. This method is applied here to the monthly global maps of the photosynthetic pigments that were derived from the coastal zone color scanner (CZCS) data archive [Feldman et al., 1989]. The photosynthetically active radiation (PAR) field is computed from the astronomical constant and by using an atmospheric model, thereafter combined with averaged cloud information, derived from the International Satellite Cloud Climatology Project (ISCCP). The aim is to assess the seasonal evolution, as well as the spatial distribution of the photosynthetic carbon fLxation within the world ocean and for a "climatological year", to the extent that both the chlorophyll information and the cloud coverage statistics actually are averages obtained over several years. The computed global annual production actually ranges between 36.5 and 45.6 Gt C yr -1 according to the assumption which is made (0.8 or 1) about the ratio of active-to-total pigments (recall that chlorophyll and pheopigments are not radiometrically resolved by CZCS). The relative contributions to the global productivity of the various oceans and zonal belts are examined. By considering the hypotheses needed in such computations, the nature of the data used as inputs, and the results of the sensitivity studies, the global numbers have to be cautiously considered. Improving the reliability of the primary production estimates implies (1) new global data sets allowing a higher temporal resolution and a better coverage, (2) progress in the knowledge of physiological responses of phytoplankton and therefore refinements of the time and space dependent parameterizations of these responses. Introduction Photoautotrophic production by oceanic phytoplankton is a key process for the oceanic carbon cycle. Only a small fraction of the organic matter formed by photosynthesizing unicellular algae (i.e., only a fraction of the "total" primary production) is exported from the upper lighted layers of the ocean toward the deep layers (the "export production" Eppley and Peterson, [1979]). This downward flux is however sufficient to maintain the surface CO 2 concentration lower than it would be in an abiotic ocean, as a result of the so-called "biological pump" [e.g., Sarrniento et al., 1990 Antoine and Morel, 1995b]. Before attempting to estimate this depressive effect, as well as the amount of exported carbon, a prerequisite is the determination of total primary production by the algal standing stock. The knowledge of this basic process which controls the rate of inorganic carbon fixation, indepen- denfly of its further fate, is essential 'm understanding the present carbon cycle, as well as in predicting the response of marine biota to possible changes in radiative or other physical forcing due, for instance, to global warming. The present study is specifically INow at Centre Orstom de Noumta, Noumta, Nouvelle Caltdonie. Copyright 1996 by the American Geophysical Union. Paper number 95GB02832. 0886-6236/96/95GB-02832510.00 devoted to the estimate of total primary production within the whole ocean, as it can be presently assessed from the first global view of the phytoplankton distribution, which is available thanks to the coastal zone color scanner (CZCS) data archive. The assessment of new and export production, out of the scope of the present study, would require additional information and specific works. Many uncertainties still remain about the magnitude of global primary production published values range from about 20 to 50 Gt C yr -1 (1 Gt = 1015g)[e.g., Sundquist, 1985 Berger, 1989]. in this paper it is intended (1) to demonstrate the possibility of assessing ocean primary production from satellite data, (2) to test the usefulness of a method previously developed with this aim, and (3) to examine the sensitivity of the results to various hypotheses and environmental conditions. General Outlines of the Method In essence, the method presently used [Antoine and Morel, this issue] allows the oceanic primary production, P, which is realized within the productive lighted layer, to be computed from the sea surface chlorophyll concentration (Chlsat, mg Chl m'3), as detected by an ocean color sensor. The computation is based on the following global equation [Morel and Berthon, 1989] �� P= (1/Jc)Chlto t PAR(0+) ��� * (1) 57

58 ANTOINE ET AL.: OCEANIC PRIMARY PRODUCTION where Chlto t represents the column integrated chlorophyll content (g Chl m'2), PAR(O +) is the photosynthetically available radiant energy (within the spectral range 400-700 rim) incident at the sea level per unit area and for a given lapse of time (e.g., one day, J m'2). The factor ���* has the dimension of a cross section of algae for photosynthesis, per unit of areal chlorophyll biomass, and is expressed as m 2 (g Chl) '1. Note that the numerical value given to Jc, which represents the energetic equivalent of photo- synthetic assimilate (expressed as kJ (g C)'I), has no numerical effect, provided that the same value is used when computing P and when generating the ���* values (see below). The product P.Jc provides the amount of photosynthetically stored radiant energy (PSR) [Morel, 1978]. The productive layer here considered, D, exceeds in thickness that of the euphotic layer, Z e, commonly defined as that depth where PAR is reduced to 1% of its surface value. As D extends down to the 0.1% light level, it can account for the existence of deep chlorophyll maxima, frequently, even if not systematically, located around or below Ze. The column-inte- grated chlorophyll content thereafter used, Chlto t, is defined and computed as the integral of the vertical chlorophyll profile from 0 to D. Given a Chlsat concentration field derived from space obser- vations, the computation of the photosynthetic carbon fixation rests on its pairing with "climatological" fields of the incident radiation and of the cross section ���*. The ���* field is preliminarily prepared and generated by systematically operating a spectral light-photosynthesis model [Morel, 1991] in various environmen- tal conditions selected to encompass those expected when the global ocean is concerned. For fast and practical use, this field is conveniently represented by two five-dimensional lookup tables, with, as entries, the date, latitude and cloudiness index (all three determine the amount of radiant energy and its daily course), the Chlsat concentration (from which Chlto t is derived), and finally the temperature (because the physiological parameteriza- tion includes a dependence of photosynthesis upon temperature). The first table contains the ���* and resulting P values computed when a uniform vertical biomass distribution is assumed in the second table, these values are computed by assuming structured algal profiles that may include a deep chlorophyll maximum [Morel and Berthon, 1989]. These tables also contain the Chlto t values in correspondence to each Chlsat entry and the daily PAR(O + ) values, stored for each date-latitude-cloudiness triplet. Therefore for each pixel within a satellite map, that is, for any given Chlsat concentration, with the corresponding date and latitude, and under the proviso that cloudiness and temperature are also available (through other data sources), P is straightforwardly obtained in the lookup tables with the appropriate interpolations. In essence, this method was already applied to the western and the whole Mediterranean basins [Morel and Andrg, 1991 Antoine et al., 1995], even if some hypotheses and parameters were different. In its presently generalized form, it is used to tentatively derive the photosynthetic carbon fixation and its month-by-month evolution at the scale of the various provinces of the word ocean. To prevent some misunderstanding about the meaning and capability of the present method, it must be recalled (see discus- sion by Antoine and Morel [this issue]) that such computations follow a purely diagnostic approach. The primary production actually is derived (in principle on a daily basis) from an instanta- neous chlorophyll standing stock. The actual pigment content as well as its vertical distribution reflect past influences of nutrient availability, grazing pressure, sinking, decay and physical forcing, which all have fixed and controlled the algal population at its observed level. These phenomena are in no way represented in the model itself which cannot, for this reason, be utilized as a prognostic tool in predicting the phytoplankton evolution. In practice, with the satellite data in hand for the present study, the logic of the method is somewhat twisted to cope with the lack of information on a daily basis. The chlorophyll content in each pixel must be assumed as constant over a longer period, namely over 1 month, as imposed by the availability of Chlsat information in the global CZCS composite images used thereafter. It is a limitation, which is necessarily accepted for this first attempt to derive global productivity it is anyway less severe than that resulting from the compilation of shipbound production measurements, dramatically scattered in space and time. Basic and Ancillary Data The 12 "climatological monthly mean" global chlorophyll images, as derived from the CZCS archive [Feldrnan et al., 1989] by NASA's Goddard Space Flight Center (GSFC), have been rearranged as 512x512 pixel arrays (pixel size is 78 km by 78 km at the equator, smaller at higher latitudes). Such a degraded spatial resolution, however, is much better than the resolution for the other kinds of data needed in the computation. These monthly maps, where all years (1978 until 1986) were merged, still remain incomplete because the CZCS data acquisition was not systematic, and also because some regions were persistently cloudy. The resulting coverage and sampling statistic were analyzed in a study by Yoder eta/. [ 1993], to which the reader is refered to for further information. Simple interpolation procedures have been used to fulfill these zones where information is missing. The "correction" procedure, as developed by Yoder et al. [1993] for erroneous (i.e., overestimated) Chlsat values derived at high latitudes ( 40��), has been adopted and extended to the whole year (correction is applied for the 12 months, and not only during the 5 months centered on the winter solstice of each hemisphere). Thus the CZCS data for the high latitude belts are replaced by annual time series of chlorophyll concentration. For the southern Ocean the time series of chlorophyll at the "KERFIX" station (near Kerguelen Island in the southern Indian Ocean M. Fiala and D. Ruiz Pino, personal communication, 1994) have been used, and for the northern hemisphere, time series at Ocean Weather Stations PAPA and INDIA have been used as by Yoder et al. [ 1993]. The frequency distribution of chlorophyll within the upper ocean deserves some comments. On the basis of statistical analy- ses of area versus chlorophyll concentration carried out in restricted zones, it has been recurrently stated that the upper layer pigment content would be spatially distributed according to (approximately) lognormal laws [ Campbell and O'Reilly, 1988 Yoder et al., 1993 Campbell, 1995]. It is definitely not the case when the various oceans are globally considered and the whole range of concentrations taken into account. The distributions shown in Figure la (with log-log scales) are not lognormal and rather follow power laws, so that the probability of occurence of pixels with a given Chlsat concentration is always increasing when the Chlsat value is decreasing (with perhaps an exception in the very low concentration domain). Interestingly, the slopes of these distributions in each ocean are not far from -1, so that the