Clupeoid Population Variability, the Environment and Satellite Imagery in Coastal Upwelling Systems

Reviews in Fish Biology and Fisheries 8, 445±471 (1998)

Clupeoid population variability, the environment and satellite imagery in coastal upwelling systems

JAMES COLE1Ã and JACQUELINE MCGLADE2

1Paci®c Fisheries Environmental Laboratory, NOAA±NMFS, 1352 Lighthouse Avenue, Paci®c Grove, CA 93950, USA 2Centre for Coastal and Marine Sciences (NERC), Plymouth Marine Laboratory, Prospect Place, Plymouth, PL1 3DH, United Kingdom

Contents Abstract page 445 Introduction 446 Background to coastal upwelling systems 447 Clupeoids and the environment: a general perspective 451 The management problem 454 Recruitment variability and the environment 455 Mechanistic theories Synthesis theories Problems with predicting recruitment 459 Non-linearity Scale The multiplicity and complexity of processes that in¯uence recruitment Environmental indicies Satellite oceanography: a way forward? 462 Conclusion 465 Acknowledgements 466 References 466

Abstract Sardine, pilchard and anchovy stocks form the basis of commercially important purse seine ®sheries in eastern boundary upwelling regions. High levels of environmentally driven recruitment variability have, however, made them especially dif®cult to manage. Reliable forecasts of recruitment success would greatly help with the setting of catch quotas prior to each ®shing season. Theories of how environmental conditions in¯uence recruitment success, according to survival=mortality of the early life-history stages, can be divided into mechanistic and sythesis theories. Mechanistic theories are concerned with speci®c physical processes, whereas synthesis theories attempt to unite the various

ÃAuthor to whom correspondence should be addressed (e-mail: [email protected]).

0960±3166 # 1998 Chapman & Hall 446 Cole and McGlade mechanistic processes within a single conceptual framework. Despite the successful testing of some theories, there has been little success in reliably predicting recruitment success from a knowledge of environmental conditions. Possible reasons include the following: non-linearity in the relationship between environmental parameters and recruitment; the poor spatial and temporal resolution of much oceanographic data; the wide range of different factors involved in determining recruitment success; and the choice of environmental index. The recent compilation of time series of satellite images for these regions offers a solution to some of these problems, and in doing so reopens the possibility of ®nding suf®ciently good relationships between environmental conditions and recruitment success for management purposes. In particular, the high resolution of these time series allows for the construction of environmental indices across many different spatial and temporal scales. These time series also open up the possibility of quantifying the behaviour of upwelling systems according to the evolution of their spatial structure through time, using pattern analysis techniques.

Keywords: clupeoids, coastal upwelling, population dynamics, recruitment, satellite imagery

Introduction Clupeoid ®shes are found in pelagic environments throughout the world's oceans, and are especially abundant in the productive coastal upwelling regions found along the eastern margins of the Atlantic and Paci®c oceans (Fig. 1). Species of sardine, pilchard and anchovy (Sardinops spp. and Engraulis spp.) typically form the largest stocks. The relative ease with which they are caught by purse seiners, owing to their schooling behaviour, combined with the enormous biomasses reached by some populations, has made them important economic resources. For instance, at the peak of the Peruvian anchoveta (E. ringens) ®shery, in the early 1970s, clupeoids contributed roughly a third of total world catches, which at the time were in the region of 65 million tonnes. The Atlantic herring (Clupea harengus) aside, clupeoids tend to be short lived, rarely living beyond 5±10 years, and typically recruit to the adult stock between 1 and 3 years old (Blaxter and Hunter, 1982). High levels of recruitment variability have made clupeoid ®sh especially dif®cult to manage. Because they are usually short lived, any ¯uctuations in recruitment success translate rapidly into ¯uctuations in population size, and what may be a conservative level of exploitation during years with good recruitment may during unfavourable years result in over®shing. Despite evidence that ocean conditions play a major role in determining recruitment success, according to their in¯uence on the survival of the early life-history stages (Lasker, 1975; Cury and Roy, 1989; Bakun, 1996), ®sheries scientists remain unable to anticipate environmentally driven ¯uctuations in recruitment, and to adjust management policy accordingly. This article has three main goals: ®rstly, to review the physical and ecological factors relevant to the dynamics of clupeoid populations in eastern boundary coastal upwelling systems, including a synopsis of theories which address the causes of recruitment variability according to the survival of early life-history stages; secondly, to highlight speci®c problems with the use of traditional management techniques for these stocks; Clupeoid stock dynamics in coastal upwelling areas 447

180Њ 150Њ 120Њ 90Њ 60Њ 30Њ 0Њ 30Њ 60Њ

60Њ 60Њ N

Oregon/ Northwest 30Њ California Africa 30Њ

0Њ 0Њ

Peru

30Њ Benguela 30Њ

60Њ 60Њ S 180Њ W 150Њ 120Њ 90Њ 60Њ 30Њ 0Њ 30Њ 60Њ E

Fig. 1. Eastern boundary coastal upwelling regions, adapted and redrawn from Mann and Lazier (1991). Arrows indicate prevailing winds. and ®nally, to examine why there has been widespread failure in recruitment forecasts, and to suggest new research avenues whereby this situation might be remedied. For a thorough review of clupeoid biology and ecology, please refer to Blaxter and Hunter (1982).

Background to coastal upwelling systems Coastal upwelling in eastern boundary systems (Figs 1 and 2) is driven by the action of the prevailing equatorward winds on the surface waters combined with the Coriolis effect (see Mann and Lazier, 1991 for a full discussion). Biologically it is important because often it brings fresh supplies of nutrient into the surface layers. This, in turn, fuels high levels of primary production, resulting in a productive food web which can support large populations of ®sh, marine mammals and sea birds (Lalli and Parsons, 1993). Fluctuations in the strength of upwelling favourable winds, changes in bathymetry along the coast, dynamic instabilities in current ¯ows, and remotely forced processes, such as coastally trapped waves and warm water intrusions, typically cause a high degree of spatial and temporal variability in upwelling activity such as illustrated by the three images in Fig. 3. Strong interannual variability in upwelling and current ¯ows are related to oscillations in the atmospheric pressure ®elds over the equatorial Paci®c and Atlantic Oceans (Hisard, 1988; Mann and Lazier, 1991; Bakun, 1996). 448 Cole and McGlade

Thermal upwelling SURFACE WINDS front

North

East Warm ocean water Cool LAND water

Ekman Drift Upwelling Thermocline

SEA BED

Fig. 2. Schematic representation of eastern boundary coastal upwelling in the southern hemisphere, adapted and redrawn from Mann and Lazier (1991).

El NinÄos, and the less publicised Benguela (or `Atlantic') NinÄo (Shannon et al., 1986), are the most dramatic manifestation of these atmospheric oscillations. In coastal upwelling systems they are typically characterized by intrusions of warmer water masses from the edges of the system, a deepening of the thermocline, and subsequent reductions in the upwelling of cool, nutrient-rich water from below the thermocline. A longshore intrusion of warm surface tropical Angolan water into the northern Benguela during the strong Benguela NinÄo in 1984 (Shannon et al., 1986) is shown in Fig. 3(b). The biological impacts of these events on coastal upwelling systems can be dramatic, and are often associated with greatly reduced levels of productivity throughout the entire ecosystem and the invasion of biota from neighbouring regions (Valdivia, 1978; Boyd et al.,1985). Ecologically, these systems are generally characterized by high biomasses, high productivities, low biodiversity, and a low number of trophic exchanges between

Fig. 3. Contrasting conditions in the northern Benguela: (a) strong winter upwelling activity throughout the region and a well developed upwelling ®lament between 258 and 278 S; (b) a strong intrusion of tropical Angolan water along the Namibian coast during the 1984 Benguela NinÄo event; and (c) warming off central Namibia combined with the onshore movement of South Atlantic surface water. The arrows indicate intrusions of warmer water into the system and represent potential mechanisms for the retention of clupeoid eggs and larvae, and the concentration of food particles across the associated thermal fronts. Areas contaminated by cloud cover were interpolated according to the distance-weighted mean of the surrounding non-contaminated pixels. These SST composites form part of the Cloud and Ocean Remote Sensing around Africa (CORSA) dataset, held at the Space Applications Institute of the European Commission's Joint Research Centre. Clupeoid stock dynamics in coastal upwelling areas 449 450 Cole and McGlade primary production and ®sh production (Ryther, 1969). The species composition and trophic organization of one coastal upwelling region to another is similar, and for the pelagic layer this is much as illustrated in Fig. 4. The `bottom-up' perspective on these ecosystems places high levels of ®sh production within the context of high levels of primary production and the low number of trophic exchanges between phytoplankton and ®sh. According to this view, physical variability can propagate rapidly through the food web due to the close relationship between levels of primary production and environmental conditions, the relatively small number of trophic exchanges, and low levels of energy and material recycling within the web itself (e.g. Baird et al., 1991). In contrast, the trophic dynamics of these regions have recently been characterized as being under a substantial amount control from the middle, in both a `top down' and `bottom up' sense (Rice, 1995; Bakun, 1996). Typically the mid-trophic layers form a `wasp's waist', insofar as there are fewer species here than either towards the top or bottom of the food chain. These mid-trophic levels are in turn dominated by highly variable populations of one or two clupeoid species, whose abundance not only exerts a large in¯uence on the amount of food available to higher trophic levels, but simultaneously may also control the amount of grazing pressure exerted on lower levels. Some control can genuinely originate from the middle given that the physical

MACROPHYTOPLANKTON e.g. diatoms

MACROZOOPLANKTON MEGAZOOPLANKTON e.g. copepods e.g. euphausiids

CLUPEOIDS TOP PREDATORS e.g. anchovy, pilchard, e.g. mackerals, dolphins, sardines seals, gannets, and cormorants

Fig. 4. Major pelagic feeding relationships in an eastern boundary upwelling system. Clupeoid stock dynamics in coastal upwelling areas 451 factors thought to directly in¯uence the abundance of clupeoids extend beyond those that simply determine gross levels of primary production, as will be discussed below.

Clupeoids and the environment: a general perspective Variations in physical forcing are increasingly considered to underpin natural ¯uctuations in ®sh populations in terms of their biomass, recruitment success, and distribution (Hartline, 1980; Crawford et al., 1991; Mann, 1993; Sharp, 1995; Bakun, 1996). This section takes a broad look at the way in which the environment is known to in¯uence clupeoid populations. The problems involved in managing these species, and the ways in which physical conditions determine reproductive success, according to the differential mortality of the planktonic early life-history stages, will be addressed in greater detail in the following two sections. Ways in which environmental conditions may directly and indirectly in¯uence clupeoid populations in coastal upwelling regions are illustrated in Fig. 5. The distribution of the adult population may be directly in¯uenced by factors such as temperature, oxygen concentrations and feeding conditions. Although it is often dif®cult to isolate one particular environmental factor controlling distribution (except in cases of severe anoxia or pollution), the following observations provide general evidence for the role played by the environment: during warm events, anchovy and sardine=pilchard stocks are known to contract to small areas adjacent to the coast where some localized upwelling is maintained (Thomas and Boyd, 1985; Longhurst and Pauly, 1987); decadal shifts in the distribution of catches may be associated with decadal-scale ¯uctuations in environmental conditions (Crawford and Shannon, 1988); and lastly, adults often migrate to relatively sheltered areas to spawn, where SSTs (sea surface temperatures) are at a local maximum, and offshore transport and turbulent mixing at a minimum (Parrish et al., 1983; Le Clus, 1990, 1991). Recruitment success can also be directly in¯uenced by the environment according to how environmental factors, such as offshore transport and nutrient enrichment, in¯uence the survival and mortality of the vulnerable planktonic early life-history stages (Fig. 6). For instance, offshore transport and surface currents determine whether eggs and larvae are advected towards or away from suitable nursery areas (Parrish et al., 1983), whilst levels of nutrient enrichment and turbulent mixing in the water column determine the suitability of feeding conditions for the larvae (Lasker, 1975, 1978; Rothschild et al., 1989). Insofar as environmental conditions can directly affect the distribution of adults and predators they can also indirectly in¯uence juvenile and adult mortality via cannibalism, prediction and ®shing pressure. For instance, the extent to which adult horse mackerel (Trachurus murphyi, Carangidae) feed on anchoveta in the Peruvian system is related to upwelling conditions and sea surface temperatures. Under normal upwelling conditions, the ranges of the two species do not overlap, but when there is warming, either in summer or due to an El NinÄo event, the horse mackerel move shoreward where they can feed on the anchoveta (Muck and Sanchez, 1987; Muck, 1989). Likewise, contraction of clupeoid stocks into coastal concentrations during warm events not only makes them easier targets for ®sherman, thus causing higher catches per unit effort (Thomas and Boyd, 1985), but may also result in increased competition for food and higher levels of cannibalism of eggs and larvae. 452 Cole and McGlade

ENVIRONMENT ADULT POPULATION e.g. e.g. upwelling intensity, biomass, location, temperature, turbulence, Temperature and oxygen age structure, intrusions of warm water preferences size structure

ADVECTION

Retention vs. offshore loss of eggs and larvae

FEEDING CONDITIONS

larvae, juveniles, and adults

CANNABILISM

PREDATION

FISHING ACTIVITY

intensity location

LIFE HISTORY STRATEGIES

e.g. timing and frequency of spawing, egg size, age at maturity, fecundity

Fig. 5. How environmental conditions can affect clupeoid populations in eastern boundary upwelling systems. ÐÐÐÐ direct interactions; ÐÐÐÐ indirect interactions; ± ± ± feedback mechanisms. (See text for more explanation.) Clupeoid stock dynamics in coastal upwelling areas 453

PLANKTONIC STAGES Offshore Advection

Spawning EGGS

Predation/Cannibalism

Hatching

Offshore Advection Fishing

Predation ADULT POPULATION LARVAE Starvation

Starvation Predation/Cannibalism

Metamorphosis

Recruitment PRE-RECRUITS Starvation

Predation

Fig. 6. Generalized clupeoid life cycle, and major sources of mortality.

From an evolutionary perspective, the interaction between environmental conditions and reproductive success plays an important role in shaping the life-history strategies of these species. Pilchard, sardine and anchovy species in coastal upwelling regions all display similar life-history traits, such as serial spawning and early maturity, which allow `bet-hedging' solutions (Stearns, 1976) to the unpredictable survival of their early life-history stages. In other words, by maturing early and spreading their spawning activity over much of the year, they greatly increase their chances of hitting a sequence of environmental events favourable to the survival and growth of the early life-history stages. Furthermore, evidence from simulation models has shown that herrings, sardines, and anchovies each have reproductive strategies which are best geared to breeding under differing temporal regimes of environmental variability (Armstrong and Shelton, 1990), and that in reality we might expect to see a correspondence between the predominant spectra of environmental variability (i.e. weekly, seasonal, or interannual) in a particular region and the dominant clupeoid species. The role that environmental variability is thought to play in driving ¯uctuations in clupeoid recruitment success and biomass is substantiated by the existence of good relationships between these ¯uctuations and various environmental indices, both at regional (Shannon et al., 1988; Cury and Roy, 1989; Waldron et al., 1998) and trans- oceanic scales (Kawasaki and Omori, 1988; Lluch-Belda et al., 1989). Furthermore, the fact that different clupeoid species have different environmental `preferences' for successful reproduction means that physical conditions also play a role in mediating the 454 Cole and McGlade relative dominance of different clupeoid populations in a given area. For instance, in the eastern Paci®c, the relative dominance of anchovies versus sardines appears to correspond to inter-decadal trends in temperature conditions (Sharp and McLain, 1993); during warm periods with reduced upwelling, sardines tend to dominate, whereas during cool periods with enhanced upwelling anchovies dominate (see also YanÄez et al., 1998). Since the end of the Second World War, ®shing pressure has had tremendous impact on the population dynamics of these species, and has often made it dif®cult for ®sheries scientists to distinguish between the relative importance of ®shing versus non-®shing- related factors in driving population ¯uctuations. Nonetheless, independent evidence for large natural variations in population size prior to the onset of industrialized ®shing is provided by long-term records of guano harvests and by the relative abundance of clupeoid scales at different strata in coastal sediments (Soutar and Isaacs, 1974; De Vries and Pearcy, 1982; Crawford et al., 1987; Shackleton, 1988). The latter also con®rms reversals in the relative dominance of anchovy and sardine=pilchard populations prior to the onset of heavy industrial ®shing. Whilst density-independent factors may be primarily responsible for driving the year- to-year population dynamics and distribution of clupeoids, density-dependent processes may also have some regulatory role, and are particularly likely to come into play after favourable environmental conditions have caused population booms. For instance, Kawasaki and Omori (1995) provide evidence for the far-eastern (i.e. Japanese) sardine population being regulated by both density-independent and density-dependent processes depending on what `phase' the population was in. They argue that good environmental conditions may be what trigger population explosions when the population is small and located in coastal waters, whereas, when the population is large and distributed in less productive oceanic waters, density-dependent competition for food between the adults leads to poor-quality eggs which in turn leads to poor recruitment and population decline. In a similar fashion, predation and cannibalism of eggs and larvae have been proposed as density-dependent mechanisms providing an upper limit to clupeoid population sizes off South Africa (ValdeÂs et al., 1987).

The management problem The history of the major clupeoid ®sheries throughout the latter part of this century has been one of `boom and bust' with sharp increases in catches often being followed by dramatic stock collapses. The collapse of the Californian sardine in the 1950s was the ®rst to occur in a coastal upwelling system and was followed by the South African pilchard in the 1960s, the Namibian pilchard in the late 1960s and the Peruvian anchoveta in the early 1970s. Leaving aside the various human and political problems associated with ®sheries management in general (Arnasson, 1993; Anon., 1994), the sustainable management of clupeoid stocks in upwelling regions poses a particular challenge to ®sheries scientists. Large natural population ¯uctuations, due to high levels of environmentally driven recruitment success coupled with their short life spans, is a dif®cult backdrop against which to make management decisions. Most managers currently have no way of reliably predicting recruitment success, and what may be a conservative level of exploitation during years with good recruitment may prove disatrous in years with poor recruitment. The problem is exacerbated by over®shing which, via reducing the number of age classes Clupeoid stock dynamics in coastal upwelling areas 455 in a population, further increases its susceptibility to recruitment variability, and ultimately stock collapse, if the number of age classes becomes so reduced that the population is unable to bridge a string of years with poor recruitment success (Butterworth, 1983). Traditional ®sheries management techniques for estimating sustainable yields and population size from catch statistics have proved to be of little worth. For instance the Schaefer surplus production model used in the Peruvian anchoveta ®shery (Pitcher and Hart, 1982) and the virtual population analysis (VPA) estimates of `current biomass' used for the Namibian pilchard (Butterworth, 1980) both failed to register substantial stock declines whilst the populations were in a state of collapse. These techniques were originally developed for long-lived ground ®sh with high levels of density-dependent population regulation, and as it turned out the assumptions from which they worked proved inappropriate when applied to heavily exploited clupeoid populations. In the case of surplus production models the problem lies with their assumption of a steady-state population primarily regulated by density-dependent interactions. For current biomass estimates from VPA, the problem is with the estimation of `current year ®shing mortality' using ®shing effort as a scaler. Fishing mortality has typically been assumed to be linearly proportional to effort, which in turn assumes a homogeneous distribution of ®sh and randomly directed ®shing effort ± assumptions which are quite wrong for clupeoid ®sheries. In reality the fact that clupeoids are tightly shoaling and that purse seiners do not operate randomly leads to Palheimo± Dickie effects (Palheimo and Dickie, 1964), in which catch rates do not fall as quickly as stock size. Hence, during periods of stock decline, the use of ®shing effort as a linear scalar can underestimate ®shing mortality, and consequently overestimate stock size. Moreover, dif®culties in quantifying these Palheimo±Dickie effects have precluded their incorporation into management models which utilize catch statistics. There have been two responses to these problems with traditional techniques. The ®rst is an increased reliance on direct surveys for estimates of current biomass prior to quota setting. Hydro-acoustic surveys are now the most widely used, although biomass estimates have also been derived from egg counts using the egg-production method (Butterworth, 1983; Hampton et al., 1990; Hampton, 1992). The second, as touched on in the previous section, is an increase in the research effort directed at uncovering how environmental factors in¯uence the population dynamics of these ®sh, particularly as regards driving recruitment variability. Reliable forecasts of year-class strength based on a clear understanding of the relationship between environmental conditions and recruitment success would be of great bene®t in managing these volatile stocks. Not only would they improve the con®dence with which ®sheries managers could set quotas, but higher average catches from healthier stocks would also clearly bene®t the ®shing industry (Cochrane and Star®eld, 1992). The next section will take a closer look at some of the mechanisms whereby environmental conditions may in¯uence recruitment, and will discuss reasons why to date there has been a more or less universal failure in the actual prediction of recruitment success.

Recruitment variability and the environment Recruitment success can be thought of as an integrated function of processes acting across a wide range of life-history stages, from the size and condition of the spawning 456 Cole and McGlade population at one end to pre-recruit survival rates at the other. Rudimentary stock±recruit relationships do exist for clupeoids (Lasker, 1985), but, in practice, variable mortality of the vulnerable planktonic eggs and larvae is thought to be the main cause of year-to-year recruitment variability. Theories of how recruitment success is in¯uenced by the survival of these early life- history stages fall into two broad categories, as illustrated in Fig. 7: `mechanistic theories' which deal with speci®c sources of egg and larval mortality, and `synthesis theories' which attempt to unite the various mechanistic theories within a larger framework. Mechanistic theories can, in turn, be divided into three further categories: predation±cannibalism based, starvation based, and advection based (Hutchings, 1992).

MECHANISTIC THEORIES Starvation theories are based on the premise that if larvae do not encounter suf®cient quanities of food after yolk-sac absorption they will die (Blaxter and Hunter, 1982). Hjort's `critical period' hypothesis was the ®rst to work from this supposition, and prescribed the need for suitable food during the critical transition between internal and external sources of energy, i.e. soon after yolk-sac absorption, for successful recruitment (Hjort, 1914). Exactly what is the most critical stage of larval development, in terms of ®nal recruitment success, remains a matter of speculation, although for marine ®sheries in general there is increasing evidence it is the later larval stages, and not the ®rst- feeding stages, which are the most critical (Cushing, 1996). Nonetheless, the underlying

RECRUITMENT THEORIES

MECHANISTIC SYNTHESIS

Cannibalism/ Advection Triad Optimal Environmental Window predation (Parrish et al 1983) Starvation (Bakun 1993) (Cury & Roy 1989)

Temperature and growth rates

Critical Period Match-Mismatch Vertical Stability Turbulence (Hjort 1914) (Cushing 1975) (Lasker 1975) (Rothschild et al 1989)

Fig. 7. A classi®cation of theories which address the causes of variable recruitment in pelagic ®sh as a result of the differential mortality of the early life-history stages. Clupeoid stock dynamics in coastal upwelling areas 457 premise of larval starvation has been expanded upon by a number of more modern theories. Cushing's match±mismatch theory (Cushing, 1975, 1996) developed from observations on ®sh stocks in the north-east Atlantic and North Sea, and concerns the timing of reproductive cycles with local productivity cycles, particularly with regards to larval development. The contention is that the greater the overlap between peak larval abundance and peak productivity, the more larvae that will survive to metamorphosis. Lasker's stability hypothesis was speci®cally tailored for pelagic ®sh in upwelling regions and prescribed the need for upwelling activity to be balanced by suf®ciently calm conditions so that thermoclines and maximum chlorophyll layers (MCL) could form. The argument was that only in these MCLs would food concentrations, especially of dino¯agellates, be high enough to enable successful larval feeding (Lasker, 1975, 1978). Finally, turbulence has also been hypothesized as mediating the feeding success of ®sh larvae according to its in¯uence on encounter rates between larvae and food particles (Rothschild et al., 1989; Rothschild, 1991). `Advection'-based theories concern themselves with the transport of eggs and larvae towards or away from suitable nursery areas (Iles and Sinclair, 1982). They are of special concern in coastal upwelling regions due to the large offshore movement of water associated with Ekman drift. Comparative studies of these regions have shown that clupeoids tend to avoid spawning in areas and at times with strong upwelling when there would be a high risk of eggs and larvae being transported into unproductive oceanic waters (Parrish et al., 1983). In practice, starvation and advection-based theories both prescribe similar environmental conditions for recruitment success in these regions; namely quiescent upwelling conditions to prevent the offshore loss of eggs and larvae, and to allow the development of chlorophyll maximum layers for successful larval feeding. Nonetheless, it is important that calm periods are balanced with suf®cient levels of upwelling. Poor recruitment has often occurred during warm events owing to impoverished feeding conditions resulting from the suppressed upwelling of nutrient-rich water (e.g. Valdivia, 1978; Boyd et al., 1985; Le Clus, 1985). Predation and cannibalism are also important sources of egg and larval mortality, and have received some attention as density-dependent mechanisms for maintaining the dominance of one clupeoid species over another and in placing upper limits on population size (ValdeÂs et al., 1987; ValdeÂs and Cochrane, 1992). As regards year-to- year recruitment variability, however, problems in quantifying their impact makes it dif®cult to assess whether these processes do in fact differ enough between years to explain this variability. Furthermore, even these `density-dependent' sources of mortality are far from independent of environmental factors. Temperature is likely to be of particular importance, insofar as it affects egg hatching and larval development rates. For instance, given that susceptibility to predation amongst marine organisms is largely size dependent, one might expect that higher temperatures would promote the survival of early life-history stages by inducing quick hatching and increasing the rate at which ®sh larvae can outgrow their predators (Laurence, 1990).

SYNTHESIS THEORIES None of the mechanisms described above will be the sole process determining recruitment success in any particular system. For instance in an upwelling system, 458 Cole and McGlade advective processes, vertical stability of the water column, temperature, food production and turbulence are all likely to some extent or another to be involved in in¯uencing recruitment success according to the feeding, growth and retention of the early life- history stages. For this reason synthesis theories have been developed which attempt to bring the various mechanistic theories together within a single conceptual framework. The optimal environmental window (OEW) theory stresses the importance of a suf®cient balance between the upwelling of nutrient-rich water and calm conditions for encouraging successful clupeoid recruitment (Cury and Roy, 1989), as illustrated by Fig. 8. It has been successfully tested on clupeoid populations of California, Peru, Chile, the Iberian peninsula, and north-west Africa where it was found that recruitment success was maximized at intermediate levels of Ekman transport and turbulence (Cury and Roy, 1989; Roy et al., 1995; Serra et al., 1998). In addition, there is also evidence for the existence of an optimal environmental window between anchovy stocks and new production in the southern Benguela (Waldron et al., 1998), and between SST and anchovy recruitment in the northern Benguela (Cole, 1997). The argument behind the theory is that at low levels of upwelling and turbulence, there is neither suf®cient primary productivity nor high enough encounter rates between larvae and prey to enable successful larval feeding, whilst at high levels eggs and larvae are swept offshore, and the lack of any vertical strati®cation prevents the formation of maximum chlorophyll layers. Hence, it is at intermediate levels of upwelling activity

LIMITING FACTORS

LOW FOOD HIGH OFFSHORE PRODUCTION ADVECTION RECRUITMENT OPTIMAL ENVIRONMENTAL WINDOW LOW ENCOUNTER NO VERTICAL RATES STRATIFICATION

UPWELLING INTENSITY LOW MODERATE HIGH TURBULENCE LOW MODERATE HIGH SEA SURFACE TEMPERATURE HIGH MEDIUM LOW Fig. 8. Schematic representation of the optimal environmental window (OEW) theory of clupeoid recruitment success in coastal upwelling regions. Adapted and redrawn from Cury and Roy (1989). Clupeoid stock dynamics in coastal upwelling areas 459 where the `optimal' trade-off between the various physical processes in¯uencing recruitment success lies. Bakun's triad hypothesis (Bakun, 1993, 1996) generalizes three broad categories of ocenographic process thought to be important in in¯uencing recruitment success; namely enrichment of the food chain, retention of the eggs and larvae within suitable nursery areas, and concentration of food particles for the ®rst-feeding larvae and subsequent development stages. Within a coastal upwelling system, enrichment will result from the upwelling of nutrient-rich water from below the pycnocline; retention from a reduction in offshore transport and the advection of other water masses into the system; and concentration according the formation of thermoclines and presence of fronts where food particles can be concentrated. Given that the physical conditions leading to enrichment versus retention and concentration are often mutually exclusive, spatio-temporal variability in the physical dynamics of these systems will be important in determining whether or not there is a suf®cient balance between these processes for recruitment success or not. The triad hypothesis arose from observations from a wide variety of ®sheries and marine systems, whereas the OEW hypothesis is speci®cally tailored toward clupeoid in coastal upwelling systems. Nevertheless, within the context of these eastern boundary systems both theories attempt to unite, or synthesize, the very same underlying mechanistic processes thought to in¯uence recruitment success. The OEW hypothesis remains the only one of the two to have been empirically tested, however. In spite of a large body of anecdotal evidence gathered from climatic, oceanographic and ®sheries data from around the globe (Bakun, 1996), the triad theory still remains untested in any strict quanititative way. The triad theory does, however, provide three broad, and conceptually clear classes of process that can be referred back to when thinking about how best to measure the environmental dynamics of a system for recruitment studies. The current problem is that of the three processes enrichment is the only one to have been suf®ciently well quanti®ed (according to SST or wind-based measures of upwelling activity) for inclusion in empirical recruitment studies. In order for the theory to be empirically tested in the future, and in order to ®nd the `optimal' trade-offs between the three processes for recruitment success, reasonably consistent ways of quantifying retention and concentration need to be found. One way this might be achieved is suggested further on.

Problems with predicting recruitment Assuming that reasonably consistent relationships do exist between environmental conditions and recruitment success, there are three main considerations when attempting to successfully uncover these relationships for the purposes of making recruitment forecasts. Firstly, given the complex chain of causal relationships between environmental events and their impact on clupeoid populations (Sharp, 1987), we need to identify those processes which have the smallest `distance' (i.e. the smallest number of causal links) to the proximal factors affecting the mortality=survival of the early life-history stages, but which are also easily measurable across the entire range of the population under consideration. Secondly, we need to choose the most appropriate indices for measuring these processes. Thirdly, an appropriate model is needed for testing=investigating the nature of the relationship. 460 Cole and McGlade Why then, in spite of the successful testing of some of the theories described above, the identi®cation of environmental processes likely to have an important impact on recruitment, and the existence of long-term relationships between environmental indices and year-class strength, has there been repeated failure in the reliable year-to-year prediction of recruitment success? Four possible reasons are put forward below.

NON-LINEARITY Non-linearity in the relationship between environmental conditions and recruitment success may sometimes explain the breakdown of linear correlations. Dealing with this problem requires either a priori assumption as to the type of non-linear relationship, or else an algorithm which is able to iteratively `explore' the data in such a way as to uncover the nature of any non-linearities (for example, the Alternating Conditional Expectation algorithm, used by Cury and Roy (1989), to uncover OEW relationships between upwelling activity and recruitment).

SCALE Environmental forcing acts on populations of marine organisms across a wide range of different temporal and spatial scales. The fact that ecological phenomena at different scales in the ocean are often caused by different environmental factors (Legendre and Demers, 1984) means that unambiguously linking physical and ecological processes can be dif®cult. Given the current interest in understanding physical forcing and biological response in marine environments, the whole issue of scale and physical±biological processes in the oceans has consequently received fairly extensive coverage in recent years (Powell, 1989; Ricklefs, 1990; Levin, 1992; Mullin, 1993). With regard to the failure to adequately predict recruitment success, ®sheries and environmental parameters have often been measured and compared to each at different, or inappropriate, spatial and temporal scales (Taggart and Frank, 1990; Parrish, 1997). For instance, in studies which have found good long-term relationships between single environmental parameters and recruitment, the environmental parameter has often been averaged over relatively long time intervals (such as a year), and measured from areas which are different in size or location from the distribution of the population under consideration (Shannon et al., 1988; Cury and Roy, 1989; Lluch-Belda et al., 1989). Whilst this may reveal general relationships and provide insight into the impact of remote forcing on ®sh populations, it will also mask variations in local environmental conditions, and consequently is unlikely to provide a robust basis for recruitment prediction. Ultimately it is local conditions, irrespective of how they relate to wider processes, which determine the survival of the early life-history stages and hence recruitment success. It should be remembered, nonetheless, that historically there have been enormous constraints on where, when and at what resolution environmental and biological data could be collected. Prior to the advent of marine satellite remote sensing, biologists and oceanographers were in the main limited to measurements from ships of opportunity, coastal observations, weather reports, and the frequency with research ships could be deployed. This has been a particular handicap in coastal upwelling systems where high- resolution environmental monitoring is required to adequately capture their physical and biological variability. Clupeoid stock dynamics in coastal upwelling areas 461

THE MULTIPLICITY AND COMPLEXITY OF PROCESSES THAT INFLUENCE RECRUITMENT The wide number of different factors which can potentially determine recruitment success, and the complexity of their interactions, may simply preclude the effective prediction of recruitment success from single environmental indices. This may explain why good correlations between recruitment success and environmental factors may sometimes break down (Shannon et al., 1988; Lawson, 1997) as might occur, for instance, if a factor which is not re¯ected by the environmental index being used, suddenly assumes a more dominant role in determining ®nal year-class strength. Current approaches to dealing with this problem include the construction of multiparameter models, which are based on a good understanding of the conditions under which different factors can in¯uence recruitment (Cochrane and Hutchings, 1995). Neural networks may, in the future, also prove to be particularly useful, given their ability to `learn' what the deterministic relationships are between multiparameter inputs and single outputs (Jarre- Teichmann et al., 1995).

ENVIRONMENTAL INDICES In choosing environmental indicies for ®sheries research there are four considerations: what are the environmental processes and features relevant to the question being asked; what, in theory, is the best way to measure them; how easily and cheaply can they be measured; and ®nally, what indices have been collected in the past. Because non- experimental ®sheries research is by nature retrospective, and often needs long time series, this ®nal consideration also acts as an important constraint. In practice, sea surface temperatures and wind-based indices, especially Ekman transport, have been the most commonly used physical indices for looking at the relationship between the environment and the behaviour of clupeoid populations in these regions. The reasons for this are as follows. Firstly, they are cheaply and regularly collected, both at sea by research vessels and ships of opportunity, and from coastal monitoring stations. Secondly, records of these environmenal parameters are typically very long, and may stretch back as far as the last century (Woodruff et al., 1987; Roy and Mendelssohn, 1998). Thirdly, they often directly, or indirectly, represent environmental processes thought to in¯uence the dynamics and distribution of ®sh populations. In environment±recruitment studies, SSTs and Ekman transport have often been used as proxies for upwelling activity and offshore transport. Nevertheless, it is worth examining in greater detail what exactly it is that these two indices measure. Although wind-forcing is the main factor driving offshore transport and coastal upwelling in eastern boundary regions, the use of Ekman transport as an index of offshore transport and biologically-enriching upwelling activity is limited by the following assumptions. Namely, that surface ¯ow is determined soley by the action of local wind drag on the ocean surface combined with the Coriolis effect, and that there is a consistent relationship between offshore transport and the upwelling of cool, nutrient-rich water from below the pycnocline. In reality, upwelling activity and surface circulation in these regions are also subject to a number of other local and remote factors, examples of which include bathymetry, coastally trapped waves (Boyd, 1987), and the remotely forced intrusion of warmer water masses. These factors not only disrupt the relationship between wind-forcing and 462 Cole and McGlade upwelling activity by introducing time lags and non-linearities, but in extreme cases, for instance during El NinÄos and Benguela NinÄo's, may block the upwelling of cool, nutrient-rich water altogether. Furthermore, this warm event `blocking'may occur even when there are increases in equatorward wind stress (Shannon et al., 1986; Bakun, 1987). Biologically-enriching upwelling activity can, therefore, only be partially represented by wind-based indices. A further problem with Ekman transport is the fact that part of its calculation involves multiplying wind velocity by the `alongshore equatorward component of wind velocity' (Peixoto and Oort, 1992). Thus, in these regions where the prevailing coastal winds are more or less parallel to the coast in the equatorward direction, this more or less amounts to wind speed squared. Given that wind is probably the most variable of all `ocean-relevant' parameters, this squaring may result in inaccurate estimates of mean Ekman transport when wind speed is irregularly and infrequently measured. Sea surface temperature, in contrast, is temporally much less variable than wind speed, and does not require any mathematical treatment after it has been measured. Furthermore, it is an emergent property of a system's oceanography, and as such re¯ects the dynamic balance between a wide spectrum of different forcing factors. In eastern boundary systems, SSTs will largely be determined by the play between the upwelling of cold water, solar warming of the surface layers, and the advection of warmer water masses from the edges of the system. Hence, given suf®cient spatio-temporal coverage, SSTs may provide a better representation of a coastal upwelling system's physical state than wind-based indices, and consequently may be more useful for studies of environmental±®sheries interactions. This is not to say that there are no problems with the use of SST as an environmental index for recruitment studies, but in this case the problems are related more to the spatio-temporal scales at which SST has been measured and averaged across (as discussed above), rather than with what the index actually represents. One inherent problem with SSTs, however, is that they cannot directly re¯ect the presence or otherwise of sub-surface features such as thermoclines. Nonetheless, given that the formation of thermoclines in these regions is largely a function of solar heating, upwelling activity (and the turbulent mixing associated with it) and intrusions of warmer water masses (Boyd, 1987), there is likely to be some kind of reasonably consistent relationship between surface temperatures and levels of vertical strati®cation (Cole, 1997).

Satellite oceanography: a way forward? The advent of remote sensing marked a new era in the observation and monitoring of the marine environment. The ®rst remote observations of the ocean surface from space were from the TIROS-2 satellite during the early 1960s, and in 1978 the ®rst sensors speci®cally designed for marine research were deployed with the launching of the Seasat, TIROS-N, and Nimbus satellites (Sherman, 1985). Currently there are three different types of sensor commonly used for marine research: those which measure re¯ected radiation, emitted radiation and back-scattered radiation, as illustrated in Fig. 9. Sensors that measure re¯ected radiation operate in the visible portion of the electomagnetic spectrum (EMS) and are used for monitoring chlorophyll concentrations and other factors which affect `ocean colour', such as coastal sediments and pollutants. Clupeoid stock dynamics in coastal upwelling areas 463

Electromagnetic Spectrum

Ϫ Ϫ Ϫ Ϫ Ϫ Ϫ Ϫ Wavelength 10 710 610 510 410 310 210 1 110102 103 104 105 106 107 108 109 (micrometres) cosmic rays gamma rays X-rays Ultraviolet Visible Infra Red Microwave Radar & radio Television

SENSOR SENSOR SENSOR RADAR RADIATION REFLECTED EMITTED RADIATION RADIATION BACKSCATTERED

Less than 50 metres

Max depth from which reflected light can be detected

Fig. 9. Schematic representation of the three classes of sensor most commonly used for marine remote sensing. The electromagnetic spectrum included in the inset is modi®ed from Lillesand and Kiefer (1987).

Sensors that measure emitted radiation operate in the infra-red and microwave portions of the EMS, and are typically used for deriving sea surface temperature estimates. Sensors that measure back-scattered radiation operate in the radar section of the EMS. They are used for directly measuring sea level, wave height and surface roughness, and as such may be used for estimating related phenomena, such as surface wind stress and wind-driven surface ¯ows (Carter, 1992). For a more complete description of the various different sensors used for marine monitoring and their applications the following publications can be referred to: Robinson (1985); Salzman (1985); Robinson and Guymer (1996); Victorov (1996). In spite of potential problems with cloud cover and limitations on the depth to which satellite sensors can record information (Flament et al., 1994), the main advantage of using satellites for marine research is their ability to sample large areas of ocean on a considerably more frequent basis, and with better spatial coverage than other forms of sampling. 464 Cole and McGlade Ocean colour images of chlororphyll concentration, and sea surface temperature (SST) images have been the most commonly used remotely sensed products used in ®sheries research. From these images it is possible to identify particular features and processes known to in¯uence the distribution and abundance of ®sh stocks, such as upwelling activity, frontal systems, and patterns of food abundance. To date, though, it is the distribution of ®sh populations and their planktonic life-history stages, not their year-to-year ¯uctuations in abundance, which have received the greatest attention (see Laurs, 1997 for a recent overview of the applications of satellite remote sensing in ®sheries research). For example, satellite images have been used to investigate anchovy spawning off southern California in relation to upwelling activity and phytoplankton concentration (Lasker and PelaÂez, 1981; Fielder, 1983), the distribution of temperature fronts and albacore tuna in the California Current system (Laurs and Brucks, 1985), the relationship between thermal fronts, mesoscale eddies and the distribution of blue®n and albacore tuna off Tasmania (Reddy et al., 1995), and for forecasting the distribution of commercial ®sheries in the Arabian Sea off north-west India with respect to thermal variability (Kumari et al., 1994). The potential for using remotely sensed environmental data for recruitment studies has been greatly expanded in recent years; mainly due to the processing and compilation of SST time series of satellite images by organizations such as NASA (Holt and Digby, 1997), and the JRC (European Commission Joint Research Centre). Poor-resolution environmental data collected, or averaged, over inappropriate spatio- temporal scales was one of the reasons discussed earlier for the failure to predict recruitment. These image time series offer a solution, insofar as they allow for the construction of environmental indices across a wide range of temporal and spatial scales. The areas and times from which indices are extracted can, for example, be matched to the range of the population under question, and to those times of the year when most of the spawning and juvenile development takes place. The high resolution of the data also offers the opportunity to experiment with alternative ways of treating it. Instead of deriving `annual' indices of SST conditions by averaging SSTs over time and space, one could, for example, count the `number of mean weekly SST events' above a certain temperature within subdivisions of the area in question over the course of an entire reproductive season (Cole, 1998). In this way, local and short-term events which may in¯uence recruitment are more likely to be fully integrated into the index and not be masked out. Another advantage of satellite image time series is their ability to provide detailed spatial representations of the physical state of marine systems through time, as opposed to just acting as high-resolution databases for environmental parameters. This is clearly illustrated in Fig. 3(a±c) where patterns in the gradient structure of the images allow us to visually identify particular features and processes associated with enrichment, retention, and concentration, such as upwelling activity, upwelling ®laments and intrusions of warmer water masses. The application of pattern analysis techniques to these time series has now opened up the possibility of deriving meta-level indices of a system's physical behaviour for use in ®sheries and ecological research. The difference from more traditional approaches is that a system's behaviour is classi®ed according to the evolution of its spatial structure, Clupeoid stock dynamics in coastal upwelling areas 465 rather than according to the absolute value of indices extracted from speci®c areas or locations. Principal components analysis (PCA, also known as empirical orthogonal analysis) is one technique which has begun to be used for just such a purpose. For instance, it has been successfully used to evaluate changes in vegetation across Africa (Eastman and Fulk, 1993), the behaviour of the Californian Current off Baja California from SST images (Gallaudet and Simpson, 1994), the behaviour of the coastal upwelling system off Mauritania from SST images (HernaÂndez-Guerra and Nykjaer, 1997; Maus, 1997), and the behaviour of the northern Benguela upwelling system, also from SST images (Cole and McGlade, 1998). The dif®culty of measuring retention and concentration, and how this has been an obstacle to empirically testing the triad theory, was mentioned earlier. One bene®t of using a spatial analysis technique, such as PCA, might be the ability to derive indices of retention and concentration according to the distribution and orientation of SST and ocean colour gradients. For example, from the PCA of a time series of SST images for the northern Benguela, Cole and McGlade (1998) found that principal components (PCs) II and III provided a good representation of processes likely to affect the retention of clupeoid eggs and larvae, and the concentration of suitable food particles for the developing larvae across thermal fronts and thermoclines. Cole (1998) subsequently found a good correspondence between environmental conditions and clupeoid recruitment (R2 values of 0.96 and 0.88 for anchovy and pilchard respectively) when using the time-varying `loadings' on PCs II and III from Cole and McGlade (1998), in combination with the `number of coastal SST events' over 19 8C, as environmental indices. Although the results of this particular study need to be treated with caution, given that there was only 6 years' overlap between the available recruitment estimates and the satellite data, it does illustrate ways in which satellite image time series could pro®tably be used for recruitment studies. Satellite image time series clearly offer an important additional data source for studies into the effects of environmental variability on the behaviour of ®sh stocks and recruitment success. Nevertheless, we must not lose sight of theory and mechanism. Although the shear volume of data contained in these time series offers great potential for exploratory data analysis, a purely data-driven approach would inevitably result in an increase in the number of spurious correlations between the `environment' and ®sh stocks. Given this danger and the fact that what satellite images record (SST, spatial patterns, etc.) are often proxies for the factors directly affecting the survival of larval ®sh, it is important that investigations take place ®rmly within the context of hypothesized relationships, both as regards how the environment acts on ®sh stocks, and as to the relationship between what a satellite image represents and the proximal factors acting on the ®sh.

Conclusion Understanding recruitment well enough to enable reliable predictions of year-class strength, both in and out of eastern boundary regions, remains one of the main challenges of ®sheries science. Given that we are unable to keep track of the unique sequence of environmental conditions each developing ®sh larvae is exposed to, we inevitably have to investigate environment±recruitment linkages at scales that are far removed from the immediate `cause and effect' levels of larval mortality. When choosing environmental 466 Cole and McGlade indices for studying recruitment, it is crucial that we are aware of what bio-physical processes they are likely to be measuring, to tailor their spatio-temporal resolution as necessary, and ®nally to choose a suitable model for investigating the linkages. The recent compilation of high-resolution satellite image time series offers a way forward, insofar as they provide solutions to some of the historial problems with deriving appropriate environmental indices at suitable spatio-temporal scales for clupeoid recruitment studies in eastern boundary upwelling regions.

Acknowledgements Most of the work towards this paper was conducted whilst the ®rst author held a research studentship from the United Kingdom's Biotechnology and Biological Sciences Research Council, at the University of Warwick. Nick Mountford and Andrew Yool are thanked for reading and commenting on the draft manuscipt.

References Anon. (1994) The tragedy of the oceans. The Economist, 19 March 1994, pp. 21±24. Armstrong, M.J. and Shelton, P.A. (1990) Clupeoid life-history styles in variable environments. Env. Biol. Fishes 28, 77±85. Arnasson, R. (1993) Ocean ®sheries management: recent international developments. Mar. Policy 17, 334±339. Baird, D., McGlade, J.M. and Ulanowicz, R.E. (1991) The comparative ecology of six marine ecosystems. Phil. Trans. R. Soc. B 333, 15±29. Bakun, A. (1987) Monthly variability in the ocean habitat off Peru as deduced from maritime observations (Proc. The Peruvian Anchoveta and its Upwelling Ecosystem: Three Decades of Change). Manila: ICLARM (cont. no. 391), pp. 46±74. Bakun, A. (1993) The California Current, Benguela Current, and Southwest Atlantic Shelf ecosystems: a comparative approach to identifying factors regulating biomass yields. In Sherman, K., Alexander, L.M. and Gold, B.D. eds. Large Marine Ecosystems: Stress, Mitigation, and Sustainability. Washington, DC: Am. Ass. Advancement of Science, pp. 199±224. Bakun, A. (1996) Patterns in the Ocean: Ocean Processes and Marine Population Dynamics. La Jolla, CA: California Sea Grant. 323 pp. Blaxter, J.H.S. and Hunter, J.R. (1982) The biology of clupeoid ®shes. Adv. Mar. Biol. 20, 1±223. Boyd, A.J. (1987) The oceanography of the Namibian shelf. PhD thesis. Univ. Cape Town. 191 pp. Boyd, A.J., Hewitson, J.D., Kruger, I. and Le Clus, F. (1985) Temperature and salinity trends off Namibia from August 1982 to August 1984, and their relation to the spawning success of pelagic ®sh. Colln Scient. Pap. Int. Commn SE Atl. Fish. 12, 53±58. Butterworth, D.S. (1980) The value of catch statistics based management techniques for heavily ®shed pelagic stocks with special reference to the recent decline of the South West African pilchard stock. Colln Scient. Pap. Int. Commn SE Atl. Fish. 7(II), 69±84. Butterworth, D.S. (1983) Assessment and management of pelagic stocks in the southern Benguela region (Proc. The Expert Consultation to Examine Changes in Abundance and Species Composition of Neritic Fish Resources, San Jose, Costa Rica). FAO Fish. Rep. No. 291, 2, pp. 329±405. Carter, D.J.T. (1992) Wave height and wind speed from satellite radar altimeters. In Guide to Satellite Remote Sensing of the Marine Environment (Intergovernmental Oceanographic Commission (IOC) manuals and guides No. 24). Paris: UNESCO, 33±81. Cochrane, K.L. and Hutchings, L. (1995) A structured approach to using biological and environmental parameters to forecast anchovy recruitment. Fish. Oceanogr. 4, 102±127. Clupeoid stock dynamics in coastal upwelling areas 467

Cochrane, K.L. and Star®eld, A.M. (1992) The potential use of predictions of recruitment success in the management of South Africa anchovy resource (Proc. Benguela Trophic Functioning, Cape Town). S. Afr. J. Mar. Sci. 12, 891±902. Cole, J. (1997) The surface dynamics of the northern Benguela upwelling system and its relationship to patterns of clupeoid production. PhD thesis, Univ. Warwick (http://www.oikos.warwick.ac.uk/ ,wupert/index.html). 208 pp. Cole, J. (1998) Environmental conditions, satellite imagery, and clupeoid recruitment in the northern Benguela upwelling region. Fish. Oceanogr. (in press). Cole, J. and McGlade, J. (1998) Temporal and spatial patterning of sea surface temperatures in the northern Benguela; possible environmental indicators of clupeoid production (Proc. Benguela Dynamics, November 1996, Cape Town). S. Afr. J. Mar. Sci. (in press). Crawford, R.J.M. and Shannon, L.V. (1988) Long-term changes in the distribution of ®sh catches in the Benguela (Proc. Long Term Changes in Marine Fish Populations). Vigo, Spain: Institutio de Investigaciones Marinas de Vigo, pp. 449±480. Crawford, R.J.M., Shannon, L.V. and Pollock, D.E. (1987) The Benguela ecosystem IV, the major ®sh and invertebrate resources. Oceanogr. Mar. Biol. Ann. Rev. 25, 353±505. Crawford, R.J.M., Underhill, L.G., Shannon, L.V., Lluch-Belda, D., Siegfried, W.R. and Villacastin- Herrero, C.A. (1991) An empirical investigation of trans-ocean linkages between areas of high abundance of sardine (Proc. Long-Term Variability of Pelagic Fish Populations and their Environment, Sendai, Japan). Toyko: Pergamon Press, pp. 319±332. Cury, P. and Roy, C. (1989) Optimal environmental window and pelagic ®sh recruitment success in upwelling areas. Can. J. Fish. Aquat. Sci. 46, 670±680. Cushing, D.H. (1975) Marine Ecology and Fisheries. Cambridge, UK: Cambridge University Press. 278 pp. Cushing, D.H. (1996) Towards a Science of Recruitment in Fish Populations. Oldendorf, Germany: Ecology Institute, 175 pp. De Vries, T.J. and Pearcy, W.G. (1982) Fish debris in sediments of the upwelling zone off central Peru: a late Quaternary record. Deep-Sea Res. 28, 87±109. Eastman, J.R. and Fulk, M. (1993) Long time series evaluation using standardized principal components analysis. Photogrammetric Engineering and Remote Sensing 59, 1307±1312. Fielder, P.C. (1983) Satellite remote sensing of the habitat of spawning anchovy in the Souther Californian Bight. CalCOFI Rep. 24, 202±209. Flament, P., Firing, J., Sawyer, M. and Trefois, C. (1994) Amplitude and horizontal structure of a large diurnal sea surface warming event during the Coastal Ocean Dynamics Experiment. J Physical Oceanogr. 24, 124±139. Gallaudet, T.C. and Simpson, J.J. (1994) An empirical orthogonal function-analysis of remotely-sensed sea-surface temperature variability and its relation to interior oceanic processes off Baja- California. Remote Sensing Env. 47, 375±389. Hampton, I. (1992) The role of acoustic surveys in the assessment of pelagic ®sh resources on the South African continental shelf (Proc. Benguela Trophic Functioning, Cape Town). S. Afr. J. Mar. Sci. 12, 1031±1050. Hampton, I., Armstrong, M.J., Jolly, G.M. and Shelton, P. (1990) Assessment of anchovy spawner biomass off South Africa through combined acoustic and egg production surveys. Rapp. P.-v. ReÂun. Cons. Int. Explor. Mer 189, 18±32. Hartline, B.K. (1980) Coastal upwelling: physical factors feed ®sh. Science 208, 38±40. HernaÂndez-Guerra, A. and Nykjaer, L. (1997) Sea surface temperature variablity off north-west Africa: 1981±1989. Int. J. Remote Sensing 18, 2539±2558. Hisard, P. (1988) El NinÄo response of the tropical Atlantic ocean during the 1984 year (Proc. International Symposium into Long Term Changes in Marine Fish Populations). Vigo, Spain: Institutio de Investigaciones Marinas de Vigo, pp. 273±290. Hjort, J. (1914) Fluctuations in the great ®sheries of northern Europe viewed in the light of biological research. Rapp. P.-v ReÂun. Cons. Int. Explor. Mer 20, 1±228. 468 Cole and McGlade

Holt, B. and Digby, S.A. (1997) NASA's satellite oceanography archives (Proc. Changing Oceans and Changing Fisheries: Environmental Data for Fisheries Research and Management ± A Workshop, Paci®c Fisheries Environmental Group, Paci®c Grove, CA). NOAA Tech. Memo. NMFS-SWFSC 239, pp. 36±49. Hutchings, L. (1992) Fish harvesting in a variable productive environment ± searching for rules or searching for exceptions? (Proc. Benguela Trophic Functioning, Cape Town). S. Afr. J. Mar. Sci. 12, pp. 297±318. Iles, T.D. and Sinclair, M. (1982) Atlantic herring: stock discreteness and abundance. Science 215, 627±632. Jarre-Teichmann, A., Brey, T. and Haltof, H. (1995) Exploring the use of neural networks for biomass estimates in the Peruvian Upwelling Ecosystem. NAGA, ICLARM Q. 18(4), 38±40. Kawasaki, T. and Omori, M. (1988) Fluctuations in the three major sardine stocks in the Paci®c and the global trend in temperature (Proc. Long Term Changes in Marine Fish Populations). Vigo, Spain: Institutio de Investigaciones Marinas de Vigo, pp. 37±53. Kawasaki, T. and Omori, M. (1995) Possible mechanisms underlying ¯uctuations in the Far East- ern sardine population inferred from time series of two biological traits. Fish. Oceanogr. 4, 238±242. Kumari, B., Solanki, H.U., Raman, M. and Narain, A. (1994) Role of remote sensing in effective utilisation of oceanographic features for long term ®sheries forecast: a case study in the northwestern waters of India (Proc. Remote Sensing for Marine and Coastal Environments,New Orleans, USA). Ann Arbor, MI: ERIM Conferences, vol. 1, pp. 105±111. Lalli, C.R. and Parsons, T.R. (1993) Biological Oceanography: An Introduction. Oxford: Pergamon. 301 pp. Lasker, R. (1975) Field criteria for survival of anchovy larvae: the relation between inshore chlorophyll maximum layers and successful ®rst feeding. Fish. Bull. US 73, 453±462. Lasker, R. (1978) The relation between oceanographic conditions and larval anchovy food in the California current: identi®cation of factors contributing to recruitment failure. Rapp. P.-v ReÂun. Cons. Int. Explor. Mer 173, 212±230. Lasker, R. (1985) What limits clupeoid production? Can. J. Fish. Aquat. Sci. 42, 31±38. Lasker, R. and PelaÂez, J. (1981) The use of satellite infrared imagery for describing ocean processes in relation to spawning of the Northern Anchovy (Engraulis mordax). Remote Sensing Env. 11, 439±453. Laurence, G.C. (1990) Growth, survival and recruitment in large marine ecosystems. In Sherman, K., Alexander, L.M. and Gold, B.D. eds. Large Marine Ecosystems: Patterns, Processes and Yields. Washington, DC: Am. Ass. Advancement of Science, pp. 132±150. Laurs, R.M. (1997) Overview of satellite remote sensing applications in ®sheries research (Proc. Changing Oceans and Changing Fisheries: Environmental Data for Fisheries Research and Management ± A Workshop, Paci®c Fisheries Environmental Group, Paci®c Grove, CA). NOAA Tech. Memo. NMFS-SWFSC 239, pp. 9±16. Laurs, R.M. and Brucks, J.T. (1985) Living marine resource applications. In Salzman, B. ed. Satellite Oceanic Remote Sensing. Adv. Geophysics 27, 419±452. Lawson, P.W. (1997) Interannual variability in growth and survival of chinook and coho salmon (Proc. Estuarine and Ocean Survival of Northeastern Paci®c Salmon, A Workshop.). NOAA Tech. Memo. NMFS-NWSFC 29, pp. 81±92. Le Clus, F. (1985) Effect of a warm water intrusion on the anchovy stock off South West Africa to pre-1983 levels. Colln Scient. Pap. Int. Commn SE Atl. Fish. 11, 99±106. Le Clus, F. (1990) Impact and implications of large-scale environmental anomalies on the spatial distribution of spawning of the Namibian pilchard and anchovy populations. S. Afr. J. Mar. Sci. 9, 141±159. Le Clus, F. (1991) Hydrographic features related to pilchard and anchovy spawning in the northern Benguela system, comparing three environmental regimes. S. Afr. J. Mar. Sci. 10, 103±124. Clupeoid stock dynamics in coastal upwelling areas 469

Legendre, L. and Demers, S. (1984) Towards dynamic biological oceanography and limnology. Can. J. Fish. Aquat. Sci. 41, 2±19. Levin, S.A. (1992) The problem of pattern and scale in ecology. Ecology 73, 1943±1967. Lillesand, T.M. and Keifer, T.M. (1987) Remote Sensing and Image Interpretation, 2nd edn. New York: John Wiley & Sons. 721 pp. Lluch-Belda, D., Crawfod, R.J.M., Kawasaki, T., MacCall, A.D., Parrish, R.H., Schwartzlose, R.A. and Smith, P.E. (1989) World wide ¯uctuations of sardine and anchovy stocks: the regime problem. S. Afr. J. Mar. Sci. 8, 195±205. Longhurst, A.R. and Pauly, D. (1987) Ecology of Tropical Oceans. San Diego: Academic Press. 407 pp. Mann, K.H. (1993) Physical oceanography, food chains and ®sh stocks: a review. ICES J. Mar. Sci. 50, 105±119. Mann, K.H. and Lazier, J.R.N. (1991) Dynamics of Marine Ecosystems: Biological±Physical Interactions in the Oceans. Boston: Blackwell Scienti®c Publications. 466 pp. Maus, J. (1997) Sustainable ®sheries information management in Mauritania; implications of institutional linkages and the use of remote sensing for improving the quality and interpretation of ®sheries and biophysical data. PhD thesis, Univ. Warwick. 259 pp. Muck, P. (1989) Major trends in the pelagic ecosystem off Peru and their implications for management (Proc. The Peruvian Upwelling Ecosystem: Dynamics and Interactions). ICLARM Conf. Proc. 12, 386±403. Muck, P. and Sanchez, G. (1987) The importance of mackerel and horse mackerel predation for the Peruvian anchoveta stock (a population and feeding model) (Proc. The Peruvian Anchoveta and Its Upwelling System: Three Decades of Change, Callao, Peru). ICLARM Stud. Rev. 15, 276±293. Mullin, M.M. (1993) Webs and Scales: Physical and Ecological Processs in Marine Fish Recruitment. Seattle: Univ. Washington Press, Washington Sea Grant Program. 135 pp. Palheimo, J.E. and Dickie, J.M. (1964) Abundance and ®shing success. Rapp. P.-v ReÂun. Cons. Int. Explor. Mer 155, 152±163. Parrish, R.H. (1997) Time, space, and ®sh scales: applications of retrospective environmental data to ®sheries research (Proc. Changing Oceans and Changing Fisheries: Environmental Data for Fisheries Research and Management ± A Workshop, Paci®c Fisheries Environmental Group, Paci®c Grove, CA). NOAA Tech. Memo. NMFS-SWFSC 239, pp. 24±29. Parrish, R.H., Bakun, A., Husby, D.M. and Nelson, C.S. (1983) Comparative climatology of selected environmental processes in relation to eastern boundary current pelagic ®sh reproduction. (Proc. The Expert Consultation to Examine Changes in Abundance and Species Composition of Neritic Fish Resources, San JoseÂ, Costa Rica). FAO Fish. Rep. 293(3), 731±777. Peixoto, J.P. and Oort, A.H. (1992) Physics of Climate. New York: Am. Ins. Physics. 520 pp. Pitcher, T.J. and Hart, P.J.B. (1982) Fisheries Ecology. London: Chapman and Hall. 416 pp. Powell, T.M. (1989) Physical and biological scales of variability in lakes, estuaries, and the coastal ocean. In Roughgarden, J., May, R.M. and Levin, S.A., eds. Perspectives in Ecological Theory. Princeton, NJ: Princeton Univ. Press, pp. 157±176. Reddy, R.V., Lyne, R., Gray, R., Easton, A. and Clarke, S. (1995) An application of satellite-derived sea surface temperatures to southern blue®n tuna and albacore off Tasmania. Australia. Scientia Mar. 59, 445±454. Rice, J. (1995) Food web theory, marine food webs, and what climate change may do to northern marine ®sh populations. In Beamish, R.J., ed. Climate Change and Northern Fish Populations. Can. Spec. Publ. Fish. Aquat. Sci. 121, pp. 561±568. Ricklefs, R.E. (1990) Scaling patterns and process in marine ecosystems. In Sherman, K., Alexander, L.M. and Gold, B.D. eds. Large Marine Ecosystems, Patterns, Processes and Yields. Washington, DC: Am. Ass. Adv. Science, pp. 169±178. Robinson, I.S. (1985) Satellite Oceanography ± An Introduction for Oceanographers and Remote Sensing Scientists. Chichester, UK: Ellis Horwood. 455 pp. 470 Cole and McGlade

Robinson, I.S. and Guymer, T. (1996) Observing oceans from spcae. In Summerhayes, C.P. and Thorpe, S.A., eds. Oceanography: An Illustrated Guide. London: Manson Publishing, pp. 69±88. Rothschild, B.J. (1991) On the causes for variability of ®sh populations ± the linkage between large and small scales. In Kawasaki, T., Tanaka, S., Toba, Y. and Taniguchi, A., eds. Long-Term Variability of Pelagic Fish Populations and Their Environment. Oxford: Pergamon Press, pp. 367±376. Rothschild, B.J., Osborn, T.R., Dickey, T.D. and Farmer, D.M. (1989) The physical basis for recruitment variability in ®sh populations. J. Cons. Int. Explor. Mer 45, 136±145. Roy, C. and Mendelssohn, R. (1998) The development and use of a climatic database for CEOS using the COADS database. In Durand, M.-H., Cury, P., Mendelssohn, R., Roy, C., Bakun, A. and Pauly, D., eds. Global Versus Local Changes in Upwelling Systems. Paris: ORSTOM, pp. 27±44. Roy, C., Porteiro, C. and Cabanas, J. (1995) The optimal environmental window hypothesis in the ICES area: the example of the Iberian sardine. ICES Coop. Res. Rep. 206, 57±65. Ryther, J.H. (1969) Photosynthesis and ®sh production in the sea. Science 166, 72. Salzman, B. (ed.) (1985) Satellite Ocean Remote Sensing. Adv. Geophysics, 27, 511 pp. Serra, R., Cury, P. and Roy, C. (1998) The recruitment of the Chilean sardine (Sardinops sagax) and the `Optimal Environmental Window' (Proc. Global Versus Local Changes in Upwelling Systems, Paris). Paris: ORSTOM, pp. 267±274. Shackleton, L.Y. (1988) Fossil pilchard and anchovy scales; indicators of past ®sh populations off Namibia (Proc. International Symposium on Long Term Changes in Marine Fish Populations). Vigo, Spain: Instituto de Investigaciones Marinas de Vigo, pp. 55±68. Shannon, L.V., Boyd, A.J., Brundrit, G.B. and Taunton-Clark, J. (1986) On the existence of an El NinÄo type phenomenon in the Benguela system. J. Mar. Res. 44, 495±520. Shannon, L.V., Crawford, R.J.M., Brundrit, G.B. and Underhill, L.G. (1988) Reponse of ®sh populations in the Benguela ecosystem to environmental change. J. Cons. Int. Explor. Mer 45, 5±12. Sharp, G.D. (1987) Climate and ®sheries: cause and effect or managing the long and short of it all (Proc. The Benguela and Comparable Ecosystems, Cape Town). S. Afr. J. Mar. Sci. 5, 811±838. Sharp, G.D. (1995) It's about time: new beginnings and old good ideas in ®sheries science. Fish. Oceanogr. 4, 324±341. Sharp, G.D. and McLain, D.R. (1993) Fisheries, El NinÄo±Southern Oscillation and upper-ocean temperature records: an eastern Paci®c example. Oceanography 6, 13±22. Sherman, J.W.I. (1985) Introduction. In Salzman, B., ed. Satellite Oceanic Remote Sensing. Adv. Geophysics 27, pp. 1±9. Sherman, K. (1991) The large marine ecosystem concept: research and management strategy for living marine resources. Ecol. Appl. 1, 349±360. Soutar, A. and Isaacs, J.D. (1974) Abundance of pelagic ®sh during the 19th and 20th centuries as recorded in anaerobic sediment off the Californias. Fish. Bull. US 72, 257±273. Stearns, S.C. (1976) Life history tactics: a review of the ideas. Q. Rev. Biol. 51, 3±47. Taggart, C.T. and Frank, K.Y. (1990) Perspectives on larval ®sh ecology and recruitment processes, probing the scales of relationships. In Sherman, K., Alexander, L.M. and Gold, B.D., eds. Large Marine Ecosystems, Patterns, Processes and Yields. Washington, DC: Am. Ass. Adv. Science, pp. 151±164. Thomas, R.M. and Boyd, A.J. (1985) Trends in the Namibian pilchard Sardinops ocellata ®shery in 1984. Colln Scient. Pap. Int. Commn SE Atl. Fish. 12, 181±191. ValdeÂs Szeinfeld, E.S. and Cochrane, K. (1992) The potential effects of canabalism and intraguild predation on anchovy recruitment and clupeoid ¯uctuations (Proc. Benguela Trophic Functioning, Cape Town). S. Afr. J. Mar. Sci. 12, 695±702. ValdeÂs Szeinfeld, E.S., Shelton, P.A., Armstrong, M.J. and Field, J.G. (1987) Cannabalism in South African anchovy: egg mortality and egg consumption rates (Proc. The Benguela and Comparable Ecosystems, Cape Town). S. Afr. J. Mar. Sci. 5, 613±622. Clupeoid stock dynamics in coastal upwelling areas 471

Valdivia, J.E.G. (1978) The anchoveta and El NinÄo. Rapp. P.-v.ReÂun. Cons. Int. Explor. Mer 173, 196±202. Victorov, S. (1996) Regional Satellite Oceanography. London: Taylor & Francis. 312 pp. Waldron, H.N., Brundrit, G.B. and Probyn, T.A. (1998) The relationship between annual potential new production in the southern Benguela upwelling system and its pelagic ®shery. S. Afr. J. Mar. Sci. 18 (in press). Woodruff, S.D., Slutz, R.J., Jenne, R.L. and Steurer, P.M. (1987) A comprehensive ocean±atmosphere data set. Bull. Am. Met. Soc. 68, 1239±1250. Ya nÄez, E., Garcia, M. and Barbieri, M.A. (1998) Pelagic ®sh stocks and environmental changes in the South East Paci®c (Proc. Global Versus Local Changes in Upwelling Systems, Paris). Paris: ORSTOM, pp. 274±291.

Accepted 3 July 1998