Predicting climate is a major scientific challenge. Society has a strong interest because natural climate variations have large impacts and because there are indications of a human influence on global climate.

Climate researchers study the complex atmosphere/ocean/sea-ice system. These studies have been stimulated in many countries. At a European level, funding came from the Fourth Framework Programme, through the Environment and Climate Programme, and - to a lesser extent - through the Marine Science and Technology programme. These European programmes have contributed to worldwide progress. However, this very progress also revealed the enormous complexity of the problem.

Climate variability and climate change are global phenomena and international coordination of climate research is essential. This has been realised at an early stage by organisations such as the World Meteorological Organisation (WMO), the International Council of Scientific Unions (ICSU), the United Nations Environment Programme (UNEP) and the Intergovernmental Oceanographic Commission (IOC), who started a World Climate Research Programme (WCRP) with the Tropical Ocean Global Atmosphere (TOGA) programme and the World Ocean Circulation Experiment (WOCE) as major elements. Although these programmes were quite successful by themselves, a certain amount of integration was lacking. When WCRP started to define a follow-up for WOCE and TOGA, it aimed at a broad programme, with input from theorists, modellers and observationalists from different disciplines (e.g. meteorology, oceanography and palaeoclimatology). The result is the new Climate Variability and Predictability (CLIVAR) programme. Central to CLIVAR is the problem of climate variability and predictability. The design of CLIVAR has taken many years, but now (1998) it has matured, and the programme is in the process of worldwide implementation. An outline of CLIVAR will be given in section 2.

In 1996, Euroclivar was established to help implement CLIVAR in Europe. The recommendations in this document are based on discussions in the Euroclivar committee (Annex D), where the following questions have been addressed:

• What is the relevance for Europe and for European policy making?
• How do the recommended activities fit in with worldwide climate research efforts, and what do they contribute to the CLIVAR programme?
• Do the proposed ideas make efficient use of available European expertise, and are they feasible and sufficiently mature?
• Is there added value to enhanced European collaboration? What can be gained from collaboration with groups outside Europe?

Amongst the numerous topics recognised by CLIVAR as necessary to better understand the climate variability, we selected a few key actions which may be studied efficiently by the European climate research community. In formulating our recommendations we took full advantage of the outcomes of the specialised Euroclivar workshops that have been held in the past few years. A list of these workshops is given in Annex B.

CLIVAR is a fifteen-year research programme. Its aim is to extend our understanding of climate variability and predictability, on time scales from seasons up to a century. The CLIVAR Science Plan (CLIVAR, 1995) gives the following specific objectives:

• To describe and understand physical processes responsible for climate variability and predictability on seasonal, interannual, decadal and centennial time scales through the collection and analysis of observations and the development and application of models of the coupled climate system,
• To extend the record of climate variability over time scales of interest through the assembly of high-quality palaeoclimatic and instrumental datasets,
• To extend the range and accuracy of seasonal to interannual climate prediction through the development of global, coupled predictive models,
• To understand and predict the response of the climate system to the growth of the radiatively active gases and aerosols and to compare these predictions to the observed climate record in order to detect the anthropogenic modification of the natural climate signal.

The CLIVAR programme is initially organised in three component programmes:

• CLIVAR-GOALS: a study of natural seasonal-to-interannual global climate variability and predictability [GOALS = Global Ocean Atmosphere Land System],
• CLIVAR-DecCen: a study of natural decadal-to-centennial global climate variability and predictability,
• CLIVAR-ACC: a study of the response of the climate system to the addition of radiatively active gases and aerosols to the atmosphere [ACC = Anthropogenic Climate Change].

However, it is intended that these component programmes merge when CLIVAR evolves, with projects that cut through time scales and programme components. Indeed, many tools and techniques are common to the different programme components. These are

• observing systems (both in situ and remote),
• numerical modelling and prediction,
• analytical and diagnostic studies,
• dataset development (including historic data and palaeo-data).

CLIVAR is administered by an International Project Office, which is located in Southampton. The CLIVAR Initial Implementation Plan (CLIVAR, 1998) was published in June 1998. A summary document (CLIVAR, 1997) appeared already earlier. The Implementation Plan identifies a number of Principal Research Areas. CLIVAR-GOALS consists of four areas. The first is "El Niño/Southern Oscillation (ENSO): Extending and Improving the Predictions". Two other areas are concerned with the interannual variability of the Monsoon Systems (Asian/Australian and American). The fourth area focuses on the "Variability of the African Climate System". CLIVAR-DecCen considers five Principal Research Areas. Four are related to the dominant modes of decadal variability: the Pacific-Indian mode, the Tropical Atlantic Variability, the Southern Ocean Climate Variability and the North Atlantic Oscillation. A fifth focus is on the Variability of the Atlantic Thermohaline Circulation. CLIVAR-ACC consists of two Principal Research Areas, namely Climate Change Prediction and Climate Change Detection and Attribution.

Common elements of all of these areas are:

• the identification of patterns and mechanisms of variability,
• extension of the data record of past climate data to study natural climate variability and to validate the models,
• the implementation of observing, modelling, computing and data collection systems to understand and predict climate variability and climate change,
• focused process studies both in the atmosphere (clouds, radiative feedbacks, etc) and in the ocean (watermass transformation, boundary currents, interbasin exchange, mixing, the role of eddies, deep convection, the role of sea ice, etc).

We refer to the Initial CLIVAR Implementation Plan (CLIVAR, 1998) for more details.

A better understanding of climate is of great importance for Europe. There are three broad categories of reasons:

• Natural fluctuations of the climate of Europe have many very significant consequences. Nearly all fields of human activity are affected: safety, health, infrastructure, agriculture, energy, economy . . . . 
• Natural climate fluctuations elsewhere in the world will have global implications through geophysical, socio-economic and political mechanisms. The geophysical relation is through so-called teleconnections which may link climate variability outside Europe with the climate of Europe.
• Improved prediction and detection of anthropogenic climate change will provide a firmer scientific basis for an active emission policy.

We will discuss each of these issues in some more detail.

The climate of Europe exhibits considerable variability on a wide range of time scales. Improved understanding of this variability is essential to assess

• the likely range of future climate fluctuations and the extent to which these fluctuations are predictable,
• the potential impact on Europe of climate change due to changing levels of greenhouse gases.

Many processes influence the climate of Europe. Although the chaotic component may be strong some understanding of more coherent behaviour in the Atlantic sector has started to emerge over the last few years. The best known pattern is the North Atlantic Oscillation (NAO) pattern, which is associated with the NAO Index, often defined as the pressure difference between the Azores and Iceland. When pressure is high in the Azores it tends to be low over Iceland, and vice versa. The NAO varies on many time scales but of particular interest are time scales of a season and longer. The NAO is associated with changes in climate throughout the Atlantic sector from Greenland to the equator. Positive anomalies of the NAO are associated with stronger than normal westerly winds, which tend to give higher than normal temperatures over Europe. Changes in the storm tracks and related moisture fluxes are associated with anomalous rainfall: a high NAO index is statistically associated with dry conditions over much of central and southern Europe. Indeed very many features of European climate can be related to the NAO.

The NAO also affects the oceans by its influence on the formation of deep water in the Greenland/Iceland/Norwegian Sea and Labrador Sea. This in turn influences the strength of the Atlantic Meridional Overturning Circulation (MOC), also loosely known as the thermohaline circulation. Variability in the MOC is of considerable potential importance to Europe, be it through variability around a stable mean or for its potential to undergo sudden changes. While it is likely that climatic feedbacks of the MOC variability, caused by atmospheric variability in the watermass transformation regions, occur on fairly long time scales, the possibility of sudden changes in MOC from one stable state to another needs investigation.

The NAO is however not the only source of atmospheric variability. In the tropical Atlantic, too, there is significant variability in Sea Surface Temperature (SST), winds and precipitation, on both interannual and decadal timescales. Although this variability may seem remote from Europe, it is not clear how variability in one region (the tropical Atlantic) affects that in another (Europe) either directly through atmospheric teleconnections or indirectly through processes involving the ocean. It is necessary therefore to take a whole-body approach to understand how the Atlantic as a whole influences European climate.

In summary, it is clear that the CLIVAR Principal Research Areas: North Atlantic Oscillation, Variability of the Atlantic Thermohaline Circulation and Tropical Atlantic Variability are of direct importance for Europe. It seems obvious that Europe must make a significant contribution to these research areas.

Life in tropical countries often revolves around subsistence farming of rain-fed crops, which renders its rapidly growing population vulnerable to climatic fluctuations. Tropical countries may be far away from Europe but the economical system is now global. If the rains destroy the growth of coffee in Central America or when the drought kills cattle in Australia, it will affect European people. The direct influence of ENSO on the European climate is still under debate, but there are indications that La Niña years are seen in the European climate. Teleconnections affecting the Mediterranean and European climate variability seem also to involve the Asian Summer Monsoon and the African climate. More investigations with seasonal forecast experiments will help understand if and how European climate is influenced. A successful seasonal forecast is not expected for every European winter, but it will be useful to get at least the extreme events, which could affect precipitation and temperature over Europe.

ENSO, the Asian Summer Monsoon and African climate are active areas of research throughout the European modelling community. There is also expertise in modelling of the tropical oceans. This is important because the oceans have a significant influence. The processes which determine the state of the atmosphere are varied and interactive. Correct simulation of these processes poses a demanding test for a General Circulation Model (GCM). Research in this area will lead to substantial benefits in terms of model development. An accurate simulation of tropical variability relies on the description of processes on a wide range of space and time scales. Advances in our understanding of these processes and our ability to simulate them will be equally relevant to weather forecasting, seasonal prediction and climate change studies. Europe has a strong tradition in assisting the monitoring of African climate, in process studies of African climate variability and in predictability studies for African climate. European initiatives have assisted in the building of indigenous national station datasets (especially through ORSTOM), even at the daily time scale. Through collaboration with local African scientists and institutes, datasets for the whole of the continent have been built. These greatly contributed to improving our knowledge of African climate and climate variability.

Current evidence shows that Europe is vulnerable to the anticipated anthropogenic climate change in several respects. The effects of reduced precipitation would be felt strongly in the south; rises in sea level would affect low-lying coastal areas (Venice, for example, and the Netherlands). In the absence of emission reductions, much firmer regional predictions are required to guide long term planning, particularly for water resources and sea defences. If emissions policies are to be put in place, then we need to know what reductions are needed to produce a given amelioration of change, and in the longer term, confirmation that policies are being effective. Simulated global mean temperature changes are needed to determine the required level of stabilisation. Currently there is an uncertainty of two degrees or more in the simulated global temperature change for the end of the next century. It is obvious that work needs to be done to improve projected temperature changes and stabilisation estimates. Europe is taking the lead on emission policies. As the implications of such policies become more apparent, the need to demonstrate real costs in avoiding emission reductions, and a cost benefit due to stabilisation scenarios will become more acute, both to carry policies within Europe, and persuading the rest of the world to follow.

	European and Atlantic climate variability	Global teleconnections	Anthropogenic climate change
CLIVAR GOALS
ENSO: Extending and Improving Predictions		X
Variability of the Asian-Australian Monsoon System		X
Variability of the American Monsoon System
Variability of the African Climate System	X	X
CLIVAR DecCen
North Atlantic Oscillation	X
Tropical Atlantic Decadal Variability	X
Atlantic Thermohaline Circulation	X
Indo-Pacific Decadal Variability		X
Southern Ocean Climate Variability
CLIVAR Anthropogenic Climate Change
Climate Change Prediction			X
Climate Change Detection and Attribution			X

We have organised our recommendations along lines that follow naturally from the above arguments:

• European and Atlantic climate variability,
• Global teleconnections,
• Anthropogenic climate change.

The relation with the CLIVAR plan has been indicated in table 1. Elements that are of particular interest for Europe have been marked. These elements will be worked out in the next three chapters. In general, we will closely follow the CLIVAR plan. However, the proposed studies of the North Atlantic Oscillation will comprise a wider range of time scales than envisaged under the CLIVAR DecCen plan; and the proposed global teleconnection studies will also address decadal time scales. A cross-cutting summary of observational and modelling aspects together with recommendations for their implementation is given in chapters 7 and 8.

This chapter consists of three sections. In the first section we give an outline of our present understanding of European and Atlantic climate variability. In section 4.2 we formulate our recommendations for a concerted observational programme. Section 4.3 recommends a number of modelling studies, with emphasis on better understanding of natural climate variations. Modelling of anthropogenic effects will be discussed in more detail in chapter 6.

A substantial proportion of the climate variability in the Atlantic/European region is associated with the NAO pattern. Although the NAO is a natural mode of variability of the atmosphere, its phase may be influenced by surface, anthropogenic or even stratospheric processes. The NAO Index has undergone major low frequency variations this century. Its phase was primarily positive (associated with strong westerlies) in the early part of the century, negative from the 1950s to 1970s, while record positive values were recorded in the 1980s and 90s. The change in the NAO over the past twenty years has been accompanied by a strengthening of the Pacific North American (PNA) pattern over the Pacific Ocean, cooling of the midlatitudinal oceans and warming over Northern Hemisphere land masses.

Several explanations have been proposed to explain the observed low frequency variability in the NAO. One hypothesis is that the NAO fluctuations are a response to changes in the oceans, in particular changes in SSTs. Although plausible, this suggestion has been hard to demonstrate. In fact, the largest variation in the NAO, the winter change from the negative values of the 1960s to the highly positive phase since the 1980s has not been reproduced in atmospheric models forced with the observed SST patterns, with one possible exception. Another hypothesis is that the low frequency NAO fluctuations have been a response to changing external forcings such as solar insolation, volcanic eruptions, or anthropogenic emissions of climatically important trace gases. A third hypothesis is that the NAO variability is simply generated internally in the atmosphere. Distinguishing between these hypotheses is a high priority goal. One of the difficulties is the short nature of the observational record. It is important, therefore, to extend the record of Atlantic climate variability by building on a variety of palaeoclimatic archives.

Low frequency variability in the North Atlantic Ocean is also poorly understood. Some of this variability arises as a response to fluctuations in the NAO. SST anomalies that have been observed to propagate slowly in the upper ocean gyre circulation may be formed in response to a change in the NAO. The NAO also influences the formation of deep water in the Greenland/Iceland/Norwegian (GIN) and Labrador seas, which in turn influence the strength of the Atlantic MOC. Although the MOC is a phenomenon of basin-to-global scale, its dynamics and in particular its variability may be controlled by rather small-scale processes associated with the formation and spreading of deep waters. Model results and palaeo-oceanographic data indicate long-term fluctuations in the overall strength of the MOC with transitions between fundamentally different states of the MOC sometimes taking place within decades. Although it appears that changes in the MOC can occur naturally, there is also the possibility that they could be triggered by anthropogenic emissions. A breakdown of the MOC could lead to drastic changes for global climate and especially European climate. The main objectives of a programme investigating the role of the Atlantic MOC should be to understand those oceanic and ice processes which are critical for the dynamics of the MOC, through modelling and observation, including analysis of palaeoclimatic data and the early instrumental record.

In the tropical Atlantic, there is substantial variability on interannual and decadal time scales. A possible dipolar structure in the tropical Atlantic SSTs, with opposite anomalies north and south of the equator has been noted. There is ample evidence, however, that SST fluctuations in these regions are not temporally coherent on all time scales. Although the tropical and higher latitudes are frequently considered separately, they overlap and are unlikely to be independent. There is also a relation between the Atlantic and the Pacific and its ENSO. Some interaction takes place via teleconnections within the tropics, and some via the higher latitude PNA pattern. Some of the interannual variability in the equatorial Atlantic may arise in response to tropical Pacific variability. Interannual variability in the tropical Atlantic is however weaker than in the tropical Pacific.

The predictability of climate fluctuations in the Atlantic/European region needs to be established. Although much of the variability appears random, not all the causes are understood, and so some predictability may exist. In particular, there is interannual predictability deriving from the influence of ENSO. Even if the predictability is limited, improved understanding of climate variability in the Atlantic/European region is essential so that we can better assess the potential impact on Europe of climate change due to changing levels of greenhouse gases, and the likelihood of rapid climate change associated with changes in the MOC.

A concerted modelling and observational programme is needed to observe, understand and predict variability in the Atlantic on time scales from seasonal to decadal. In the next section we propose a coordinated observing programme capable of addressing equatorial, tropical and midlatitude variability, and external influences such as ENSO.

The processes to be observed can be classified as upper-ocean (above 1000 m, though in parts of the ocean, such as the tropics perhaps considerably shallower) and intermediate and deep processes. In general, the observational systems required for the upper and lower part of the water column are at different stages of development. For this reason, it is convenient to separate the proposals for the upper ocean from those specifically for the MOC. In certain locations. however, it is possible to share instrumentation. It is almost always the case that the upper ocean observations can assist in observing the MOC, but not vice versa.

A permanent upper-ocean network is needed to document the equatorial and tropical anomalies as well as the generation and further development of the upper-layer anomalies like the Great Salinity Anomaly or the subtropical thermal anomalies which are apparently not just advected along with the mean circulation but involve air/sea interactions. An upper-ocean observational network for these anomalies would consist of the elements articulated below. Some of these elements are already operational at this time. To optimise the system it is recommended that design studies for a cost-efficient combined in situ and satellite upper-layer network for the Atlantic are carried out.

PIRATA
The Pilot Research Moored Array in the Tropical Atlantic (PIRATA) provides a basic observational system in the tropical Atlantic, which plays a similar central role to the Tropical Atmosphere Ocean (TAO) array in the Pacific. PIRATA is a pilot project, in which Brazil, France and the United States participate. It is funded only until 2000. It is recommended that PIRATA be maintained and expanded beyond the end of the pilot phase and that international participation be increased. The proposed extension would involve five additional stations and is necessary to better cover anomaly patterns in the southeastern and northeastern upwelling regimes as well as in the northern tropics. Western boundary observations are also needed to measure the equatorward flow of thermocline water; this flow is one branch of the shallow meridional circulation cell that links the subtropical and tropical regions of the South Atlantic, the southern Subtropical Cell (STC). The proposed additional stations are:

• P1 and P2 off the Mauritania-Senegal coast, with P1 in the vicinity of the Canaries current and P2 in the Guinea dome,
• P3 and P4 off the Angola coast in the vicinity of the Angola dome; P3 at about 5 E-5 S, P4 at about 10 W-10 S,
• P5 in the northwestern area of the present PIRATA array.

An extensive profiling float network
An extensive network A proposal (ARGO, Array for Real-time Geostrophic Oceanography) has been made to deploy a global network of autonomous profiling floats. For the Atlantic a deployment of ~1000 floats is envisaged. These should be deployed in both the North and South Atlantic (to 25 S in order to observe the variability associated with the southern part of the tropical 'dipole' as well as the interior structure of the subsurface, equatorward-flowing branch of both the northern and southern STCs). Salinity observations are particularly important in the subpolar North Atlantic, so the floats should have salinity sensors where possible. The problem of present grid-deployed float seedings, where floats may vacate some areas and cluster in others could be overcome in the future by new types of floats (e.g. gliders) that can actively navigate and keep a preselected position.

Voluntary Observing Ships
Voluntary Observing Ships (VOS) will continue to play an important role. A continuation of the VOS-XBT lines with coverage at WOCE intensity is required. Analyses for pattern coverage requirements have to be carried out to determine where high-density sampling is necessary. In the subpolar North Atlantic the depth of T7 XBT probes is not always sufficient to measure all of the mixed layer. Salinity observations are particularly important in the subpolar North Atlantic. Recording surface salinities by thermosalinograph on VOSs and along XBT lines would be important. The TOGA lines provide a useful complement to the PIRATA array. They should be maintained until any redundancy is established.

In addition to the oceanographic observations, the VOS meteorological measurements are also of great importance. The North Atlantic is well sampled, but improvements to the accuracy of the data are required. For example, the present VOS data may underestimate the heat fluxes during winter time cold air outbreaks over the western North Atlantic. Improved instrumentation (which is being developed both in the USA and Europe) should be implemented on a subset of the Atlantic VOS.

Time series stations
Continuous records at a relatively small set of stations have proved to be very effective in the past. These important stations must be continued. In particular:

• It is essential to continue stations Bravo and Panulirus/Bermuda in the centres of the subpolar and subtropical cells, respectively, which provide valuable information on decadal heat storage and circulation anomalies.
• Station Mike in the Norwegian Sea has given insight into decadal warming of the Norwegian Sea Deep Water and propagation of the Great Salinity Anomaly. Mike is scheduled to be discontinued in 2000. Observations at this site should be continued.
• In the eastern subtropics, station ESTOC was started a few years ago. Together with Bermuda it yields a baroclinic circulation index across the Subtropical Gyre.
• Two stations are required at the latitude of the tropical/subtropical exchanges, i.e. near 10-15 N to provide integral measurements of the zonal baroclinic variability gradient. One in the eastern interior might be a PIRATA station. A second one should be located close to the western boundary, near the Lesser Antilles.
• An expansion of the time series network into the South Atlantic should be considered.
• Other stations are under consideration for providing baroclinic transport time series for key advection elements in the subpolar and subtropical domains.

Time series stations of the future will be of the autonomous type, e.g. cycling CTDs or other types of moored stations or even be advanced free-floating devices (see also above).

Surface drifters
Surface drifters are required to determine upper-layer circulation and Ekman transports. These can be upgraded to measure winds, atmospheric pressure and salinity.

Remote sensing of SST and sea level
Satellite SST and altimetry analyses will add significant information on the patterns of currents and SST variations.

The primary source of atmospheric surface and upper-air observations for CLIVAR will be the Global Climate Observing System (GCOS) which is built upon the observing system that is maintained in support of operational numerical weather prediction. The atmospheric observation network for CLIVAR is predicated upon

• the maintenance of continuous operation of critically important ground and space-based components of GCOS,
• continuing improvements in the performance of four-dimensional data assimilation schemes that operate in real-time or near real-time,
• the availability of a succession of upgraded reanalysis products from ECMWF and NCEP, and
• the free and open exchange of observational datasets.

None of the above can necessarily be taken for granted.

Special atmospheric observations (localised in space and time) may be required to address specific issues relating to the treatment of boundary layer processes in coupled atmosphere/ocean/land models or for intercomparison of measurements made with different kinds of instrumentation.

Surface flux observations
Of special interest to oceanographers are the surface flux fields of momentum, heat and fresh water. Improving these requires improvements to atmospheric models used for weather analyses, and improvements in the way surface data are assimilated in such models as well as targeted flux studies.

Obtaining better surface fluxes remains a daunting challenge for CLIVAR. Both reanalysis products from Numerical Weather Prediction (NWP) centres, and satellite observations will be the primary data available for this purpose. The latter, however, require extensive "ground truth" and significant, large-scale biases have been identified in the former. Improvement of this situation requires focused air/sea flux studies in selected regions. Such regions would include various regimes where biases have not yet been determined, where ground truth has not been obtained, and where air/sea fluxes are too important to rely on NWP and satellite products (e.g. Labrador Sea, subpolar subduction). Such studies would include shipboard experiments and long-term (for an annual cycle at least) deployments of moored surface buoys. They would focus on improving and enhancing the NWP and satellite products, but would not be widespread enough to contribute directly to the global coverage of surface fluxes.

It is important to determine how realistic model simulations of variability in the thermohaline circulation are at short (decadal) time scales under imaginable anthropogenic changes. For comparison with model simulations an essential parameter is measurement of the rate of change of the MOC and associated heat and fresh water transports. Coverage for a particular latitude range would require a combination of the upper-ocean monitoring network, discussed above, with:

• repeated WOCE-type zonal hydrographic and current profiling ship sections,
• a moored current-meter array at the western boundary and additional stations for integrating baroclinic transports in the interior,
• bottom-moored devices to interpolate between full-depth stations, such as barotropic flow meters and inverted echo sounders.

A study for optimum coverage needs to be carried out analysing variability patterns of WOCE sections and the output of high-resolution numerical models. The 48 N section is a priority candidate for a MOC index section, for a number of reasons. First, it is the division line of subpolar/subtropical exchange. Second, anomaly patterns suggest a North-South heat storage anomaly dipole across this latitude. Third, a number of WOCE sections have already been obtained and can serve for pattern analysis. Finally, intent has been expressed by BSH (Germany) to repeat this section at WOCE quality as a contribution to GOOS every three years.

Several other sections with repeat top-to-bottom hydrographic coverage merit further consideration:

• 38 S, to measure the MOC, heat and freshwater exchange of the entire Atlantic with the global circulation system,
• 26.5 N, close to the heat transport maximum, where considerable prior knowledge on the variability has been built up,
• 10 S, to measure the combined effects of the STC and the top-to-bottom MOC,
• a meridional section to further subdivide the basin (Cape Farewell to Brazil).

If possible these sections should be combined with western boundary arrays.

In conjunction with the MOC section measurements, inventory observations of water mass changes and anthropogenic CO₂ uptake should be carried out in appropriate time intervals (to be determined from the WOCE/AIMS analysis).

Tracers
Much has been learned about ventilation rates and pathways from transient tracers (CFCs and anthropogenic tritium/helium), especially in the subpolar Atlantic and Nordic Seas. The decadal time scales of these substances permit a revealing focus on the integrated ocean circulation and its low-frequency variability. Transient tracers also provide powerful information on the rates of subduction processes which are very difficult to measure using other techniques. In addition, deliberate tracers (SF₆) have proved extremely effective in process studies for marking a specific volume of water and monitoring its evolution.

Transient tracers should be included in the suite of routine hydrographic measurements. Comparison of tracer data from sections repeated over several years will be particularly revealing. For example, the changes in Greenland Sea convection have been strikingly highlighted using these methods. Transient and deliberate tracers are also recommended for process studies where detailed knowledge of mixing and transport are required. Finally, comparison of tracer data with model predictions is a very good test of the realism of the modelled ventilation mechanisms and its variability. Systematic analysis of the extant transient tracer data is only just beginning and is actively encouraged.

There are strong links which exist between the NAO and the Arctic such as the observed correlations between the NAO and Fram Straits ice export and the NAO with warm water inflow into the Arctic. Therefore, it is important to extend observations well into the Nordic seas.

An important part of a CLIVAR observational network will be to monitor the Arctic freshwater export through the Denmark Straits and Baffin Bay as northern boundary conditions for North Atlantic anomalies. Close cooperation with ACSYS (joint participation in implementation working groups) is recommended.

In a wider context, monitoring the overflows over the Greenland-Iceland-Scotland sills is considered very important in order to supply a northern boundary condition for Atlantic Ocean modelling efforts.

Besides setting up quasi-permanent observations as summarised above, a set of process studies is needed in CLIVAR to improve climate variability in models and make parametrisation of certain poorly known processes in these models more realistic. These studies would involve observations, models and assimilation, as well as theory. We briefly outline them here, although they pertain also to many of the modelling issues discussed in section 4.3.

An anomaly experiment
After identification by the observing system, developing upper-ocean heat content anomalies will be sampled intensively to determine their evolution and propagation characteristics. In the light of their possible contribution to decadal atmospheric variability, anomalies in the upper layers of the ocean, circulating in both the subtropical and subpolar gyres, should be sampled intensively over several annual cycles, to determine the manner in which their associated SST and thermal anomalies are modified by air/sea interaction, and the cycle of mixing and restratification. Fluid passing through the Gulf Stream system is exposed to the influence of strong NAO forcing at the start of the Atlantic storm track, has its properties changed by convection, and then "sprays out" flooding the subpolar and subtropical gyres. We need to determine the longevity and characteristic pathways of these anomalies, and their potential influence on the atmosphere above. The upper-ocean observing network would be augmented with additional floats and surveys in key areas (e.g. the region of 18 degree water under the influence of strong NAO forcing). When used in combination with models assimilating altimetric (and other) data, the fate of the anomalies, their longevity and characteristic pathways can be mapped out. We would also like to quantify, as a function of time scale, the role of the ocean in the advection and transfer of heat to and from the surface, through subduction, Ekman layers, geostrophic eddies etc.

Subpolar mode water
Subpolar mode water process studies should be carried out to determine the mechanisms controlling the maintenance, evolution, and subduction of water directly modified by air/sea fluxes.

Subtropical mode water
Similarly, studies should be carried out to determine the formation and evolution of subtropical mode water

Equatorial processes
There is need for a dedicated process study designed to improve our understanding of equatorial mixing and transport processes. These processes, peculiar to the equatorial zone, play a large role in the tropical heat and fresh water balance, and are important for exchange to and from the subtropical oceans of each hemisphere.

Deep-circulation studies
These studies should be designed specifically to address constraints on the control of the MOC. For instance, the strength and distribution of mixing, or recirculation in deep basins should be studied.

A selected number of proxy data is needed to reconstruct decadal climate variability in the Atlantic sector over longer time periods than can be inferred from instrumental data. The most promising archives presently are tree rings and ice cores, in which the NAO has been detected. Varved lake deposits and historical reconstructions of weather patterns are further candidates for such reconstructions. These palaeoclimatic data need to be calibrated both with observations and reanalysis data sets (ERA40, NCEP). Such data will also indicate regions in which the signal of natural variability is particularly high and efforts of palaeoclimatic reconstruction must be enhanced. This task relies on the improvement and availability of the designated CLIVAR proxy data centres and on collaboration with ongoing activities in PAGES (CLIVAR-PAGES Intersection). A close link must be established also with the forecast centres (e.g., ECMWF, NCEP) who produce the atmospheric reanalyses.

GOOS
For decadal observations, government agency commitments are ultimately needed. These will be organised under the Global Ocean Observing System (GOOS). A GOOS Commitments Conference is planned for the fall of 1999. CLIVAR has to make sure that its essential needs for continued observations such as time-series stations and moored arrays, XBT lines, repeat hydrography sections and contributions to a profiling float network, will be taken up by agency commitments.

New Technology
In some cases, Europe is developing its own technology, in others (e.g. PIRATA) it is using USA instrumentation without developing its own expertise. Where appropriate Europe should develop the necessary expertise needed to implement a sustainable observing programme for the Atlantic from the tropics to the high latitudes. For a modern ocean observing system a combination of conventional and still to be developed technology will be required. A fruitful cooperation between the research community and industrial development firms in Europe can be anticipated. A start has been made to develop a European profiling float. This float, PROVOR, is a modified version of the French MARVOR.

Ocean analyses
It is also recommended that the WOCE measurements for the Atlantic for 97/98 be interpreted using dynamical models and a special data assimilation effort.

Organisation/coordination
The implementation of the suggested observational network needs to be done jointly by groups from both sides of the Atlantic. An overview panel needs to be established. It is recommended that such a panel is set up by CLIVAR. This panel should also be involved with the numerical modelling activities discussed below.

Many atmospheric GCMs still have major problems simulating synoptic and lower-frequency variability in the North Atlantic region. For example, simulated storm tracks are often too zonal, instead of curving northeastward, and the occurrence of Atlantic blocking events is often significantly underestimated. Furthermore, lower-frequency variability such as that of the stationary waves is often poorly represented. The difficulty experienced in trying to remedy these model problems reflects poor understanding of the processes involved. The biases in atmospheric GCMs are particularly evident in the Atlantic sector. These biases could lead to poor representation of interannual and interdecadal variability in such models as well as in the fluxes used to force ocean components in coupled models. These uncertainties need to be reduced in order to give more credibility to seasonal, decadal and anthropogenic forecasts made using GCMs.

Recommendations
The following studies are recommended:

• Studies with a hierarchy of atmosphere models ranging from state-of-the-art GCMs to simple idealised models, designed to investigate the mechanisms responsible for atmospheric variability on time scales from days to decades, and the processes that determine the spatial coherence of this variability, in particular with regard to the North Atlantic Oscillation.
• Studies to investigate the mechanisms behind atmospheric teleconnections into and out of the Atlantic sector. For example, how can a diabatic heating anomaly in the tropics lead to dynamical changes in the Atlantic sector? Do changes in the Atlantic subtropical jet lead to changes in the Atlantic intertropical convergence zone and vice versa?

Long-range forecasting relies heavily on the influence of SST anomalies on the atmosphere. Yet the nature of this influence, especially outside the tropics, is poorly understood. Experiments in which extratropical SST anomalies are imposed in GCMs generally indicate a response that is weak by comparison with the internal variability, but these results could be misleading. It appears that the response is unlikely to be through a simple local enhanced mean heating, as is often the case in the tropics, but rather through the impact on the development of individual weather systems which tend to grow over the western North Atlantic and decay over the eastern North Atlantic/Western Europe. It is not known to what space and time scales the atmosphere responds and in particular the extent to which the structure of the Gulf Stream is important to the atmosphere. These issues are vital for the consideration of possible interactive modes of the atmosphere/ocean system and to coupled modelling of the North Atlantic in general. An issue of particular importance is whether the recent observed trend in the NAO Index arose in response to changes in SST or not.

Recommendations
In order to address these issues, the following numerical experiments are required:

• Studies to investigate the sensitivity of individual weather systems to SST anomalies. Such studies would initially use high resolution forecasting models and cases for which considerable additional observational data exist (for example, FASTEX IOPs). The structure of the boundary layer, the magnitude of the surface fluxes, and the development of the whole system should be compared with observations. The sensitivity to various parametrisations (of the ocean mixed layer, for example) and to the specified SST anomalies should be determined. Additionally, the sensitivity of solutions to model resolution should be investigated.
• Studies to investigate the scales and patterns of Atlantic SST to which the atmosphere is most sensitive, and the atmospheric predictability that derives from changing SST. Ensemble integrations of atmospheric models forced with both observed and idealised SST anomalies are needed. Particular attention should be focused on: 1. whether the observed multidecadal fluctuation in the NAO index can be reliably reproduced and if so, which regions of SST are important; 2. the influence of tropical and subtropical Atlantic SST anomalies on the tropical and extratropical atmosphere. Carefully controlled studies in which several different models are forced with the same SST anomalies would be useful to check the robustness of results and increase ensemble sizes.
• Studies to investigate the usefulness and limitations of atmosphere model experiments with prescribed SST anomalies. Under what circumstances could the results of such experiments be misleading? To what extent can the results be related to the variability that arises when the same model is coupled to an ocean mixed layer or to a dynamical ocean model? What is the role of interactions between the atmosphere and land surface?
• Further development of data sets, for example reanalyses such as ERA40 and surface and subsurface temperature and salinity data sets, to improve characterisation of the observed atmosphere and ocean variability.

Recent work suggests that a large part of the variability in the North Atlantic Ocean can be understood as the ocean's response to atmospheric fluctuations. These fluctuations, which are dominated by large-scale patterns such as the NAO, influence the wind and buoyancy driven ocean circulations as well as the convection and subduction processes that determine the properties of the sub-mixed layer ocean. We are, however, far from possessing a detailed understanding of the relationships between the surface fluxes of heat, fresh water and momentum and temporal variation of the ocean state.

Of particular importance is improving understanding of the oceanic teleconnection pathways via which fluctuations in surface exchanges in one region can influence the ocean, and especially SST, in another region at a later time. Anomalous water masses formed along the northern flank of the Subtropical Gyre may re-emerge to influence SST in the Gulf Stream. Fluctuations in wind stress curl over the Subtropical Gyre may promote the development of heat-content anomalies that propagate northward to affect deep convection in the Greenland or Labrador Seas. Changes in deep convection may impact the overturning circulation, including its contribution to the Gulf Stream transport. Fluctuations in wind or buoyancy fluxes in the subtropics may influence equatorial SST via meridional circulation cells that feed equatorial upwelling.

Recommendations
We recommend a wide range of studies with ocean models of varying complexity to investigate the local and non-local relationships between temporal variations in the surface fluxes of heat, fresh water and momentum and temporal variations in the ocean state. In the design of models and experiments special attention should be paid to the following:

• Key processes in the "warm water transformation" pathway. These processes include: 1. the mechanisms responsible for the development, propagation, and decay of heat content anomalies; 2. the interplay of processes that influence oceanic convection, such as sea-ice formation and melting, run-off, advection of anomalous temperature and salinity from lower latitudes and from the Arctic, and variability in local surface buoyancy fluxes; 3. the processes that modulate the overturning circulation; and 4. the impact of changes in the overturning circulation on SST.
• Mechanisms that determine SST variability in the tropical Atlantic, in particular the role of ocean dynamics. Also mechanisms linking the tropical Atlantic with higher latitudes including 1. the modulation of meridional circulation cells by subtropical winds or buoyancy fluxes and 2. the influence of the tropical Atlantic variability on higher latitudes through, for example, modulation of the western boundary current input.
• The importance on decadal time scales of internal ocean variability.

In addition we recommend the following studies:

• Studies with Ocean General Circulation Models (OGCMs) to investigate how well the observed evolution of the Atlantic ocean, including the tropical Atlantic, can be simulated given observed surface fluxes or related quantities. Careful thought should be given to the formulation of the surface boundary condition in these studies. Multidecadal model simulations should be rigorously evaluated against available surface and subsurface observations, with particular attention to poorly understood phenomena such as the development, propagation and decay of heat content anomalies and to basic features such as the distribution of principle water masses. A comparison between different GCMs would help to establish which errors in simulations are associated with surface flux errors and which with model errors. Model experiments could be designed to address which aspects of the surface forcing are essential to reproduce key features of the observed variability.
• Studies to investigate the nature and the role in the large-scale circulation of key physical processes that are poorly represented in OGCMs, in particular the role of topography (especially "choke points") and of turbulence.
• Development of an ocean data-assimilation system. An active programme in data-model synthesis must be a component of the programme, both to help plan it and to interpret the observations taken. The combination of model and data fields, presuming that both contain elements of the same variability, allows for a better definition of what took place than either by itself. The comparisons of model simulations without data assimilation to these analyses (as well as to analyses based solely on data) allow model deficiencies to be identified and hence point to directions for model improvements. If predictability is established, these analyses then serve as the initial conditions for forecasts of the coupled system.

Coupled ocean/atmosphere GCMs are powerful tools for understanding the climate system. However, due to biases in both the atmospheric and the oceanic components, such models often have unrealistic mean states, annual cycles, and interannual variabilities. Much work remains to be done in documenting, understanding and alleviating such problems.

Coupled GCMs are the central tools for seasonal and longer time scale forecasting. There is evidence of seasonal predictability over Europe in spring and summer but the mechanisms that give rise to this predictability are not understood. High priority must be given to research directed at improving the realism of coupled forecast models, improving understanding of the basis for predictability, and improving forecasting techniques.

Recommendations
The following studies are recommended:

• Studies to evaluate and improve the skill with which coupled ocean/atmosphere models are able to simulate the observed mean climate, including the seasonal cycle. Development of new parametrisations for key processes in the coupled ocean/atmosphere system.
• Studies to elucidate the mechanisms responsible for variability in the Atlantic/European sector in coupled GCMs, and to evaluate the realism with which observed seasonal to decadal variability is simulated. Insight into mechanisms may be gained through experiments in which ocean/atmosphere interactions are restricted to certain regions (such as the Atlantic). The role of interactions between the atmosphere and the land-surface must also be investigated.
• Studies to elucidate the influence of ENSO on the Atlantic/European sector. What SST anomalies does ENSO force in the Atlantic and do these anomalies significantly influence the atmosphere? How predictable is the Atlantic response to ENSO? (This will be discussed in more detail in sections 5.1 and 5.2.)
• Studies to investigate the mechanisms responsible for variability in the tropical Atlantic, including the "Atlantic ENSO" and the influence through teleconnections. (See also sections 5.1 and 5.2.)
• Studies using coupled models to investigate the predictability of natural climate fluctuations in the Atlantic/European sector on both seasonal and decadal time scales. Such studies should consider the role of land surface processes as well as ocean/atmosphere interactions.
• Research into the use of coupled models for forecasting, including the problems of initialising and assimilating data into coupled models, and the sensitivity of forecasts to model formulation.
• Studies to investigate how changing levels of greenhouse gases may influence climate variability in the Atlantic/European sector, and how natural variability influences the response of climate system to such changes. (See also chapter 6.)

There is considerable evidence for the existence of patterns called "teleconnections" as the major modes of climate variability at the seasonal to the decadal time scales. Teleconnections extend over the entire globe and, as already briefly discussed in the previous chapter, several of them involve the European continent. Major teleconnection systems that are potentially relevant to Europe and the Mediterranean are linked to phenomena in the tropical atmosphere/ocean system. Specifically, these involve ENSO, the Asian Summer Monsoon, and several components of the African and American climate system. Thus it is necessary that these components of the climate system are studied in their own right. In this chapter we will first discuss some general aspects of teleconnections. This will be followed by sections devoted to ENSO, the Asian Summer Monsoon and African climate variability, three areas in which European researchers have considerable expertise. The variability of the American Monsoon Systems (VAMOS), another key area of CLIVAR, is covered by a Pan-American initiative.

It is important to realise that teleconnections basically represent a statistical relationship (i.e. correlation) from which causality cannot necessarily be inferred. Therefore sensitivity experiments with coupled and uncoupled models are essential for identifying the key regions that give rise to these teleconnection patterns.

Seasonal prediction for the European/Atlantic sector is likely to depend heavily on the quality of the simulation of the teleconnection patterns. A similar argument may also apply for climate change prediction. There are three distinct elements to this problem. If we consider that ENSO is the major mode of interannual variability, then the first question is whether a model produces a realistic simulation of tropical SST variability. This may depend on, for example, interactions with intraseasonal variability. If the SST variability is taken as given (e.g. as in atmosphere-only AMIP-type runs), then the second important question relates to the skill of the model in translating the SST anomaly into an atmospheric diabatic heating anomaly. This is by no means straightforward and will be sensitive, particularly, to details of the boundary-layer and convection schemes. The third element involves the teleconnection of that diabatic heating anomaly to remote areas through the global circulation. Even if the model successfully translates the SST anomaly into a diabatic heating anomaly, the teleconnection will depend on the model's basic state. In all these elements the seasonal cycle is of fundamental importance. SST variability, such as in El Niño, may have a seasonality, and the translation and teleconnection characteristics will depend strongly on the phase of the seasonal cycle through the basic state.

Much of the energy which drives the circulation of the atmosphere is absorbed in the tropics in processes, which are strongly influenced by tropical convection. Errors in the simulated tropical circulation can often be related to errors in the mean distribution of the heating and hence in the representation of the latent heating by cumulus convection. The direct response of the tropical atmosphere to El Niño is manifested in an eastwards shift in the main area of convection over the Pacific Ocean. It is this displacement in the heating pattern that gives rise to the teleconnection patterns which communicate El Niño to the rest of the globe. Thus cumulus convection is of fundamental importance in our understanding and ability to simulate El Niño and its global teleconnections.

It is clear from satellite observations that the seasonal mean distribution of diabatic heating is built up from a variety of space and time scales. These range from the individual clouds, to the cloud clusters associated with synoptic scale disturbances, to the super clusters or ensembles of clusters. In turn, the mesoscale (sub-grid scale) organisation of convection can significantly affect the surface turbulent fluxes through wind gustiness, which may have important implications for the hydrological budget in the tropics. It is becoming increasingly clear that it is not sufficient for a GCM just to simulate reasonable seasonal or monthly mean fields. Indeed, greater emphasis is now being placed on the need for understanding the various spatial and temporal scales that make up the climatic mean and the ability of a GCM to represent these scales. Consequently, convective and surface flux parametrisations must be developed in the light of the interactions between the wide range of space and time scales of cumulus convection. A key question is whether convection acts to force the large scale circulation, whether it is purely a slave of the large scale circulation or whether it involves feedbacks between the two.

• The signature, in space and time, of the major teleconnection patterns relevant to Europe and the Mediterranean should be established using reanalyses. Their key source regions and the mechanisms (coupled and uncoupled) by which they are communicated to remote regions should be investigated using diagnostic studies of reanalyses and model simulations, as well as sensitivity experiments with coupled and uncoupled models.
• The impact of the quality of the simulated teleconnection patterns on the level of skill for seasonal prediction models should be assessed. The interaction between systematic errors in the model's basic state and the quality of the teleconnection patterns should be investigated through sensitivity experiments.
• The interaction between the various scales of convective organisation should be simulated using a hierarchy of models ranging from cloud resolving models and single column models through to the complex GCMs. The parametrisation of cumulus convection in these models needs to be improved, including the effects of subgridscale convective organisation on surface fluxes. TOGA-COARE data should be exploited to investigate the interaction between various scales of tropical convection. This research is closely allied to the aims of GEWEX, particularly the GEWEX Cloud Systems Study (GCSS) in which European scientists actively participate.
• Continued assessment should be made of the ability of GCMs to represent the various spatial and temporal scales of convection, particularly those associated with convectively coupled equatorial waves, and including the influence of the extratropics on tropical convection and circulation. This should be a standard diagnostic procedure for validating convective parametrisations. Appropriate validation datasets based on reanalyses, satellite data and in situ observations should be compiled. Existing datasets, such as the high resolution global cloud imagery archive developed by the EU Project CLAUS (Cloud Archive User Service) should be exploited.

The El Niño/Southern Oscillation (ENSO) is the dominant mode of coupled ocean/atmosphere variability at interannual scales. As such it warrants a continuing research effort within Europe to understand the processes involved in ENSO (including the relationship with decadal variability), to improve its simulation and to investigate its impact on European climate variability.

During the last decade, a dense observation network has been deployed in the tropical Pacific by the TOGA programme, composed of tide gauges, ships of opportunity and TAO moorings. The full network was completed in 1992 and, with the help of satellites, a survey of anomalous events is now possible. Maps in real time are available to follow the development of anomalies in the atmosphere and ocean, and were particularly effective in monitoring the growth of the 1997/98 El Niño. Recent observations have highlighted the difficulty in defining a canonical pattern for El Niño since its amplitude, its phase and its structure during its mature phase may vary from one event to another. The success of simple dynamical models to forecast ENSO during the 1980s was subsequently overturned by the prolonged Pacific Ocean warm event of the early 1990s and by the rapid growth of the 1997/98 El Niño.

Although significant progress has been made in understanding and modelling the dynamics of El Niño, the 1997/98 event raised some important issues. In particular, the possibility that coupling between the atmosphere and the ocean at all scales may be important for determining the mean climate and its low frequency variability is an area of increasing interest. For example, westerly wind bursts (WWB),which are known to excite oceanic Kelvin waves, themselves potential players in El Niño, appear to be closely related to the active phase of the intraseasonal or Madden-Julian Oscillation (MJO). The MJO propagates from the Indian Ocean to the Pacific Ocean and strengthens considerably over the western Pacific. It shows both pronounced seasonality and strong interannual variability in its occurrence. The general relationship between the activity of the MJO and WWBs, equatorial ocean waves and El Niño is not fully understood but, as recent events suggest, may be very important.

A primary aim of CLIVAR is to extend and improve the predictions of El Niño. As part of achieving this goal, it is particularly important to clarify the early stages of the development of an El Niño event. While the development phase is characterised by the eastward propagation of the thermocline anomaly, it is not clear why and when the propagation starts. Its beginning seems to coincide with the occurrence of the WWBs over the western Pacific. These wind bursts deepen the upper-ocean mixed layer and generate eastward jets in the vicinity of the equator. Whether this advective process can provide a significant pulse for the instability to grow has to be explored.

The response of the ocean to atmospheric subseasonal variability is complex and depends crucially on the internal structure of the upper layers of the ocean. Whether this local complex response is important for the large-scale and low-frequency adjustment is still an open question, although there is evidence to suggest that mechanisms linked to salinity could potentially affect the large scale response. In the western Pacific where fresh water input is high, salinity is especially important because it limits the depth of the mixed layer and therefore influences the momentum and heat budgets of the upper-ocean mixed layer. The transition zone between the warm pool and the central Pacific is marked by a strong salinity front, associated with a convergence of zonal currents. A WWB blowing above the eastern part of the fresh pool destroys the local barrier layer by entrainment, generates an eastward oceanic surface jet which advects the salinity front eastward, thus creating a new barrier layer further east. Its presence tends to increase the warming in the upper-ocean mixed layer, which may, in a coupled framework, favour the eastward propagation of the WWB, thus potentially acting as an air/sea instability which might favour the development of El Niño.

The classic description of El Niño is that it behaves as a delayed oscillator. TOPEX/POSEIDON sea level data have been shown to accurately observe the sea surface height variability over the globe and, more particularly, in the tropical Pacific, the core region of ENSO. Various studies have already provided descriptions of long equatorial wave propagation as well as estimates of their reflection at both the eastern and western boundaries. The observations during the 1997/98 event suggest that once the event was initiated (likely by westerly wind anomalies in the western Pacific and the generation of ocean Kelvin waves), the development and decay phases of the warm conditions during the 1997/98 El Niño event were in good agreement with the delayed action oscillator mechanism. Finally, ocean/atmosphere coupled feedbacks (e.g. location of westerly wind anomalies west of the 29^o C isotherm, development of easterly wind anomalies in the western Pacific in 1998) will need to be investigated in the context of a coupled model to fully address the coupled nature of ENSO.

• Greater understanding of the phenomenology of sub-seasonal atmospheric variability including its mean behaviour and its seasonal and interannual variability needs to be developed. This can be achieved within the existing resources using reanalyses, satellite and ground-based observations. Assessment of the representation of subseasonal atmospheric variability in seasonal and climate prediction models, coupled and uncoupled, should be a standard diagnostic procedure and should form part of the intercomparison for atmosphere-only (e.g. AMIP II) and coupled models (e.g. CLIVAR WGCM).
• The basic simulation of the MJO should be improved. Further research into the influence of an interactive ocean surface and the extratropics on the characteristics of subseasonal atmospheric variability should be carried out.
• The communication of atmospheric subseasonal variability to the ocean is through wind stress and surface heat fluxes. A consistent dataset of wind stresses, turbulent and radiative fluxes should be developed; this should involve the development of a single flux algorithm for estimating turbulent fluxes and consolidation of existing estimates of radiative fluxes; and it should include an estimation of uncertainties. The impact of these uncertainties on the ocean response should be investigated through sensitivity experiments with OGCMs.
• The importance of the internal structure of the upper layers of the ocean, particularly salinity and currents, in determining the response of the ocean to subseasonal atmospheric forcing should be investigated. Diagnostic studies of detailed in situ observations, particularly during TOGA-COARE, should be used to validate OGCM structures and to guide the development of subgridscale parametrisations.
• As a continuation of the initial studies within the EU project on Scale Interactions (SINTEX), the impact of subseasonal atmospheric variability on the simulation of the mean state and El Niño in coupled seasonal and climate prediction models should be investigated. Appropriate validation datasets should be developed which describe the behaviour of the upper ocean and the characteristics of ocean waves, such as those from the TAO buoy data and from TOPEX/POSEIDON observations of sea surface height.
• The implications of scale interactions for the limits of predictability for El Niño should be assessed. The spatial patterns of stochastic forcing that are optimal for exciting or disrupting ENSO should be investigated in a range of coupled models. The results may be used to provide guidance on where/what additional observations would be beneficial. The possibility that the apparently chaotic interannual behaviour of the MJO may act as a limit to seasonal and interannual prediction of ENSO should be investigated further with reanalyses, satellite data and ensembles of model integrations.

Historical data analysis has revealed decadal and multi-decadal patterns in ENSO and other modes of tropical variability. This may be related to slow stochastic variations in the ENSO system and/or to slow variations in the base states of the coupled climate system which modulates variation on interannual time scales. The analysis of observations is often limited by the difficulty in constructing long climatic records and in separating the interannual variability from a slowly evolving basic state. It is necessary to compare the recent decades to much longer time series, going back to the beginning of industrialisation, and to diagnose the evolution of ENSO and its teleconnections on these long time scales. This is necessary to determine whether recent changes in the behaviour of ENSO are linked to natural decadal variability, or whether ENSO is potentially being modified by global warming.

New theories have been proposed to explain the decadal modulation of the ENSO variability. Most of them are based on the idea that the structure of the equatorial thermocline undergoes decadal changes. It has been hypothesised that these are caused by coupling of the central North Pacific to the equatorial Pacific via thermal anomalies that propagate equatorward in the oceanic thermocline. Whether upper ocean temperature observations can be interpreted in support of this hypothesis is controversial. Numerical experiments using ocean circulation models of different complexity alternatively indicate that decadal tropical variability is mainly driven by decadal changes in low-latitude winds.

An interdecadal trend in MJO activity has been noted in the NCEP/NCAR 40-year reanalysis, which has been confirmed in model results using an ensemble of atmosphere-only integrations forced with observed SSTs. It is suggested that the increase in MJO activity during the 1980s and 90s might be related to a long-term warming of tropical SSTs during the period 1950 to present. It is interesting to note that the increased activity of the MJO coincides with the more active El Niño cycle during the 1980s and 90s. The decadal changes in MJO activity also suggest that if tropical SSTs continue to warm, the activity of the MJO may tend to increase further which then might have implications for the future behaviour of El Niño.

• Extended records of El Niño and tropical SSTs should be developed using conventional and proxy data. Estimation of uncertainties in decadal variability and long term trends should be included. The sensitivity of the record to time variations in the observational database, particularly the introduction of satellite data in the late 1970s, should be investigated. Sophisticated statistical tools should be developed to separate interannual variability from the slowly evolving background state.
• The extent to which decadal variability in the tropical/equatorial regions is dominated by tropical wind forcing and not by thermal anomalies arriving from midlatitudes should be assessed in long integrations of coupled and ocean-only models. Furthermore, these integrations shall be used to establish how decadal changes of the equatorial thermocline structure are related to a decadal modulation of ENSO variability.
• The possible interdecadal trend in MJO activity should be confirmed using further model studies and future reanalysis projects e.g. the ECMWF 40-year reanalysis (ERA40). This is necessary to determine whether this trend is real or whether it is an artefact of the changing observational database through the reanalysis record. Reanalysis experiments to ascertain the sensitivity to changes in the observational database (e.g. introduction of satellite winds) should be encouraged.

The possible relation between ENSO and European climate variations was already briefly discussed in chapter 4. There is increasing evidence that during major ENSO events one can attribute climatic anomalies in the European/Atlantic sector to the remote effects of El Niño. Recent success at ECMWF in predicting the seasonal mean anomalies for Europe during the strong El Niño of 1997/98 has emphasised the need for understanding the processes by which ENSO can influence the circulation in the Atlantic/European sector. Firstly, the Pacific/North American (PNA) pattern has a robust signal which extends over the east coast of North America and has a considerable influence on the regional circulation over the western North Atlantic. Secondly, ENSO may induce a remote response in the SSTs of the tropical and subtropical Atlantic Ocean through the modulation of the trade winds. tropical Atlantic SSTs are known to be correlated with rainfall in the Sahel and north-east Brazil (Nordeste), and have been implicated in recent active hurricane seasons. Teleconnections between the tropical Atlantic and higher latitudes may also exist.

• The relationship between ENSO and the variability in the Atlantic/European sector should be studied using reanalyses and model diagnostics. Experimentation with coupled models should be carried out to investigate the mechanisms through which variability in the Atlantic coupled system is sensitive to the effects of ENSO. What SST anomalies does ENSO force in the Atlantic and do these anomalies significantly influence the atmosphere?
• The importance of the response of the tropical Atlantic ocean to ENSO forcing should be assessed. Essential to this is the continuation of the PIRATA array, which was already recommended in chapter 4.
• The influence of ENSO on the seasonal predictability of the European/Atlantic sector should continue to be assessed in seasonal forecast ensembles with atmosphere-only and coupled models, as is already in progress through the EU Project on Seasonal Prediction (PROVOST). Research to improve the basic simulation of El Niño in seasonal forecast and climate models should be on-going. The sensitivity of the prediction to the definition of the initial state of the ocean should be investigated in collaboration with ocean data assimilation studies, such as those initiated within the EU Projects AGORA and PROVOST.

The Asian Summer Monsoon dominates the tropical circulation of the Eastern Hemisphere; due to the greater continentality of the Northern Hemisphere and the unique orographic configuration of the East African Highlands and the Tibetan Plateau, the Asian Summer Monsoon is the most vigorous and influential of all the monsoon circulations. The influence of the Asian Summer Monsoon extends to many regions remote from Southeast Asia. Specifically, recent studies have suggested that the arid regions of North Africa and the dry summers of the Eastern Mediterranean may be a direct consequence of the Asian Summer Monsoon. The seasonal cycle of the Eastern Mediterranean, particularly the commencement of dry, settled summer weather, is closely tied to the seasonal cycle of the Asian Summer Monsoon, specifically the onset of the monsoon in May.

The Asian Summer Monsoon exhibits substantial interannual variability which is related to the slowly varying boundary forcing, particularly the phase of ENSO, indicating that there may be potential for predictability at seasonal and interannual time scales. This has already been exploited in the statistical forecasts for India, which show considerable skill. This skill has not been matched, however, by dynamical forecasts.

Considerable progress has been made through the EU Project on the Asian Summer Monsoon (SHIVA) in understanding the factors that determine monsoon variability and the interactions between intraseasonal and interannual time scales. SHIVA has contributed significantly to model development and to improvements in various aspects of monsoon simulation. Nevertheless there remain many outstanding questions such as the reasons for the continuing lack of seasonal predictability with dynamical methods. One of the future major challenges will be to understand why numerical models show so little skill and to identify those key physical processes that are crucial for improving model performance. Specifically the role of land-surface processes and the interaction with the Indian Ocean at subseasonal and interannual time scales needs to be studied.

The Asian Summer Monsoon is also believed to play an active role in the interannual variability of the coupled ocean/atmosphere system, specifically in the evolution of ENSO and the Tropical Biennial Oscillation (TBO). Future research should concentrate on the simulation of the monsoon in coupled models and in clarifying its role as either a broadcaster or receptor of ENSO.

The extended record of All India Rainfall displays multi-decadal behaviour in which there is a clustering of wet or dry anomalies. It is clear that the epochal behaviour of the Indian Summer Monsoon rainfall is not associated with similar variations in El Niño, i.e. that dry epochs are not associated with a tendency for a clustering of warm El Niño events. The mechanisms involved in this decadal variability are not understood, but the influence of decadal time scale fluctuations in SST cannot be ruled out.

• Improvements in the simulation of the monsoon's mean seasonal evolution, its intraseasonal characteristics and its response to changes in the boundary forcing are still needed. This will rely on the observational database, particularly reanalyses, to validate and identify deficiencies in models. Field data from the GEWEX Asian Monsoon Experiment (GAME) should be exploited. Intercomparison of monsoon simulations in atmosphere-only (AMIP II) and coupled models (CLIVAR WGCM) should be supported.
• The key regions of the climate system where perturbations (e.g. SST or land-surface anomalies) can influence the variability of the monsoon on seasonal to interannual time scales with some predictability need to be identified. This will involve well-designed forcing experiments with primarily atmosphere-only models. It will include identification of the relationship between interannual and intraseasonal variability and the extent to which the intraseasonal variability is influenced by the slowly varying boundary conditions.
• The observing systems that need to be put in place to monitor the perturbations in these climatically sensitive regions should be identified. The development of a predictive capability for seasonal to interannual time scales will require initialisation of ocean/atmosphere models with real-time ocean observations. The placement of the TAO array in the tropical Pacific has had an enormous impact on the predictive skill for El Niño; for other parts of the ocean the observational base remains extremely limited. In particular, the development of a buoy network in the tropical Indian Ocean should be encouraged and supported.
• The role of the Asian Summer Monsoon in the global circulation, and its potential to influence the interannual variability of the coupled ocean/atmosphere/land system should be investigated further using long integrations with coupled models. The decadal variability of the monsoon in these integrations should be diagnosed and mechanisms to explain this variability should be sought.
• Links with scientists in monsoon-affected countries should be encouraged.

The African continent is the largest of all tropical land masses and with a rapidly growing population it clearly has social and economical importance as well as meteorological. Due to the severe and sometimes tragic social and economical impacts of interannual variability of rainfall in the semi-arid and tropical rainy zones of both hemispheres most scientific interest to date has focused on these regions. However climate variability has a major impact on the economies of most African nations, and a programme to provide the scientific basis for improved prediction can be expected to play a substantial role in sustainable development for the continent.

A primary concern is predictability, on seasonal-to-interannual and intraseasonal timescales. Dynamical models often have difficulty in predicting the rainfall variability and in previous intercomparisons (AMIP-I) considerable differences in seasonal rainfall predictions for some African regions (e.g. the Sahel) have been found between models forced with the same SSTs.

The problems models have in predicting rainfall in this region are likely to be due to a combination of poorly simulated seasonal cycles associated with a poor representation of deep cumulus convection and land-surface interactions together with poor representations of the important global teleconnections.

Our understanding of how deep cumulus convection interacts with the land surface, including vegetation, is particularly poor. The skill of a GCM in the semi-arid zones will depend very strongly on this. It is essential that we develop our understanding and modelling capabilities in this regard. Any investigation of these processes must also involve an examination of the scale interactions between the convection and the rain-producing weather systems themselves including the squall lines and easterly waves. Early results from the EU-funded West African Monsoon Project (WAMP) have suggested that the reanalysis datasets may be inadequate for some applications due to lack of observations in critical regions, though this is not true for all regions of Africa. This evaluation of reanalysis products should continue. We require an integrated approach to improving model simulations, linking a hierarchy of modelling efforts together with observations. It is also increasingly important to link this work with hydrology and agriculture.

Evidence suggests that African climate variability is linked to major global modes of variability. This includes links with the NAO, ENSO as well as tropical Atlantic SST variability. While a coupled modelling effort is clearly required to predict the global SSTs, the mechanisms through which remote regions affect the African climate are far from understood and require investigation. Evidence also suggests that the decadal rainfall variability observed in the Sahel may be linked to variability in the nature of these teleconnections, in particular with ENSO and the Atlantic Ocean. Further work is required to unravel the variability in the nature of these teleconnections by combining model simulations with available observations.

It is also important to remember that these interactions are two-way. In particular, the tropical and subtropical troposphere over the Atlantic is influenced by African climate variability. A clear example of this is the observed correlation between Atlantic tropical cyclone activity and west African rainfall. The mechanisms for this, which may involve large-scale teleconnections as well as easterly waves, have yet to be determined.

Understanding teleconnections on hourly, daily, weekly and monthly-seasonal timescales is critical to underpinning predictability. The teleconnections include links from tropical convection to the atmosphere across surrounding tropical oceans, and also teleconnections from tropical Africa to midlatitudes, including Europe. Lag responses of Indian and tropical Atlantic oceans to ENSO complicate the ENSO/Africa teleconnection chain, since these oceans also influence the climates of Africa directly. Indeed, we need to know more about how these oceans generate their own internal coupled ocean/atmosphere variability, and the extent to which the key Atlantic and Indian Ocean SSTs are predictable.

Our knowledge of African climate has benefited from a number of field experiments and from routine upper-air and surface observations. In recent years though the spatial and temporal resolution of routine observations has deteriorated. It is essential that efforts are made to improve this situation for climate monitoring and routine forecasting. Extending the PIRATA oceanic array across the Atlantic should also be encouraged, as well as developing an array for the Indian Ocean.

• Improvements in our understanding and simulation of the two-way teleconnections between the climate variability across the African continent and coupled ocean/atmosphere variations around the tropics is required. A key question is why ENSO effects are transmitted to Africa in some years, but not in others. In addition, teleconnections between African climate variability and variations at higher latitudes need to be better identified and understood, especially the role of the Mediterranean region in African climate. These teleconnections are again two-way. The research methodology should involve a combination of model intercomparison, idealised modelling experiments, and diagnostic study of reanalysis data and available observations.
• Studies of subtropical African climates and their interactions with low-frequency modes of midlatitude variability like the NAO are needed.
• The source of decadal variability in the African climate system and its interaction with interannual time scales needs to be addressed. An important issue is the extent to which decadal variability has been generated by internal land/atmosphere interaction over Africa, or whether coupled ocean/atmosphere processes dominate.
• The spatial and temporal extent of African seasonal predictability should be estimated using dynamical models and statistical relationships. Predictability studies should include estimates of the likelihood of major weather events (e.g. tropical cyclones and dry spell lengths). The key SSTs in the Atlantic and Indian Ocean should be evaluated and integrated into overall predictability estimates for African climate. Dynamical and statistical methods of downscaling GCM simulations and predictions need to be developed and evaluated.
• Improvements in the simulation of the mean seasonal cycle of rainfall in tropical and subtropical Africa and its intraseasonal characteristics is required. The interaction of deep cumulus convection with the land surface and evolving vegetation should be a crucial part of this together with the weather systems themselves to address scale interaction issues. This should involve the use of a hierarchy of different types of model together with observations.
• An observational programme in West Africa is recommended to support the seasonal cycle work and scale interactions. This should focus on the evolving land surface and its interaction with the deep cumulus convection and hydrology. This observational programme should build upon the CATCH hydrological experiment which should become a GEWEX continental scale experiment.
• Efforts should be made to maintain and increase our routine monitoring network through expanding the upper-air network. Efforts should be made to ensure that existing observational data is not lost so that it can be used in future reanalyses. Studies with existing reanalysis products should provide suggestions on where and what types of additional observations should have the best impact on improved analysis products, leading to better understanding and prediction.
• Extending the PIRATA oceanic array across the Atlantic and including a buoy network in the Indian Ocean should also be encouraged.
• Scientific collaboration with African scientists should be encouraged. Alongside this, the role of the African regional centres as data centres should be supported.

This chapter focuses on understanding and predicting anthropogenic climate change. It consists of three parts. In the first section prediction of human-induced climate change is discussed. The treatment of clouds in climate models is an important source of uncertainty. Therefore, section 6.2 is dedicated to cloud representation in climate models. The last section is primarily concerned with the detection and attribution of climate change, but it also contains several suggestions for model improvement.

The scientific basis of prediction of climate change must be improved. This is important in order to guide adaptation to human-induced climate change and also to assess the requirement for stabilisation scenarios and their effectiveness. Improvements should derive from advances in modelling and observations. The emphasis should be on quantifying uncertainty through sensitivity studies, and on reducing uncertainty with improved representation of physical processes.

Europe is currently leading the way in the prediction of climate change. This is being continued through modelling programmes at a national level, with support from the EU, such as SIDDACLICH or SINDICATE, and through specific national programmes. Results from different groups have been compared in model intercomparisons, in an attempt to understand differences that were found, for example, in cloud feedbacks. Europe has also considerable expertise in modelling the conversion of emissions to concentrations and in integrated assessment. Assessment studies are often made with simplified climate models. These will need to be revised as revised climate scenarios become available from the more realistic models. Coordination of work in this area is undertaken by WCRP through CLIVAR, in collaboration with supporting programmes (GEWEX, ACSYS, WGNE etc) and by IGBP (IGAC, GAIM etc).

Work to date has found indications for a human effect on global climate. Future human effects may be substantial, though there is considerable uncertainty in the magnitude and distribution of changes. In arriving at these conclusions various shortcuts have been made by making simplifying assumptions about the emission scenarios used, the conversion of emissions to concentrations, and the derivation of radiative forcing given changes in concentrations. The possibility of surprises has been raised, but not been investigated systematically.

The uncertainty due to climate feedbacks (especially those associated with cloud and water vapour, see next section, for more details) needs to be quantified and understood through sensitivity experiments, and where possible reduced by reference observational and field studies. This work has tended not to attract funding in the past as it involves improving existing models which can be time consuming, and cannot be guaranteed to produce short-term gains. As a result, the uncertainty in climate sensitivity has not been reliably quantified, let alone reduced in the last two decades. From the factor of three uncertainty often quoted, this remains the largest uncertainty for climate predictions. Europe is in a particularly good position to address this problem given its expertise in modelling, observations and field studies. The large uncertainty is such that some diversity in modelling is desirable. Focused intercomparisons of simulations from a small number of models (and comparisons with real data and smaller-scale models) are the most promising way forward - and an approach which Europe has already started and is well placed to exploit. Some groups are beginning to include biological and chemical processes into models. Crude sensitivity studies indicate that these may have considerable effect. Further studies may show that uncertainties have been underestimated, and can only be reduced by including these processes.

The effect of aerosols (direct and indirect, all varieties) is not yet well known. There is also uncertainty in other aspects of tropospheric chemistry, particularly ozone. It has been shown that European climate is sensitive to aerosol forcing, especially over the next few decades. Given our rudimentary understanding of these forcing agents, there is a need first to quantify the range of uncertainty and then reduce it. In addition, estimates of future aerosol emissions, especially for sulphur, are not well established. It is expected that the next set of IPCC sulphur emission scenarios will be much lower than those issued in 1992.

We are also almost at the stage where coupled ocean/atmosphere models can be run for extended periods without artificial flux adjustments. This will allow a more realistic assessment of long term effects, including changes in sea level and possible collapse of the North Atlantic deep circulation. They will also allow a more realistic assessment of "surprises". However, the simulation of such "surprises" is likely to be model dependent, so some diversity of models is desirable.

Many of the feedbacks and new factors being considered are still very uncertain, it is unlikely that this work will lead to a decrease in the range of uncertainties over the next two or three years. Uncertainties will be assessed in detail in the IPCC 2000 report, for which drafts will be available in late 1999. This will provide an opportunity to produce a well-defined EU programme to reduce uncertainties in estimation of human-induced climate change and the benefits and effectiveness of emissions policies to limit undesirable climate change. Hence, it does not make sense to produce a detailed work plan at present. However, one can list the main areas where further research is needed and likely to prove beneficial. Work will be needed to

• Advance understanding of physical processes which govern the climate response in order to reduce the current uncertainty in climate response.
• Advance understanding of the chemical processes in the atmosphere (with IGBP) in order to obtain improved simulations of the concentrations of radiatively active gases and aerosols, the resulting radiative forcing and hence of the response to human emissions.
• Develop more comprehensive climate models, including a dynamical biosphere and cryosphere for example.

In addition, there is a need for

• More realistic emission scenarios, especially for sulphur. More carefully considered stabilisation scenarios will need to be assessed in climate simulations in order to inform and defend EU emissions policies into the next millennium.

Clouds have intricate feedback effects on the climate system. These feedbacks are responsible for a large part of the uncertainty affecting evaluations of future climate change. This has significant implications for global and especially regional predictions of climate change. Within cloud research there are different groups, each with there own expertise, for example: General Circulation Models (GCMs) modellers, cloud scale modellers and those involved in making and interpreting observations of cloud properties on a range of spatial scales. The shared expertise of these communities should be focused on improving cloud/climate feedbacks in GCMs. A long-term objective is to identify those cloud processes that are the most important in the determination of climate sensitivity in order to reduce the uncertainties in GCM predictions of climate change.

European scientists have expertise in many areas of the clouds and climate issue. Specific examples are: aircraft and surface observations of clouds (e.g., groups at the Meteorological Research Flight (MRF) based at the UKMO and UMIST); manipulation of satellite data (e.g., LMD and GKSS); cloud process studies (e.g., JCMM, UKMO, Meteo France and the University of Lille) through to GCM modelling (Hadley Centre, ECMWF, LMD, MPIM and other places).

There have been several large international model intercomparison projects, such as AMIP and FANGIO, which have been useful in diagnosing the range of cloud feedback in a large number of model results and some more detailed comparison exercises between a limited number of models (e.g., as part of the EU funded project on "Cloud feedback and validation", and the intercomparison of 2x CO₂ equilibrium experiments under the WGCM). These will help to understand how changes in detailed aspects of the cloud parametrisations may affect the model representation of the present climate and its sensitivity to external forcing. However, confidence in model parametrisations can only be achieved through validation of the physical processes and fundamental properties of the model simulations using observational data. Projects such as the GEWEX Cloud System Study (GCSS) and EU funded EUCREM are primarily aimed at comparisons of cloud scale models with data on individual clouds such as is obtained from aircraft measurements observations. The observational programme ACE-2 utilises combinations of in-situ and remotely sensed data to better describe in-cloud relationships of microphysics and cloud morphology with optical properties. However, direct comparison of GCM model simulations of clouds on typically 200 km sized grids with such small-scale data is difficult. More directly comparable are satellite measurements of cloud properties and radiative fluxes. Several climatologies of such measurements (e.g. ERBE, SSMI and ISCCP) are now widely used in the GCM community. Such data has proved invaluable for model validation, however attention needs to be focused on optimising the overlap of such programmes with each other and with periods of climatological interest. In addition, there is little or no data on some cloud properties (e.g. ice water paths) that can be usefully compared to GCM simulations.

The following general recommendations are made

• More attention should be given to those cloud processes that have an important influence on climate sensitivity.
• All useful data for model validation should be made available. Modellers should be made more aware of the existence of these data. Expertise with comparing observations with model output should be shared within the modelling community and experience gained by observationalists in overcoming problems such as sampling should be learned by the modelling community.
• Future observational programmes should be developed in-line with priorities in climate research. They should target measurements of key variables.
• The connection between GEWEX and CLIVAR cloud research should be strengthened

In addition, we make more specific recommendations in a number of areas.

Mechanisms of cloud feedbacks:
Modelling and observational groups should put much more emphasis on studies to isolate specific cloud processes and mechanisms of cloud feedback.

• "Synoptic type" or "cloud system" composites should be used to aid observational-data-to-model comparisons.
• It is important to retain temporal and spatial information in models and observations, because this allows diagnosis of processes and relationships between variables. Information is often averaged (both spatially and temporally) for comparisons, but a lot of the complexity of cloud systems is lost in this approach making a correct interpretation of the comparison difficult.
• GCM experiments should be designed to produce diagnostics and relationships between variables that are directly parallel to observables (for example the variation of cloud optical thickness with temperature).
• Similarly, diagnoses of observational data should aim to produce hypotheses which are testable in models (another example is the variation of cloud water path with droplet size).
• In trying to overcome the difficulty in model to model comparisons, the experience gained in observational data to data comparisons should be taken into account. This includes how to account for sampling effects and examining distribution statistics.
• Cloud radiative feedbacks cannot be understood by considering top of the atmosphere fluxes alone. Cloud changes affect the radiative heating of the atmosphere and surface and hence change the atmospheric circulation leading to further changes in cloud. Each part of this relational loop must be considered.
• Studies of the mechanisms of cloud feedback will require 1. an understanding of model sensitivities and modelling needs; and 2. an understanding of what can be observed and what is required from observations. As few people currently have both skills, there is a need to focus education and training to develop such cross-expertise.

Integrated cloud studies
Greater use should be made of the hierarchy of model and observations to integrate cloud studies across the range of scales. Model scales range from GCMs through Single Column Models (SCMs) to Cloud Resolving Models (CRMs). Observations vary from in-situ aircraft measurements for individual clouds through lidar and radar measurements of synoptic systems to global satellite measurements at around the GCM grid scale. For the purposes of parametrisation development, model validation and testing sensitivities and ultimately linking detailed observations on the cloud scale to cloud parametrisations on the GCM scale, a hierarchical approach to both modelling and observations is needed. For example:

• Aircraft in-situ measurements of cloud microphysics are needed to verify remote sensing measurements, such as dual wavelength radar observations of size and concentration of ice cloud particles.
• Cloud Resolving Models (CRMs) with explicit representation of microphysical processes may be verified using both in-situ aircraft data and remote sensing measurements (for example, combinations of radar, lidar and passive radiometry).
• CRMs can be used to test parametrisations in SCMs and ultimately in GCMs. For example, testing mixed phase parametrisation via cloud temperature could involve comparisons of in-situ observations, CRMs, and GCMs. Existing Programmes such as GCSS and EUCREM compare detailed observations with CRMs and SCMs for particular case studies. However this could be expanded to use CRMs to explore the effect of changes in dynamical forcing outside of the current observational parameter space.
• Testing of cloud parameters and processes in models is needed on as many temporal and spatial scales as possible, for example, to understand the association of cloud systems with particular large-scale dynamical conditions.

Intercomparisons
It is important that the research community reaches agreement about research strategies for observation-to-model comparisons and for model-to-model comparisons. As an important prerequisite for such comparisons agreement should be reached about certain basic definitions, such as the definition of cloud feedback.

Reanalysis
Reanalysis groups are urged to pay more attention to cloud and hydrological data. Reanalysis groups should establish better connections with groups such as GEWEX and CLIVAR, working on hydrological issues, to ensure that there will be as much attention for the correct production of cloud and related hydrological diagnostics in new reanalysis projects as for the more basic model variables. Results from existing projects would suggest that this is not the case at the moment.

The central objective here is an improvement of the scientific basis of detection and attribution of climate change and improvement of uncertainty estimates, taking advantage of advances in modelling and observations. This will require an assessment of credibility of physical processes as well as advanced statistical tests.

A successful detection and attribution research programme requires solid grounding from other areas of climate research. In particular it needs continuous, high quality, homogeneous observations of climate over at least the last 50 years, accurate estimates of forcing changes over this period, good simulations of the response to these changes, and good knowledge of internal climate variability, which needs to be separated from the forced changes. Homogeneous observations of climate greater than 50 years would allow better comparison of model simulated variability with observed variability.

Europe is currently leading the way in the detection and attribution of climate change. This is being continued through modelling programmes at a national level, with support from the EU (e.g., SIDDACLICH) and specific detection and attribution programmes (e.g., QUARCC). Current work indicates that uncertainties in forcing and climate sensitivity (showing up in model-model differences) confuse the attribution issue.

Several centres have now made estimates of the contribution of individual greenhouse gases, of a variety of aerosols (including indirect aerosol effects) and of natural forcings (due to changes in solar output and volcanoes). At this stage, it is difficult to predict precisely what the outcome of including these new factors will be. Many of the new factors being considered are very uncertain. Nevertheless, it is likely that this work will lead to an improved estimate of uncertainties (which may well be larger than previous estimates). The following are key scientific questions:

• Does the balance of evidence still support a discernible human influence on climate?
• If so, can we quantify the anthropogenic contribution to observed climate change?
• What does successfully detecting climate change and attributing it to anthropogenic effects tell us about future climate change?
• What are the main sources of uncertainty in the detection and attribution of climate change? What can be done to reduce these uncertainties and in what order of priority should it be done?

To obtain answers to these questions a comprehensive set of recommendations has been formulated.

First, we make the following general recommendations

• The uncertainties in observations, radiative forcing and model response need to be quantified and methodologies developed to take these uncertainties into account in future detection and attribution studies.
• The continuation of current observing programmes is essential in order to build up sufficiently long time series for detection and attribution of climate change. This must involve both conventional and remote observations. New observations must be made in a manner that preserves the temporal homogeneity of existing records.
• More good quality historical data must be amassed and research continued into their correction and homogenisation. Examples of such data are the archives of temperature observations neither yet digitised nor corrected for instrumental bias. This will require collection of historical metadata describing the observing systems.
• The main modes of internal variability in models should be compared and assessed against observations.
• The indirect aerosol effect, of which the largest contribution is probably from sulphates, gives rise to the biggest uncertainty in estimates of historical forcing. Reducing this uncertainty is a main priority for future research.
• Detection studies should be made using simulations of 20th century climate. In view of the small signal to noise ratio of changes over this period, ensembles of model simulations are required. This will require greatly increased computing capacity.

More specifically we make the following recommendations in the different areas:

• The GCOS recommendations for the current observing network should be implemented.
• Future satellite missions should overlap for at least one year, preferably two, and be maintained in stabilised orbiting environments.
• A continuous space-borne measurement of solar irradiance needs to be maintained. Good knowledge of solar irradiance changes over decadal time scales will require improved radiometer capabilities.
• "New" measurements (e.g. aircraft, sub-surface ocean, acoustic tomography, GPS occultation, and humidity data) should be added to the climate observing system with the aim of using them for detection and attribution studies. Note that metadata from these systems should be collected and archived.
• The measurement of the atmospheric sulphate burden needs be improved in order to allow adequate validation of sulphur cycle models.
• Radiation transfer codes should be validated against observations as well as laboratory data.

• Records of the time histories of the factors affecting climate and the associated radiative forcings should be improved with the aim of using them to simulate the 20th century climate.
• A set of recommended forcings should be developed for the period 1900 to 2000. This should include concentrations of well-mixed greenhouse gases (CO₂, CH₄, CFCs), emissions of SO₂ (to be used by models containing a sulphur cycle), concentrations of tropospheric ozone and the SPARC recommendations for solar irradiance changes (as a function of wavenumber), possible volcanic aerosol and stratospheric ozone concentrations.
• The reasons for jumps in reanalysis time series, and for differences between reanalyses, should be understood. A reanalysis intercomparison project, in which each reanalysis was performed with identical SSTs, corrected radiosonde and satellite data should be carried out.
• The reasons for the poor comparison between many proxy data sources and temperature measurements should be explored with a view to improving the proxy data. The development and expansion of existing datasets of tree-ring density and tree-ring width should be continued and these data made easily available to the scientific community in order to allow validation of model variability.

• The physical reasons for the modelled response need to be understood and verified against observations. Possible techniques are sensitivity studies, inter-model comparison and validation against observations including volcanic eruptions.
• The differences in GCM "signal" estimates should be assessed and their effects on detection and attribution studies examined.
• The understanding of current indirect aerosol forcing and its representation in models should be improved to permit representation of indirect forcing over the last century.
• A CMIP subproject should compare and assess the main modes of internal variability in models. The regime behaviour in models should be compared and assessed against observations. The datasets from the AMIP II and CMIP projects may allow this.
• The vertical coherence of temperature and humidity in models should be compared with those observed, in order to assess the models' simulation of water vapour and water vapour feedback. The data sets from AMIP II may allow this.
• Well-mixed greenhouse gases should be represented explicitly rather than as equivalent CO₂.
• The response of stratospheric ozone to solar and volcanic influences should be explored and documented, and the subsequent influence on climate assessed.
• The need to model, within GCMs, the chemical processes that determine concentrations of tropospheric, and possibly stratospheric, ozone, rather than use off-line calculations, should be explored.
• Radiation schemes used in GCMs should be improved so that the radiative impact of thin aerosol clouds such as those due to volcanic aerosols is accurately computed.
• Mechanisms, other than total solar irradiance changes, for possible solar/climate relationships should be examined.

• Algorithms should be developed and used that take account of the presence of noise in signals. Methodologies should be developed that take account of observational, model and forcing error.
• Detection and attribution methodologies should be tested in systems, which are simple enough to understand what is happening. Encouragement should be given to the statistical community to become involved in detection and attribution studies.

• The signals used in detection and attribution studies should be computed from an ensemble of simulations of the 20th century. Ensemble sizes should be at least four. To carry out this requires a large increase in computing power.
• Detection and attribution studies should be carried out, bearing in mind the uncertainties in present forcings and model responses. Research aimed at improving model and forcings should be carried out in parallel with these studies. The utility of multivariable studies should be explored.
• Model uncertainty in fingerprint detection results needs to be explicitly addressed, for example, by using different model's fingerprints in parallel, searching for fingerprints from one model simulation in the transient simulation of a different model.
• Other variables, apart from mean temperature, should be used in detection and attribution studies. Possibilities include diurnal temperature range, sea level changes and perhaps atmospheric circulation changes. "Perfect model" studies should also be carried out to establish variables, on the basis of signal-to-noise ratio, that could be considered for detection studies.
• To test the sensitivity of existing results, detection studies using restricted datasets (e.g., land-only) and with the global mean removed should be carried out to allow independent estimates and identify what is critical for detection studies.
• Detection studies using an AGCM forced with historical SSTs should be accompanied by studies using a coupled model. The forcings and AGCM used should be the same as those used in the coupled model. This will allow intercomparison and interpretation of results. The veracity of the atmospheric model's variability needs to be carefully assessed on the time scales relevant to detection and attribution of climate change.
• Results of detection and attribution studies must be presented in a form intelligible to policy makers, by simplifying results from fundamental basic science.

Chapter 7: Climate Observations

Chapter 8: Climate Modelling

Annexes