The world faces the dual challenges of feeding a burgeoning 21^st century population of perhaps 9 billion, while at the same time sustaining life support systems (Natl. Res. Council 1999). In recent decades, agricultural output outpaced human population growth and reduced famine. Nevertheless, in the near future the food supply must continue to expand, but must do so with reduced environmental consequences (Natl. Res. Council 1999). Innovative environmental information will be central to this expansion.

Agroecosystems play a central role in world food production and food security. Managed agroecosystems combine the complexity, multiplicity of scales, and feedbacks of biophysical interactions in natural ecosystems with the additional intricacies of human decision-making (Dalgaard et al 2003).

The overarching goal of this project is to understand and model the dynamic interactions of natural and human components in agroecosystems, with special emphasis on assessing the scope for active adaptive management in response to newly available knowledge (eg, climate information, insights on human decision-making) and ensuring sustainability of life support systems.

Agricultural stakeholders consistently rank climate variability among the top sources of risk to production or profits, and the top risk that can be reduced. Climate/crop interactions show marked non-linearities (eg, nonlinear growth responses to temperature and radiation), and threshold responses (increasing rainfall is generally favorable, but an excess may cause total loss of crops). Climate fluctuations occur on a range of temporal scales, from intra-seasonal to centennial and longer. We focus on two scales–seasonal-to-interannual and interdecadal– important for decision-making, resource management, and infrastructure planning.

On seasonal-to-interannual scales, the El Niño-Southern Oscillation (ENSO) phenomenon is the major single source of climate variability in many parts of the world (Trenberth & Stepaniak 2001). Advances in understanding and observations of the oceans and atmosphere have made it possible to predict with imperfect but usable skill ENSO-related sea surface temperature (SST) anomalies months in advance (Latif et al 1998; Goddard et al 2001). In turn, predicted SSTs and atmospheric general circulation models are used to predict regional seasonal mean precipitation and temperature (Mason et al 1999). Seasonal climate forecasts are being disseminated by several agencies around the world under the largely untested assertion that they are usable knowledge.

The emerging ability to forecast regional climate is a hallmark achievement of the first ten years of the U.S. Global Change Research Program (Stern & Easterling 1999) and creates an exciting natural laboratory to learn how an important and prevalent complex system such as agriculture may respond. Agricultural decision-makers can use seasonal climate forecasts to mitigate unwanted impacts and take advantage of favorable conditions (Hammer et al 2001).

Nevertheless, several studies have identified theoretical and practical obstacles to the use of climate information and forecasts (Pulwarty and Redmond 1997; Orlove & Tosteson 1999; Stern & Easterling 1999; Broad & Agrawala 2000; Glantz 2001; Broad et al 2002; Hartmann et al 2002; Lemos et al 2002; Patt & Gwata 2002). Some obstacles stem from limitations inherent to the climate system’s complexities: forecasts have coarse spatial and temporal resolution, not all relevant climate variables can be predicted, the skill of forecasts is not well characterized or understood, contradictory predictions may coexist. Other obstacles include procedural, institutional, and cognitive difficulties in receiving/understanding information, or in the ability/willingness of decision-makers to modify their actions. Successful use of seasonal climate forecasts in agriculture must be based on understanding these constraints and how to overcome them.

Inter-decadal climate variability can have dramatic consequences on agroecosystems. A marked increase since the 1970s in spring-summer precipitation in the Pampas of central-eastern Argentina, our study area, has contributed to significant changes in land use patterns. Continuous cropping has replaced agriculture-pasture rotations in many places. More dramatically, areas that were climatically marginal (only fit for grazing) have become 100% agricultural. In contrast, the period 1930-1960 was much drier in the Pampas: in the 1930s wire fences were buried by storms like those in the Dust Bowl. Similar inter-decadal shifts between dry and wet epochs have been reported for the US (Lettenmaier et al 1994; Kunkel & Changnon 2003; Mauget 2003). Our study on the Pampas will provide insights from another realization of such shifts, allowing comparisons grounded in common experiences.

Climate variability and adaptive responses must be assessed within the specific technological, economic, institutional, and land tenure contexts in which they take place. Argentine agriculture underwent major changes in the last decades, particularly since the early 1990s (Satorre 2001; Schnepf et al 2001):

Technological innovations such as short-cycle wheat varieties allowing a wheat/soybean double crop, no-tillage planting, and genetically modified varieties have played a major role in land-use changes in the Pampas (Senigagliesi et al 1997).

Economic drivers also favored agriculturalization. Most of Argentina’s agricultural production is exported; demand for animal protein especially in fast-growing economies-in-transition created a large market for grains. After elimination of taxes in the early 1990s, farmers received international commodity prices. This fostered investment in technology (larger farming machinery, greatly increased use of fertilizers and biocides). Despite re-introduction of crop export taxes in 2002, the devaluation of Argentine currency favored agriculture overall.

The creation of governmental and stakeholder institutions for agricultural research and extension enhanced dissemination of technologies and allowed agricultural expansion.

Finally, changes in land tenure regime, in turn, influence production systems. Half of the area currently planted in the Pampas is not owned by farmers exploiting it. Short leases (usually one year) provide incentives to maximize short-term profits via agriculture.

The intertwined effects of climate, economic and social drivers have large consequences on complex natural/human systems. One example to which we return throughout the text is the expansion of soybean in the Argentine Pampas.

Read more on the expansion of soybean in the Pampas...

Adaptive responses to climate and other risk factors require salient information to support decisions. Agricultural outcomes of decisions are more relevant to stakeholders than raw climate information: a farmer is more interested in receiving likely distributions of crop yields or economic returns than a seasonal precipitation forecast (see also Hammer et al 2001). A greater capacity is needed to convert raw climate information (seasonal forecasts, decadal climate projections) into distributions of outcomes for risk assessment and management.

To do so, outcomes of alternative decisions in agriculture can be simulated through process (crop growth) models that often require daily weather as input (Ritchie et al 1998). Historical daily series can be used, but often are short or unavailable. Alternatively, crop models could be driven with output from numerical ocean-atmosphere models used to simulate climate. Unfortunately, spatially coarse output from these models does not accurately represent conditions (particularly precipitation) at the scales of decisions (Easterling 1997). Models with higher spatial resolution still do not produce daily values with realistic temporal structure (Mearns & Giorgi 1999), whereas plant growth shows a highly nonlinear sensitivity to the arrangement of daily weather (eg, lengths of dry or wet spells). We are developing disaggregation tools to bridge the typically coarse spatial and temporal scales of climate scenarios and the smaller scales of process models (agronomic, hydrological) used to explore outcomes under diverse climate scenarios (Wilby et al 1998; Corte-Real et al 1999; Palutikof et al 2002).

Scenario-building and simulation-based research can provide useful insights. However, strategies to deal with climate risks also must be grounded on a firm understanding of human decision-making under uncertainty within the complex context of agricultural production. Actual use of climate information in agricultural decisions and the production decisions themselves will most likely deviate from frequently used prescriptions (eg, maximization of subjective expected utility, SEU).

Decision-makers can pursue different non-standard decision goals. For example, our work with farmers in the Pampas indicates that minimization of decision regret (Loomes & Sugden 1982) is a goal frequently observed. The anticipation of looking “foolish” makes many farmers reluctant to act on probabilistic forecasts of climate conditions that may not materialize, even if the expected value of such action is shown to be positive. Another common goal is aiming for satisfactory target levels of returns, rather than profit maximization, reflecting the desire for cognitive simplification (Simon 1956; Payne et al 1990). Adequate simulation and prediction of responses to uncertain climate scenarios requires realistic models of risky decision making and probabilistic information use closely linked to observed decision processes (Stewart 1997).

Decadal climate variability and other drivers have contributed to a large increase in Argentine agricultural output (crop production has doubled in the last two decades). Nevertheless, the environmental consequences and the sustainability of production and life support systems are receiving increasing attention. Some relevant issues include:

• Despite widespread no-tillage planting intensive cropping already is having consequences on the environment, such as erosion, and loss of nutrients and organic matter (Casas 1998);
• Production systems that have evolved partly in response to increased rainfall may not be viable if (as is entirely possible) climate reverts to a drier epoch;
• Soybean currently covers almost half of the total cropped area in Argentina. Such a system may be brittle to shocks or surprises such as large climate anomalies or price fluctuations.

Unavoidable tradeoffs between productivity, stability, and sustainability will need to be addressed for agroecosystems in the Pampas (Viglizzo & Roberto 1998). On one hand, the Argentine economy is enjoying the competitive advantages of soybean production: this crop is the country’s main export (AACREA 2003). On the other hand, are growing concerns about implications of the so-called “soybean monoculture.” A bill recently submitted to the Argentine Congress calls the soybean expansion “alarming” (Huergo 2003) and proposes disincentives. The growing tension between objectives offers a unique opportunity for salient, credible scientific knowledge to inform policy-making.

A distinctive component of our framework is the informational domain, as we argue that it plays a central role in inducing active adjustments and adaptive behaviors in complex systems. Attributes of the informational domain include not only procedural aspects such as salience, legitimacy, credibility, and access (Cash et al 2003), but also proximate characteristics such as compatibility (match between information and needs) and place.

Although there are other interactions between domains (externalities, lighter arrows in Fig. 1), we focus here on decision-making as the major process (from a stakeholder viewpoint) mediating between domains. This component involves common human limitations in information processing (Plous 1993; Nicholls 1999; Stewart 2000), individual characteristics along cognitive or affective dimensions (“personality” variables and risk attitudes, Weber 2001), and individual differences in goals for decisions, i.e., different objective functions. For adaptive responses, the decision-making process mediates and “filters” linkages (dark arrows) among domains: the result is a set of subjective perceptions of the values and likelihood of decision inputs and outcomes.