Causes of variability in agricultural statistics
Figure 1 provides a theoretical illustration of some of the factors which affect the variability of agricultural yields (heavy read line). They can be roughly grouped in three categories according to their variations over time:

• growing smoothly, such as more or less regular technology and management trends (i.e. mechanization, varieties, irrigation, and the farmers know-how), but growing more abruptly for several years in succession in the case of innovations. A typical innovation would be the new introduction of irrigation;
• discontinuous, like extreme factors of various origins and policy decisions which affect management (for instance, farmers may decide to use less fertilizer if it is no longer subsidized) or infrastructure, such as the construction of a road which provides access to new markets;
• pseudo-cyclic, like weather

The result may be a complex curve where it is often rather difficult to separate weather from other factors, particularly if weather itself is affected by a trend... In addition, the figure assumes that the function linking yield and the factors is known (a simple addition in this case).

Figure 1: Some of the factors affecting agricultural yields (represented by a heavy red line: technology and management trends, innovation, policy, extreme factors and weather)

Although it is generally agreed that, under normal circumstances (see note 1), weather variability remains one of the main factors behind the inter-annual variability of agricultural production, it is very difficult to estimate how much production is lost due to variability.

This is due to several factors, including the following

• the fact that agricultural statistics are reported by administrative areas, which are rarely homogeneous from an agricultural and agro-ecological (including climatic) point of view. The area-wide average therefore cancels local variability;
• weather can impact agricultural production systems at many levels (production, harvest, storage, transportation...) both directly and indirectly through diseases, pests, damage to infrastructure, etc. This subject has received a lot of attention and it is not necessary to repeat it here.
• not only are agricultural statistics spatially aggregated values, but some practices of national services, like for instance the reporting of harvested rather than planted areas also tend to depress the effect of weather variability.

We simply underline that climate variability impacts agricultural output (production) through its effects on yields and areas planted. Yields are affected, as indicated, by weather as the main "random" factor, but also by mostly continuous technological trends (including new varieties and management), innovations (including management innovations), agricultural policies (mostly national policies) and extreme factors of various origins.

Areas depend more on economic factors; planted areas vary according to labour availability, level of mechanization and expected return (prices). Areas harvested are often strongly linked to environmental conditions, including poor weather during the cycle, damage to infrastructure due to extreme conditions, or simply very low yields for which it is not economical to harvest at all.

Quantification of loss due to variability, and how much is due to climate?

An attempt can be made to estimate how much production is lost currently because of the variability of climate. The following methodology was followed:

• take a national production time series
• for each year Y, take the maximum production value Pm in the 7-year interval from Y-3 to Y+3
• compute the difference between the production P of year Y and Pm, and express it as a percentage "loss": (Pm-P)/Pm*100%.

The approach assumes that no marked technological progress took place in the seven year period. An example for Thailand is shown in Figure 2: the "loss" varies between 0 and about 25% and shows a slight downward trend probably due to stagnating productions since 1980.

Figure 2: Total cereal production and production loss due to inter-annual variability in Thailand between 1961 and 1994 (based on FAO statistics). The 7-year moving interval used for defining the "maximum yield" becomes an asymmetrical interval at the end of the series, i.e. 1991 (1988-94) is the last complete interval, followed by 1989-94 for 1992 etc.

The same approach can be applied to areas harvested and to yields, leading to Figure 3. Logically, production undergoes the largest fluctuations. The comparison of areas and yields is interesting, as the two curves keep crossing, suggesting that, according to the years, areas planted (i.e. mainly socio-economic factors) or yields (see note 2) (i.e. mainly environmental factors) have played the main part. It can be assumed that a large fraction of the "losses" thus put into evidence are directly ascribable to weather.

Figure 3: Percent loss of cereal production, yield and areas in Thailand between 1961 and 1994 (based on FAO statistics)

Figure 4 indicates that, in the case of Niger, the crude above-mentioned National Rainfall index accounts for about 25% of the loss. This is clearly an underestimate, as rainfall was affected by a negative trend during much of the period under consideration. The second example below (wheat in Italy, Figure 5) shows a marked dependence of yield on radiation (solar energy) which accounts for about half the non-technology variability.

Figure 4: Percent "loss" of total cereal production in Niger as a function of National Rainfall Index (mm). R=0.63 for the figured regression line (heat capacity model)

Figure 5: De-trended national wheat yield (see note 3) in Italy (1951-86) as a function of estimated September-March global radiation, together with quadratic trend.

Although this is still only an assumption, it seems thus that at least 50% of the variability of agricultural production is due to weather, in developing and developed countries alike.

Table 1 and Table 2 illustrate the "average loss" (as defined above) over the period from 1964 to 1991 in several countries and groups of countries.

As noted above in the case of Thailand, production shows the highest average losses, while the relative importance of area and yield varies, with area being usually highest. It is also worth observing that some very high values occasionally occur in developing countries. At the global scale, values tend to be lower due to the averaging effects.

Regarding extreme factors, their impact can be shown to have marked effects at the national level only in some particularly disaster-prone countries, like Bangladesh (Figure 5) where the losses due to the extreme conditions can reach 2.5 million tons (in 1988, according to Hofer and Messerli (1997)), or about 15% of the total rice production (see note 4). In comparison, factors like hail, frost or fires are only of marginal significance.

Figure 6: Production of three rice typologies in Bangladesh. Boro is a dry season (winter) irrigated crop; aus is pre-monsoon crop (Apr.-Aug.) and Aman is a late-monsoon crop (June-Dec.). Extreme conditions are indicated as F for floods, D for drought, C for tropical cyclones and W for war. After Gommes, 1992.

To conclude...
In summary, it appears to be rather difficult, to quantify the impact of climate variability on agricultural production. In agreement with Palm and Dagnelie (1993) and Palm (1997), it may be concluded that, at regional scale, the largest fraction of inter-annual variability of crop yields and production can be ascribed technology and management in developed countries (see note 5). In Europe and for annual crops, non-weather factors account for more than 50% of the variance in 75% of the cases studied, and for more than 75% in 40% of the cases. Roughly 20% of the variability is due to other factors, of which at least half is weather dependent.

According to Oerke et al. (1994), production losses losses due to pests, diseases and weeds amount to 15%, 14% and 13 % on average for the main cereals and potatoes, in the absence of control measures. This refers to actual conditions. When compared with potential yields, the loss reduction is roughly 70 % about equally distributed between pests, diseases and weeds.

In developing countries, particularly in semi-arid areas, the fraction of variability due to weather is at least of the same order of magnitude, but because technology trends tend to be much less marked, the actual role of climate appears to be much larger, up to 100% in extreme cases of virtual complete crop loss due to generalized drought.

Notes
Note 1. Thus excluding war, major epidemics, etc.

Note 2. If reported areas correspond to harvested ones rather than to planted ones, the effect of weather variability on yields is artificially depressed.

Note 3. The detrended yield is the departure of yield values from the time-trend, assumed to take into account the technology and management component of yield. Like in many developed countries, radiation, not rainfall, is the main limiting factor to agricultural production. The time-trend accounts for about 80% of the total variation of yields, leaving about 20% to be accounted for by other factors.

Note 4. Hofer and Messerli see the erosion of river banks (loss of land) as one of the main problems in Bangladesh.

Note 5. For perennial crops, the trends account for between 25 and 75% of the variance in about half the countries studied by Palm and Dagnelie. 36 % show no trend at all.

References
FAO, 1989. Rainfed agriculture in Asia and the Pacific region. FAO Regional Office for Asia and the Pacific, Bangkok, 224 pp.

Gommes, R. 1992. Current climate and populations constraints on agriculture, chapter 4 (p. 67-86) in Kaiser and Drennen, 1993.. Cornell University, Ithaca (N.Y.), 8-9 Oct. 1992: St. Lucie Press, Delray Beach, Florida.

Gommes, R. 1997. Prévision agrométéorologique des rendements: quelques moyens et méthodes utilisés par la FAO dans un contexte de sécurité alimentaire. In: Tychon and Tonnard, (Eds), 1997, 145-175.

Hofer, T., and B. Messerli, 1997. Floods in Bangladesh. Report prepared for the Swiss Agency for Development and Cooperation by Inst. Geography, Univ. Bern. 32 pp.

Oerke, E.C., H.W. Dehne, F. Schoenbeck and A. Weber, 1994. Crop production and crop protection. Elsevier, Amsterdam.

Palm, R., 1997. Les modèles de prévision statistique: cas du modèle Eurostat-Agromet. In: Tychon and Tonnard, (Eds), 1997, 85-108.

Palm, R., and P. Dagnelie, 1993. Tendance générale et effets du climat dans la prevision des rendements agricoles des différents pays des C.E. Official Publications of the EU, EUR 15106, Luxembourg. 132 pp.

Tychon, B., and V. Tonnard, 1997. Estimation de la production agricole à une échelle régionale. Official Publications of the EU, EUR 17663, Luxembourg. 202 pp.