Annex 6. RISK MANAGEMENT FOR DROUGHT

INTRODUCTION

The subject of this annex is the rational and efficient management of drought risk. This is implied, but apparently not worked out in any detail, in SADC’s ‘new definition of drought’, which was explained in the 1999 Donsi report thus:

“...previous definitions of drought ... tended to include what otherwise constituted conditions of normal aridity and thus unnecessarily exaggerate the level of resources for managing droughts. In addition, loose definitions of drought often resulted in hesitation to declare national disasters. ... The newly developed policies, therefore, endeavour to spell out scientific criteria for estimating the extent and severity of drought and for defining extreme rainfall conditions that constitute a “disaster drought”. A key objective of the policies is to distinguish drought conditions that are so intense and protracted that they are beyond the scope of normal risk management practices of rural communities and justify State intervention.”^[3]

This concept emphasizes that in many places moderately dry conditions, resulting in water stress in plants and hence low yields, are normal occurrences.^[4] As such they represent a risk that is best covered by the routine coping strategies of individual farm families, built into their farming systems. Rarer and more severe drought events, by contrast, are best dealt with by some form of communal or shared risk management. In both these statements the word ‘best’ can be understood in a rigorous economic sense: in the long run and over a whole nation or region, the interests of the people as a whole will be worse served if the risk management is done at the wrong level.

The rarer the event, the wider is the appropriate risk sharing. A mild drought such as occurs once in three to five years should be covered by a farmer’s own coping methods. A slightly more severe one requires communal or shared management over a considerable number of farmers, whether by their participation in an insurance scheme or by government action. A more severe drought would need the declaration of a state of ‘emergency’ or ‘disaster’ over part or the whole of a country, and the mobilisation of that country’s financial and physical reserves, perhaps with some foreign help if the reserves are not sufficient. A very rare and severe event would require and justify risk sharing over the whole human race, implemented through international relief operations.

One of the most important reasons for making clear distinctions about drought frequency and risk management is that the failure to do so distorts incentives. This may happen when local markets are distorted by the arrival of large quantities of donated food. Or it may happen when farmers make the explicit or subconscious calculation: “if I farm prudently, so as to cover the risk of moderate droughts myself, I will have my own food security but my average production and income will be lower than they might be. If on the other hand I go for high-income but drought-vulnerable crops and methods, my average income will be higher, and when my system fails in a moderately dry year someone else will rescue me from the consequences”. In effect, if people expect that wider society (government, aid agencies, NGOs) will step in even in moderate droughts, that expectation will encourage irresponsible ‘gambling’ tactics on the part of individual farmers, which will not only be risky for farm families but will also be suboptimal in a rigorous economic sense.

Of course, if a farmer is rich enough and prudent enough, he can himself spread his risk over several years, and then it may be appropriate for him to choose an inherently drought-sensitive farming system. He may be helped in spreading his risk over several years by his banker or an insurer. This scenario is practicable and sensible for large commercial farms, and is used by many of them. But it is not available to the majority of poorer people and subsistence or small farmers.

A rational drought risk management system evidently requires a probabilistic classification of droughts, which is the subject of the next section. The ways in which it can be used, to guide different sorts of risk management, are then discussed in Section 3.

PROBABILISTIC CLASSIFICATION OF AGRICULTURAL DROUGHTS

This section discusses how drought events may be classified according to their severity for agricultural purposes. It is intended to explain the concepts to readers who may not be familiar with frequency and probability analysis as used by hydrologists and others.

The literature contains references to a four-way classification of drought, ascribed to Wilhite and Glantz (1985).^[5] The four categories mix quantification of the drought (‘meteorological drought’ refers to rainfall and probabilities) and the impacts of drought (the other three categories, namely agricultural, hydrological, and socio-economic drought, describe impacts on agriculture, water resources, and human activities respectively). In the present context the important aspect is agricultural drought caused by meteorological drought, but the need is to categorise droughts by severity and frequency rather than just by their nature.

Meteorological droughts can quite easily be classified by occurrence frequency, ie probabilistically, by computing the exceedence probabilities of seasonal rainfall totals or other simple statistics.^[6] Climate change introduces an additional complication. If we describe a drought by saying, for instance (as Wilhite and Glantz do in their ‘meteorological drought’), that a certain kind of drought occurs when rainfall in a certain period is less than 70 percent of ‘normal precipitation’, then a reduction of ‘normal precipitation’ by climate change will reduce the number of mm of rain that marks the drought threshold. If on the other hand we define drought by the number of mm of rain in the period, then climate change will alter the occurrence frequency of that drought, and also alter the percentage of ‘normal precipitation’ that it represents.

The probabilistic classification of agricultural droughts is more complex because it requires a measure of the severity of the agricultural impact. The severity of a particular agricultural drought event depends on many things. Partly it depends on what hydrologists call ‘antecedent conditions’, but interpreted broadly enough to include the degree of vulnerability of farming systems, and many other kinds of preparedness. It also depends on precisely what impact is involved, which among other things will determine what period’s rainfall matters. One historical season may be critical for some crops and some people because of a seven-week dry period in December-January, but be no trouble at all for another farming system because that one has seasonal water storage and therefore reacts only to much longer dry periods. For the latter purpose the dry period might not matter much, for instance because it rained a lot in November and February, while another year, with moderately dry conditions from October to April, was critical. For yet another purpose, where over-year storage is involved, neither of those events may be as bad as a sequence of five below-average-rainfall years in quite a different decade. There are plenty of examples of this, but apparently little analysis in the region.^[7]

The critical time-scale of the drought depends on how water is stored between the moment it falls out of the sky and the moment it is taken up by crop roots, and on how much water the storage can hold. In many situations the main mode of storage is soil moisture in the root zone (or close enough to it for capillary rise), and this may be able to bridge a dry spell, within the growing season, of a few weeks. Another situation, where an irrigation system uses water from a small dam and reservoir, may be able to bridge a slightly longer dry spell, perhaps three or four months, or may be able to extend the growing season by several weeks after the effective end of rains, but will not be able to carry significant volumes of water over to the next year. The annual volume lost to evaporation and seepage from such a reservoir may be of the same order as the storage volume, especially when the average water depth is less than about 2m, which means that storing water for longer than about four months is inefficient, most of it being lost. This is what is here called seasonal storage. Over-year storage, by contrast, is achieved by a large dam whose reservoir has an average storage depth several times larger than the annual evaporation, and a live storage volume considerably larger than its mean annual inflow. Such a reservoir will easily bridge a single dry year, even a severe one, provided that the year before and the year after have good rainfall: its critical condition would be a sequence of three or more dry years, even if only moderately dry ones.

The simplest way of using a probabilistic classification of agricultural droughts is to label certain sorts of drought in ways that non-specialists can understand. For the purposes of this report only, we label four sorts of drought, two for cases where one growing season is relevant, and two for those where (because there is some over-year water storage) a sequence of dry years is critical. Our tentative label definitions are:

• A “normal dry year” is one cropping season which, for the particular farming system under consideration, is likely to be exceeded in impact severity about once in five years, or twenty times in a century (20 percent probability).

• An “extreme dry year” is one cropping season which, for the particular farming system under consideration, is likely to be exceeded in impact severity only about once in twenty years, or five times in a century (5 percent probability).

• A “normal dry sequence” is a series of two or three years which, for the particular farming system under consideration, is likely to be exceeded in impact severity about twenty times in a century (20 percent probability).

• An “extreme dry sequence” is a series of two or three years which, for the particular farming system under consideration, is likely to be exceeded in impact severity about five times in a century (5 percent probability).

Here it is deliberately emphasised that any probabilistic definition or labelling system for agricultural droughts must be specific to a particular farming system, or even to a particular crop. The steps involved in defining such a labelling system, and using it to label certain sorts of drought, might be:

1. Decide what parameters (up to about three of them) are critical for that farming system, eg:

• for a rainfed one, rainfall in a few key periods (dekads)^[8], and perhaps date of first significant rain of the usual growing season.

• For an irrigated one, rainfall over a longer period including some months or years before the growing season, the length of time depending on how much water storage the farming system has.

• For a livestock-dominated one, some period taking into account the grazing’s annual cycle and how much soil moisture storage it mobilises (over periods of a few months at least)

2. Formulate a single scalar ‘impact severity index’ using those parameters (this would always be difficult and subjective, and would need a thorough understanding of the farming system and its constraints). Alternatively, find an index that is already defined and is close enough to the needs of this farming system.

3. Compute the parameters, and hence the severity index values, for the last 20 to 40 years in a particular place (this would need a lot of data; for an already-defined index it may already have been done).

4. Assign each of those historical seasons a rank order by ‘impact severity’, and deduce a frequency distribution of the index for the historical period.

5. By adjusting if necessary for climate change or any other sort of known or predicted change, deduce a probability distribution, in terms of the defined impact severity index. This will be valid only for the particular farming system and place.

If one needs to describe agricultural seasons in simple verbal terms such as ‘normal dry year’, there are two more steps:

6. Choose one or two seasons typical of each labelled probability level.

7. Generalise these into descriptions of a typical ‘normal dry year’ and a typical ‘extreme dry year’, or whatever labels are used, for this farming system and this place only. For purposes of communication a verbal description can refer to particular recent seasons that people remember, to help them to visualise the ‘normal dry year’ etc ... for example “ a year slightly worse than 1997-98”. A description must state very clearly what farming system and place it refers to, and should warn people not to use it outside those limits.

RISK MANAGEMENT, RISK SHARING, AND THE JUSTIFICATION OF INVESTMENTS

The principles and objectives of rational risk management have been explained in Section R.1 above. Once a system of classifying agricultural seasons in terms of their drought severity for particular farming systems has been established, it can be used in various ways. Some of them are described below, starting with insurance because it is the clearest case of risk sharing, ie risk pooling and spreading.

a. Insurance

If a number of farmers all grow the same crop, they can take out insurance policies with a willing insurer in order to share their risk of low crop production due to drought (or hail, or wind, or any other natural phenomenon amenable to probability analysis). Risk is pooled over many farmers and spread over many years. The benefit to each farmer is that the farmer can go for more risky crops or farming practices that give high average income, while being protected from the consequences of occasional bad seasons. The benefit for the insurer is that, if the premiums are set right, the premium income will be higher than the aggregate cost of risk, and the difference is the insurer’s reward.^[9] For people with the necessary skills, resources and opportunities, insurance is an efficient form of risk management. It is widely used by farmers in the United States of America, for instance, and also by some commercial farmers in southern Africa. The implementation requires an accepted way of determining the severity of a particular season for the particular farming system that the farmers are using. When a growing season is determined, by this rigorous and objective procedure, to have reached an agreed threshold level of drought severity, the insurer pays out to the farmers. A particular farmer may have planted a more drought-resistant variety that year, or been lucky with the local rainfall, so that the farmer's crop actually did relatively well, in which case the insurance payout may be more than the income reduction from the crop. In another year his loss may be more than the insurance payout. So the farmer still faces a degree of year-to-year uncertainty, but it is favourably balanced for the farmer, and protects the farmer from extreme losses. The farmer therefore considers it worth-while to use the insurance scheme, even though the premium is higher than the cost of risk so that he is contributing to the income of an insurer who does not grow or produce any physical commodity. This whole set-up, involving a number of farmers and an insurer, maximises the overall income (not perfectly, but to a useful degree), and is thus economically justified and in the long-term interest of the whole society and nation.

b. National risk-sharing

A government can arrange the sharing of many people’s risks, and the spreading of risk over many years, in effect taking on the role of insurer. It can do this with an explicit “national insurance” scheme, related to specified risks and to which citizens have to pay contributions, or it can do it out of general revenue. In the latter case taxpayers are effectively paying insurance premiums, but lumped in with general taxation so that they do not know how much they are paying for the management of specific risks. When the risk-covering service is financed out of general revenue, it is not necessarily paid for by its beneficiaries, or anyway not in proportion to their benefit. The cost may be borne disproportionately by the richer citizens, if that is the government’s taxation policy. This spreads the drought risk over the whole community, not just the farmers, which may be in the national interest.

A drought risk management scheme of this type is implemented through a procedure whereby, when a drought reaches a predetermined threshold measured by an agreed severity index, an explicit state of drought is formally declared (for particular regions or crops as appropriate), and certain actions automatically follow. Having such an automatic procedure can avoid delay and ad-hoc decisions, which can make any subsequent measures like food aid, emergency measures, or special income-generating schemes quicker, more effective and more efficient (cost-effective). There can be several levels of formally declared drought, with different consequences.

c. Global or international risk-sharing

This is similar to national risk-sharing except that the community over which the risk is pooled and spread is larger, and the cost may be unequally borne, for instance rich nations effectively bearing much of the drought risk of poorer nations.

In any of these cases, whatever the group of people within which risk is to be shared, the following are needed for the design of any risk-sharing system or process:

• a way of classifying droughts for a particular purpose, such as a particular farming system (this has been described in the previous section);

• a way of fixing the probability thresholds which will trigger the automatic consequences.

This second requirement is crucial to the usefulness of the system. The economic principles underlying the design of an optimal risk management set-up, or even an acceptable suboptimal one, include:

• the aggregate of cost of risk (consequence multiplied by probability over a complete probability spectrum) should be minimised for a whole nation or other large social group;

• the various actors must face incentives that lead them to act in a way that maximises overall welfare.

If these conditions are met, even approximately, it should be possible in principle to provide a rigorous economic justification for whatever investments are needed to set up the risk-sharing system. Such an analysis would need to show that national net benefit is higher with the scheme than without it.

Even if no economic analysis is needed, the point about incentives is valuable for the design of a system. One of the more obvious signs of inappropriate risk management is that some actors face incentives to act against the common interest, for example engaging in risky farming practices at the expense of a government which, in hard times, will rescue them from the natural consequences of their actions. The other extreme is seen when, because there is no communal or government-organised risk-sharing, everyone takes a very cautious and risk-averse approach, leading to lower overall production and welfare. A well-designed risk-sharing system gives each actor an incentive to choose an appropriate (strictly speaking, an economically optimal) level of drought-vulnerability.

PRACTICAL IMPLICATIONS FOR RWCISA

The setting up in the ten countries, individually or collectively, of good risk-sharing arrangements is one of the sets of potential interventions that could form part of the Water Control Initiative.

It is not a simple undertaking; the design and then the frequency analysis of impact severity indices for individual farming systems is a massive task. The design of appropriate schemes must therefore be one of RWCISA’s projects, not a part of its preparation. There are already some schemes in place in some countries, both of the insurance type (for example, for commercial maize farmers in South Africa) and of the national type (drought management policies). If there is to be an intervention of this type, it must begin by documenting these existing schemes, and others in other regions of the world, and learning from them. But there may be much to be gained by taking a systematic approach, based on sound risk-sharing principles and including some degree of economic optimisation.