We explore the role of information in drought-related decisions in the Ebro basin. The Ebro Basin is the largest in Spain (85 600 km2) and is a highly regulated basin with 125 reservoirs (>1 Mm3) and a total storage capacity of approximately 8000 Mm3. These reservoirs are used primarily to supply water to more than 900 000 ha of irrigated agriculture and 360 hydro-electrical plants .

The larger irrigation districts are located in the north-east of the basin (Figure ). We have selected one of these, the irrigation district supplied by the Aragón and Cataluña channel (Canal de Aragón y Cataluña, CAyC), to examine the decisions made at the sub-basin scale.

Over 90 % of the water provided by CAyC is used for irrigation. The water it supplies is sourced from three reservoirs (Barasona, San Salvador and Santa Ana), and it is supplied to an irrigated area of around 98 000 ha. Two zones can be distinguished in the irrigated area: an upstream zone that can only be supplied from the Barasona reservoir as well as from the recently inaugurated San Salvador reservoir, and a downstream zone that can be supplied from all three reservoirs. These zones are 54 000 and 44 000 ha in size, respectively. The main crops grown in the area are fruit orchard (apple, pear, peach and nectarine) and extensive herbaceous crops, mainly maize, alfalfa and barley. The area cropped with wine vine surface is increasing, though it is still somewhat localised .

Three drought events that resulted in impacts to agriculture and other sectors have been recorded for the 2000–2014 period: a short drought spell in 2002, a multi-year event that lasted from the winter of 2004–2005 to the spring of 2008, and another during the years 2011 and 2012 . The impacts of the multi-year drought of 2004–2008 in the Ebro basin have been widely studied. The north-eastern part of the basin was the most impacted and agriculture was the most affected sector, with 540 million Euros of estimated losses to crop production during the hydrological year 2004–2005 and further losses of 272 million in related industries .

As the utility of information strongly depends on the particular details of the decisions and how information is used to support these, we first consulted decision-makers at the basin (Confederación Hidrográfica del Ebro, CHE), irrigation district (Comunidad de Regantes del Canal de Aragón y Cataluña) and farmer scale to better understand their decision processes, their information needs, and how they use information to support the decisions they make. For this analysis we focused on decisions regarding the allocation of water resources, in particular during drought, when curtailments may be applied (Fig. ). The methods followed for the consultation and a summary of the outputs of the interviews at each of the locations are provided in Sect. 3.

The findings of the consultation phase were used to build a model of the decisions at irrigation district scale. The design of the model from these findings is described in Sects. 4.1 (farmer decisions) and 4.2 (reservoir operator decisions). The outputs of these components are the areas of different crops that are planted during the irrigation season and the curtailments applied. The crop yield is then calculated using the area of each crop and observed meteorological data in open-source crop models (introduced in Sect. 4.3).

In order to test and quantify the effect that additional information has on the operational drought management decisions analysed, the decision model was run for two scenarios with different input information. The input data and specific parameters used in these two runs are described in Sect. 5.

The expected final results of the analysis are the relative values of information in each of the tested scenarios. We calculate this based on the total net benefit obtained by farmers during the whole period of analysis. This benefit depends on the decisions made during each season, which in turn depend on the information used to inform them. The results are described in Sect. 6 and followed by discussion and conclusions.

The stakeholders we consider are the different groups that have the capacity to modify the amount of water to be used for irrigation either by deciding on the volume of water to be supplied, or by deciding on the area and type of crops planted. This decision effectively also determines the irrigation demand. We use semi-structured interviews to develop our understanding of the decisions these stakeholders make in managing water resources and the possible adjustments to those decisions that they may make in view of water shortages. This method provides the possibility to discuss additional topics not originally envisioned by the research team in the interview guideline . and similarly apply semi-structured interviews to understand water use and management practices. Participants were asked to describe their own practices, as well as the practices of the groups they deal with in relation to drought management. A set of questions had been previously prepared, but was only used to guide the interviews and ensure that topics not mentioned were addressed and that all required details about the decision processes were collected.

One interview session was held at each of the locations (the basin authority, the irrigation association and the farm) with two or three people participating in each. In the interview that was held at the basin authority, the participants included the head of one of the basin's management units and two members of the hydrological planning office, both with expertise in drought management in the basin. In the interview at the Irrigation Association, the participants were the head of the Irrigation Association and the engineer in charge of the information service about current and expected water availability. And in the interview at the farm level, two people participated: the head of viticulture and the engineer responsible for the information service.

The information about the practices and attitudes of farmers in the study area was to a great extent obtained from the interview with the staff at the Irrigation Association who work in close collaboration with all the farmers in the area and therefore have a wider view than individual farmers. We acknowledge that the sample of interviews we held is small. Although important to the assumptions made in the development of the decision model that we constructed, we would readily agree that our representation of the farmer behaviour and the diversity of responses across farmers may be over-simplified. A larger sample of interviews would reveal more information, but we feel that is outside the scope of this paper.

The main operational decisions that the Ebro river basin authority (CHE) takes regarding drought are the declaration of drought conditions and the allocation of water in emergency situations. To guide these decisions, CHE defined a drought management plan in 2007 , which was the first of its kind in Europe. It is a very comprehensive plan and links hydro-meteorological indicators to drought severity levels. The decision to declare drought is informed by a set of indicators, derived from measurements from a dense network of in situ automatic stations. The plan establishes the main indicator to be used for each of the areas of the basin. Where applicable, water stored in the reservoir is used as the main indicator, since it is considered the most robust option. Otherwise 3-month discharge or groundwater levels are used as indicators.

In the opinion of those interviewed at CHE, the declaration of drought is currently well informed and therefore they consider that additional information would be more useful after the declaration, when conditions must be monitored closely and decisions such as selecting the most cost-effective alternative sources of water or how to secure sufficient water to guarantee environmental flows must be taken. For these decisions timing is critical and they point out that information should be available with a maximum delay of 1 week to be useful for decisions. In addition, they showed particular interest in remote sensing derived snow data to support the quantification of water availability in the basin.

General Irrigation Associations such as Canal de Aragón y Cataluña (CAyC) are responsible for the distribution of water from the reservoir to the users. In drought situations they can decide to introduce restrictions to irrigation water quotas. The decisions they make on the application of these restrictions are informed by the availability of water in the reservoirs that feed the irrigation canal system.

The main decisions that CAyC make in relation to drought is to apply restrictions (curtailments) to the maximum amount of water that irrigators can request. They make this decision when they consider that the available water resource is insufficient to reach the end of the irrigation season if full irrigation supply to meet demand is maintained. They can also decide to move water among the three reservoirs in the area. When restrictions are necessary, these are applied to all users independently of the reservoirs that they can be supplied from to ensure curtailments are applied equitably across the district. However, when water is scarce, priority is given to perennial crops such as fruit orchards and vines to ensure their survival.

To make their decisions, the reservoir operators need information both on water availability and the expected demand until the end of the season. In the interviews, they indicated that they consider that they are well informed on the availability of water given the levels in the reservoir. However, the information on water demand is limited. The difficulty of knowing the demand is due to the fact that they lack information on what crops farmers are planning on cultivating that year, and especially if the farmers will decide to plant a second crop, thus increasing the demand towards the end of the season. Currently they use historic data to estimate the demand. They are also conducting studies on the feasibility of obtaining this information from remotely sensed NDVI data described in, and although this will provide useful information on the current crops, it will not provide information on the future plans of the farmers.

Unlike the managers at the basin scale, CAyC indicated that they consider additional data on the snow cover in the headwaters to be of little use in quantifying the available resource. They argue that they tend to be cautious when accounting for snow in the estimation of total availability, since the reservoir capacity is rather small and therefore the possibility to store snowmelt runoff depends very much on the melt rate.

During the consultation, CAyC also provided details on the types of farmers present in their supply area as well as on the decisions these farmers make on what to plant. Typically, the proportion of crops planted is fruit orchards and vines for roughly a third of the area, alfalfa for another third, and annual crops for the remaining third. These annual crops are mostly winter cereals and maize. The cropping schedule adopted by farmers is mostly either a single crop of long-cycle maize or a winter cereal, or a double crop in which a winter cereal is followed by short cycle maize. The selection of one or the other by the farmers depends on the expected water availability, and is currently mainly informed by the water level in the reservoir. CAyC shares this information with them in the form of biweekly reports. Conversely, the decisions farmers make in terms of what crop to plant, and if they plan to plant a double crop, determine the demand for the season, and will therefore also have an impact on the decision to apply curtailments that are taken by the CAyC.

A prominent farmer in the supply area of CAyC is the Raimat wine producer, who also participated in the consultation process. They provided details of their information use for water resources management. Their parcels extend over 3200 ha and are highly technified, with extensive use of detailed information. In addition to in situ measurements and meteorological station data, they use Landsat satellite data and perform flight campaigns to acquire spatial NDVI and thermal data. The thermal data are used to estimate the leaf water potential and, together with temperature data, calculate a crop water stress index. This information on crop condition is used to detect spatial differences in the crops to make the most of the limited water, select the optimal moment for irrigation and prevent plagues, as well as to ensure the production is as uniform and controlled as possible.

The decisions they make are already based on high-resolution data, and therefore additional medium-resolution global data are not likely to be a valuable contribution to this type of user. However, this extensive use of information is not representative of all the farmers in the basin and other farmers may indeed benefit from additional information.

The farmers have a number of possible crop alternatives for each irrigation season. In this part of the model we simulate the decision of the farmers to follow one of the possible courses of action available to them. The result of the decision is the planted area of the selected crops. Since fruit orchards, vines and alfalfa crops are perennial crops and are typically planted for several years, their approximate areal extent is known and is considered to be constant in the model. The farmer decision model therefore focuses on determining the variable areal extent of maize and winter cereal, as well as the decision as to whether to plant a single crop or also a second crop. In the model, barley is selected to represent the winter cereal crop since it is the most common winter cereal crop in the area.

The courses of action represented in the model consist of a series of decisions made during the irrigation season. The possible actions are depicted in Fig. , which shows the choices that can be made at each decision stage in the calendar. At each of these decision points, the option that farmers would prefer to take if they perceive there are sufficient water resources available is indicated by a blue A. The preferred decision(s) if there are insufficient resources is marked with a red A. The choices and the calendar are based on the information provided by the stakeholders we interviewed, supported by literature sources . We consider two types of farmers in the model, with different options available to them: technified farmers managing large plots and smaller-scale farmers. In the model, technified farmers (marked with T1 in Fig. ) can support a double crop, and this is always their preferred option because of its higher productivity. However, in years of low water availability (red A in the figure) they may decide to leave the land fallow instead of planting a second crop. Smaller-scale farmers (T2) can only manage a single crop and in this case long-cycle maize is the most productive option. In years of low water availability, their decision will depend on the level of risk they are willing to take. We consider three levels of risk aversion (R1, R2 and R3 in the figure). The safest option to secure a crop is to plant long-cycle barley at the beginning of the season (R1), but they can also decide to wait for conditions to improve, taking the risk of having to leave the land fallow if there is no improvement (R>1). If, however, water availability increases by February, they can decide to plant long-cycle maize. If availability is still low, they can secure a crop by planting a short-cycle barley (a less productive option than the long-cycle version) (R2), or they can decide to wait longer (R3). In April, they can still plant a long-cycle maize crop if availability has improved, but, if it has not, then it is too late to plant barley and they have no other option than to leave the land fallow.

The choice of a particular course of action in the model is based on the water availability at the moment of the decision and is used as an indicator of the expected availability of water during the remaining season. For the decisions made before the start of the irrigation season (November and February), the availability is based on the observed volume in the reservoir since these decisions have no influence on the level of the reservoir until the actual start of the irrigation season. For the decisions after the start of the irrigation season (decision points in April and May) a simulated volume is used (Fig. ). The simulated volume is based on the observed volume at the beginning of the irrigation season and the accumulated inflow from the beginning of the irrigation season until the decision date. From the start of the irrigation season the decisions made will have an influence on the level of the reservoir and therefore observed levels are not representative as an input for further decisions. Note that the barley crop is not irrigated and therefore does not pose a demand on the available water resource. However, as can be seen in Fig.  the choice for planting a barley crop will have implications for the available options later in the season and therefore indirectly influences the irrigation demand.

Every 2 weeks during the irrigation season (March–October), the decision on whether to apply curtailments to the maximum amount of water that irrigators can request is re-examined. This decision requires an estimate of the total amount of available water during the season and the total water demand. To estimate the total amount of water that will be available during the season, the inflow into the reservoir from the beginning of the hydrological year in October up to the week of the decision (represented as F[h:i] in Fig. ) is compared with the percentiles of historic data. Data of accumulated inflow in the reservoir are preferred to reservoir levels as input information for the reservoir operator decision model to avoid the influence of the actual decisions of the managers and simulate them independently. The percentile curve in which the value of the current year is positioned is then used to sample from the climatological record a projection of the inflow series into the reservoir until the end of the season.

The total water demand for irrigation until the end of the season (D[s:n]) is calculated as the sum of the demand until the decision day (D[s:i]) and the expected demand from the decision day until the end of the season (D[i:n]). The first is the product of the crop surfaces already planted (Cs[i-1]), which is the output of the farmer decision model, and the resulting crop demand (CD[s:i]) obtained from the crop models using observed meteorological data. The latter (D[i:n]) is unknown to the managers. In the model, an average demand per unit area of crop calculated with the crop model data for the period 2000–2014 is used as an estimate to inform their decision. This is a simplification, as the demand to the end of season will depend on the expected climatological conditions and the crop surfaces planted. The actual demand up to the decision day could be used as an estimate of the expected demand until the end of the season. However, while that could provide an indication of the climatological conditions, the decisions made by farmers on which crops to plant at future decision moments are unknown.

When the estimated total amount of available water during the season is insufficient to fulfil the total demand, curtailments are applied. Conversely, if in a later week the expected total available water is found to be enough to fulfil the total demand, restrictions are lifted.

The output of the farmer model are the areas of each of the crops planted each year by the farmers. These crop areas determine the demand in the reservoir operator decision model. The outputs of the operator decision model on the other hand are the curtailments posed on the water supplied to the farmers until the end of the season, which is determined every 2 weeks. These curtailments may reduce the yield of the crops if these are already planted and the demand cannot be satisfied, but will also influence the farmer decisions within the irrigation season. If curtailments are in force when the farmers are deciding what to plant (April and May), then the model assumes that the farmers do not consider water availability to be good, leading to decisions commensurate with low water availability being made. These decisions will consequently influence the demand.

AquaCrop-OS was used to simulate barley and maize yields. These crops are the main focus of the analysis and they require a more detailed and flexible simulation to differentiate the different growing cycles and planting dates. Default parameters for maize and barley were adapted for the diverse cycles using data from , and .

CropWat 8.0 was used for alfalfa and fruit orchards, the crops that are considered to have a constant crop surface in the analysis. Default parameters were used but were adapted to the cropping calendar in the Ebro basin. Peach tree was selected as the representative fruit orchard crop. An irrigation calendar of 14 days was selected to match the reservoir operators' decision.

The percentage of reduction in crop yield was calculated as the maximum percentage of unsatisfied demand during the season. The reason for this is that, when there is insufficient water, farmers prefer to stop watering a part of the area, rather than apply insufficient water to the whole area. These percentages were calculated using the same 2-week time step of the operator decision. The areas in which irrigation was stopped were considered to have no yield and their contribution was subtracted from the full supply yield values derived from the crop models to obtain the final yield for each crop and year. Priority is given to the perennial crops, with the curtailments then being applied to the maize crops.

We first ran all possible decisions paths and identified the decisions that result in the highest benefits to the farmers for each season. These decisions are represented in Fig.  by the coloured points, with the volume of the reservoir on the y axis. If the point is red, then the best decision is for the farmer to follow the path marked with a red A in Fig. . If it is blue, then it is best to follow the blue A path. If the point is yellow, then it does not matter which of the paths is followed.

We use the optimal decisions based on perfect information to establish a threshold for the reservoir level to divide between good and poor water availability. A perfect threshold would be selected such that all the red points are below and all the blue points are above the threshold. This perfect threshold would always allow the farmer to make the decision that results in a higher benefit at the end of the season.

However, as can be seen in Fig. , it is not possible to obtain a perfect classification of the reservoir levels with a single threshold. The dashed lines mark the thresholds that maximise the points correctly classified. The fact that the classification cannot be perfect means that the reservoir level alone does not provide enough information on what is the best decision to make. Additional information will be valuable if it contributes to improving the classification and, therefore, results in the decision that maximises the benefits being made more often. In this case the additional information we consider is the snow cover data. This is shown in the lower part of the figure. The coloured points again indicate the decision path that would be taken based on perfect information. Again the figure shows that it is not possible to select a threshold value (dashed lines) where all the red points are below the threshold, and all blue points are above the threshold. This again indicates that this information alone does not lead to a perfect classification either.

This is different when the combined information of reservoir level and snow cover extent is used to inform the expectation of water availability. We incorporate this additional information by amending the threshold of the reservoir level. This is the solid line in the figure for the months November, February and April. Snow cover is not considered in May as that is too late in the season for snow to be of significance. When snow covers an area larger than the threshold coverage (dashed line in the lower plot), then the original threshold for the reservoir level is used. However, when the snow cover is smaller than the identified threshold, and therefore the future contribution to the reservoir volume from snowmelt is expected to be low, a second, more conservative threshold for the reservoir level is used (solid line). With the second threshold some of the red points incorrectly classified above the original threshold are now classified below the threshold.

The optimised thresholds for the reservoir level were established at 62 hm3 for November, February and April, and 82 hm3 for May; the thresholds for snow cover were set at 25 %, 35 % and 15 % for November, February and April respectively, while the increase in the reservoir level threshold for the years that are below the snow threshold was set at 20 hm3.

The high losses in some of the years are the result of the limited profit margin between the cost of planting and the selling price of the products. To illustrate further the effect of the profit margin in the decision and the value of information, we have run a series of additional simulations where the costs of planting are reduced by 50 %, 75 % and 100 % (which is the same as zero cost). The relative value of information for these simulations (shown in Fig. ) indicates there is a gradual increase in the relative value of the informed decisions as the ratio of the benefits from the crop yield to the cost of planting increases. The fully detailed gains and losses for these simulations can be found in the Supplement (Fig. S1). Relative values are still low, however, even when there is no cost for planting. This is because the uninformed decision used as a reference also improves with the reduction in the cost of planting. The course of action that performs better on average, in which the uninformed decision is based, is path 3 for the full reported cost, path 4 for the reduced costs and path 5 when no cost is considered. This means that with lower or no investment costs for planting it is better on average to plant the more water-demanding crops. These results also show that as the ratio between the profit made from the crop yield and the costs of planting increases, the relative value of the informed decisions for the years in which the optimal path is followed is also reduced.

At the reduced costs it also appears that the added value of the information from snow cover reduces and in some cases is even detrimental, particularly at the higher reservoir level thresholds. This is likely caused by the uncertainty in the relationship between snow cover and available water resources, which will be elaborated further in the discussion.

Decisions made with perfect or no information were used as reference cases. The difference in the net benefit between these two cases reveals the potential improvements that information can bring with respect to the uninformed decisions. With perfect information, losses can be avoided in seasons of water scarcity and benefits maximised when enough water is available. It should be noted that the paths for perfect information in the model maximise the benefit of the whole group of farmers, rather than that of individual farmers. In reality, the benefits and losses are not shared by the group, and individual farmers would try to optimise their individual benefit instead, though community collaboration in the form of the established user associations in the basin ensures that to an extent farmers do make decisions that contribute to a common good and the tragedy of the commons does not arise.

The uninformed case follows a conservative approach considering that for every year the available water will be limited. Although this results in high losses being avoided, the benefits are well below the potential. Using additional information to inform decisions is expected to help the decision-makers in characterising each season in terms of the water availability and selecting what and when to plant, and accordingly increasing their benefits. As additional information we test a medium-resolution snow cover product derived through remote sensing. show the value of this product in providing snow cover information at the Pyrenees range scale. Our analysis shows that the information from this product also has value at the basin headwater scale, showing improvements in the decisions made when compared to decisions informed by reservoir levels alone.

Detailed analysis of the years where there is benefit in using the additional information shows that this arises mainly from the reduction in the losses in those years in which the optimal decision to make is more uncertain. In these the years the classification of the water resource as being good or bad is difficult, and the additional information on the snow cover adds value by making it more difficult for a bad year to look good. Losses occur in 2002, 2006, 2011 and 2012 (see Fig. S2 for yearly relative value plots), which match the years for which drought impacts in irrigation agriculture have been reported . They are the consequence of an inappropriate course of action being chosen, as a result of the expected availability of water being too high, compounded by the high cost of planting relative to the return on investment of the crops planted.

To test the robustness of the observed effect of the additional information, the model was run 10 times with randomly sampled values of snow cover at each of the decision points. The results of these runs (included in Fig. S3) show that the improvement then also follows a more random pattern and in some cases the additional information is detrimental, thus supporting the hypothesis that the improvements in the decisions are indeed caused by the additional information on snow cover.

However, the results of the model indicate that selecting the option that performs better on average, as is done in the uninformed case, leads to higher benefits than when using the information on reservoir levels (either alone or supported by the MODIS snow cover information). This is in contradiction with the current practice, in which the reservoir-level information is used to support the decision and different choices are made each year. One reason that farmers do not follow this strategy may lie in the fact that not all the losses are assumed by the farmers, since there are subsidies for certain crops or for losses incurred in disastrous years. These subsidies are often based on planted surface and influence the ratio between the return from the crop yield and the investment costs incurred when planting. Additionally, the actual farmer decision on what to plant is influenced not only by water availability, but also by the market prices of the crops. Maize has a high cost of production and therefore, when its selling price is low, farmers tend to select other crops with lower production cost . In the model, however, the planting costs and selling prices were kept constant for all the years to better observe the effect of information on the selection of the crops.

The effect of the cost of planting and the profit margins on the usefulness of the information was explored by running the model for different planting costs. Changing the cost of planting modified the course of action both for the informed and uninformed decisions. The reduction in the costs results in higher relative values for the informed decisions for the period as a whole, caused again by the reduction in the net losses. The ratio between the cost of planting and the return on investment on the crop is similar to the cost–loss ratio used in evaluating the benefit of flood warnings . Where the cost–loss ratio is high, the cost of taking an action in vain (false alarm) is also high, and significant losses may be incurred. This may even result in the information being detrimental, since it does contain uncertainty and may therefore lead to wrong choices being made. Larger losses than if that information is simply ignored and the business-as-usual action is taken may then be incurred. For users with a lower cost–loss ratio, explored here by lowering the cost of planting, additional (uncertain) information becomes increasingly valuable as these users become more tolerant of making a wrong decision. The role of uncertainty in the link between the information used (reservoirs levels and snow) and the realisation of the available water resource is not directly explored in this study through for example a hydrological model, though explicitly considering the uncertainty can add further value to the information. Several authors have shown that the value of information from forecasts is always higher when these are probabilistic. In the application presented here, the relation between the reservoir levels and the available water resources is more certain than the relation between the snow cover and the available water. This may also explain the poorer performance when using snow cover information than when using only reservoir levels. This occurs in 2006. In this year the snow cover at the start of the year (February) was exceptionally high, leading to an expectation of good water resource conditions. However, this was due to widespread snowfall at the end of January just before the decision point in February. This snow melted rapidly and the snow cover in April was anomalously low, with low water resource availability for the rest of the season.

The value of the additional information is not equal to each of the different types of farmers identified. The additional information on the expected water resource provided by the snow cover is found to be relevant only to the decisions that are made in February. For the technified farmers the information is therefore of little value, as the main decisions made by them fall in November and in May, respectively, before the snow accumulation period and after the snowmelt period.

For the small-scale farmers, the additional information can be relevant for the decisions that are made when there is snow cover, primarily those made in February, but also those made in April. These small-scale farmers have only one crop. Once a decision is made to plant a crop, there is no further value to information as there is no further decision to be made. However, the benefit is again not evenly distributed. Small-scale farmers were divided here into three groups of decreasing risk averseness (R1, R2 and R3). We find that the additional information benefits the group of farmers that is willing to take more risk most. These are the farmers that decide to take the risk to wait for a possible improvement when the water availability is classified as not being good at the decision point, instead of taking the safe bet and securing a crop by planting a barley crop, which does not depend on irrigation and possible curtailments. The most risk-averse small-scale farmers (R1) do not even wait for any information on water availability and already plant barley in November. For them there is no value in the additional information. For the slightly less risk-averse farmers (R2) there is limited value in the additional information. If in January the water resource situation is expected to be good, then they will choose to plant maize, but at the first sign of it being bad they will forfeit the possible higher profits from maize and opt to take the safe bet by planting barley. The most risk-acceptant small-scale farmers benefit the most from the additional information, as it will help them make the choice between taking the gamble of waiting for the water resource availability to become better so that they can plant maize, or plant a cereal to avoid the risk of having to leave the land fallow if it does not. In this case these results show that the additional information may be beneficial to improved equity across the farmers in the irrigation district as it is most beneficial to small-scale farmers, provided they are willing to take a gamble to improve their benefits.

In this paper we model the distribution of risk averseness using only a simple percentile distribution. A more realistic distribution of risk averseness can be developed using for example the constant absolute risk aversion utility function , though this will require extensive survey data to determine how risk averseness is distributed among farmers.