Table Of ContentsNext Page

A regional commodity forecasting system for major crops in Australia

A B Potgieter, G L Hammer, P deVoil

QCCA/APSRU
Queensland Department of Primary Industries
PO Box 102 Toowoomba 4350 Queensland, Australia
Tel: 07-46881417, Fax: 07-46881193
Email: potgiea@dpi.qld.gov.au

Abstract

Queensland Department of Primary Industries has developed a regional commodity forecasting system, which integrates a shire-based stress-index wheat model with seasonal climate forecasts based on the El Nino Southern Oscillation (ENSO). It allows the examination of the likelihood of exceeding the long-term median shire yield associated with different season types at the beginning of the cropping season. This system is now run operationally for Queensland by updating the projection each month based on the actual rainfall that has occurred and any change in the ENSO phase from month to month. Although this system was principally designed to inform government in Queensland of any areas that might be more likely to experience poor crops in any year it also serves as a regional commodity forecasting system. The information generated provides an alert for exceptional circumstance issues associated with potential drought in Queensland. However, anecdotal information received from marketing agencies based on their experience with the 2000 regional wheat outlook showed that using this seasonal crop forecasting system in their decision-making processes could add value to their current approaches. Possible decisions to be taken when the outlook is for “likely to be drier (wetter) than normal” are, for instance, forward buying (selling) of grain or shifting of resources from good yielding areas to poor yielding areas.

Introduction

The logistics of handling and trading Australia’s grain commodities, such as wheat, are confounded by huge swings in production associated with climate variability. Advance information on likely production and its geographical distribution is sought by many industries, particularly in the recently deregulated marketing environment. Such information is also sought by government in relation to policy interventions triggered by the degree of exceptional circumstances (e.g. drought, bumper crops, etc.). The accessibility of up to date agricultural statistics is therefore of utmost importance to assist government in these types of decision-making processes. The downscaling of the agricultural statistics census, collated by the Australian Bureau of Statistics, from an annual to 5 yearly time scale makes the need for an objective regional commodity/crop forecasting system even of higher priority and importance.

Recent research has pointed to the potential to improve forecasting of the Australian wheat crop and its spatial distribution by integrating regional crop yield models and seasonal climate forecasts (Hammer et al. 1996, Stephens 1998 and Stephens et al. 2000). This wheat-forecasting product was based on a similar concept that had been developed from a spatial modelling system for grazing lands in Queensland (Carter et al. 2000). The motivation in both cases was largely driven by demand from government in relation to drought policy implementation. However, this information and technology are not widely available in an operational system, which can be used by government or agricultural industry in assisting their decision-making processes.

In this paper we present a regional commodity forecasting system (RCFS), examine its effectiveness in forecasting regional wheat yields for Australia for the 2000 season and consider more general possibilities for use of the information.

Method

Model selection and calibration

Hammer et al. (1996) investigated the predictive ability of different regional wheat forecasting approaches (ie. simple vs more complex) and assessed the trade-offs between accuracy and likely cost of application in real-time forecasting mode. Although, all methods achieved a useable level of skill (Table 1) the simpler empirical (e.g. rainfall regression) and agro-climatic (e.g. stress index) approaches had better predictive ability and were less demanding of input requirement than the more complex simulation / GIS approach. (e.g. APSIM-wheat). Furthermore, the agro-climatic and simulation/GIS approaches largely overcome the concern about extrapolation as they contain enough biophysical rigour to better mimic the agro-ecology outside the training period.

Table 1: R2 and mean absolute error (MAE, t/ha) values from a fit of observed weighted yield versus predicted weighted yield of each approach for each state and nationally.

Region

Rainfall Regression

Weighted Rainfall Index

Stress Index

Drought Index

APSIM-Wheat

 

R2

MAE

R2

MAE

R2

MAE

R2

MAE

R2

MAE

QLD

0.86

0.15

0.89

0.14

0.82

0.17

0.84

0.16

0.73

0.22

NSW

0.89

0.12

0.84

0.14

0.87

0.12

0.77

0.17

--

--

VIC

0.88

0.13

0.90

0.13

0.90

0.12

0.91

0.13

--

--

SA

0.88

0.11

0.76

0.16

0.85

0.13

0.92

0.08

--

--

WA

0.86

0.07

0.86

0.10

0.91

0.08

0.89

0.08

--

--

Australia

0.92

0.07

0.90

0.08

0.90

0.07

0.88

0.07

--

--

The RCFS (Figure 1) combines the simple agro-climatic model for wheat, with near real-time climate data, and projected seasonal climate based on the SOI phase system of Stone et al. (1996) to generate a crop forecast that can be updated each month through the growing season. The potential of agro-climatic yield models to explain most of the variation in wheat yield across Australia has been demonstrated by Stephens (1989 and 1995). The agro-climatic model (Stephens, 1989) uses a weekly soil water balance to determine the degree of water stress experienced by the crop. This stress index (SI) is used in a simple regression model to predict wheat yield for each wheat-producing shire (local government area) in Australia. The index is similar in concept to that proposed by Nix and Fitzpatrick (1969). It utilises biophysical knowledge of the crop, allows consideration of soil type effects, and derives the stress index by contrasting soil water supply with crop demand.

A weighted stress index is calculated for each wheat-producing shire from the values for points falling inside that shire. Point values were weighted by the relative area of the shire they represented by constructing Thiessen polygons around each point. The shire SI was then related to 19 years of historical actual shire wheat yields (Australian Bureau of Statistics: 1975-1993) by linear regression analysis. A term was included in the linear regression to account for any technology trend over time in the shire yield data. The resultant technology trend was examined by Cornish et al. (1998) and is not considered in this paper. The variance of wheat yield explained, using the stress index, ranged between 82 - 91% at state level and was 90% at national level (Table 1).

Figure 1: Depicting the derivation of the regional commodity forecasting system (RCFS)

Outlook

Near real-time daily rainfall data sets for over 800 recording stations, at national scale, are collated and used in running the model from the start of October the previous year up to the end of a particular month. Thus, starting soil water at the end of a summer fallow period is simulated. Historical climate data (ie. climatology) are used to project likely future weather data to produce likely future yield outcomes for each wheat-growing shire. These future yield outcomes are called yield plumes (Figure 2). The forecast distribution is created based on the SOI-phase analogue years (ie. years in history most like the current year) and may involve from 15-20 individual projections. Summary statistics (e.g. median, 10th/90th percentiles, etc.) are derived from these projections and contrast against the simulated long-term median shire wheat yield (ie. bench mark) within the broad cropping region of Australia. The median yield is based on predicted performance over the past 99 years using the agro-climatic model for wheat with long-term rainfall records (Figure 3). The outlook is given as the probability of exceeding the long-term median wheat yield of each shire based on analog years in history of which the SOI phase for April/May was similar to the prevailing phase at the time of the forecast. Areas coloured in yellow to red have a low chance of exceeding the median yield, whereas areas coloured in green to blue have a high chance. The grey shading indicates those areas where it is equally likely (ie. 50/50) to be above or below the long-term median given current circumstances. The calculation of benchmark yields and outlook chances do not take into account damage due to pests, diseases or frost.

Figure 2: Diagram explaining the creating of the forecast based on actual climate and projected climate

Figure 3: Simulated long-term median shire wheat yield for Australia (1901 - 1999)

Results and discussion

The map in Figure 4 shows the probability of exceeding median wheat yield for each wheat-producing shire within Australia as predicted at the beginning of the wheat season (end of May 2000) using the RCFS. This shows poor chance of above median yield in parts of Western Australia (WA; e.g. inland areas) and good chance of above median yield in much of New South Wales (NSW), Queensland (QLD; e.g Central and western), Victoria (VIC) and South Australia (SA). However, a definite di-pole is visible in Queensland with low probabilities in the southeastern parts and high probabilities in the central and western parts of QLD. This map is indicating the likely size of the total crop and highlights those areas where production has greatest chance of being abnormally high or low. This provides forward warning in relation to logistics for transport and quantifies the potential need for exceptional circumstance support from government for producers adversely effected by drought.

Figure 4: Probability of exceeding the long-term simulated shire wheat median for Australia for the 2000 season, given the SOI phase was ‘rapidly falling’ in April/May.

Potential of seasonal climate forecasting

To examine the potential utility of seasonal climate forecasting, the monthly update throughout the 2000 season was done using projections based on all years as well as on the subset of years associated with the prevailing SOI phase at the time of the forecast. The distributions generated were aggregated and compared at state and national levels (Figure 5). Although, there was no strong ENSO signal in effect during 2000, the projections based on the historical analogue years associated with SOI phases tended to have narrower distributions early in the season and shifted towards the final outcome sooner than the projections based on use of all years for a number of states. Anecdotal information received from marketing agencies based on their experience of the 2000 regional wheat outlook showed that seasonal crop forecasting can add value to their decision-making processes when it is used in addition to their current approaches. Possible decisions to be taken when the outlook is for “likely to be drier (wetter) than normal” are, for instance, forward buying (selling) of grain or shifting of resources from good yielding areas to poor yielding areas. A greater sample of seasons is required before any general conclusions about utility of forecasts can be drawn. Hindcasting studies are currently in progress so that a more robust comparison of the system’s skill and value can be made.

Historically, forecasts have not been produced in a probabilistic manner. Agencies responsible derive single estimates based on adjustment from knowledge of previous year(s) or from projections of median rainfall. This approach retains simplicity, but at the expense of any information on production risk.

Implications and lessons

This case study indicates potential for use of seasonal forecasts in commodity forecasting for government policy support and for decision making in industry. The system was designed primarily in relation to policy needs and is now operational for this purpose. It provides the quantification in time and space required too assist government decision-making in relation to implementing drought policy. More specific products, such as maps identifying areas at extremely high risk will likely be developed as the interaction with policy users develops. The presentation and interpretation of information on production risk and degree of ‘exceptionality’ is of prime importance. In addition, credibility of the forecasting/information system must be high for its effective use in this arena. This requires scientific rigour, accuracy and repeatability.

Although this system has not yet been interfaced with industry decision-makers in any detailed manner, it is sufficiently advanced to provide a useful basis for dialogue. One of the issues to emerge in preliminary discussions relates to the output of distributions. There is a range of awareness of risk management concepts in the business sphere. Where this is well developed, rapid utilisation of possibilities seems most likely. Otherwise, greater effort is required to develop awareness of the gap between what is possible and what is expected.

Figure 5. Wheat yield predictions for Australia (AUS) and each state in Australia in 2000. Predictions are updated at the end of each month. The solid lines show predicted median and top and bottom decile for predictions based on the remainder of the season projected using the SOI-based forecast. The dashed lines show the top and bottom decile of the predicted values at the end of each month when all years in the historical record are used for projection. The horizontal line is the simulated long-term median yield (over all years).

References

Cornish, P.S., Ridge, P., Hammer, G., Butler, D., Moll, J. and Macrow, I. (1998) Wheat yield trends in the northern grains region – a sustainable industry? In Proceedings 9th Australian Agronomy Conference. Wagga Wagga 1998 pp. 649-652. (Australian Society of Agronomy: Wagga Wagga, News South Wales).

Carter, J.O., Hall, W.B., Brook, K.D., McKeon, G.M., Day, K.A., Paull, C.J., (2000) Aussie grass: Australian rangeland and grassland assessment by spatial simulation. In Applications of Seasonal Climate Forecasting in Agricultural and Natural Ecosystems - The Australian Experience (Eds GL Hammer, N Nicholls and C Mitchell) pp. 329-349. (Kluwer Academic, The Netherlands)

Hammer, G.L., Stephens, D. and Butler, D. (1996) Development of a national drought alert strategic information system. Volume 6a, Wheat Modelling Sub-Project – Development of Predictive Models of Wheat Production. In Report to the Land and Water Resources Research and Development Corporation.

Nix, H.A., Fitzpatrick, E.A. (1969) An index of crop water stress related to wheat and grain sorghum yields. Agric. Meteor. 6:321-327.

Stephens, D.J, Lyons, T.J and Lamond, M.H. (1989) A simple model to forecast wheat yield in Western Australia. Journal of the Royal Society of Western Australia 71, 77-81

Stephens, D.J. (1995) Crop yield forecasting over large areas in Australia. In Unpublished Ph.D Thesis. (Murdoch University: Perth, Western Australia)

Stephens, D.J. and Lyons, T.J. (1998) Rainfall-yield relationships across the Australian wheatbelt. Australian Journal of Agricultural Research 49, 211-223.

Stephens, D.J., Butler, D.G. and Hammer, G.L. (2000) Using seasonal climate forecasts in forecasting the Australian wheat crop, In Applications of Seasonal Climate Forecasting in Agricultural and Natural Ecosystems - The Australian Experience (Eds GL Hammer, N Nicholls and C Mitchell) pp. 351-366. (Kluwer Academic, The Netherlands)

Stone, R.C., Hammer, G.L. and Marcussen, T. (1996) Prediction of global rainfall probabilities using phases of the Southern Oscillation Index Nature 384, 252-255.

Top Of PageNext Page