Abstract

Use of seasonal forecasts to improve management of land and water for agriculture and the general community is sometimes limited by lack of historical data for the user's location. Customised forecasts for multiple locations may be used to examine spatial coherence of ENSO effects on streamflow in order to make maximum use of information in a region.

This paper examines the implications for availability of water for irrigated cotton production by analysing historical records from five stream gauging stations in the Condamine-Balonne Basin of southern Queensland. This basin constitutes one quarter of the Murray-Darling Basin.

The overall methods were: to examine the forecasting skill available for streamflow gauging stations relevant to the irrigated cotton industry in the region; to use customised spatial seasonal forecasts in order to examine the coherence of the climatic effects in the region, and thus to make the best possible assessment of the coming season; to draw some conclusions about likely water availability from the analyses; and to examine some implications at farm level. In practice these implications will be tempered by water allocations and licence conditions.

Customised geospatial forecasts of streamflow on any scale chosen by the user can improve land and water management by providing advance warning of likely water availability, particularly during periods when the SOI is in a negative phase by early winter. Forecasts using persistence of streamflow provided useful additional skill in summer. Spatial coherence was a useful concept that improved confidence in the reliability of the forecasts. Further development of the AUSTRALIAN RAINMAN computer package will provide a convenient tool to carry out such analyses. Forecasts of streamflow can assist decision making in the irrigated cotton industry. This work could be extended to other industries and regions.

Introduction

The highly variable climate in Australia continues to be a major impediment to successful agriculture and to sustainable land and water management. Forecasts of seasonal conditions, particularly rainfall, providing there is acceptable skill (statistically significant test), may assist primary producers, agri-business and governments to improve their management of climate risks and opportunities. Forecasting skill is needed for a broad range of management decisions e.g. cropping, stocking rates, irrigation scheduling, and sustainable management of the land and water environment. The need to improve management of land and water is highlighted by over-commitment of water allocations in the Murray-Darling Basin (Anonymous 1996) and the attendant problems of land clearing and salinity (Anonymous 1999).

To take advantage of recent improvements in seasonal forecasting of rainfall or streamflow, e.g. by using the El Niño/Southern Oscillation (ENSO) phenomenon via the Southern Oscillation Index (SOI) (McBride and Nicholls 1983) or patterns of sea surface temperatures (Drosdowsky and Chambers 1998), primary producers need historical climatic records for their specific location. When this is not available, the question is how to make the best use of what we have. The Bureau of Meteorology's Australian collection of long-term rainfall stations (over 3800 prepared for the AUSTRALIAN RAINMAN computer package (Owens et al. 1996) provides acceptable geographical coverage for most agricultural areas. However records of streamflow at stream gauging stations are poor by comparison, being far more difficult to collect; there are far fewer of them, and they are often short (less than 30 years). This issue and its effect on our ability to analyse and forecast climate variability affecting land and water management are examined in more detail elsewhere (Clarkson et al. 2000, Clewett et al. 2000).

Climatic signals in the Pacific Ocean in the autumn of 2001 have indicated that another El Niño episode may be imminent (R. Stone, pers. comm.). As this may cause major deficits of rainfall and therefore streamflow for irrigated agriculture, this paper will consider, as an example, how customised spatial forecasting of streamflow may assist cotton growers in the Condamine-Balonne Basin (Darling Downs and Maranoa region) of Queensland to understand the impacts of ENSO on river flows.

Methods

The overall methods were: to examine the forecasting skill available for streamflow gauging stations relevant to the irrigated cotton industry in the region; to use customised spatial seasonal forecasts in order to examine the coherence of the climatic effects in the region, and thus to make the best possible assessment of the coming season; to draw some conclusions about likely water availability from the analyses; and to examine the implications for enterprise decisions.

Historical records of streamflow for 14 gauging stations in the Condamine-Balonne Basin, which occupies the northern 25% of the Murray-Darling Basin, were obtained from the Queensland Department of Natural Resources and Mines. The study concentrated on the five stations most relevant to the cotton industry in the St George area. Four of these (Chinchilla, Cotswold, Weribone and St George) are located on the Condamine-Balonne river system between Chinchilla and St George, and the fifth (Cashmere) is on the Maranoa River, a tributary of the Balonne River. Water for irrigation is obtained either by licensed water harvesting into earthen ring tanks during times of high river flow or from the Beardmore Dam and the Jack Taylor Weir near St George via irrigation supply channels. This dam is small (capacity approximately 82,000 megalitres (ML)) and inflows are gauged at Weribone and Cashmere. Cotton requires two to seven irrigations between September and February.

To assess the extent of ENSO effects on streamflows, customised concurrent analyses using three-month averages of the SOI were carried out for the winter, spring and summer periods - being the seasons from which water harvesting and inflows to dams might provide water for the next cotton crop.

Forecasting skill was then examined using the five-phase SOI system (Stone et al. 1997) in the two-month period before each season. Because no significant SOI effects were found for seasons ending later than October, a new alternative forecasting system was also examined in which average streamflow in the previous season was used to estimate flows in the next one (persistence) (Chiew et al. 2000, Clarkson 2000). In that method terciles of prior streamflow were used to divide subsequent flows into three corresponding groups.

Two statistical tests for significant shifts in frequency distributions of streamflow caused by ENSO were carried out. The Kruskal-Wallis (KW) test (Conover 1980) tested for changes in the overall ranks of streamflow among the SOI phase categories; the Kolmogorov-Smirnov (KS) test (Conover 1980) compared the distribution from one SOI group with the pooled distribution of the other groups. The same principles applied to tests involving persistence as a forecasting tool.

In the context of geospatial information for agriculture, spatial coherence was taken to mean the consistency and uniformity in streamflow responses to climatic processes under examination. Examples included presence of statistical significance in forecasting relationships, consistency in the type of response to positive and negative ENSO episodes, percentage changes in streamflow, and the chance of median streamflow related to different phases of the SOI.

Calculation of means of streamflow and analyses of persistence were carried out with the computer package AUSTRALIAN RAINMAN Version 3.3 (Clewett et al. 1994, 1999), combined with a prototype plug-in STREAMFLOW supplement (Clarkson et al. 2001). A prototype of RAINMAN Version 4, incorporating selection of multiple locations and production of maps, was used to examine spatial coherence of ENSO effects, particularly KS significance tests across the 14 gauged locations. The RAINMAN packages together enabled customising of: forecasting tools, choice of season, lag interval between forecast period and season, spatial grouping, and choice of output as tables, graphs or maps.

Results

Mean annual streamflow at the five key gauging stations ranged from 138,000 ML per year at Cashmere to 1.2 million ML at St George (Table 1). Only two of the five stations close to St George had long records (over 70 years) but all showed high coefficients of variation in annual flow from 99% to 139%. Corresponding coefficients for individual months were commonly around 300% (data not shown). Seasonal streamflow was highest in late summer and lowest in early spring (Figure 1).

Table 1. Annual streamflows ('000 Megalitres) and length of record for five gauging stations in the Condamine-Balonne Basin of Queensland.

Stream gauge	River	Mean	Median	CV(%)	No. of years
St George	Balonne	1266	859	130	76
Cashmere	Maranoa	138	76	139	26
Weribone	Balonne	1367	824	99	25
Cotswold	Balonne	797	350	135	24
Chinchilla	Condamine	521	283	124	76

Figure 1. Seasonal distribution (January to December) of streamflow in the Balonne River at St George for the period 1922-2000.

The concurrent analysis showed that when the seasonal average SOI was below -5, the chances of at least median streamflow in the same season were significantly reduced by 20 to 30% at four sites, particularly at St George (Table 2). The effects were strongest in winter (June to August). The opposite effects occurred when the SOI was above +5, although fewer of them were significant in spring (September to November) and summer (December to February).

Table 2. Concurrent effect of the SOI (averaged over three months) on the percent chance of median seasonal flow for stream gauges in the Condamine-Balonne Basin.

Season SOI class		Stream gauge and river
		St George Balonne	Cashmere Maranoa	Weribone Balonne	Cotswold Balonne	Chinchilla Condamine
Jun-Aug	below -5 -5 to +5 above +5	20 *** 56 71 **	22 54 43	22 ** 50 75 *	22 * 54 75 #	29 ** 51 82 **
Sep-Nov	below -5 -5 to +5 above +5	29 ** 54 65	30 54 67	30 * 55 71 #	44 43 80	29 * 51 67
Dec-Feb	below -5 -5 to +5 above +5	32 * 58 52	25 50 80	10 ** 69 80 #	27 # 69 50	26 * 63 * 45

[Significance of KS test: # indicates P <0.10, * indicates P <0.05, ** indicates P <0.01, and *** indicates P <0.001].

Forecasts of streamflow using SOI phases in the two-month period before the season, showed that the last season for which SOI effects were statistically significant was from August to October, which is shown in Table 3. For this period the negative SOI phase indicated reductions of 20 to 50% in the percent chance of median streamflow; positive responses to the positive SOI phase were smaller and again less frequent. Responses to the other SOI phases were not significant.

The spatial coherence of negative responses to the negative SOI phase was clearly demonstrated across most of the 14 stream gauges in the maps of the KS test (Figure 2).

Table 3. Forecast percent chance of median streamflow in the season August to October, using five phases of the SOI in the previous June-July period. [Significance indicated as for Table 2].

Stream gauge	No. of years	SOI phase
		falling	negative	neutral	rising	positive
St George	78	46	0 ***	53	53	74 *
Cashmere	27	40	29	75	50	43
Weribone	28	40	14 **	50	50	86 *
Cotswold	28	50	29 #	60	60	57
Chinchilla	77	54	30 #	38	53	63 *

Figure 2. Spatial view of statistical significance of forecast skill (KS test) for seasonal streamflow in the Condamine-Balonne Basin of Queensland, produced by a prototype of AUSTRALIAN RAINMAN Version 4.

Forecasting streamflow beyond October by use of streamflow persistence indicated that, following periods of low streamflow, the chance of at least median flow in the next season was significantly lower at three locations, particularly Chinchilla where it was reduced to 24% in the season November to January (Table 4). No significant effects were found after January. The percent chances of streamflow at St George show graphically the size of the negative effect during the period November to January (Figure 3).

Table 4. Forecast percent chance of at least median streamflow in the season November to January, using terciles of streamflow in the previous August to October (persistence). [Significance indicated as for Table 2].

Stream gauge	Prior streamflow (August to October)
	Lowest tercile	Middle tercile	Highest tercile
St George	32 *	62	56
Cashmere	31 *	100 *	44
Weribone	33	50	63
Cotswold	33	60	56
Chinchilla	24 **	63	60 *

Figure 3. Forecast chance of streamflow at St George in November to January using terciles of streamflow in the pervious August to October (persistence).

Discussion

The results indicate that ENSO greatly affects the outlook for streamflow in the Condamine-Balonne Basin with a high degree of spatial coherence, particularly during negative episodes of the SOI. Once the SOI falls into the negative phase by the end of June, there is a high probability of large reductions in streamflow and therefore water availability in spring. Although significant forecasting skill for the negative SOI phase does not extend to summer in the examples studied, use of streamflow persistence as a forecasting tool can provide useful additional insights up to the end of January when flows in the lowest tercile occur during spring.

Use of customised spatial forecasts confirmed that the effects of ENSO could extend over a large drainage basin. Because human activities in diverting water in one part of a river system directly affect water availability downstream, it is important to know that ENSO effects are spatially coherent (similar in type, strength and direction) on a catchment scale.

There are many implications of negative effects of ENSO on streamflow for farmers growing irrigated cotton. These include how much area to plant, thereby making best use of scarce water and also minimising costs of establishing crops that may fail from lack of water. In this age of forward selling contracts, advance warning of a bad season can enable suitable caution in signing further contracts, as well as financial planning to cover shortfalls on existing ones.

Spatial analyses of rainfall in data-sparse areas such as the southern brigalow region in Queensland could have similar implications for grazing and dryland cropping enterprises and the overall food and fibre chain. Possible applications include: decisions about stocks of seed and fertiliser (farmers and agri-business), stocking rates of sheep and cattle for sustainability and profit, use by governments for policy on drought relief, and regional buying of produce by large food retailers.

Conclusion

Customised spatial forecasts of streamflow on any scale chosen by the user can improve land and water management by providing advance warning of likely water availability, particularly during periods when the SOI is in a negative phase by early winter. Persistence of streamflow provided useful additional forecasting skill in summer. Spatial coherence was a useful concept that improved confidence in the reliability of the forecasts. Further development of the AUSTRALIAN RAINMAN computer package will provide a convenient tool to carry out such analyses. Forecasts of streamflow can assist decision making in the irrigated cotton industry, however in practice the implications of this work will be tempered by water allocations and licence conditions. This work could be extended into other industries and regions.

Acknowledgments

Historical records of streamflow used in this study were provided by the Queensland Department of Natural Resources and Mines as part of collaboration in the Rainman Streamflow Project, which received financial support from Land & Water Australia.

Development of AUSTRALIAN RAINMAN to incorporate a plug-in STREAMFLOW supplement, new forecasting tools and mapping, received financial support from the Rural Industries Research & Development Corporation and Land & Water Australia.

References

Anonymous (1996). Setting the Cap – Report of the Independent Audit Group. Murray-Darling Basin Commission, Canberra, November 1996.

Anonymous (1999). The Salinity Audit of the Murray-Darling Basin – a 100-year perspective, 1999. Murray-Darling Basin Commission, Canberra, 1999.

Chiew, F.H.S., McMahon, T.A. Zhou, S.L. and Piechota, T.C. (2000). Streamflow variability, seasonal forecasting and water resources systems. In: Applications of Seasonal Climate Forecasting in Agricultural and Natural Ecosystems – The Australian Experience (Editors: G. Hammer, N. Nicholls and C. Mitchell), Kluwer Academic.

Clarkson, N.M., (2000). Rainman Streamflow Project Final Report to LWRRDC. Land and Water Resources Research and Development Corporation, Canberra.

Clarkson, N.M., Owens, D.T., James, R.A. and Chiew, F.H.S. (2000). Seasonal streamflow forecasts to improve management of water resources: 3. Issues in assembling an adequate set of Australian historical streamflow data for forecasting.

Clarkson, N.M., Clewett, J.F. and Owens, D.T. (2001). STREAMFLOW: a supplement to AUSTRALIAN RAINMAN to improve management of climatic impacts on water resources. Queensland Government Department of Primary Industries, Toowoomba (in preparation).

Clewett, J.F., Clarkson, N.M, Owens, D.T. and Abrecht, D.G. (1994). Australian Rainman: Rainfall Information for Better Management. Department of Primary Industries, Brisbane.

Clewett, J.F., Smith, P.G., Partridge, I.J., George, D.A. and Peacock, A. (1999). AUSTRALIAN RAINMAN Version 3: An integrated software package of Rainfall Information for Better Management. QI98071, Department of Primary Industries Queensland.

Clewett, J.F., Clarkson, N.M., and Chiew, F.H.S. (2000). Seasonal streamflow forecasts to improve management of water resources: 2. Impact of climate variability and ENSO on streamflow and runoff in Australia. Proceedings of the 3rd International Hydrology and Water Resources Symposium of The Institution of Engineers, Australia. Perth, November 2000. Volume 2: 637-641.

Conover, W.J. (1980). Practical Nonparametric Statistics (2^nd ed.). New York: John Wiley and Sons.

Drosdowsky, W. and Chambers, L. (1998). Near global sea surface temperature anomalies as predictors of Australian seasonal rainfall. Bureau of Meteorology Research Centre Report No. 65, Bureau of Meteorology, Melbourne.

McBride, J.L. and Nicholls, N. (1983). Seasonal relationship between Australian rainfall and the Southern Oscillation. Monthly Weather Review. 111, 1198-2004.

Owens, D.T., Clarkson, N.M. and Flood, N. (1996). Development of a long-term daily and monthly rainfall collection for Australia. Proceedings of the Second Australian Conference on Agricultural Meteorology. The Impact of Weather and Climate on Agriculture. The University of Queensland, Brisbane, 1-4 October 1996. pp 235-239.

Stone, R.C., Hammer, G.L., and Marcussen, T. (1997). Prediction of global rainfall probabilities using phases of the Southern Oscillation Index. Nature, 384: 252-255.