EVALUATION OF CROP GROWTH MODELLING TO SUPPORT ESTIMATES OF SEASONAL CANE YIELD FOR THREE MILL ZONES IN SOUTHERN QUEENSLAND

摘要

Forecasts of mill zone seasonal cane yield are required by sugar industry sectors at a range of lead times before the crushing season to facilitate forward sale of sugar and for planning the logistics of the crushing season. There have been significant reductions in milling staff allocations to yield estimation in recent decades and mills are seeking additional objective tools to support subjective yield estimates. Simulation modelling has been used for routine estimation of seasonal yield in South Africa since 2001 and at research level to selected regions in Australia since 2005. This paper documents relevant research undertaken for the Bundaberg, Isis and Maryborough mill zones. APSIM 7.0 was used to simulate yields of irrigated autumn and springplant cane and first ratoon cane from an autumn plant crop grown on two soil types. Simulated yield for assessment dates of 28 February, 30 March, 30 April and 15 September were correlated with actual seasonal yield. Simulation for one soil type and only the average of irrigated autumn and spring planting sequences provided a simple and effective procedure for yield indexing. Initial results showed higher correlations between simulated and actual yield at Bundaberg than for the other two regions. The final predictive model was based on correlations of simulated yield for Bundaberg with actual yields in all regions. Coefficients of determination for regression of simulated yield with actual yield ranged between 0.79 and 0.90 for March and April assessment dates, with root mean square errors for this period ranging between 3.1 and 4.9 tonnes cane/ha for 1999 to 2008 seasons. Model-based estimates were 83 to 100% successful in indexing the magnitude of seasonal yield relative to actual yield of the previous year between 1999 and 2010 at the end of February when only semi-quantitative yield estimates are required. Simple model based applications may therefore be further developed to support current subjective yield estimating techniques.