Everingham, Y.L. (2007) Comparing phase based seasonal climate forecasting methods for sugarcane growing regions. In: Proceedings of the International Congress on Modelling and Simulation. pp. 574-581. From: MODSIM07 International Congress on Modelling and Simulation: Land, Water & Environmental Management: Integrated Systems for Sustainability, 10-13 December 2007, Christchurch, New Zealand.

PDF Restricted to Repository staff only

 

4

Abstract

Climate forecasting systems that group years on the basis of a climate forecasting index like the Southern Oscillation Index (SOI) or sea surface temperatures (SSTs) are quite simple to explain to industry personnel. Phase systems identify a subset of years (analogues) that have the same phase for a particular month. Industries can then investigate how the response of interest varied historically by the SOI or SST phase and self-validate the system. This is possible because industry members will remember the big wet and big dry years. Phase systems also allow industry personnel to visualise distributional shifts in rainfall and other responses (e.g. yield) between the different phases. These components spark a great deal of interest and enthusiasm at case study meetings. The simplicity of phase systems contributes to increased understanding of the forecasting approach, and highlights both the strengths and limitations associated with seasonal climate forecasting. Given that climate forecasts are not a perfect science, it is important that industries understand the risks and probability concepts so they can better integrate forecasts into a decision-making framework.

The Australian sugar industry has predominantly used the five-phase SOI climate forecasting system as its benchmark in recent years. The purpose of this paper is to compare the performance of the benchmark system with other phase-based climate forecasting systems. Three-phase and nine-phase SST forecasting systems and a three-phase SOI system formed part of the investigation. An assessment is made across the sugarcane growing regions and across the calendar year, simultaneously. This is done for seven sugar growing regions that collectively produce approximately 90% of Australia's sugar. A methodology that enables a fair comparison of the systems is presented. This methodology caters for the different number of phases with each forecasting system. We consider three performance measures: P-values of (i) the Kruskal-Wallis (KW) test statistic, (ii) a linear error in probability space (LEPS) skill score and (iii) a relative operating characteric (ROC) skill score for above and below median rainfall. P-values are used to overcome obstacles associated with the different numbers of phases. This is important since, by chance alone, it is easier to get a higher or better categorical LEPS score for systems that have more phases.

Results can vary with the performance measure. If ROC- and LEPS-based performance measures were preferred, then the three-phase SST system produced a higher number of significant results across the regions and three-month rolling periods. If performance measures that reflect the degree of distributional shifts or discriminatory ability between phases are preferred, then the five-phase SOI system produced the highest number of significant fields. Taking into consideration dependencies and auto-correlations associated with the response measurements across the calendar year and across coastal regions which essentially differ in latitudinal positioning, it is important to assess the likelihood that the number of significant fields could have occurred purely by chance.

Whilst a methodology for comparing different phase systems, where the number of phases varies from system to system is presented, the dilemma as to which performance measures to base decisions remains. Users must carefully consider which performance measures are most appropriate for their investigation.

Actions (Repository Staff Only)

Item Control Page