May 21 - July 30
Warming Southeast Australian Climate: The Effects of Sea Surface Temperatures (SSTs)
Kwanshae Flenory, Michael Richman,and Lance Leslie
What is already known:
- The Australian climate is warming at an alarming rate, with a higher frequency of, and more intense, hot months.
- Impacts have been widespread and severe including habitat reduction, crop damage and human loss.
- Past work has used linear regression techniques with limited climate driver attributes to establish predictive relationships to warming.
What this study adds:
- A combination of linear regression and bagging trees used to select from a large pool of sea-surface temperature and climate drivers.
- Predictions of January mean maximum temperatures were made at four sites using kernel regression techniques.
- At the four sites, correlation between predicted and observed air temperature ranged from 0.67 to 0.84, and the mean absolute error 0.65 to 1.18°C.
- In comparison to previous studies, these predictions were more accurate.
Abstract:
In the past few decades, the climate in Australian has been warming at an alarming rate when compared to historical variations. Associated with that warming, extended heat events, lasting for weeks to months have plagued the country. Climate model projections suggest that such events will occur more frequently and intensify in the future. The extreme temperatures have damages ecosystems through droughts and fire and resulted in the loss of human life.
This study examines how the combination of sea surface temperatures (SSTs) and climate drivers predict summer mean maximum temperature at selected locations in SE Australia. Ninety-one ocean grid boxes of SST surrounding Australia were used for simultaneous and lag1 relations as well as 42 climate drivers, creating a suite of 224 potential predictors. Variable reduction using 5-fold cross validated linear regression and bagging, resulted in ~ 90% reduction in the number of variables passed to the final prediction equations. Linear multiple and nonlinear kernel regression methods were applied to predict the January anomalies of maximum temperature using this reduced set of predictors. For the nonlinear regressions, two kernels were evaluated: polynomial and radial basis function. The polynomial degree and radial basis function kernel width were optimized for sea surface temperatures and climate drivers by maximizing their 10-fold cross validated correlations with the air temperatures at the various locations in SE Australia. The key findings were (1) climate drivers had as much significant influence on the prediction accuracy as SSTs and (2) the combination of the reduced sets of SSTs and climate drivers often accounted for 40-60% of the January mean maximum temperature variance. Such a large percentage of predictable variance is expected to lead to more effective monthly temperature predictions.