Jump to Main Content
Assessing the reliability of dynamical and historical climate forecasts in simulating hindcast pasture growth rates
- Harrison, Matthew T., Christie, Karen M., Rawnsley, Richard P.
- Animal production science 2017 v.57 no.7 pp. 1525-1535
- climate, cumulative distribution, farmers, income, livestock, meteorological data, models, pastures, prediction, records, uncertainty, weather forecasting, Tasmania
- A priori knowledge of seasonal pasture growth rates helps livestock farmers plan with pasture supply and feed budgeting. Longer forecasts may allow managers more lead time, yet inaccurate forecasts could lead to counterproductive decisions and foregone income. By using climate forecasts generated from historical archives or the global circulation model (GCM) called the Predictive Ocean Atmosphere Model for Australia (POAMA), we simulated pasture growth rates in a whole-farm model and compared growth-rate forecasts with growth-rate hindcasts (viz. retrospective forecasts). Hindcast pasture growth rates were generated using posterior weather data measured at two sites in north-western Tasmania, Australia. Forecasts were made on a monthly basis for durations of 30, 60 and 90 days. Across sites, forecasting approaches and durations, there were no significant differences between simulated growth-rate forecasts and hindcasts when our statistical inference was conducted using either the Kolmogorov–Smirnov statistic or empirical cumulative distribution functions. However, given that both of these tests were calculated by comparing growth-rate hindcasts with monthly distributions of forecasts, we also examined linear correlations between monthly hindcast values and median monthly growth-rate forecasts. Using this approach, we found a higher correlation between hindcasts and median monthly forecasts for 30 days than for 60 or 90 days, suggesting that monthly growth-rate forecasts provide more skilful predictions than forecast durations of 2 or 3 months. The range in monthly growth-rate forecasts at 30 days was less than that at 60 or 90 days, further reinfocing the aforementioned result. The strength of the correlation between growth-rate hindcasts and median monthly forecasts from the historical approach was similar to that generated using POAMA data. Overall, the present study found that (1) statistical methods of comparing forecast data with hindcast data are important, particularly if the former is a distribution whereas the latter is a single value, (2) 1-month growth-rate forecasts have less uncertainty than forecast durations of 2 or 3 months, and (3) there is little difference between pasture growth rates simulated using climate data from either historical records or from GCMs. To test the generality of these conclusions, the study should be extended to other dairy regions. Including more regions would both enable studies of sites with greater intra-seasonal climate variability, but also better highlight the impact of seasonal and regional variation in forecast skill of POAMA as applied in our forecasting methods.