There is a growing desire for near-term ecological forecasts, but there are few implementations that make regularly updated forecasts. We implemented a continuously updated forecasting system for the flower and leaf phenology of 66 plant species throughout the United States. Species specific phenology models were built using data from the National Phenology Network. Daily forecasts are made using temperature from the most current NOAA climate forecasts, which are themselves updated every 6 hours. Visual maps of forecasted dates across species ranges are available at http://phenology.naturecast.org. Using observations from 2018 we can calculate the skill of the forecasts compared to baselines, the contribution of variance from forecasted mean temperature, and the forecast horizon of plant phenology.
Results/Conclusions
Preliminary results are available with forecasts from Jan. 5, 2018 thru mid-February 2018. Almost 60% of forecasts exhibit significant (p<0.05) trends through time in in the estimated DOY for the phenological events (43% increasing and 16% decreasing). This demonstrates that as forecasts are updated with the newest information, estimates shift, often directionally, for most species. In 25% of forecasts these shifts also influence the 95% prediction interval of the forecasted event, showing that regularly updating forecasts can influence knowledge of uncertainty as well as the point estimates for the forecasts. The general expectation is that uncertainty should decrease as more data is added and the date of the forecasted event nears, so it is interesting that the prediction internals are not changing in the majority of forecasts, and in 15% of cases the uncertainty is actually increasing overall (ie. a higher 95% prediction interval). However, as of Feb 15, 2018 the majority of forecasts have events predicted to happen 2.5 months away, allowing ample time for forecast confidence to improve.