During the evaluation period (2023), submitted forecast models were most useful when in-lake conditions were most different to historical observations.
As conditions continue to move beyond what can be predicted from past observations, these forecasting methods will be more and more valuable! 📈

Skillful forecasts were generated using four model classes - process models, machine learning, empirical, and multi-model ensembles.

35-40% of models outperformed null models and the most successful models accounted for more sources of uncertainty (process, parameter, initial conditions, observational, drivers) with poor models often over-confident in their forecasts.

The tutorial can be expanded from the original 90-minute version to include materials on workflow automation⏲️▶️and forecast evaluation 🏅❌!
Come and try #EcologicalForecasting!

This tutorial introduces a simple near-term, iterative forecasting workflow using #NEON data to get you started thinking about how to implement real-time forecasts that include uncertainty!
github.com/eco4cast/NEO...