Current approaches to forecasting the impacts of climate variability focus primarily on using model hindcasts to understand and calibrate previous extreme events and then using models to explore future scenarios. At the same time, there have been advances in the use of ground, tower, and remotely-sensed data to monitor forest health, at time scaled increasingly close to real time. However, the models we use to explore hypothetical responses are rarely employed in real-time to predict responses to climate variability as it is occurring. Near-term iterative forecasting has the potential to address this and bridge between monitoring and modeling efforts.
Here we report on efforts within the Near-term Ecological Forecasting Initiative (NEFI) and the PEcAn project to improve our ability to assimilate a wide range of forest data into models to constrain forest state variables in models. Specifically, we have produced open-source multi-model tools that implement a range ensemble-based state assimilation approaches, and coupled these to workflows for ingesting a wide range of data and estimating input uncertainties. We also introduce our efforts to extend this state data assimilation system into an iterative forecast system and discuss what steps are required to make such a system operational.
Results/Conclusions
In terms of forecasting forest responses, phenology forecasts are currently most advanced, with multiple real-time forecasts in operations, and results demonstrating our ability to modify these forecasts to incorporate new information about the state of the system, rather than just updating climate drivers. Second, we demonstrate near-term forecasts of forest carbon and water fluxes, driven by NOAA ensemble weather forecasts and assimilating near real-time eddy-flux observations. Initial forecasts have been produced using statistical models, and we are working on extending this to process-based land surface models. Third, we demonstrate efforts to assimilate remotely-sensed data, focusing first on airborne imaging spectroscopy, which is able to detect changes in forest composition, leaf area, and foliar water content. Ongoing work is extending this to a broader range of remote sensing products, with a particular emphasis on forest disturbance. Finally, we report on efforts to assimilate forest inventory and tree-ring data into process-based forest models, updating interannual forecasts with observations of forest growth, mortality, and recruitment. Moving forward, we are on the cusp of being able to produce iterative forecasts at the site-scale, with future efforts focused on fusing multiple observations simultaneously and extending to large spatial scales.