The data revolution in ecology is providing the essential fuel that is driving ecology toward being a more robust and predictive science while at the same time better responding to societal needs. This revolution is not just one of sensors and computers, but also one of larger-scale coordinated research and observations. Still, for most ecological systems no one type of information provides a complete picture, which is driving an increased need for data fusion across networks, with iterative data assimilation being particularly important for improving ecological prediction. This will be illustrated with examples of cross-network data fusion across a wide range of ecological systems and the discussion of both the recent advances and continued challenges in statistical approaches for data fusion. At the same time, effectively leveraging the data revolution requires the creation of community tools and networks that address analysis and modeling needs beyond data production. We will also report on efforts by the Ecological Forecasting Initiative to build a community of practice.
Results/Conclusions
Highlighted examples of near-term iterative forecasts driven by cross-network data fusion include: land-surface carbon and water pools and fluxes (NEON, Fluxnet, USFS FIA, NASA), vegetation phenology (Phenocam, NEON, USA-NPN, NASA, NOAA), soil microbiome (NEON, UNITE, GreenGenes, FUNGuild), and lake algal blooms (GLEON, NEON). Through these examples and simulation experiments we demonstrate the critical need to account for systematic errors in observations and models when fusing data sources, a challenge which is always present but frequently goes unnoticed in the absence of multiple constraints and true independent validation. We present methods for accommodating biases in statistical likelihoods and for relaxing assumptions of traditional data assimilation approaches, such as our derivation of the Tobit-Wishart Ensemble Filter. At a broader-scale, we also present the community-building work being done by the Ecological Forecasting Initiative’s working groups and Research Coordination Network, focusing on the cross-cutting interdisciplinary challenges shared by members of the forecasting community regardless of their focal systems. A key part of this is to develop community standards for forecast archiving and metadata, which allow the development of shared tools, 3rd party validation, and cross-forecast synthesis. Finally, we describe the launch of our open community-wide NEON forecasting challenge and the protocols for participating.