Near-term ecological forecasting is an emerging necessity for supporting environmental decision making and adaptive management. However, creating and maintaining automatic, iterative forecasts can be computationally demanding and prohibitive for researchers. Many forecasting workflows have been developed and maintained independently, resulting in a lack of interoperability that creates a barrier of entry for new forecasters.
One objective of the Near Term Ecological Forecasting Initiative (NEFI) is to develop tools and workflows to make forecasting accessible to researchers who lack advanced technical backgrounds or the computation resources to create iterative near term forecasts. This objective is being accomplished through the creation of a scaleable, cloud-based system for submitting, generating, archiving, and disseminating multi-model ecological forecasts. Our workflow starts with a web front end where users can request packages, drop R scripts into the web portal, and upload supporting files. Apache OpenWhisk is then used to create containers with the appropriate dependencies to run the code. The OpenWhisk functions are run on virtual machines on GENI (Global Environment for Network Innovations) and the results are stored using MongoDB. Forecast results are returned to the website after model runs are completed and a R Shiny website is used to explore results and model error.
Results/Conclusions
Multiple forecasts have been deployed through the system and user experience has been tested. The system was tested by numerous new users in April with projects by graduate students in an Ecological Forecasting semester course and again in July by NEFI summer course participants. At each testing stage, the system and workflow were improved through surveying users about ease of use, accessibility, and overall experience with the system.
The NEFI server makes it easier for researchers to implement, archive, and share iterative forecasts. Any group can develop a forecast and automatically run it in our system. Beyond making forecasting accessible, this initiative has led to the development of standards for open archiving and sharing of workflows that allows for the development of community tools. Additionally, by updating forecasts with new data iteratively, we are able to compare forecasts to new data and create hypotheses about predictability in ecology with regards to uncertainty and error. Our system makes forecasting accessible for all ecological researchers which in turn will advance the field as we share tools, data, and forecasts.