Community Agricultural Model Platform (CAMP) for multi-model ensemble simulations of crop growth and development

Uncertainty in crop growth simulations arises from model structure, input data, parameter values, and calibration procedures. Multi-model ensembles can reduce the influence of individual model-structure errors and extreme single-model predictions, but their application is often hindered by the need to prepare different computing environments for each crop model. Here we propose the framework where containers are used to provide a uniform interface for a diverse set of crop models in order to support the multi-model ensemble within a computing platform. This platform was referred to as Community Agricultural Model Platform (CAMP). CAMP was designed to launch individual crop models through an application programming interface (API) implemented within the container. The platform consisted of subsystems for management, storage and administration of containers. In particular, a web based interface for data exchange was provided within each container where an object-oriented programming approach was used to facilitate the implementation of the API for individual crop models. As a case study, the CAMP was used to estimate the heading dates of rice using five models that had different computing environments such as operating system and programming languages. The ensemble estimate showed errors comparable to the better-performing individual models and reduced the influence of large errors produced by individual members although calibration-related uncertainty remained. Our results demonstrated that CAMP would facilitate the application of multi-model ensemble simulations using crop models developed and shared by crop model developers. This would benefit a wide range of stakeholders allowing for the ensemble simulations to make reasonable decisions on crop management practices as well as climate change impact assessment, which merits further studies to extend functionality of CAMP for Model as a Service (MaaS).