Bayes Linear Emulation of Simulated Crop Yield

TL;DR

This paper develops a Bayes linear emulator to efficiently predict crop yields from complex, high-dimensional simulation data, enabling faster analysis of environmental and agricultural scenarios.

Contribution

It introduces a novel application of Bayes linear methods to emulate high-dimensional crop yield simulations, improving computational efficiency and diagnostic capabilities.

Findings

Successful emulation of crop yield data

Effective diagnostics for emulator validation

Potential for large-scale environmental modeling

Abstract

The analysis of the output from a large scale computer simulation experiment can pose a challenging problem in terms of size and computation. We consider output in the form of simulated crop yields from the Environmental Policy Integrated Climate (EPIC) model, which requires a large number of inputs such as fertiliser levels, weather conditions, and crop rotations inducing a high dimensional input space. In this paper, we adopt a Bayes linear approach to efficiently emulate crop yield as a function of the simulator fertiliser inputs. We explore emulator diagnostics and present the results from emulation of a subset of the simulated EPIC data output.