Porting a process-based crop model to a high-performance computing environment for plant simulation

Increasing concerns about food security have stimulated integrated assessment of the sustainability of agricultural systems at regional, national and global scales with high-resolution. Traditionally, the process-based agricultural models are designed for field scale studies that obtain inputs, run the simulations and provide outputs through the graphic interface. The graphic interface based model dose not suit for modelling practices requiring a large number of simulations. Here, we developed a high performance approach which concurrently executed the Agricultural Production Systems sIMulator (APSIM) simulations using parallel programming techniques. In this approach, an APSIM simulation template with replaceable parameters was firstly designed, and new simulations based on the template was then constructed by dynamically replacing parameters of climate, soil and management options. We parallelized the batched running method in a shared-memory multiprocessor system using Python's Multiprocessing module. We tested the approach with a case study that simulated the productivity of continuous wheat cropping system during 20 years period along the Australian cereal-growing regions under management practices of 5 levels nitrogen application and 3 stubble management practices. More than 170 K runs were finished in 43h by using 64 workers, achieved a speedup ratio of 60. The parallelized method proposed in this study makes large-scale and high-resolution agricultural systems assessment possible.