Scaling up SWAP: multi-instance parallel execution for soil-water–atmosphere-plant simulations

The wide availability of high-performance computing resources has opened possibilities to employ complex, physically-based soil-water flow models at larger scales. While detailed 1D models like SWAP (Soil-Water-Atmosphere-Plant) are ideal for modelling unsaturated flow and crop growth at field scale, their traditional design tailored for single-column applications limits their scalability. This legacy structure creates a significant challenge when attempting to perform quasi-3D distributed simulations or when coupling with regional groundwater models such as MODFLOW, where thousands of interacting columns must be solved simultaneously.

This poster presents a refactoring effort that enables SWAP to run many independent 1D columns truly in parallel within one process, supporting high-throughput simulations with minimal I/O bottlenecks. This modernization targets the needs of modern environmental data science, such as (i) memory-efficient integration with spatial libraries (e.g., NetCDF), and (ii) seamless compatibility with parameter estimation and uncertainty frameworks (e.g., PEST) which require thousands of iterative model calls. By enabling direct memory access to state variables, the updated framework removes the overhead of disk operations, facilitating autocalibration and sensitivity analysis even at scale.

A motivating application is the Kinrooi subirrigation experiment, where spatial variability in soil hydraulic properties and boundary conditions influences the effective recharge rate. The poster shows how the updated parallel approach makes it practical to explore these spatial patterns while preserving the mechanistic detail of the original model.