Why PlantSimEngine?​#

PlantSimEngine helps you build plant simulations from process models that you can inspect, test, reuse, and replace. You choose the scientific equations, the plants, organs or other entities they describe, and the exchanges between them. The engine connects their inputs and outputs and runs the resulting simulation in Julia.

This is useful when your research question requires a particular combination of processes or assumptions:

• Crop modelling: assemble development, light interception, biomass, or water-balance processes at the resolution your question needs. Start with a coupled simulation for a canopy.

• Canopy ecophysiology: connect weather, local conditions and organ processes, including calculations that must iterate together. The MAESPA-style synthesis demonstrates how these calculations can work together, using teaching models.

• Functional–structural plant modelling: apply processes to organs, connect their results to the plant, and update the structure during growth. Start with one multiscale plant.

The scientific models, parameter sets, input data, and validation for your species or experiment come from your own work or model packages such as PlantBiophysics.jl. PlantSimEngine provides the tools for connecting and running them.

Compare hypotheses without rewriting the whole simulation​#

A process can have several model implementations. For example, you may want to compare two biomass-production equations or introduce water limitation into a previously radiation-driven model. Each implementation keeps its own parameters and equations; the scenario specifies where it runs and how it receives inputs.

Replacement depends on the model's interface: the values it needs and produces, and what those values mean. The new model must provide the outputs other models need, and its own inputs must be available. A variable contract records information such as units, whether a quantity is per plant or per unit area, and whether it is a rate or an accumulated amount. The same variable name alone is insufficient. Moving from a quantity per unit ground area to a total per plant, for example, needs an explicit conversion.

To replace a model only for selected plants, its inputs, outputs, and other requirements must match the original model. A replacement that needs an extra input, such as soil water content, requires changes to the simulation setup. See model compatibility and replacement.

Model authors can read each equation together with its inputs and outputs. Researchers assembling models can see which equations and assumptions were chosen in the simulation setup. Both can test a component before investigating its behaviour in the full system.

Choose the representation that answers your question​#

An object can represent a leaf, a plant, a canopy layer, a soil compartment, or another entity you define. You can work with one object, several objects, or a hierarchy imported from a multiscale tree graph. Geometry is optional. Different models can be applied to different selections of these objects.

The same process equation can be reused over compatible objects while the scenario handles their selection and connections. A plant-level model can, for example, read values from its own leaves. The modeller decides how to combine these values, for example by summing leaf areas, and supplies any required area, mass, or time conversion. Changing resolution also requires checking the assumptions and validity of the chosen models. See value coupling across objects.

When organs appear, disappear, or move within the plant structure, PlantSimEngine updates the affected model selections and connections. This lets growth change which organs contribute to a plant-level calculation while retaining their identities and historical outputs. Follow Modify Plant Structure.

Connect different time steps and control scientific iteration​#

Canopy exchange may run hourly while development runs daily. Each model can run at its own time step, with explicit rules for reading values produced at another time step. The appropriate averaging, accumulation, or rate-to-amount conversion is a scientific choice. The cadence tutorial makes those choices explicit.

Some calculations also need a controller: an energy-balance algorithm may call gas-exchange models repeatedly while finding an accepted leaf temperature. Explicit model calls let the controller manage that iteration and record the accepted result once. See advanced execution.

Understand what the simulation will do​#

The Authoring and Diagnostics interfaces expose model declarations, missing inputs, the sources of values, execution order, local environmental conditions, and which results are saved. You can inspect why a particular leaf receives a value, which model supplies it, and when it is updated. The graph viewer provides another view of how the models are connected.

These reports help separate a coupling problem from a problem in an equation or its assumptions. Declared units and physical meanings help reveal incompatible connections; scientific validation still needs suitable observations, reference results, and tests. See the model authoring API and model testing guide.

Keep repeated simulations practical​#

PlantSimEngine prepares model selections and connections before repeated execution, avoids copying input values where possible, and processes groups of similar objects together. Model equations remain ordinary Julia calculations whose performance can be measured and improved.

There is application-level performance evidence: the 2025 PlantBiophysics.jl paper reports a median of 5.3 microseconds for one leaf and one time step of its coupled energy-balance, photosynthesis, and stomatal-conductance benchmark. That result concerns the implementations, versions, inputs, and hardware used in the study. For your scenario, measure initialization, repeated execution, structural changes, and saving results separately using the benchmarking guidance. A public parallel or distributed executor remains planned work.

AI-assisted model development​#

An AI coding agent is software that can read and edit code and run tests with your development tools. Small process models, explicit inputs and outputs, declared units, and reports on model connections give it concrete information to read and check. An agent can help draft a model, compare declared interfaces, assemble a scenario, inspect its connections, and run tests with the tools available in your development environment.

PlantSimEngine provides a versioned AI agent skill, a catalog of available models, and the same authoring tools used by people. You supply the coding agent and its execution environment. Model assumptions, physical conversions, supporting references, and scientific validation remain the responsibility of the researcher. Begin with Implement a basic model to see the complete model, an independent equation test, and its use in a simulation.

How this fits among plant-modelling tools​#

Modularity and multiscale modelling have a substantial history. APSIM supports replaceable model interfaces; DSSAT combines crop models with data and experimental workflows; OpenAlea provides components, multiscale structures, and plant-geometry tools; and GroIMP integrates FSPM modelling and visualisation. Cropbox also uses Julia and dependency analysis for declarative crop modelling.

PlantSimEngine's contribution is the combination described here: readable process equations, explicit inputs and outputs, a choice of how to represent plants, different time steps, controlled iteration, changing structures, and calculations you can inspect. It is a useful fit when you want to develop or assemble that scientific model combination yourself, while keeping its choices visible and its components independently testable.