Agronomic Modelling Is Not Just About Building a Good Model
At Armosys, we work closely with Marco Acutis and Alessia Perego, the researchers behind ARMOSA, a Tier 3 process-based agronomic model originally developed in academia and now applied across real-world agricultural and carbon-related contexts.
One thing has become increasingly clear through this collaboration: agronomic modelling is not simply about creating a “good model.”
Representing how crops, soils, climate, water, nitrogen, and management practices interact under real farming conditions requires far more than equations alone. The real challenge lies in transforming the complexity of agricultural systems into a scientifically robust and operationally usable methodology.
This is precisely where the distinction, and complementarity, between scientific research and operational implementation becomes essential.
Agricultural systems are inherently complex
Real agricultural systems are not controlled environments. They are dynamic ecosystems where biological, physical, chemical, and management-related processes continuously interact.
Climate acts as a driving variable that cannot be controlled but whose effects must be represented. At the same time, crops are not generic entities: each species, variety, and genotype responds differently to environmental conditions. Soil itself introduces another layer of complexity through its physical, chemical, and biological properties. Then come agronomic decisions, tillage, fertilization, irrigation, crop rotations, weed management, all of which directly influence system behaviour.
These interactions ultimately generate outputs such as biomass production, soil organic carbon dynamics, CO₂ emissions, N₂O emissions, methane fluxes, and nutrient balances.
A process-based model such as ARMOSA attempts to represent these interactions not empirically, but through the numerical description of the underlying biophysical processes.
Photosynthesis, radiation interception, biomass accumulation, nitrogen cycling, soil water dynamics, organic matter decomposition: each process is translated into mathematical relationships capable of reproducing how the agroecosystem evolves over time.
The goal is not simply to produce an output, but to create a reproducible quantitative representation of the system itself.
A model alone is never enough
One of the most important lessons from agronomic modelling is that possessing a sophisticated model is not sufficient.
Even the most advanced process-based models cannot rely purely on theory. Agricultural systems always contain elements of variability that require empirical adjustment. Certain coefficients and parameters depend on local conditions, management practices, crop varieties, or environmental contexts.
This is where methodology becomes fundamental.
Calibration is not a secondary technical step; it is an essential scientific process. It involves optimizing biophysical parameters so that model simulations align with measured experimental observations.
But calibration alone is not enough.
A model can easily become overfitted, highly accurate for the specific dataset used during calibration, yet unreliable when applied elsewhere. This is why validation is equally critical. Independent datasets must confirm that the model remains consistent across different years, sites, and conditions.
The scientific robustness of modelling therefore depends not only on equations, but on rigorous methodologies governing:
- calibration,
- validation,
- data independence,
- uncertainty assessment,
- interpretation of outputs,
- and operational deployment.
Only after this process can simulations become scientifically reliable enough to support practical decision-making.
From academic research to real-world applications
The transition from academic research to operational implementation revealed another crucial dimension of modelling.
In research environments, models are typically developed using highly controlled experimental datasets: precise meteorological measurements, detailed soil analyses, carefully monitored management records, and extensive field observations.
Real-world agricultural applications are very different. Farm-level data are often incomplete, uncertain, or heterogeneous. Meteorological stations may not exist near the fields being simulated. Soil properties may only be partially known. Management histories can be fragmented or inconsistently recorded.
Moving ARMOSA into operational contexts therefore required not only scaling the model itself, but also developing methodologies capable of handling uncertainty in input data. This became particularly important in carbon-related applications, where outputs may contribute to carbon accounting, reporting frameworks, MRV systems, or audit processes. In these contexts, uncertainty is not a side issue. It must be explicitly quantified, understood, and communicated.
Why process-based modelling matters
One of the key strengths of ARMOSA lies in its process-based structure.
Empirical models often work well only within the specific conditions where they were calibrated. Outside those conditions, their behaviour becomes difficult to justify scientifically because they rely primarily on statistical correlations between inputs and outputs.
Process-based models operate differently. Because they explicitly represent the underlying biophysical mechanisms driving the system, they are inherently more transferable across different geographies, farming systems, climates, and management practices.
The adaptability does not come from forcing outputs to match observations through arbitrary fitting. It comes from ensuring that the relevant processes themselves are represented correctly.
As Prof. Acutis explains, the crucial point is not simply connecting inputs to outputs, but representing the physical and biological processes occurring in between.
This is what allows the same modelling framework to remain scientifically coherent across highly diverse agricultural contexts.
Scientific rigour in operational environments
Applying a model operationally also requires preserving scientific interpretability. Model outputs cannot become detached from the assumptions and processes that generated them. This is why a large part of the work in operational modelling focuses on data preparation, input consistency, and methodological transparency. The quality of outputs depends directly on the quality and understanding of the underlying inputs.
At Armosys, this means treating modelling not as a black-box software exercise, but as a structured scientific workflow where assumptions, limitations, uncertainty, and data quality remain visible throughout the process. This is especially important in carbon projects, where outputs may influence reporting, verification, or financial decisions.
Continuous evolution through real-world application
One of the most valuable aspects of the collaboration between academia and Armosys is the continuous feedback loop between scientific research and operational deployment. Real-world implementation exposes the model to a scale and diversity of conditions that would be impossible to reproduce within a single research institution alone.
Operational applications generate new datasets, new calibration opportunities, new validation scenarios, and new challenges related to scalability, APIs, remote sensing integration, and heterogeneous agricultural systems. These experiences continuously feed back into the scientific evolution of ARMOSA itself. In this sense, the model is never truly “finished.” Agronomic modelling should not be understood as a static tool, but as an evolving scientific process that improves through continuous interaction between research, field observations, operational implementation, and methodological refinement.
Beyond outputs: modelling as scientific methodology
Ultimately, the value of agronomic modelling does not lie solely in producing numbers.
Its value lies in the scientific methodology surrounding those numbers:
how processes are represented,
how uncertainty is treated,
how models are calibrated and validated,
how assumptions are maintained,
and how outputs remain scientifically interpretable when applied to real agricultural systems.
This is the philosophy behind ARMOSA and the approach we pursue at Armosys. What we provide is not simply a model output. It is a structured scientific methodology designed to represent, interpret, and support decision-making within the complexity of real-world agricultural systems.
Join the conversation on LinkedIn