Redesigning a nutrient model to enable faster model development

Abstract

: Farming systems modelling is a well-established science and an important tool for analysing complex farming systems problems. An ability to model the major components of a farming system lies at the core of this approach. As a result, accurate and effective modelling of important Soil Organic Matter (SOM) stocks and processes is important in many applications of Farming Systems Models. These include various SOM and mineral pools, flows of carbon and nutrients between these pools, and losses from these processes. The SoilNitrogen model within APSIM (Holzworth et al, 2018) provides such a capability. Many researchers using the SoilNitrogen model have explored changes for improving model performance. Such efforts include i) changes in the number of soil pools, ii) flows between these pools, iii) functional forms used in calculating flows, and iv) parameters values used in these functions. However, implementation of such changes was not always straightforward because of the way models were implemented. Though there was some consideration of object-oriented approaches for abstracting the science of carbon and nutrient dynamics into instantiable models, the science was mostly captured using traditional procedural programming approaches in FORTRAN or C#. Though there was a clear conceptual model behind the specification of the model (e.g. pools, flows, algorithms), the use of traditional programming approaches did not reflect this way of thinking. Furthermore, it was difficult for researchers to understand or change model design. We used an approach, already found to be widely successful within APSIM for crops (Brown et al, 2014), to capture the needs of researchers. Classes were developed for pools, flows and functions. Users can alter the number of pools, flows between pools, the functional forms used to calculate flows, and parameter values used in functions. All can be altered via the user interface or command-line tools, thus facilitating alternate model configurations or designs within simulations to compare model structures and parameterisations (ie model comparison). This new functionality is referred to as the APSIM Nutrient model. We