What is commonly referred as either crop or cropping system simulation model is a set of interlinked mathematical representations of approaches which are abstractions of a single biological or physical process. The methodology for their evaluation has evolved in time, but it has always targeted models as unique and immutable, except for versioning, discrete units. This paper has explored aspects both of model composition in the perspective of evaluating alternate modelling approaches, and of modelling solutions evaluation. Soil temperature was chosen as case study, evaluating nine modelling solutions against a multi-year, multi-location database of field recorded time series of data. Multi-metric indices were also developed to quantify different aspects of model performance and to get a better insight on the impact of sub-model replacement. Results showed that the hybrid solution implementing the cascading model (soil water redistribution), Parton’s approach (surface temperature), and SWAT (temperature along the soil profile) led to the best compromise between agreement and robustness under the explored conditions. The model libraries used to run the analysis, in form of extensible model components, are freely available for download, and they allow for further extension of the composition exercise.