Grapesoil: An integrated model to simulate water and nitrogen fluxes in diversified vineyards

Abstract

Cover crops in vineyards bring numerous benefits, including enhanced soil health, improved water infiltration, and potential pest reduction. However, they also present risks, such as reduced vine vigour and yield due to competition for water and nutrients (Celette & Gary 2013, Garcia et al., 2018). Understanding these interactions is crucial for vineyard management. To address this challenge, the GrapeSoil model was recently developed to simulate the effects of various practices on grapevine growth and yield, vineyard water balance and N balance at field scale. The vineyard and cover crop water balance is based on Walis, which was initially developed by Celette et al. (2010) and subsequently validated and utilized (Guilpart et al. (2014) and Delpuech & Metay (2018)). Constructed in R, the GrapeSoil model is built on independent functions and requires approximately 250 parameters and 20 inputs. These inputs may be sourced from surveys with winegrowers, field measurements, and literature reviews. The present study evaluates the consistency and sensitivity to key soil and plant parameters of the preliminary GrapeSoil model. This evaluation used a dataset collected by Nicolas Guilpart (2014) from a Mediterranean vineyard (South of France). Key model outputs, such as grapevine growth, yield, and soil water and N levels, were compared to observations using statistical indicators (Root Mean Square Error ‘RMSE’, Relative Root Mean Square Error ‘RRMSE’, and Efficiency). Sensitivity analysis explored the influence of parameters like berry number, canopy height, soil water holding capacity, and initial soil inorganic N levels on model outcomes. Results showed that GrapeSoil properly simulated the Fraction of Transpirable Soil Water (FTSW). Yet, the RMSE was ~ 0.1 across all modalities, representing less than 20% of variations, with efficiency around 0.9 for most modalities. All modalities combined, the RMSE remained at 0.1, representing an RRMSE of 21% and a model efficiency of 0.79. Simulations of soil inorganic N content lacked accuracy due to limited and inconsistent data. For most modalities, the RMSE was ~ 10 kg N/ha, which represented a low N amount. The RRMSE was 91% across all modalities and model efficiency -1.06. Lastly, the model permitted to predict yield and biomass with an error of 66% (from 29 to 91%). In the actual version of the model berry number is fixed. Further improvement of the model is required to simulate berry numbers and berry fresh weight.