Most current crop models are point-based models, i.e. they simulate agronomic variables on a spatial footprint on which they were initially designed (e.g. plant, field, region scale). To assess their performances, many indicators based on the comparison of estimated vs observed data, can be used such as root mean square error (RMSE) or Willmott index of agreement (D-index) among others. However, shifting model use from a strategic objective to tactical in-season management is becoming a significant issue for the agronomic community, especially in a precision agriculture context. Spatialization (i.e. using point-based crop models on a different scale than its native spatial footprint) represents a solution to tackle this tactical issue. However, the use of classical indicators to evaluate spatialized point-based crop model performance raises questions, and they may not be relevant to evaluate spatialized crop model performances.

The objective of this work is to compare the evaluation of spatialized crop model performances using different indicators (aspatial : e.g. RMSE or D-index; and spatial : e.g. Cambardella index, Moran index or Geary index) for different simulation scales. Model performance is defined by an adequacy between the observed data and the simulated data and the preservation of the spatial structure of the simulated variable in relation to the observed variable. It should be noted that the preservation of the spatial structure of the simulated variable is important in a precision agriculture objective. This work focuses on a case study. The crop model WaLIS (Water baLance for Intercropped Systems) has been used to simulated vine water restriction (estimated through the predawn leaf water potential - PLWP) for a vineyard in the South of France. WaLIS was employed at different scales (field, site, zone, etc…) to generate PLWP maps and various performance indicators were applied to these maps to assess model and map accuracy. Preliminary results showed differences with model evaluation and interpretation between aspatial and spatial indicators, which lead to different indicators promoting different maps and different modelling scales as optimal. Further results will be presented in the final paper and indicator behavior with respect to the change of modelling scale will be discussed further.