Guillaume Jégo1, René Morissette2, Gilles Bélanger1, Athyna N. Cambouris3, Nyiraneza Judith4 and Bernie Zebarth5, (1)Agriculture & Agri-Food Canada, Quebec, QC, Canada (2)Agriculture Canada, Québec, QC, Canada (3)Quebec Research and Development Centre, Agriculture & Agri-Food Canada, Québec, QC, Canada (4)Charlottetown Research and Development Centre, Agriculture & Agri-Food Canada, Charlottetown, PE, CANADA (5)Fredericton Research and Development Centre, Agriculture and Agri-Food Canada, Fredericton, NB, Canada
The ability of process-based crop models to adequately predict yield and N uptake of potato (Solanum tuberosum L.) for a large range of fertilization rates under the conditions of eastern Canada has never been tested. Our objectives were 1) to calibrate and validate the STICS crop model for the cultivars Shepody (determinate, mid-season maturity) and Russet Burbank (indeterminate, late maturity) with cultivar-specific N dilution curves, and 2) to quantify the gain of model performance associated with the use of specific N dilution curves instead of a single generic curve. Several datasets including leaf area index (LAI), plant biomass, tuber biomass, and N uptake for several N rates (0 to 280 kg N ha-1) and collected at Charlottetown (PEI, Canada), Fredericton (NB, Canada), and Québec (QC, Canada) were used. Model calibration was done with one dataset from Charlottetown for Shepody and one dataset from Québec for Russet Burbank, while all other datasets were used to validate STICS predictions. For Shepody, LAI validation was good with a normalized root mean square error (NRMSE) of 27 %, while it was about 30 % for plant and tuber biomass. NRMSE for plant and tuber N uptake were 25 % and 31 % respectively. For Russet Burbank, LAI predictions were better than for Shepody with a NRMSE of 23 %. However, plant biomass predictions were similar with a NRMSE of 29 %. Tuber biomass prediction was less efficient with a NRMSE of 35 %, as well as plant and tuber N uptake with 41 %. For Shepody, differences in model performance were relatively small with the use of cultivar-specific N dilution curve instead of a single generic one (relative RMSE variation lower than 9 %). The model performance improvement with specific dilution curves was larger for Russet Burbank with a RMSE decrease of 35 % for LAI and 19 % for tuber N content.