“…Various combinations of runs were used to calibrate model parameters, k w and S, for each dimension and each PC. The combined runs used for calibration were as follows: all (1 model with 24 runs used for calibration, 0 runs to predict); the leave-one-out approach with all runs but one (24 models with each 23 runs used for calibration, 24 runs to predict); the split-half approach with calibration performed using either run 3,5,6,7,10,11,13,14,18,19,22,23 (1 model with 12 runs used for calibration, 12 other runs to predict), or run 1,2,4,8,9,12,15,16,17,20,21,24 (1 model with 12 runs used for calibration, other 12 runs to predict); six runs, namely, 1, 20-24 (1 model with 6 runs used for calibration, 18 runs to predict), or two runs namely 1 and 20 (1 model with two runs used for calibration, 22 runs to predict). The root mean square error (RMSE) or the root mean square error of prediction (RMSEP) was calculated for all six approaches, both approaches essentially estimate the average distance from the modeled (RMSE) or predicted (RMSEP) f cov to the measured f cov .…”