2.5 Model calibration and validation
The full dataset (n=372) was split into two as calibration dataset with
70% (n=260) samples for model development and validation dataset with
30% (n=112) samples to assess the performance of the model
independently. The spectral data were calibrated using the laboratory
estimated soil properties by five multivariate techniques: (a) linear -
PCR and PLSR and (b) non-linear - MARS, RF and support vector regression
(SVR). The models were evaluated for prediction accuracy using model
evaluation parameters like R2, mean bias error (MBE),
RMSE and RPD. The prediction accuracy of different models was
categorized based on RPD as excellent (RPD > 2), acceptable
(2 ≥ RPD ≥ 1.4) and non-reliable (RPD > 1.40) (Changet al. , 2001). Generally, high R2 and RPD and
low RMSE indicate a model with good predictive ability. We, in the
present study, calculated ranks for different models considering the
R2, MBE, RMSE and RPD. Based on the ranking for
calibration and validation, an average rank was calculated to interpret
the prediction performance of the model. Lower the value of the rank,
better was the prediction performance.