Table 3.

Calibration and validation statistics of the best models obtained to predict the midday stem water potential (ψ_s) in grapevines under field conditions from on-the-go NIR spectroscopy at the sub-replicate unit, and field replicate scales.

Modeling variables and scale			Calibrationa		Cross validationa				Predictionc
					10-fold		LODOb
Season	Canopy side	Spectral treatment	RMSE	${\text{R}}_{\text{c}}^2$	CV-RMSE	${\text{R}}_{\text{c}\text{v}}^2$	CV-RMSE	${\text{R}}_{\text{c}\text{v}}^2$	RMSE	${\text{R}}_{\text{p}}^2$
	Sub Replicate Unit
	East	SNV+D1W15	0.156	0.86	0.171	0.83	0.192	0.77	0.151	0.86
	West	SNV+D1W15	0.195	0.78	0.214	0.73	0.251	0.71	0.188	0.78
	East & West	SNV+D1W7	0.168	0.83	0.190	0.79	0.253	0.72	0.173	0.81
	Field Replicate
2015	East	D1W15	0.173	0.90	0.171	0.82	0.203	0.79	0.150	0.85
	West	SNV+D1W15	0.160	0.85	0.207	0.74	0.222	0.82	0.194	0.74
	East & West	D1W7	0.132	0.89	0.189	0.79	0.230	0.81	0.167	0.84
	Sub Replicate Unit
2016	East	SNV+D1W15	0.103	0.79	0.119	0.71	–	–	0.132	0.68
	West	D1W7	0.111	0.77	0.131	0.68	–	–	0.131	0.54
	East & West	D1W7	0.106	0.78	0.128	0.68	–	–	0.133	0.62
	Sub Replicate Unit
2015 & 2016	East	SNV+D1W15	0.178	0.74	0.187	0.71	0.227	0.59	0.191	0.69

^aNumber of samples (n) used for the development of calibration and cross validation (10-fold) models. Season 2015: 234 for East and East & West, and 238 for West models at the sub replicate unit scale. At the field replicate level, 84 data were used for East and 86 for West, and East & West models. Season 2016: 165 samples for East, West and East &West models. Seasons 2015 & 2016: 384 samples.

^bNumber of samples (n) used for the development of cross validation models using the LODO approach. Season 2015: 318 for East and East and East & West, and 324 for West models at the sub replicate unit scale. At the field replicate level, 102 data were used for East and 104 for West, and East & West models. Seasons 2015 & 2016: 496 samples.

^cNumber of samples (n) used for prediction or external validation. Season 2015: 54 for all canopy side models at the sub replicate unit scale, and 18 at the field replicate level. Season 2016: 43 samples for East, West and East &West models. Seasons 2015 & 2016: 97 samples.

SNV, standard normal variate; DnWm, Savitzky-Golay filter with n-degree derivative, window size of m; RMSE, root mean square error (MPa); ${\text{R}}_{\text{c}}^2$, determination coefficient of calibration; CV-RMSE, root mean square error of cross-validation (MPa); ${\text{R}}_{\text{cv}}^2$, determination coefficient of cross-validation; ${\text{R}}_{\text{p}}^2$, determination coefficient of prediction; 10-fold, 10-fold venetian blind cross validation; LODO, leave one day out cross validation.