Figures (8)  Tables (6)

    • Figure 1. 

      Location of study vineyards (Source: Esri, USGS).

    • Figure 2. 

      Total daily potential evapotranspiration and total daily rainfall provided by an on-site weather station, with cultivation practices implemented during the two growing seasons.

    • Figure 3. 

      Overview of the workflow. UAV = uncrewed aerial vehicle; DOY = day of the year; NDVI = normalized difference vegetation index; ECa = apparent electrical conductivity; PET = potential evapotranspiration.

    • Figure 4. 

      An illustration of generating a downscaled 3 m UAV image for each UAV survey date. (a) UAV image is presented as RGB composites, with measured Ѱstem presented as yellow points. (b) Calibrated 0.05 m UAV image exhibiting Ѱstem estimation. (c) Green pixels are pure pixels representing grapevines after clipping pixels of interrow, overlaid with 3 m red grids. (d) Downscaled 3 m UAV image after spatial aggregation of pixels in red grids.

    • Figure 5. 

      Box plot of measured Ψstem collected during two growing seasons at Pencarrow (n = 86) and Wharekauhau (n = 62) vineyards. X symbols refer to the average values on the survey dates. Horizontal lines in the boxes refer to median values on the survey dates.

    • Figure 6. 

      Scatter plots between predicted Ѱstem and measured or reference Ѱstem (kPa) for the training and test sets of (a) the calibration of UAV images acquired in 2020/2021. (b) Calibration of UAV images acquired in 2021/2022. (c) Calibration of satellite (PS) images. (d) Prediction of Ѱstem in 2020/2021. Red dashed lines are 1:1 lines. Samples from the two study vineyards are considered collectively in each regression model.

    • Figure 7. 

      The temporal patterns of normalized Ψstem prediction (lines) during the growing season in 2021/2022, with the measured Ψstem measurements (points) for (a) Pencarrow and (b) Wharekauhau. The shaded bands bordering the lines encompass one standard deviation.

    • Figure 8. 

      Box plots of predicted Ψstem values, Ψstem values acquired from 3 m calibrated UAV images, and measured Ψstem values for each survey date at the (a) Pencarrow and (b) Wharekauhau vineyards. UAV = uncrewed aerial vehicle, and Ψstem = stem water potential.

    • Acquisition dateData sourceTime gap (d)
      15/11/2020Satellite−
      16/11/2020Satellite−
      27/11/2020Measured/UAV−
      02/12/2020Satellite−
      04/12/2020Measured/UAV1
      05/12/2020Satellite
      15/12/2020Satellite−
      17/12/2020Satellite−
      31/12/2020Satellite−
      04/01/2021Satellite−
      14/01/2021Measured/UAV/
      Satellite      		0
      15/01/2021Satellite−
      22/01/2021Measured/UAV1
      23/01/2021Satellite
      26/01/2021Satellite−
      01/02/2021Measured/UAV1
      02/02/2021Satellite
      23/11/2021UAV−
      29/11/2021Measured/UAV−
      09/12/2021Measured/UAV−
      11/01/2022Measured/UAV−
      21/01/2022Measured/UAV−

      Table 1. 

      Details of data acquisition for measured Ψstem, DJI Phantom 4 multispectral UAV imagery, and PlanetScope satellite imagery. The orange letterings indicate the UAV-satellite image pairs used in the second calibration. UAV = uncrewed aerial vehicle.

    • Vegetation indexFormulaReference
      Red Edge Chlorophyll Index(NIR/Red edge) − 1[39]
      Difference Vegetation IndexNIR − Red[40]
      Enhanced Vegetation Index2.5 × (NIR − Red)/(NIR + 6 * Red − 7.5 × Blue + 1)[41]
      Excess Green Index2 × Green − Red − Blue[42]
      Green Normalized Difference
      Vegetation Index      		(NIR − Green)/(NIR + Green)[43]
      Modified Chlorophyll Absorption Ratio Index((Red edge − Red) − 0.2 × (Red edge − Green)) ×
      (Red edge/Red)      		[44]
      Modified Soil Adjusted
      Vegetation Index      		(2 × NIR+1 − ((2 × NIR + 1)2 − 8 × (NIR − Red))1/2)/2[45]
      Modified Triangular
      Vegetation Index      		1.2 × (1.2 × (NIR − Green) − 2.5 × (Red − Green))[46]
      Normalized Difference
      Red Edge Index      		(NIR − Red edge)/(NIR + Red edge)[47]
      Normalized Difference
      Vegetation Index      		(NIR − Red)/(NIR + Red)[48]
      Normalized Difference
      Green/Red Index      		(Green − Red)/(Green + Red)[40]
      Optimized Soil Adjusted
      Vegetation Index      		(NIR − RED)/(NIR + Red + 0.16)[49]
      Red: Green RatioRed/Green[50]
      Simple RatioNIR/Red[51]
      Transformed Chlorophyll
      Absorption Reflectance Index      		3 × ((Red edge − Red) − 0.2 × (Red edge − Green) ×
      (Red edge/Red))      		[52]
      Visible Atmospherically
      Resistant Index      		(Green − Red)/(Green + Red − Blue)[53]

      Table 2. 

      List of vegetation indices used in this study.

    • PredictorNoteNumber
      DOYThe number is added on along the growing
      season.      		1
      NDVICollected in late November for the 2020/2021 and 2021/2022 seasons separately.1
      ECa−1
      Elevation−1
      Slope−1
      Total daily rainfallThe rainfall amounts on each day of the 30-day
      period beforehand was used as a predictor.      		30
      Total daily PETThe PET amounts on each day of the 30-day
      period beforehand was used as a predictor.      		30
      Irrigation and
      Fertigation      		The water input amounts, either sourced from
      irrigation or fertigation, on each day of the 30-day period beforehand was used as a predictor.      		30
      Plucking and TrimmingThe occurrence of the events on each day of the 30-day period beforehand was used as a
      predictor.      		30

      Table 3. 

      A summary for the predictors used in developing the Ѱstem prediction model. DOY = day of the year; NDVI = normalized difference vegetation index; ECa = apparent electrical conductivity; PET = potential evapotranspiration.

    • Modeling purposeRegression modelData size of measured
      or reference Ψstem      		RMSE of the
      training set (kPa)      		RMSE of the
      test set (kPa)
      Calibration of UAV images
      acquired in 2020/2021
      (Fig. 3a)      		Multilayer
      perceptron      		8596113
      Calibration of UAV images
      acquired in 2021/2022
      (Fig. 3a)      		Random forest
      regression      		63121106
      Calibration of
      Satellite images (Fig. 3e)      		Random forest
      regression      		42,2344759
      Prediction of Ψstem (Fig. 3g)Random forest
      regression      		151,5802531

      Table 4. 

      Results of modeling performance. UAV = uncrewed aerial vehicle; Ψstem = stem water potential; RMSE = root mean square error.

    • PencarrowWharekauhau
      r0.89*0.87*
      Significance levels are noted as * when p ≤ 0.001.

      Table 5. 

      Results of similarity analysis, presented by the Pearson correlation coefficient (r), represent the consistency between predicted and reference Ψstem maps across four survey dates for the two study vineyards.

    • Survey dateVineyardData sourceMeanSDCV
      29th November 2021PencarrowPredicted Ψstem6095.590.92
      Ψstem from 3 m calibrated UAV56218.113.22
      Measured Ψstem585113.9319.47
      WharekauhauPredicted Ψstem6039.261.53
      Ψstem from 3 m calibrated UAV56521.863.87
      Measured Ψstem52887.3116.55
      9th December 2021PencarrowPredicted Ψstem5935.030.85
      Ψstem from 3 m calibrated UAV54715.192.78
      Measured Ψstem40064.2516.05
      WharekauhauPredicted Ψstem5808.951.54
      Ψstem from 3 m calibrated UAV54817.423.18
      Measured Ψstem45382.1818.14
      11th January 2022PencarrowPredicted Ψstem6624.840.73
      Ψstem from 3 m calibrated UAV68119.532.87
      Measured Ψstem672204.2730.39
      21st January 2021PencarrowPredicted Ψstem6706.731.01
      Ψstem from 3 m calibrated UAV71722.863.19
      Measured Ψstem803233.2729.03
      WharekauhauPredicted Ψstem66810.351.55
      Ψstem from 3 m calibrated UAV74024.673.34
      Measured Ψstem95799.3510.39

      Table 6. 

      Summary statistics for predicted Ψstem, Ψstem acquired from 3 m calibrated UAV images, and measured Ψstem for each survey date at the Pencarrow and Wharekauhau vineyards. UAV = uncrewed aerial vehicle; SD = standard deviation; CV = coefficient of variation.