Article Figures & Data
Figures
- Figure 1
Location of the midpoint of the sampled panels within the 2.6 ha study block at the Cornell Lake Erie Research and Extension Laboratory, Portland, NY.
- Figure 2
Maps of some key dependent and independent model variables to illustrate spatio-temporal patterning in the block. All data are presented on a common standardized (0 - 1) legend based on the maximum and minimum values in each layer.
Tables
- Table 1
Day of the year (and date) for three key phenological stages in 2019 to 2021 at the Lake Erie Research and Extension Laboratory. Bloom +30 is same date in July from the June date.
- Table 2
Vegetative indices (VIs) calculated from the three available bands of the CropCircle 430 canopy sensor.
- Table 3
Dates of canopy sensing surveys during the three years of the study translated into phenological time (before or after budbreak, floraison, and veraison) to indicate the asynchronicity of vine phenology between years. DABB, days after budbreak; DBF, days before floraison (bloom); DAF, days after floraison; DBV, days before veraison; DAV, days after veraison.
- Table 4
Explained variance from cross-validation of four different models using different available inputs applied to two different regression approaches (stepwise-multivariate linear regression [S-MLR] and random forest regression [RFR]) across three years (2019 to 2021). The models were recalibrated for each year before cross-validation using relevant available variables. The best-performing model in each year is indicated in bold; RFR results are in italics.
- Table 5
Mean average error (MAE) (Mg/ha for yield and kg/vine) from cross-validation of four different models that used different inputs (M1 to M4) applied to two different regression approaches (stepwise-multivariate linear regression [S-MLR] and random forest regression [RFR]) across three years (2019 to 2021). The models were recalibrated for each year using the relevant available variables. The best-performing model in each year is indicated in bold; RFR results are in italics. The higher yield MAE in 2021 is associated with a much higher mean yield in this year.
- Table 6
The key predictors and timing of data acquisition (expressed in phenological time) in each year for the best-performing models identified from Tables 4 and 5. For the random forest regression (RFR), the first five predictors are shown, followed by their predictive power from the cross-validation in parentheses. For the stepwise multi-linear regression (S-MLR), the order reflects the stepwise progression, with the dominant predictor at each step given along with the number of times (out of 10) it was selected during cross-validation. Acronyms for vegetative indices (VIs) are the same as in Table 2.
Supplemental Table 1 The key predictors and timing of data acquisition (expressed as phenological time) in each year from all models generated in the study. For the random forest regression (RFR), the first five predictors are shown, with the prediction power from the cross-validation given in parentheses. For the stepwise multi-linear regression (S-MLR), the order reflects the stepwise progression, with the dominant predictor at each step given along with the number of times (out of 10) it was selected in the cross-validation process. Acronyms for vegetative indices (VIs) are the same as in Table 2 in the main manuscript. Acronyms for phenological stages are the same as in Table 3 in the main manuscript.
Vol 74 Issue 1
Jump to section
Related Articles
Cited By...
More from this TOC section
Similar Articles
