Table 6 Statistics on quality of spatial inference methods.

From: PEST-CHEMGRIDS, global gridded maps of the top 20 crop-specific pesticide application rates from 2015 to 2025

Method Order of polynomials Fraction of data for calibration Number of points for calibration Number of points for validation Correlation
coefficient
R
Normal. Root mean square error
NRMSE (%)
Rank
(1 − |R|) * 100 + NRMSE
Minimum = best
Monovariate
with weighted linear combination
1
(linear)
0.07 49 656 0.701 10.33 40.2
0.2 141 564 0.674 10.80 43.4
0.5 352 353 0.649 12.90 48.0
2
(quadratic)
0.07 49 656 0.775 9.10 31.6
0.2 141 564 0.709 10.31 39.4
0.5 352 353 0.689 12.34 43.4
3
(cubic)
0.07 49 656 0.774 9.16 31.8
0.2 141 564 0.716 10.26 38.7
0.5 352 353 0.696 12.24 42.6
Multivariate
without interaction products
1
(linear)
0.07 49 656 0.719 10.74 38.8
0.2 141 564 0.752 9.25 34.1
0.5 352 353 0.772 10.48 33.3
2
(quadratic)
0.07 49 656 0.437 13.65 70.0
0.2 141 564 −0.173 15.51 98.2
0.5 352 353 −0.177 17.66 100.0
Multivariate
with interaction products
1
(linear)
0.07 49 656 0.719 10.95 39.1
0.2 141 564 0.760 9.39 33.4
0.5 352 353 0.842 8.83 24.6
2
(quadratic)
0.07 49 656 0.741 10.88 36.8
0.2 141 564 0.740 9.51 35.5
0.5 352 353 0.776 10.35 32.8
  1. Monovariate polynomials were calculated using the Matlab function polyfit, while multivariate polynomials were calculated with the Matlab function fitlm.