The authors have declared that no competing interests exist.
In precision agriculture (PA) fertilizing, based on soil testing, production maps and crop nitrogen (N) demand, is the key to maximizing yields and tempering fertilizer costs. A trend study has considered the output / input relationships performed on a farm that has progressively adapted to PA procedures over two decades. The evolutions of the variability parameters of yield, comprising the repeatability coefficient of repeated plots, the vegetative vigour (NDRE) at the panicle initiation (pi) stage, and the nitrogen utilization efficiency (NUE) were monitored and compared by means of mixed linear models over a sixyear period, after the variable nitrogen (N) fertilization rate (VNFR) had been enlarged to the whole 230 ha of one farm. At pi key fertilization stage, a corrective dose was applied by tacking the correlation between Npi and the measured NDRE in strong negative mode. The evolution of the yield, for the 20122017 interval, based on 1165 ha^{1} parceldata, showed a significant yearly increase of 2.3% more than the regional trend (+0.5%). The variability parameters of the yield, that is, the standard deviation (+7.3%), range (+7.1%), coefficient of variation (+5.4%) and maximum (+2.1%) were enhanced over the years, but the minimum remained stable. The repeatability of the parcel yield generally appeared low (r = +0.31), but it tended to increase by 8.3% year^{1} (P = 0.018). At the same time, the vegetational vigour also showed significant increases of the NDRE means (+3.0%) as well as of the maximum (+0.8%), but also large oscillations in the standard deviation and in the coefficient of variation. No significant regression of the NDRE on the coefficient of variation of the yield was established. The favorable increase in yield was found to be independent of the distributed Ntotal. A strong negative correlation (imposed) between Npi and NDRE (0.90) and a negative correlation with production were observed for a sample field (but in the area of maximum production). It is recommended that a partial correlation between Yield and Ntot should be considered in the I /O features for a parity of NDRE, which apparently decreases the negativity of the relationship. In short: with the same total input of N, the PA increased the yield, but also its variability – and it did not reduce the variability as predicted by the theory  by strengthening the repeatability. This is an evidence that in many of the parcels with minimum yield the limiting factors cannot be referred to the N availability.
After long and detailed studies focused on production models developed in universities
The progress in yield raising and homogenizing by use of the fertilization maps may be quickly achieved at the beginning of the process
Since the PA is welcome in rice cultivation, in this work we aimed to study in retrospect the productive consequences of its application on a PA Italian rice farm, oriented to obtain maximum yields. In this paper we are concerned in featuring the trends in the variability parameters, obtaining some lessons from fields.
A trend study has considered the output / input relationships performed on a farm that has progressively adapted to PA procedures over two decades. The evolutions of the variability parameters of yield, comprising the repeatability coefficient of repeated plots, the vegetative vigour (NDRE) at the panicle initiation (pi) stage, and the nitrogen utilization efficiency (NUE) were monitored and compared by means of mixed linear models over a sixyear period, after the variable nitrogen (N) fertilization rate (VNFR) had been enlarged to the whole farm.
The PA experience started in 1998 with a harvester which was equipped with a yield meter, and then with a yield monitor, a satellite receiver and a light bar to assist manual drive. The farm now spreads fertilizers using a VRT technology, following prescription maps or real time signals of vigor sensors using an OptRx by AgLeader (
Before Npi administration, a drone was used to assess the mean and probable range of the NDRE readings of the sensor, and to fix the thresholds. Following an RTK signal, herbicides are sprayed by a VRT sprayer, which controls the opening and closing of the nozzles lead from a GPS signal, and yield maps are obtained by an harvester equipped with an AgLeader yield meter. All the technology had to be adapted to the soil (
Items  Units  Worst  Best 
Average Yields  t ha^{1}  7.30  8.29 
Gravel (>2mm)  mg kg dry sub.^{1}.  12.57  16.00 
Sand (20.5 mm)  mg kg dry sub.^{ 1}  312.71  332.71 
Silt (0.50.02 mm)  mg kg dry sub.^{ 1}  582.29  565.14 
Clay (<0.02 mm)  mg kg dry sub.^{ 1}  105.00  102.14 
pH H_{2}O  pH units  5.70  5.77 
pH CaCl2  pH units  4.97  5.01 
CaCO_{3} TOT  mg kg dry sub.^{ 1}  0.00  0.00 
CaCO_{3} ACTIVE  mg kg dry sub.^{ 1}  0.00  0.00 
Organic Matter (calculated)  mg kg dry sub.^{ 1}  22.57  22.00 
Organic C (Dumas)  mg kg dry sub.^{ 1}  13.11  12.76 
N TOT  mg kg dry sub.^{ 1}  1.30  1.24 
C/N rate  9.91  10.23  
Cationic exchange capacity  meq 100g dry sub.^{ 1}  13.46  13.11 
Exchangeable Ca  meq 100g dry sub.^{ 1}  4.43  4.22 
Exchangeable Mg  meq 100g dry sub.^{ 1}  0.86  0.82 
Exchangeable K  meq 100g dry sub.^{ 1}  0.24  0.20 
Exchangeable Na  meq 100g dry sub.^{ 1}  0.04  0.04 
Basis saturation  %  41.29  40.14 
Ca/Mg rate  5.17  5.20  
Mg/K rate  3.87  3.99  
Assimilable P  mg kg dry sub.^{ 1}  27.00  29.86 
Yield data from each 19x19m2 mesh plot have been elaborated for the 201217 interval (No. 1165 ha^{1}). Yield repeatability was calculated as a Pearson correlation of the plotyields within field between consecutive years. Several mixed models (PROC MIXED by SAS 9.0), concerning the yield and NDRE, were applied to the statistical parameters of the fields, whereby the effects of the field or cultivar were considered fixed and the year was taken as the random factor in order to estimate the linear trend. Conversely, when the years were considered fixed, and the fields were considered as the random factor, a least squares solution was found. The covariation of the variability between the different traits was enhanced by fitting a linear or nonlinear regression of the estimates. A Friedman’s test for paired samples was used to compare the yield averages of the farm
Place: r12; r13 and r23 then r12.3 = (r12 – r13 * r23) / ((1r13^2)^.5* (1r23^2)^.5)).
Place: regression b1/2 = covariance12 / stdev2 ^2; regression b2/1 = covariance 12 / stdev1 ^2 then (b1 * b2) = r12^2.
A yearly increase of +2.3% more than the +0.5% of the regional features (
Year  Ntotal  Yield  NUE  NW Italy yield 
kg ha^{1}  t ha^{1}  kg N kg^{1}  t ha^{1}  
2012  165.0  7.47  45.30  6.80 
2013  167.0  7.62  45.63  6.63 
2014  165.0  7.65  46.38  6.44 
2015  160.8  7.70  47.88  6.67 
2016  162.2  8.21  50.60  6.77 
2017  184.3  8.36  45.39  6.92 
Mean  167.4  7.84 a  46.9  6.70 b 
Trend\year  1.58  1.63  0.8  0.8 
Linear trend %  0.9%  2.3%  1.7%  0.5% 
P.value linear  0.053  0.009  0.063  0.213 
P.value quadratic  0.033  0.067  0.214  
P.value cubic  0.023  0.192 
The statistics of the yield of the single plots confirmed the +2.3% year^{1 }trend(
Mean  St.dev  Coef.var  Max  Min  Range  Repeatability  
2012  7.48  0.85  0.10  8.91  5.19  3.58  0.183 
2013  7.62  1.05  0.12  9.24  4.94  4.13  0.273 
2014  7.65  0.82  0.09  8.86  5.16  3.56  0.445 
2015  7.70  1.02  0.10  8.93  4.62  4.28  0.284 
2016  8.21  1.17  0.12  9.61  4.94  4.67  0.325 
2017  8.36  1.26  0.13  10.01  4.84  5.20  0.370 
Mean  7.84  1.03  0.11  9.26  4.95  4.24  0.313 
P.value Means  0.0073  0.0004  0.0247  0.0212  0.7794  0.0032  0.002 
Trend year^{1}  0.179  0.075  0.006  0.191  0.065  0.299  0.026 
Trend %  2.3%  7.3%  5.4%  2.1%  1.3%  7.1%  8.3% 
P.value linear  0.0002  0.0003  0.0164  0.0071  0.3633  0.0003  0.0184 
P.value quadratic  0.224  0.147  0.200  
R 
0.824  0.525  0.463  0.494  0.164  0.712  0.125 
The NDRE trend grew by +3.0% year^{1} (P =0.044), with significantly different annual values (P =0.000;
Year  Mean 
St. dev  Coef. Var 
Max  Min  Range 
2012  0.254  0.032  0.136  0.319  0.194  0.157 
2013  0.295  0.033  0.140  0.375  0.227  0.215 
2014  0.311  0.031  0.100  0.361  0.231  0.150 
2015  0.316  0.037  0.119  0.370  0.212  0.214 
2016  0.290  0.032  0.113  0.347  0.195  0.163 
2017  0.317  0.035  0.115  0.354  0.236  0.181 
Mean  0.297  0.033  0.121  0.354  0.216  0.180 
P.value Mean  0.000  0.032  0.290  0.027  0.195  0.027 
Trend year^{1}  0.009  0.001  0.005  0.003  0.003  0.001 
Trend %  3.0%  3.0%  4.1%  0.8%  1.4%  0.6% 
P.value  0.044  0.240  0.833  0.054  0.355  0.077 
: correlation (mean, coef. var) =0.33
The SD spanned the years (P =0.032) and rose by +3% (not significant) with an important rise in 2015.
The coefficient of variation did not fluctuate over the years, but highlighted a descending parabolic trend (P =0.05.
The Minimum remained stable over the years, while the range and the maximum oscillated (P =0.027), but tended to increase by 1.2% year^{1}, following a nonlinear shape.
The coevolution of the two variables was apparently null (
The vegetation vigour NDRE statistics of fifteen cultivars grown in Italy are reported in
Cultivar  Mean  St. dev  Max  Min  Range  Coef. variation 
ALLEGRO  0.201  0.025  0.283  0.137  0.183  13% 
CARNISE_  0.262  0.024  0.341  0.168  0.141  10% 
RONALDO  0.285  0.033  0.381  0.142  0.202  12% 
LEONARDO  0.299  0.029  0.398  0.231  0.199  10% 
GLORIA  0.307  0.027  0.371  0.197  0.166  9% 
MARE  0.312  0.028  0.375  0.239  0.117  10% 
DUCATO  0.315  0.028  0.352  0.212  0.148  9% 
CL26  0.319  0.038  0.407  0.188  0.224  12% 
TERRA  0.321  0.031  0.387  0.199  0.219  10% 
KRISTALL  0.325  0.028  0.380  0.221  0.159  9% 
DARDO  0.333  0.036  0.409  0.197  0.192  11% 
SIRIO_CL  0.333  0.035  0.419  0.193  0.199  11% 
SOLE  0.334  0.028  0.403  0.272  0.116  9% 
SELENIO  0.336  0.109  0.397  0.223  0.189  31% 
CENTAURO  0.346  0.034  0.431  0.178  0.206  10% 
Prob.  <.0001  <.0001  0.0008  0.0084  0.1262  <.0001 
As shown in
Items  Means  St. dev.  Coef. Var  Pearson Correlations  
2017  N0  NPI  NTot  NDRE  
Yield  9.5  0.753  7.9%  0.02  0.11  0.04  0.14 
NDRE  0.318  0.026  8%  0.23  0.93  0.25  1 
Npi  63  6.9  11%  1  0.30  
N0  143  13.1  9%  
Ntot  206  13.4  7%  
Npi/N0  30%  
NUE  21.70  
Yield,N .NDRE  0.06  0.01  
2018  Yield  N0  Npi  Ntot  NDRE  
Yield  8.7  1.013  11.6%  0.19  0.58  0.38  0.54 
NDRE  0.319  0.045  14%  0.16  0.92  0.47  1 
Npi  47  8.7  19%  1  0.50  
N0  118  20.7  18% 


Ntot  164  23.7  14%  
Npi/N0  28%  
NUE  18.85  
Yield,N .NDRE  0.22  0.17 

In
The first lesson learned from the farm was that there was a low repeatability of parcel yields in subsequent years. Since this parameter is independent of the operators, it is generally neglected. The PA capabilities can now be used to monitor the parcel responses at a capillary level, thus parcel repeatability could be considered by means of algorithms, especially considering that its positive evolution seems to follow the increase in variability. In fact, a perfect PA aims to a strong homogeneity in parcel production, ergo a zerocorrelation between the O/I framework. The germinability of the seeds for flooded rice production is a very critical point, and a variable steam density may affect the yield
The second lesson pertained to the increase in variability of the yield and, to a lesser extent, in NDRE after that a strong negative correlation (< 0.90) was imposed on the Npi variable fertilization, as inversely proportional to the vegetative vigour attained near PI. Cordero et al
The improvement of the mean and maximum yields, and not of the minimum, depends on several plots where the limiting factor cannot be referred to the N availability. A next challenge will be to check the repeatability for yield in those critical plots and to find which is the limiting factor, or, more simply, to reduce the N rate, avoiding a waste.
The third lesson concerned the negative correlation between Ntot and yield, which resulted even after absorbing the Npi in the Ntot. In the experimental plot studies, the correlations were generally positive, and even very high, until the maximum yield was reached
The fourth lesson is related the uncertainty of the reference parcels which identified at zero input
The NDRE variability also depends to a great extent on the cultivar, and on the evolution of the temperature and light intensity of the growing season, and it is therefore not automatic to apply the correlation between NDRE and the N rate for PI fertilization of the previous years. The empirical method applied till now on the farm is: i) to consider the prescription threshold table applied in the same cultivar in the previous years; ii) to detect by a drone the gross average NDRE for this cultivar in the current year, and iii) to adapt the prescription threshold table to this parameter. It could also be useful to adapt the N rates to the weather forecast for the following 40 days, if it could be considered reliable.
The study is detached from the usual basic models and applications of the PA, but unfortunately there are no detailed references of similar duration and width to be compared. However, PA is about, and interesting lessons, derived from real and not from model domain, featured the evolution of high production rice fields. The PA on this farm increased the yield for the same total input of N, but also its variability was increased – and not reduced, as predicted by the theory – but strengthening the repeatability, a neglected parameter.
The present elaborations of field parcel data suggest that the actual applied N rates are near, or over, the maximum of the parabolic curve that represents the nonlinear correlation between Ntot and Yield. Slightly reducing the maximum N rates would probably not affect the maximum yields.
In order to reduce the variability of NDRE and of the yields, the first factor that must be considered is the plant density: having an even density of between 200 and 250 plants m^{2} should be the key to improving the results (
An advantage of the reduction in the N rates at PI, where NDRE is already high, is a reduction in the risk of attacks of
All over the paper we have deliberately avoided representing maps of vigor and minute details of production and Nsupply. The fact that the repeatability was barely +0.31 over 6 years indicates that the soil factor & Nprepi explains less than 10% of the variability, so we would have shown a 90% noise in the maps
This paper underlines that PA still has a huge potential that still must be explored.
The authors wish to thank the Fondazione CRT, Torino – Italy, for the financial support to the scientific activities of the “Accademia di Agricoltura di Torino”.