Table 1.
Summary information of the three datasets used in this analysis.
Fig 1.
Maize productivity is highly dependent on nitrogen fertilization.
Contrasted across the three datasets: (A) EONR, (B) yield increase or delta yield from EONR fertilization, (C) yield without nitrogen fertilization, and (D) the difference between optimal nitrogen rate and EONR. Box limits indicate the first and third quartile, whiskers indicate 1.5 times the interquartile range, points indicate outliers, and box line is the median.
Fig 2.
For maize, iNUE relative to increasing nitrogen fertilizer rate is not static.
(A-C) iNUE shown for three datasets. For the NA dataset, iNUE at low nitrogen rates was higher for medium-textured vs fine-textured soils (D). For the NE dataset, iNUE was higher for continuous maize compared to maize rotated with soybean (E). With each graph, a line represents one site-year. The black and solid colored lines indicate the best-fit linear lines.
Fig 3.
Parameters of the quadratic response model are valuable for explaining weather and soil influence on iNUE.
(A) Two examples from the same site (NE dataset) of maize yield response to nitrogen fertilization (black squares = 1999, blue circles = 2000). For both, the quadratic-plateau model fits the measured yield response well, yet differently; they have similar EONR values but different quadratic-plateau model coefficients. (B) Bubble plot of all three datasets combined showing the relationships between the quadratic-plateau model terms (a, b, and c) and EONR. Low values for “a” coincide with low values for “b” (r2 = 0.46); and low values for “a” are related to high EONR (r2 = 0.50). (C) The response model terms are valuable for understanding causal relationships of weather, soil, and management to iNUE. Using a random forest model, important weather and soil variables were identified that predict the rate of iNUE decrease (i.e., slope coefficients), classed as “High” or “Low”. Rates of decrease were derived using the 1st derivative of the quadratic-plateau model and converted into grain-N units (2 x “a” quadric term x 0.0115) and classified as “High” or “Low” based on the median value of -0.608. Weather variables were calculated across different time periods: total season (April 15-September 15), emergence (April 15-June 1), growth (June 2-July 15), and grain fill (July 16-September 15). (D) An example of decision tree predicting slopes (68% accuracy on a withheld testing portion of the data) as “High” or “Low” using weather information, which included the PPT (cumulative precipitation) during establishment and for the entire season, GDD (growing degree days) during establishment, and AWDR (abundant and well distributed rainfall) during establishment.
Fig 4.
Foregone profit (relative to EONR) is minimal when iNUE is low.
(A-C) Foregone profit for each of the three datasets, calculated as the profit difference at iNUE and economically optimal nitrogen rate (EONR). iNUE was limited to < 60%. Each line represents one site-year.
Fig 5.
Realize large improvements in average NUE by removing low iNUE.
(A) Incremental NUE for the three datasets shown corresponding to foregone profit. (B) Average NUE [(grain Nx–grain N0)/x; x = N rate)] shown relative to foregone profit. (C) Nitrogen saved (EONR- N rate) shown relative to foregone profit. (D) Yield reduction (or loss) with sub-EONR fertilization shown relative to foregone profit. With each graph, box limits are the first and third quartile, whiskers indicate 1.5 times the interquartile range, circles indicate outliers, the box line is the median, and the black circle is the mean.
Fig 6.
With modest sub-EONR fertilization, average NUE improvements of ~10% can be realized.
(A) Averaged over the three studies, this equation shows the average NUE shown relative to foregone profit when N fertilization is less than EONR. (B) Model showing the average N saved with sub-EONR fertilization relative to foregone profit.