January 2010
PERSPECTIVES ON THE MODERN N CYCLE
Ecological Applications, 20(1), 2010, pp. 91–100
Ó 2010 by the Ecological Society of America
Narrowing the agronomic yield gap with improved
nitrogen use efficiency: a modeling approach
T. D. AHRENS,1,5 D. B. LOBELL,2 J. I. ORTIZ-MONASTERIO,3 Y. LI,4,6
AND
P. A. MATSON1
1
Department of Geological and Environmental Sciences, Stanford University, Stanford, California 94305 USA
2
Program on Food Security and the Environment, Stanford University, Stanford, California 94305 USA
3
International Maize and Wheat Improvement Center (CIMMYT), Wheat Program, Apartado Postal 6-641,
06600 Mexico D.F., Mexico
4
Institute of Land and Food Resources, University of Melbourne, Victoria 3010 Australia
Abstract. Improving nitrogen use efficiency (NUE) in the major cereals is critical for more
sustainable nitrogen use in high-input agriculture, but our understanding of the potential for
NUE improvement is limited by a paucity of reliable on-farm measurements. Limited on-farm
data suggest that agronomic NUE (AEN) is lower and more variable than data from trials
conducted at research stations, on which much of our understanding of AEN has been built.
The purpose of this study was to determine the magnitude and causes of variability in AEN
across an agricultural region, which we refer to as the achievement distribution of AEN. The
distribution of simulated AEN in 80 farmers’ fields in an irrigated wheat system in the Yaqui
Valley, Mexico, was compared with trials at a local research center (International Wheat and
Maize Improvement Center; CIMMYT). An agroecosystem simulation model WNMM was
used to understand factors controlling yield, AEN, gaseous N emissions, and nitrate leaching
in the region. Simulated AEN in the Yaqui Valley was highly variable, and mean on-farm AEN
was 44% lower than trials with similar fertilization rates at CIMMYT. Variability in residual
N supply was the most important factor determining simulated AEN. Better split applications
of N fertilizer led to almost a doubling of AEN, increased profit, and reduced N pollution, and
even larger improvements were possible with technologies that allow for direct measurement
of soil N supply and plant N demand, such as site-specific nitrogen management.
Key words: agronomic efficiency; gaseous N emissions; nitrate leaching; nitrogen use efficiency; sitespecific nitrogen management; wheat; WNMM model; yield.
determining the size of an analogous agronomic nitrogen-use efficiency (AEN) gap and to understand factors
limiting higher on-farm AEN achievement.
Concerns for sustaining increasing levels of crop
production include a broad spectrum of issues, including
loss of biodiversity, greenhouse gas emissions, and major
alterations in the global nitrogen (N) cycle. Increased
reactive N in the environment has been linked to surface
water eutrophication, nitrate contamination of groundwater, local and global atmospheric pollution, and
adverse impacts on human health (Vitousek et al. 1997,
Galloway et al. 2003, Townsend et al. 2003). Here we use
AEN as a proxy for sustainable N utilization in a given
agroecosystem. Nitrogen use efficiency (NUE) is a
measure of the proportion of available N taken up by a
crop, and has been measured and expressed in various
ways (Dobermann 2007). We use AEN as our metric
throughout this paper, defined as the increase in crop
yield per unit of applied N:
INTRODUCTION
Two of the most important challenges for global food
security in the near future are increasing crop production to meet food and energy demands with plateauing
yield potentials of major cereals, and managing the
increased production in a sustainable manner (Matson
et al. 1997, Tilman et al. 2002, Cassman et al. 2003,
Robertson and Swinton 2005). A majority of the
increased supply is expected to come from increasing
yields on existing farmland, and much attention has
been paid to increasing regional production through
decreasing the yield gap (the difference between the
highest-yielding farms in a given region and the averageyielding farms; (Cassman 1999, Cassman et al. 2003,
Lobell et al. 2005). The focus of this study was on
Manuscript received 1 April 2008; revised 29 September
2008; accepted 12 November 2008. Corresponding Editor: N. B.
Grimm. For reprints of this Invited Feature, see footnote 1,
p. 3.
5
E-mail: [email protected]
6
Present address: Institute of Subtropical Agriculture, The
Chinese Academy of Sciences, Changsha, Hunan 410125,
China.
AEN ¼ ðYf Y0 Þ=Nrate
ð1Þ
where Yf is the yield in a fertilized field, Y0 is the yield in an
unfertilized subplot in the same field, and Nrate is the
fertilization rate in the fertilized field. This definition,
91
92
Ecological Applications
Vol. 20, No. 1
INVITED FEATURE
along with the similar REN (recovery efficiency, defined as
the difference in N uptake in aboveground biomass
between a fertilized and unfertilized plot, divided by the
fertilization rate in the fertilized plot; REN differs from
AEN only by using the difference in aboveground N
uptake in the numerator instead of the difference in yield),
has been used in a number of recent field studies of sitespecific N management (Dobermann et al. 2002, Peng et
al. 2006, Khurana et al. 2007) and global syntheses (Ladha
et al. 2005, Dobermann 2007). Surprisingly, very few
published studies described the distribution of AEN
achieved in farmer fields in a given region. One of the
major reasons that few data exist for on-farm AEN is that
such data are time consuming and costly to obtain.
Determination of AEN either requires farmers to leave a
subplot of a given field unfertilized (a loss of potential
income) or costly 15N tracer methods.
Limited data from rice systems suggests that on-farm
AEN tend to be lower than efficiencies achieved in nearby
research plots (Dobermann et al. 2002, Ladha et al. 2005,
Peng et al. 2006, Dobermann 2007). Two explanations for
the difference include better management in research plots
(where pests, weeds, water supply, and other nonfertilization management practices can be controlled
more easily) or bias in research plot trials (Cassman et
al. 2002, Dobermann 2007). Differences in management
history can influence residual N reserves even under
situations where the farmer’s management practices are
being mimicked, resulting in misrepresentative AEN
measurements. For example, trials conducted in the same
stationary plots year after year lead to artificially high
AEN measurements due to residual soil N pools that are
only depleted in less fertilized plots (Dobermann 2007). In
comparison, on-farm AEN may be lower due to higher
fertilization rates in previous years, resulting in higher
residual N pools, high unfertilized yields, similar fertilized
yields, and lower overall AEN. Research-station AEN
measurements could also be influenced in the opposite
direction if plots have high soil N reserves from overfertilized crops before establishing the fertilizer trial.
Neither case is representative of farmers’ fields.
AEN is associated with strong spatial variability in
agroecosystems. Field studies in irrigated rice found that
on-farm AEN was characterized by high field-to-field
variability, and site-specific nitrogen management resulted in significant increases in AEN (Dobermann et al.
2002). Cassman et al. (2002) found large amounts of
variability in AEN in smaller data sets of on-farm trials
in wheat and maize in the midwestern United States and
India. Cui et al. (2008) also found high field-to-field
variability in indigenous N supply in wheat and maize
systems in the North China Plain, suggesting a large
potential for increased AEN through site-specific nitrogen management.
Recent site-specific nitrogen management field studies
demonstrated that accounting for differences in residual
N and improved timing of N applications are two of the
primary factors driving increases in AEN and economic
and environmental benefits to the farmer (Dobermann et
al. 2002, Peng et al. 2006, Khurana et al. 2007, 2008). Peng
et al. (2006) reported that a doubled AEN was achievable
with site-specific nitrogen management adoption in onfarm rice trials in China, and similarly impressive gains in
AEN were reported in Punjabi rice (83% increase)
(Khurana et al. 2007) and wheat systems (63% increase;
Khurana et al. 2008). By allowing farmers to make inseason estimation of crop N requirement, site-specific
nitrogen management enables them to overcome two
important technological constraints to higher AEN
achievement: directly measuring soil N supply and inseason crop demand. In a modeling study, Lobell (2007)
found that when uncertainties in soil N supply and crop N
demand were reduced simultaneously, optimal N rate
could be reduced by as much as by 24%, 35%, and 56% in
rice, wheat, and maize, respectively, to maximize profit.
Economic incentives represent another important
constraint to managing for high AEN. All farmers face
an economic constraint to high AEN achievement to
some degree because it is neither profitable nor desirable
for global food production to manage too far on the
low-N end of the N response curve to achieve high AEN.
An exception to this rule occurs when crops are grown
without fertilizer (and consequently have an infinitesimal AEN), but such crops mine soil nutrients and cannot
be grown under such management forever (Sheldrick
and Lingard 2004). Additionally, distortions in the price
of N and commodities may negatively influence AEN by
affecting changes in optimal fertilization rates. Examples
include subsidized inputs (Naylor et al. 2001), commodity price supports (Pingali and Shah 1999), or failure to
account for environmental externalities (Vitousek et al.
1997, Townsend et al. 2003). Technological, economic,
and other constraints have been explored in detail for
increasing crop yield (Mosher 1978), but never for AEN.
Understanding the factors that contribute to AEN
achievement in individual farms is critical for determining
how much of the efficiency gap can be decreased through
changes in management (e.g., improving fertilization
timing and delivery, accounting for residual soil N) and
how much is unavoidable (e.g., year-to-year climate
variability). Simulation modeling is ideally suited for this
objective because individual factors can be varied
independently and sensitivity analyses (SA) techniques
used to evaluate controls on on-farm AEN achievement.
In this study, we combine field data and a locally tested,
process-based N simulation model for agroecosystems to
investigate causes of spatial variability in AEN in the highinput irrigated wheat systems of the Yaqui Valley,
Mexico. We refer to this variability in AEN as the
achievement distribution, analogous to the achievement
distribution of yield and profits studied by Mosher (1978).
MATERIALS
AND
METHODS
Site description
The Yaqui Valley is situated between the Sierra
Madre Mountains and the Gulf of California in Sonora,
January 2010
PERSPECTIVES ON THE MODERN N CYCLE
93
FIG. 1. The Yaqui Valley, Sonora, Mexico. The locations of the 80 farms surveyed in the Valley in 2003 are shown in black on
the inset.
Mexico (Fig. 1), and is agro-climatically representative
of 40% of the developing world’s wheat-growing
regions. The Valley contains ;233 000 ha of irrigated
agriculture, and is home to the International Wheat and
Maize Improvement Center (CIMMYT) research station. Mean annual precipitation is 320 mm over the last
25 years and averages of daily minimum and maximum
air temperatures during the primary growing season
(January–April) are 9.68C and 27.38C, respectively.
Irrigated, winter-sown spring wheat (Triticum turgidum
L.) is the most widely planted crop in the valley, and
typical wheat management practices are described in the
definition of spatially variable inputs section below.
Beginning in 1994, a series of field experiments were
established to investigate N cycling dynamics in the
Valley’s irrigated wheat fields as part of a collaboration
between Stanford University and CIMMYT researchers
to try to understand the environmental, economic, and
social sustainability of the Yaqui Valley. Findings in these
studies have included high emission rates (6.6–11.3 kg
Nha1cycle1) of NO and N2O (Matson et al. 1998);
leaching rates of 2–26% (Riley et al. 2001); and a clear
connection between on-farm management practices and
coastal algal blooms (Beman et al. 2005). Data from the
original (1994–1996) and subsequent field experiments
and soil surveys were used here for model parameterization and testing.
Method overview
We employed a locally tested agroecosystem model to
establish achievement distribution curves (defined here
as the distribution of a dependent variable ordered from
best to worst) and to determine how farm-to-farm
variability contributes to regional AEN, leaching losses,
and gaseous N emissions. Simulations of crop growth
and N fates in individual farms involved three principal
steps: (1) parameterization and testing of the Water and
Nitrogen Management Model (WNMM) (Li et al. 2007)
simulation model at the research-plot level; (2) definition
and testing of spatially variably inputs; and (3)
simulation of individual farms using farmer-reported
management practices with local climate and soil data.
The simulation model was then used to determine best
management practices (BMPs) by optimizing simulations for the split application of N that maximized the
difference between income from crop yield and costs for
N fertilizer. site-specific nitrogen management was
simulated by altering the second N application based
on residual soil N supply. Finally, a Monte Carlo
sensitivity analysis was performed for overall importance assessment of management, soil and climatic input
variables in determining yield, AEN, and N fates.
Model description and data used for parameterization
WNMM is a spatially referenced coupled biogeochemistry-crop growth model (Li et al. 2007). Briefly,
WNMM uses farm management data with meteorological inputs to simulate N and C dynamics, crop growth,
and soil moisture in one dimension on a daily time step.
WNMM simulates organic matter dynamics in fresh
organic matter, microbial biomass (live and dead) and
humus (active and passive) pools. Decomposition of
each pool is simulated independently using first-order
kinetics. Decomposed C is partitioned among emitted
CO2, uptake by microbial biomass and active humus,
and the fate of liberated N is determined by C:N ratios
of new microbial biomass, fresh organic matter, and
active humus. Denitrification, nitrification, volatiliza-
94
INVITED FEATURE
Ecological Applications
Vol. 20, No. 1
FIG. 2. Seasonal (a) soil moisture dynamics, (b) NH4þ and (c) NO3 pools, (d) crop biomass, and (e) NO and (f ) N2O emissions
simulated by the Water and Nitrogen Management Model (WNMM; solid line) and measured in the field (open circles) in the
reduced N treatment during the 1994–1995 crop cycle.
tion, and urea hydrolysis are simulated with first-order
kinetics and regulated by environmental conditions.
Volatilization and its associated pH effects are calculated on an hourly time step. Emissions of NO and N2O
are calculated as a fixed percentage of nitrification and
denitrification rates and regulated by soil temperature,
inorganic N concentrations, and moisture status. Ammonium adsorption is simulated with a Freundlich
isotherm to allow for simultaneous simulation of
exchange reactions on particle surfaces and clay-fixed
ammonium dynamics. Nitrate leaching uses the convection–dispersion equation. Soil moisture is modeled
with a tipping-bucket approach. Crop growth progresses
based on user-defined biomass growth at developmental
stages, and harvest indexing is used for crop yield.
WNMM is explained in detail by Li et al. (2007).
WNMM was parameterized and tested with detailed
crop growth data, soil N pools and processing rates,
trace gas emissions measurements (NO, N2O), and soil
moisture dynamics collected in two management regimes (typical farmer practice and a reduced-N alternative) throughout two crop cycles (October 1994–
March 1995, October 1995–March 1996) (Matson et al.
1998). Nitrate leaching was measured in 1995–1996 and
1997–1998 (Riley et al. 2001). WNMM was able to
reproduce trends in crop growth, soil moisture, and N
dynamics throughout the growing season (Fig. 2).
Seasonally summed gaseous emissions (NO, N2O) were
also similar to field measurements (Fig. 3), and
simulated dry matter production and crop yield were
within 10% of field measurements.
FIG. 3. Simulated seasonal fluxes of NO (open circles) and
N2O (solid circles) compared to field-measured fluxes in two
seasons (1994–1995 and 1995–1996) and treatments (typical
farmer practice and best management practice) at the International Maize and Wheat Improvement Center (CIMMYT)
research station in Sonora, Mexico.
January 2010
PERSPECTIVES ON THE MODERN N CYCLE
95
TABLE 1. Values and distributions used for management, soil, and climate variables in the
Monte Carlo simulations.
Parameter
Units
Values Total N fertilization
Timing of first fertilization
Timing of second fertilization
Fertilizer type, first fertilization
kg N/ha
days before sowing
days after sowing
fertilizer type
Irrigation number
Sowing date
number of irrigations
day of the year
(1 January ¼ 1)
year
%
%
pH units
mm/h
g/cm3
%
cmolkg1ds
mm
1 ¼ perfect mixing
mg Nkg1ds
mg Nkg1ds
d1
Climate
Initial total soil N
Soil organic matter
pH
Saturated hydraulic conductivity
Bulk density
Clay content
Cation exchange capacity
Tillage depth
Tillage effectiveness
Initial ammonium content
Initial nitrate content
Rate constant for active soil
organic C decomposition
251 6 50
27 6 18
55 6 9
broadcast urea (0.12),
anhydrous ammonia (0.24),
fertigation (0.74)
3 (0.3), 4 (0.675), 5 (0.025)
331 6 13
1984–2004 (equal weighting)
0.084 6 0.015
1.10 6 0.23
8.58 6 0.43
6.0 6 0.6à
1.14 6 0.11
34 6 4.0
22.0 6 0.71
100–150 (uniform distribution)
0.45 6 0.5
5.0–35.0 (uniform distribution)
1.0–19.0 (uniform distribution)
0.0001–0.002 (uniform distribution)
Values are either mean 6 SD (normal distributions are assumed unless noted) or discrete
values (with weighting in parentheses).
à Initial values for saturated hydraulic conductivity are shown for 0–15 cm. Conductivities
in deeper layers were a third of initial values in the 15–30 cm layer, 0.2 mm/h in 30–60 cm,
and 0.7 mm/h for 60–90 cm.
Definition of spatially variable input data
Management practices on 80 farms (Fig. 1) were
documented in a survey conducted in 2003 (Lobell et al.
2005), and WNMM was used to simulate crop growth
and N dynamics at each farm with local management,
soil, and climate data. Management variables used in
WNMM simulations included N fertilization amount
and timing, fertilizer type, sowing date, and number and
timing of irrigations, and are summarized in Table 1. At
each farm, the closest meteorological station of a
network of 14 stations distributed throughout the valley
was used for daily minimum and maximum temperature,
solar radiation, and precipitation (data available online).7
Initial values for soil properties (texture, soil organic
matter, N percentage [%N], cation exchange capacity,
pH) were based on a survey of the valley’s five major soil
types in 2005. In each soil type, three different farmer’s
fields were tested in five locations in each field, and in
both the bed and furrow at each location. Within-field
samples were homogenized and analyzed for variables
including pH, texture, %N and %C. Soil organic matter
was calculated as %C divided by 0.58. Simulations chose
randomly from a normal distribution around the mean
(6SD) for each soil parameter for the soil type on which
each farm was situated. Initial NH4þ and NO3 pools
were drawn randomly from uniform distributions of 5–
35 and 1–19 mg N/kg, respectively, based on mean prefertilization NH4þ (20 mg N/kg) and NO3 pools (10 mg
7
hhttp://www.pieaes.org.mx/datos.htmi
N/kg) measured during the 1994–1995 field experiments
at CIMMYT. Pre-planting inorganic N concentrations
were not collected in the 2005 soil survey. Locally tested
pedotransfer functions were used to generate values for
soil hydraulic parameters (soil moisture at saturation,
field capacity and wilting point) from soil texture,
organic matter and bulk density (Lobell and OrtizMonasterio 2006).
WNMM simulations performed well when tested
against regional data on residual N supply and remotely
sensed yield. WNMM predicted soil N supply of 115 6
24 kg N/ha (mean 6 SD) in the 80 surveyed farms
compared to measured valley-wide values of 110 6 50 kg
N/ha (Lobell et al. 2004). Mean simulated yield was 5.43
6 0.53 Mg/ha, similar to remotely sensed yield (5.16 6
0.78 Mg/ha) in the 80 farms for the year in which the 80
farms were surveyed.
Determination of BMP and site-specific
nitrogen management
A series of simulations was conducted to determine
the optimal rate of N application. The BMP was chosen
as the rate that maximized the mean marginal rate of
return for N application for the 80 farms. Marginal rate
of return (MRR) was calculated as
MRR ¼ ðYi; j 3 Pw Þ ðNiþj 3 Pn Þ
ð2Þ
where i and j represent N application rates at the first
and second applications, with each equal to 50, 75, 100,
125, or 150 kg N/ha. May 2006 prices were used for the
96
Ecological Applications
Vol. 20, No. 1
INVITED FEATURE
tion method over a four-year period for all 80 farms.
The amount of fertilizer in each split application was
varied for determination of economically optimal
practices, and the sowing date at all farms was set to
the long-term optimal sowing date (December 1; D. B.
Lobell, personal communication). Farmer-reported practices were used for all other management variables.
Site-specific nitrogen management was simulated
using the same management variables used in the BMP
determination, but initial fertilization rates were set to
50 kg N/ha, and the application rate resulting in the
highest average profit over four individually simulated
years (2001–2005) was used for the second fertilization.
The highest profit was determined independently in each
field. For comparison, typical pre-sowing fertilization
rates in the Valley are 160 kg N/ha.
Sensitivity analysis methods
Sensitivity analyses of biogeochemical and agricultural models are often used to identify the relative
importance of input parameters on variance in model
outputs. Identification of these factors can aid in
prioritizing efforts in research or decision support.
Many types of SA are available, ranging from simple
one-at-a-time (OAT) screening exercises to more complex local and global variance-based methods (see
Saltelli et al. 2004). We carried out a SA here for
importance assessment using Monte Carlo (MC) methods for determining the input variables most responsible
for simulated AEN. Correlation between input variables
and model outputs was estimated using Spearman’s
rank correlation coefficient (q) with the statistical
software R (version 2.5.0; R Project for Statistical
Computing 2007).
Distributions for 19 input variables were defined by
results from the farmer surveys and randomly sampled for
simulations (n ¼ 500 simulations; Table 1). The distributions were randomly sampled for each model run using
the rnorm and runif functions in the statistical software R
(V. 2.5.0). Twenty years of climate data (1985–2004)
measured at CIMMYT were used for meteorological
inputs (data available online, see footnote 7). Input values
for soil texture, N pools, and organic matter were based
on the 2005 soil survey for all five soil types.
FIG. 4. (a) Yield, (b) agronomic nitrogen use efficiency
(AEN), and (c) profit simulated in the 80 surveyed farms for the
2002–2003 crop cycle (crossed bars indicate mean 6 SD)
compared with a N rate trial at CIMMYT managed with
typical farmer practices. Dashed lines indicate the greatest
marginal rate of return (Eq. 2) using the N response curve for
CIMMYT.
price of wheat (Pw; 0.191 US$/kg) and N fertilizer (Pn;
US$ 0.82 for urea and US$ 0.78 for anhydrous
ammonia) (Ortiz-Monasterio and Raun 2007). The
BMP is expected to vary from year to year due to
climate variability and residual N supply, and to
account for this, the optimum fertilization practice was
calculated as the single most economical split-applica-
RESULTS
AND
DISCUSSION
On-farm NUE achievement
Limited field measurements of AEN reported in the
literature suggest that on-farm AEN is lower and more
variable than similar trials in local research stations
(Ladha et al. 2005, Peng et al. 2006). We also found this
to be the case in the Yaqui Valley, as mean simulated onfarm AEN in this study was 44% lower than AEN in trials
at CIMMYT managed with similar fertilization practices
(J. I. Ortiz-Monasterio, unpublished data). The distributions of on-farm yield, AEN, and profit (defined as the
marginal rate of return; Eq. 2) are shown in Fig. 4a–c
together with similar measurements from N rate trials at
January 2010
PERSPECTIVES ON THE MODERN N CYCLE
FIG. 5. The achievement distribution curve for AEN in the
80 simulated farms in the Yaqui Valley in 2002–2003. Gray
circles show the farmer surveys, black circles (which appear as a
line) show best management practices (BMP). The distribution
of efficiencies in 297 research station trials is shown in the box
plot superimposed on the right (Ladha et al. 2005). Lower and
upper limits of the box signify 25th and 75th percentiles, and
error bar boundaries signify 10th and 90th percentiles.
CIMMYT. (The curves shown in Fig. 5a–c represent a
common wheat cultivar [Rayon F89] grown in field trials
at CIMMYT in 2002–2003 under typical fertilization
practices used in the valley. More efficient N response
curves have been established at CIMMYT for alternative
management regimes, which would result in even greater
differences between the response curves under alternate
management at CIMMYT and on-farm data than the
differences in Fig. 5a–c.) Lower yields for the CIMMYT
trials relative to farmers’ fields were consistent with the
lower N rates (Fig. 4a).
97
Simulated AEN in the Yaqui Valley spanned a broad
range of efficiencies, including half of the variability
reported in Ladha et al.’s (2005) global synthesis of AEN
at research stations (Fig. 5). Such variability in a single
2000-km2 region suggests that regional means can be
misleading, and understanding the causes of farm-level
variability could have important implications for regional leaching- and emission-reduction strategies. For
example, the worst 10% of farms contributed disproportionately large amounts of reactive N into the environment, emitting and leaching ;20% of regional totals.
Targeting these underperforming farms could be more
cost-effective that changing the habits of an entire
region. Understanding the social causes of this phenomenon, known as disproportionality, has been the subject
of previous interdisciplinary research (Nowak et al.
2006).
Easily obtainable yield and partial factor productivity
(PFP) statistics both downplay the importance of fieldto-field variability in AEN and N fates. The coefficient of
variation (CV) in AEN (CV ¼ 39%; Fig. 6b) was higher
than for yield (9.7%; Fig. 6a) or PFP (21%; not shown).
Sensitivity analysis
The goal of the sensitivity analysis was to elucidate
which climate, soil, and management variables had the
greatest impact on crop growth, nitrogen fates and
farmer income. One important benefit of SA is the
ability to differentiate how much of an outcome is
determined by variables that can be managed and how
FIG. 6. Achievement distribution curves for (a) yield, (b) AEN, (c) leached N, and (d) NO þ N2O emissions under typical farmer
practice (black circles), BMP management (dark gray circles), and site-specific nitrogen management (SSNM, light gray circles).
98
INVITED FEATURE
Ecological Applications
Vol. 20, No. 1
TABLE 2. The three most important input variables for each output in Monte Carlo simulation
results (n ¼ 500 simulations), as determined by Spearman’s rank correlation coefficient (q).
Output variable
Highest
rank
Second-highest
rank
Third-highest
rank
Yield
Profit
Nitrogen use efficiency (NUE)
NO emissions
N2O emissions
Leaching
sowing date
sowing date
residual N supply
fertilizer type
sowing date
climate
irrigation number
irrigation number
fertilizer amount
fertilizer amount
climate
soil hydraulics
climate
fertilizer amount
soil hydraulics
tillage depth
fertilizer amount
fertilizer amount
much is beyond the farmers’ control. The benefits of
improving management of controllable variables using
two common strategies (BMP and site-specific nitrogen
management) are explored in BMP and SSNM.
The most important input variables for all outputs in
Table 2 were management related, with the exception of
NO3 leaching. Much of the variability in AEN was due
to the variability in residual N supply, which was a far
greater predictor of AEN than total fertilization rate. SA
revealed that sowing date, irrigation strategies, and soil
properties influence yield more than N rate. Dobermann
and Cassman (2004) noted a similar disconnect between
yield and fertilization rates in on-farm trials in rice
systems. In individual high-yielding years, however,
fertilization rate may play a more significant role in
determining crop yield (Lobell et al. 2005).
BMP and SSNM
BMP and site-specific nitrogen management scenarios
were used to determine the potential benefits of two
common strategies for changes in management. Simulations using a single BMP (75–50 kg N/ha split
application, as determined by Eq. 2; Fig. 7) instead of
current farmer practices exhibited a doubling of AEN, a
47% decrease in regional NO þ N2O emissions, a 31%
reduction in N leaching, and a 12% increase in profit, all
with virtually no change in crop yield (0.3%; Fig. 6).
Site-specific nitrogen management simulations led to a
157% increase in AEN, a 59% decrease in NO þ N2O
emissions, a 36% reduction in N leaching, and a 13%
increase in profit, again with virtually no change in crop
yield.
The benefits of site-specific nitrogen management are
likely underestimated by our simulations, as crop
growth in consecutive years was not modeled. As soil
N is drawn down in subsequent crop cycles, site-specific
nitrogen management would be able to adapt N recommendations on a farm-by-farm basis, likely resulting in
even greater gains in AEN, profit, and reduced N
emissions and leaching. The benefits of BMP recommendations include saving the cost of equipment,
extension, and training costs of large-scale site-specific
nitrogen management implementation, but changes to a
uniform BMP recommendation require difficult measurements of changes in soil N supply. Matson et al.
(1998) showed that most farmers could improve their
income with a 180 kg N/ha fertilization and proper
FIG. 7. Mean (a) yield, (b) AEN, and (c) profit simulated in
the 80 surveyed farms for the 2002–2003 crop cycle under
different split fertilizer applications. The initial pre-plant
fertilization (kg basal) is shown in the symbol key, and the
total fertilization rate is indicated on the x-axis. The difference
was applied in a second application timed to coincide with the
first post-plant irrigation. The dashed line represents the mean
for simulations of the 80 surveyed farms for each output
variable.
January 2010
PERSPECTIVES ON THE MODERN N CYCLE
distribution and timing of the application. However,
these fertilization strategies were not widely adopted in
the valley, probably because of the uncertainty around
residual soil N, the perceived risk of rain causing field
inaccessibility during the second N application, or credit
agencies with lending terms dictating specific management practices (E. McCullough and P. Matson, unpublished manuscript).
The achievement distribution curves of BMPs and
site-specific nitrogen management (Fig. 6) suggested that
fields are not close to reaching an efficiency ceiling under
current management. Mean AEN increased by 91% and
157% under BMPs and site-specific nitrogen management, respectively, and the slope of the curves were
greater for these alternative management practices than
under the farmer practice. AEN achievements on the
high end of the curve (25 kg crop yield/kg N or greater)
agree with estimates of AEN in well-managed high-input
systems by Dobermann (2007), and may represent an
efficiency ceiling for fields managed under locally
available technology in the Yaqui Valley. Much of the
variability in AEN among farms was due to differences
in residual N supply, and it is expected that farms on the
low end of the curve could increase AEN to approach the
efficiency ceiling in subsequent seasons as soil N is
drawn down.
Sensor-based site-specific nitrogen management technologies are just starting to be introduced in the valley.
By allowing farmers in the valley to adjust the amount of
the second N application depending on residual N supply
and crop demand, farmers have decreased N use by 69 kg
N/ha without any decrease in yield. As a result, profits
increased by an average of 56 US$/ha (Ortiz-Monasterio
and Raun 2007). Residual N supply was the most
important determining variable for AEN in our analysis,
but it has traditionally been difficult for farmers to
measure (e.g., Dobermann and Cassman 2004). Sitespecific nitrogen management allows for indirect measurement of this pool. Reducing initial fertilization rates
would lead to greater synchrony between plant demand
and fertilizer application and lower environmental costs
(e.g., trace gas emissions during pre-planting; Matson et
al. 1998).
Economic constraints
One possible reason that farmers apply fertilizer at
rates above the profit maximum can be seen in the shape
of the N response curve and the corresponding profit
curve in Fig. 4a and c: there was less of a profit risk for
farmers to apply rates of N above the profit maximum
than below. Using the curve established at CIMMYT, it
is not surprising that risk-averse farmers would fertilize
at higher rates due to uncertainty in climate and residual
N supply at the time of planting. On the other hand, the
actual (simulated) yields and profits achieved by some
farmers followed a different curve lower than that at the
research station. These farmers applied much more N
99
than is warranted from simply being risk-averse on
CIMMYT’s profit curve (Fig. 4c).
While it is beyond the scope of this paper to perform a
full economic analysis of proper commodity and
fertilizer pricing in the Yaqui Valley, our analysis
suggests that most Yaqui farmers do not face an
economic constraint to managing for higher AEN. The
BMP was not very sensitive to wheat or fertilizer prices
within a reasonable range: a 50:50 or 75:50 kg N split
application was the most profitable option even with a
50% change in wheat or fertilizer prices, alone and in
combination. Lobell (2007) also found that fertilizer
price had little effect on N management in the Valley,
with a 20% change in price leading to less than 5%
changes in N use due to the inability for farmers to plan
for uncertainty in soil N supply or crop N demand with
typical management practices.
CONCLUSIONS
The purpose of this study was to analyze the
distribution of on-farm AEN in a high-input agricultural
region to determine the size of the AEN gap and factors
controlling farm-to-farm variability in AEN. Current
farming practices in the Yaqui Valley resulted in highly
variable simulated AEN, and more commonly used
statistics such as yield and PFP downplay this variability. AEN achievement was constrained by a combination of suboptimal timing of sowing and fertilization
in some farms and uncertainty in soil N supply and crop
N demand at the time of sowing in all farms. This
uncertainty leads to uncertainty in N fertilization
decisions under typical management practices. Simulated adoption of site-specific nitrogen management
technologies, which help reduce uncertainty in soil N
supply and early season crop demand, resulted in more
than a doubling of AEN, increased profit, and reduced
NO and N2O emissions and nitrate leaching. Farms with
low AEN emitted and leached disproportionately large
amounts of reactive N, and targeting these farms could
be a cost-effective approach to reducing regional N
pollution. Simulations and recent field trials demonstrate that site-specific nitrogen management helped
reduce technological constraints to higher AEN achievement, profit and more sustainable N management in the
Yaqui Valley.
ACKNOWLEDGMENTS
The authors thank anonymous reviewers for helpful comments and insight. Funding for this study was provided by the
David and Lucile Packard Foundation’s support of ‘‘Integrated
Studies of Sustainability: Land-Water Systems of the Yaqui
Basin,’’ an NSF Graduate Student Research Fellowship to
T. D. Ahrens, and funds from Stanford University.
LITERATURE CITED
Beman, J., K. Arrigo, and P. Matson. 2005. Agricultural runoff
fuels large phytoplankton blooms in vulnerable areas of the
ocean. Nature 434:211–214.
Cassman, K. G. 1999. Ecological intensification of cereal
production systems: yield potential, soil quality, and pre-
100
INVITED FEATURE
cision agriculture. Proceedings of the National Academy of
Sciences (USA) 96:5952–5959.
Cassman, K. G., A. Dobermann, and D. T. Walters. 2002.
Agroecosystems, nitrogen-use efficiency, and nitrogen management. Ambio 31:132–140.
Cassman, K. G., A. Dobermann, D. T. Walters, and H. Yang.
2003. Meeting cereal demand while protecting natural
resources and improving environmental quality. Annual
Review of Environment and Resources 28:315–358.
Cui, Z., F. Zhang, X. Chen, Y. Miao, J. Li, L. Shi, J. Xu, Y. Ye,
C. Liu, Z. Yang, Q. Zhang, S. Huang, and D. Bao. 2008. Onfarm estimation of indigenous nitrogen supply for sitespecific nitrogen management in the North China plain.
Nutrient Cycling in Agroecosystems 81(1):37–47.
Dobermann, A. 2007. Nitrogen use efficiency: measurement
and management. Pages 1–28 in A. Krauss, K. Isherwood,
and P. Heffer, editors. Fertilizer best management practices.
IFA, Paris, France.
Dobermann, A., and K. G. Cassman. 2004. Environmental
dimensions of fertilizer nitrogen: What can be done to
increase nitrogen use efficiency and ensure global food
security? Pages 261–278 in A. R. Mosier, J. K. Syers, and
J. R. Freney, editors. Agriculture and the nitrogen cycle:
assessing the impacts of fertilizer use on food production and
the environment. Island Press, Washington, D.C., USA.
Dobermann, A., et al. 2002. Site-specific nutrient management
for intensive rice cropping systems in Asia. Field Crops
Research 74:37–66.
Galloway, J. N., J. D. Aber, J. W. Erisman, S. P. Seitzinger,
R. W. Howarth, E. B. Cowling, and B. J. Cosby. 2003. The
nitrogen cascade. BioScience 53:341–356.
Khurana, H. S., S. B. Phillips, S. Bijay, M. M. Alley, A.
Dobermann, A. S. Sidhu, S. Yadvinder, and S. Peng. 2008.
Agronomic and economic evaluation of site-specific nutrient
management for irrigated wheat in northwest India. Nutrient
Cycling in Agroecosystems 82(1):15–31.
Khurana, H. S., S. B. Phillips, S. Bijay, A. Dobermann, A. S.
Sidhu, S. Yadvinder, and S. Peng. 2007. Performance of sitespecific nutrient management for irrigated, transplanted rice
in Northwest India. Agronomy Journal 99:1436–1447.
Ladha, J. K., H. Pathak, T. J. Krupnik, J. Six, C. van Kessel,
and L. S. Donald. 2005. Efficiency of fertilizer nitrogen in
cereal production: retrospects and prospects. Pages 85–156 in
Advances in agronomy. Academic Press, San Diego,
California, USA.
Li, Y., R. White, D. Chen, J. Zhang, B. Li, Y. Zhang, Y.
Huang, and R. Edis.. 2007. A spatially referenced water and
nitrogen management model (WNMM) for (irrigated)
intensive cropping systems in the North China Plain.
Ecological Modelling 203:395–423.
Lobell, D. B. 2007. The cost of uncertainty for nitrogen
fertilizer management: a sensitivity analysis. Field Crops
Research 100:210–217.
Lobell, D. B., and J. I. Ortiz-Monasterio. 2006. Regional
importance of crop yield constraints: Linking simulation
models and geostatistics to interpret spatial patterns.
Ecological Modelling 196:173–182.
Lobell, D. B., J. I. Ortiz-Monasterio, and G. P. Asner. 2004.
Relative importance of soil and climate variability for
nitrogen management in irrigated wheat. Field Crops
Research 87:155–165.
Ecological Applications
Vol. 20, No. 1
Lobell, D. B., J. I. Ortiz-Monasterio, G. P. Asner, R. L. Naylor,
and W. P. Falcon. 2005. Combining field surveys, remote
sensing, and regression trees to understand yield variations in
an irrigated wheat landscape. Agronomy Journal 97:241–249.
Matson, P., R. Naylor, and J. Ortiz-Monasterio. 1998.
Integration of environmental, agronomic, and economic
aspects of fertilizer management. Science 280:112–115.
Matson, P., W. Parton, A. Power, and M. Swift. 1997.
Agricultural intensification and ecosystem properties. Science
277:504–509.
Mosher, A. T. 1978. an introduction to agricultural extension.
Agricultural Development Council, Singapore University
Press, Singapore.
Naylor, R. L., W. P. Falcon, and A. Peunte-Gonzalez. 2001.
Policy reforms and Mexican agriculture: views from the
Yaqui Valley. CIMMYT Economics Program Paper No. 0101. CIMMYT, Mexico City, Mexico.
Nowak, P., S. Bowen, and P. E. Cabot. 2006. Disproportionality as a framework for linking social and biophysical
systems. Society and Natural Resources 19:153–173.
Ortiz-Monasterio, J. I., and W. Raun. 2007. Reduced nitrogen
and improved farm income for irrigated spring wheat in the
Yaqui Valley, Mexico, using sensor based nitrogen management. Journal of Agricultural Science 145:1–8.
Peng, S., R. J. Buresh, J. Huang, J. Yang, Y. Zou, X. Zhong, G.
Wang, and F. Zhang. 2006. Strategies for overcoming low
agronomic nitrogen use efficiency in irrigated rice systems in
China. Field Crops Research 96:37–47.
Pingali, P. L., and M. Shah. 1999. Rice–wheat cropping systems
in the Indo-Gangetic plains: policy re-directions for sustainable resource use. Pages 1–12 in P. L. Pingali, editor.
Sustaining rice-wheat production systems: socio-economic
and policy issues. Rice–Wheat Consortium Paper Series 5.
Rice–Wheat Consortium, Delhi, India.
R Project for Statistical Computing. 2007. R, version 2.5.0. R
Project for Statistical Computing, Vienna, Austria. hwww.
r-project.orgi
Riley, W., J. Ortiz-Monasterio, and P. Matson. 2001. Nitrogen
leaching and soil nitrate, nitrite, and ammonium levels under
irrigated wheat in Northern Mexico. Nutrient Cycling in
Agroecosystems 61:223–236.
Robertson, G. P., and S. M. Swinton. 2005. Reconciling
agricultural productivity and environmental integrity: a
grand challenge for agriculture. Frontiers in Ecology and
the Environment 3:38–46.
Saltelli, A., S. Tarantola, F. Campolongo, and M. Ratto. 2004.
Sensitivity analysis in practice: a guide to assessing scientific
models. John Wiley and Sons, New York, New York, USA.
Sheldrick, W. F., and J. Lingard. 2004. The use of nutrient
audits to determine nutrient balances in Africa. Food Policy
29:61–98.
Tilman, D., K. G. Cassman, P. A. Matson, R. Naylor, and S.
Polasky. 2002. Agricultural sustainability and intensive
production practices. Nature 418:671–677.
Townsend, A. R., et al. 2003. Human health effects of a
changing global nitrogen cycle. Frontiers in Ecology and the
Environment 1:240–246.
Vitousek, P., J. Aber, R. Howarth, G. Likens, P. Matson, D.
Schindler, W. Schlesinger, and D. Tilman. 1997. Human
alteration of the global nitrogen cycle: sources and consequences. Ecological Applications 7:737–750.