Sei sulla pagina 1di 8

Field Crops Research 100 (2007) 210–217

www.elsevier.com/locate/fcr

The cost of uncertainty for nitrogen fertilizer management:


A sensitivity analysis
David B. Lobell *
Energy and Environment Directorate, Lawrence Livermore National Laboratory, Livermore,
Box 808, L-103, Livermore, CA 94550, United States
Received 7 March 2006; received in revised form 17 July 2006; accepted 17 July 2006

Abstract
Uncertainties in soil nitrogen (N) supply and crop N demand present a challenge to farmers deciding on N fertilizer rates. While field studies
have documented the improvements in N use efficiency possible with site-specific N management approaches that address these uncertainties, a
general understanding of the importance of uncertainty across a wide range of cropping systems is yet to emerge. Here a general model of N rate
decision-making is presented which computes the optimal N rate that maximizes expected profit given uncertainties in N supply and demand. The
cost of uncertainty is measured as the difference in N rate when soil N supply and crop N demand are unknown versus known perfectly. Eliminating
uncertainty in soil N supply (but not crop demand) would reduce average N rates by 5–15% in typical irrigated rice systems, 10–30% in wheat,
and 20–40% in maize. Perfect knowledge of potential crop N demand (but not soil supply) would reduce rates by 3–10% in all systems.
Simultaneous knowledge of both factors reduced N rates by significantly more than the sum of their individual effects, reflecting important
interactions between supply and demand uncertainties. This indicates that the value of information for one factor is inversely related to the level of
uncertainty in the other. Studies that separately consider information on soil N supply or crop N demand may therefore underestimate the total
benefit of management approaches that address both these uncertainties. Site-specific N management could lead to substantial reductions of N rates
without yield loss in a wide range of cropping systems, thereby improving profitability and environmental quality.
# 2006 Elsevier B.V. All rights reserved.

Keywords: Fertilizer management; Nitrogen; Soil variability; Climate variability; Uncertainty; Simulation

1. Introduction moves beyond the farm in aqueous or gaseous forms (Matson


et al., 1997, 1998; Galloway, 1998).
Nitrogen (N) is an essential nutrient for plant growth, and Many factors that influence N use efficiency (NUE) have
often the most limiting nutrient for crop yields. The high yields been investigated, including application method, fertilizer type,
obtained with the majority of today’s crops can result in the soil and crop properties, and irrigation practices. (Cassman
removal of as much as 300 kg N ha1 in grain harvest each et al., 1993; Rauschkolb and Hornsby, 1994; Raun and Johnson,
year, necessitating substantial inputs of fertilizer N or crop 1999). One potentially important factor is the uncertainty faced
rotations with N2 fixing legumes to maintain productivity by a farmer deciding on fertilizer amounts. This includes not
(Cassman et al., 2002). Where adequate N fertilizer is not knowing the existing supply of N available in the soil, because
applied, soils will be mined of N—a common phenomenon in a of variations in factors such as soil total N, organic matter,
number of countries (Sheldrick and Lingard, 2004). Applica- aeration, and microbial activity (Cassman et al., 1996a,b), as
tion of N is fairly inefficient in most systems, with only roughly well as not knowing what crop N demand will be, because of
half of fertilizer N applied during a growing season typically variations in climatic conditions, pests, soil fertility, and other
recovered in the crop biomass that season (Smil, 1999). This factors. (Bullock and Bullock, 1994; Hammer et al., 1996;
inefficiency represents a significant cost to farmers and has Dawe and Moya, 1999). An understanding of the relative
important consequences for ecosystem and human health as N importance of these factors is needed to identify the best
options for improving NUE.
One measure of the impact or cost of uncertainty is the
* Tel.: +1 9254224148. improvement in NUE when information on soil N supply and
E-mail address: dlobell@llnl.gov. crop N demand is available. For example, Dobermann et al.
0378-4290/$ – see front matter # 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.fcr.2006.07.007
D.B. Lobell / Field Crops Research 100 (2007) 210–217 211

(2002) compared the use of site-specific nutrient management where p is profit, yield is the kg of grain produced, Nrate is the
(SSNM) to traditional farming practices for 179 irrigated rice kg of fertilizer N applied, priceGrain and priceN refer to the price
fields throughout Asia, with each site comprising 300–1000 m2. received for each kg of grain and paid for each kg of fertilizer N,
The authors found that average NUE increased by 30–40% respectively, and costsN refers to all costs other than fertilizer
under SSNM while yields increased by 7%. SSNM rates were associated with crop production (this cost is assumed hereafter
based in part on on-site measurements of the indigenous supply to be independent of Nrate). The yield, in turn, can be expressed
of N (IS), which was measured as the N uptake of a crop in a 40– as
100 m2 plot where nutrients other than N were supplied in
grain yield ¼ minðY NL ; Y L Þ (2)
ample quantities but no N was added (Janssen et al., 1990;
Cassman et al., 1998). Estimates of the yield goal, based on a where YNL refers to the attainable yield without N limitation (a
crop model driven with observed meteorology, were also used function of climate, genetic potential of the variety used, and
to determine SSNM rates (Dobermann et al., 2002). non-N constraints such as water stress or pest damage), and YL
Given the substantial improvements in NUE in the Asian represents the yield with N limitation. That is, yield will be equal
rice studies of Dobermann and colleagues, it would appear that to or less than the limit set by total N supply, but no greater. The
SSNM approaches using existing technologies hold promise for relationship between N supply and YL is determined by the
reducing N costs and N pollution while maintaining or nutrient requirements of different plant components, and is
improving farmer yields. Similar studies have been performed relatively well conserved across different management systems
in other regions (Haefele and Wopereis, 2005; Mulvaney et al., (e.g., Cassman et al., 2002; Dobermann and Cassman, 2002).
2006), yet the quantitative impacts of SSNM in many major This relationship is commonly represented by the function:
cropping systems throughout the world are poorly known, in
part because of the resources needed to conduct extensive on- Y L ¼ a þ bðN uptake Þ0:5 (3)
farm trials in multiple years. where Nuptake is the total N captured by the crop, and a and b are
While not a substitute for field trials, simulation models can crop-dependent coefficients (alternative relationships are con-
be a relatively low-cost tool for investigating potential sidered below). Fig. 1 shows the functional relationship
improvements in N management using SSNM approaches between N uptake and grain yield for three major cereal crops
(e.g., Hammer et al., 1996). For example, Lobell et al. (2004) considered here—rice, wheat, and maize (Dobermann and
utilized a model to investigate the potential value of soil Cassman, 2002; Lobell et al., 2004). N requirements increase
analysis and climate forecasts for irrigated wheat N manage- nonlinearly with increased yield, and are similar for wheat and
ment in Mexico. The model simulated the optimal rate of N rice and lower for maize at a given yield level. The lower N
given varying levels of information on IS and yield potential. requirements for maize, a C4 crop, can be attributed to the
Complete certainty for these factors improved NUE by 35% greater efficiency of CO2 uptake relative to C3 plants (Brown,
relative to the base case, where only the statistical distribution 1978; Greenwood et al., 1990).
of IS and crop yield potential were known.
An advantage of such models is that they permit sensitivity
analysis, with model parameters varied and the change in model
output observed (Saltelli et al., 2000). This allows one to
understand how well the results of a particular experiment or
simulation exercise would generalize to other cropping systems
with different soil, climate, and management conditions. In this
study, a general model of N rate decision making under
uncertainty is presented, and a sensitivity analysis is performed
to measure the influence of each model parameter. The
objectives are to assess the generality of conclusions from
localized experiments, and to identify what aspects make a
system particularly vulnerable to inefficiencies from uncer-
tainty and therefore particularly suited to investments that
would reduce these uncertainties.

2. Materials and methods

2.1. A general model of N application rates under


uncertainty
Fig. 1. Modeled relationship between N uptake and yield for rice, wheat, and
The profit a farmer realizes from growing a crop can be maize, using equation YL = a + b(Nuptake)0.5. Equations for rice (a = 1.573,
b = 0.643, R2 = 0.55) and maize (a = 3.71, b = 0.995, R2 = 0.76) are from data
expressed as across a wide range of environments synthesized in Dobermann and Cassman
(2002), and for wheat (a = 0.912, b = 0.588, R2 = 0.81) from data in North-
p ¼ priceGrain  yield  priceN  N rate  costsN (1) west Mexico presented in Lobell et al. (2004).
212 D.B. Lobell / Field Crops Research 100 (2007) 210–217

Conversely, Eq. (3) can be rearranged to determine the N Table 1


Outline of model algorithm to compute optimal N rates
uptake requirement (Nreq) to exactly match the demand for a
non-N-limited yield, YNL: Step Description
1 Choose Nrate (between 1 and 400 kg N ha1)
ðY NL  aÞ2 2 Draw value for YNL from normal distribution YNL  N(mYNL, sYNL)
N req ¼ (4)
b2 3 Draw value for IS from normal distribution IS  N(mIS, sIS)
4 Compute Nsupply = Nrate  effN + IS
Finally, total N supply can be expressed as 5 Compute yield = min(YL = a + b  (Nsupply)0.5, YNL),
where a and b are crop-specific parameters
N supply ¼ N rate  eff N þ I S (5)
6 Compute profit, p = priceGrain  yield  priceN  Nrate  costsN
where effN represents the efficiency of N application (account- 7 If IS and YNL are not known, repeat steps 4–6 1000 times,
ing for losses at the time of application caused by volatilization, using new values for IS and YNL
denitrification, leaching, and other processes), and IS represents or if only IS is not known, repeat steps 4–6 1000 times,
using new values for IS
indigenous N as described above. Importantly, effN is defined
or if only YNL is not known, repeat steps 4–6 1000 times,
here not as the overall efficiency of N use, which will depend on using new values for YNL
how much of N supply is utilized by the crop, but simply as the
8 Compute average of profits from step 7
maximum possible efficiency (i.e., efficiency threshold) given 9 Repeat steps 1–8 for each possible Nrate, and determine
the inevitable losses using current application practices. N rate that results in maximum expected profit
As discussed above, the uncertainty faced by farmers relates 10 Compute actual yield and profit, based on YNL and IS from
to both YNL (which, along with a and b, determines Nreq) and IS. steps 2 to 3 and Nrate from step 9
Here, each is represented as a normally distributed random 11 Repeat steps 1–10 300 times, to sample different
realizations of YNL and IS
variable, defined by its mean (m) and standard deviation (s): 12 Compute average Nrate, yield, and profit from step 11
Y NL  NðmYNL ; s YNL Þ (6)
I S  NðmIS ; s IS Þ (7)
simulation of YNL, (2) IS is known perfectly, for example through
If one assumes that a grower wishes to maximize expected the hypothetical use of a perfectly accurate soil testing
profit, the optimal rate of N is one that maximizes the expected procedure, and (3) both YNL and IS are known perfectly.
value of profit (farmers who are risk averse may also consider Effectively, after each draw of YNL and IS the farmer is allowed to
higher moments of profit distribution, e.g., Bullock and observe one or both values before deciding on how much N to
Bullock, 1994). To determine this value, one can draw apply. The current analysis assumes that this information is
repeatedly from the distributions for YNL and IS, compute available at no cost to the farmer. While extreme, these
the profit corresponding to all reasonable values of Nrate, and hypothetical scenarios provide a useful measure of the cost of
compute the average profit over all draws for each Nrate. The uncertainty and allow one to place a bound on the potential value
Nrate that gives the greatest average profit is then judged of climate or soil information to guide N fertilizer management.
optimal. This corresponds to the situation in which the farmer Table 1 outlines the algorithm used to implement this
has no information on YNL and IS other than their perceived conceptual model. To summarize, the simulated optimal Nrate
distributions (referred to hereafter as the control scenario). relies on the crop type, which determines the functional
The provision of climate or soil information can be viewed relationship between YNL and N supply, and on six model
simply as a change in the perceived distribution of YNL and IS. parameters: mYNL; sYNL; mIS; sIS; effN; and the ratio of N to
Here I consider three extreme examples: (1) YNL is known grain price (priceN:G). Only the ratio of prices is needed because
perfectly, for example through the hypothetical use of perfect dividing Eq. (1) by the price of grain (a constant) does not
climate forecasts combined with a perfect crop model for change the Nrate at which the expression is maximized.

Table 2
Baseline values for model parameters for each crop
Parameter Description Units Baseline value References
Rice Wheat Maize
mYNL Average non-N-limited yield t ha1 6.0 7.5 10.0 Cassman et al. (2002), Dobermann et al. (2002)
sYNL Standard deviation t ha1 0.5 0.7 1.0 and Lobell et al. (2004)
of non-N-limited yield
mIS Average soil indigenous N kg N ha1 50 110 150 Cassman et al. (2002), Dobermann et al.
sIS Standard deviation kg N ha1 20 50 50 (2002), Lobell et al. (2004),
of soil indigenous N Haefele and Wopereis (2005)
and Hossain et al. (2005)
effN N application efficiency Unit less 0.7 0.7 0.7 Cassman et al. (2002)
priceN:G Ratio of fertilizer to grain prices Unit less 2.5 2.5 2.5 Lobell et al. (2004)
See text for description of parameters. All values were perturbed by 20% in sensitivity analysis.
D.B. Lobell / Field Crops Research 100 (2007) 210–217 213

2.2. Sensitivity analysis maize simulations, with 15 and 31% reductions, respectively
(Fig. 2). The impact of knowing IS was considerably smaller in
At issue here is how the impact of information varies as a rice (7%), because of low levels of variability in IS observed in
function of model parameters. Table 2 presents baseline values studies of rice systems relative to wheat and maize (Table 2).
of parameters used for each crop, based on data selected from the Reductions of N rates when knowing both YNL and IS were
literature. (In all cases, IS was measured using N-omission plots.) much larger than the sum of reductions achieved by knowing
These values are mainly representative of irrigated systems, such either one alone (Fig. 2), reflecting an important interaction
as rice in Asia (Cassman et al., 2002; Hossain et al., 2005) and between uncertainties in crop N demand and soil N supply. This
West Africa (Haefele and Wopereis, 2005) and wheat in Mexico can be understood by considering that the impact of uncertainty
(Lobell et al., 2004), or highly productive rainfed systems, such generally scaled with the standard deviation of (Nreq  IS)
as the U.S. corn belt (Duvick and Cassman, 1999; Cassman et al., (Fig. 3). Since the variance of (Nreq  IS) can be expressed as
2002). Model output using these baseline values therefore
represents a generalized expectation for typical intensive rice, VarðN req  I S Þ ¼ VarðN req Þ þ VarðI S Þ þ CovðN req ; I S Þ (8)
wheat, and maize systems. and Nreq and IS are assumed independent in these simulations
Individual locations or years, however, will vary signifi- (covariance = 0), it follows that the standard deviation of
cantly from these values. A sensitivity analysis was therefore (Nreq  IS) is
performed, whereby each parameter was varied by 20% from
its baseline value. Ideally, each parameter would be varied S:D:ðN req  I S Þ ¼ ½VarðN req Þ þ VarðI S Þ0:5 (9)
according to its actual distribution across cropping systems;
however such information is difficult to obtain. Therefore, the Therefore, the effect of an additional source of uncertainty
sensitivity analysis presented here, while providing insight into depends inversely on the level of uncertainty already existing,
model behavior, cannot be used directly to infer the relative as dictated by the power of 0.5. Put differently, the effect of
importance of the parameters. This is because, in reality, some removing Nreq as a source of uncertainty will depend on
parameters are likely to vary more than others across locations uncertainty in IS, and vice-versa. This emphasizes the need
and cropping systems. In addition, the range of parameter to consider uncertainties in both N demand and supply, even if
values likely omit values characteristic of low-input subsistence interested in the impact of only one. It also illustrates that
systems. The assumption of a risk neutral farmer may also be SSNM will be particularly effective if it considers both climate
more problematic in low yielding systems (Dillon and and soil variability.
Scandizzo, 1978; Hardaker et al., 2004). Fig. 4 illustrates the sensitivity of optimal N rates in the
control scenario (no knowledge of YNL or IS) to each parameter.
3. Results N rates in general were quite sensitive to perturbations in the
parameter values, with values ranging by more than
Average optimal N rates for different scenarios of 100 kg N ha1 within each crop. Three parameters (mYNL,
information are displayed in Fig. 2 for the baseline parameters. sYNL, sIS) were positively correlated with N rates in all crops,
For each crop, rates were reduced by roughly 5% when values illustrating that economically optimal N rates (assuming risk
of YNL were known. Knowledge of IS led to much greater neutrality) increase in response to increased yield potential,
reductions of optimal N rates relative to YNL in the wheat and increased yield variability, and increased soil N variability.

Fig. 2. The simulated optimal N rate for representative rice, wheat, and maize systems under different scenarios of uncertainty in crop N demand (Nreq) and soil N
supply (IS): A = neither Nreq or IS is known; B = Nreq is known; C = IS is known; D = both Nreq and IS are known. Values above bars show percent reductions from rates
without any knowledge (scenario A). All values in this figure are from runs with baseline parameter values.
214 D.B. Lobell / Field Crops Research 100 (2007) 210–217

Fig. 3. Reduction of N rates (kg N ha1) when both Nreq and IS are known vs. standard deviation of (Nreq  IS). Rate reductions are roughly a linear function of S.D.
(Nreq  IS). Solid lines indicate best-fit linear regression equations (rice: y = 1.55x + 0.44, R2 = 0.98; wheat: y = 1.32x  0.07, R2 = 0.98; maize: y = 1.62x  0.96,
R2 = 0.97).

Conversely, increased values of mIS, effN, and priceN:G resulted all cases (not shown), which suggests that yield impacts of
in lower simulated N rates for all crops. Not surprisingly, N reduced uncertainty in N demand and supply are minimal for
rates were particularly sensitive to the average non-N-limited the types of systems considered here. (The same may not be
yields (mYNL), illustrating the importance of yield potential in true, for example, in systems where farmers are risk averse and
defining economically optimal rates of N. do not simply maximize expected profits.) In all simulations,
The simulated impact of knowing YNL and/or IS, in terms of constant yields and reduced N rates resulted in higher profits,
percent N rate reductions, was less sensitive to variations in with the percent increase in profits depending largely on the
parameter values than the optimal rates themselves (Fig. 5). As specified value of non-N costs.
expected, the value of knowing YNL increased as its variability The results presented here depended on the prescribed
(sYNL) increased. On a percentage basis, the impact of knowing square-root relationship between N uptake and crop yield
YNL also increased with greater values of mIS and priceN:G. This (Eq. (3)). For rice and maize, these equations were derived from
reflects the fact that absolute reductions remained fairly stable data from many sites, all of which have different non-N-limited
while the control N rates decreased. yields because of climate, soil, and management differences (K.
Percent reductions in N rates when knowing IS were greater Cassman, personal communication, 2006). The relationship at a
for increased values of mIS, sIS, and effN and for decreased specific field, the scale at which a farmer makes N management
values of mYNL, sYNL, and priceN:G (Fig. 5). The value of sIS decisions, may be different than this generalized relationship.
was the most important determinant of N rate reductions in an To test whether the results changed when using a different
absolute sense, with a 20% increase in sIS translating to a response function, the simulations were re-run after replacing
roughly 20% increase in the impact of knowing IS. The large Eq. (3) with
increase in percent rate reductions for low values of mYNL
reflected similar absolute reductions with a substantially lower Y L ¼ minða þ bðN uptake Þ0:5 ; mYNL þ s YNL Þ (10)
baseline N rate.
For the range of parameter values considered here, the total In this equation, yields are prohibited from exceeding one
impact of eliminating uncertainty in YNL and IS was nearly standard deviation above average non-N-limited yields, thereby
always at least a 20% reduction in average N rates, and in flattening the response of yield to N at high rates of N. The
extreme cases more than a 50% reduction (Fig. 5). Yield results did not differ significantly when using this flatter N
changes associated with these reductions were less than 2% in response, suggesting (though not proving) that the value of soil

Fig. 4. The sensitivity to model parameters of simulated optimal N rates. The solid vertical line shows the optimal N rate for the baseline parameter values. The bars
show the range in computed optimal rates when adjusting each parameter by 20% of its baseline value. A plus sign indicates that the higher rate was computed when
using the higher parameter value (baseline + 20%), while a negative sign indicates that the higher rate was associated with the lower parameter value.
D.B. Lobell / Field Crops Research 100 (2007) 210–217 215

Fig. 5. The sensitivity to model parameters of simulated % reduction in N rate when Nreq is known (top), when IS is known (middle), or when both Nreq and IS are
known (bottom). The solid vertical line shows the % reduction when using the baseline parameter values. The bars show the range in % reduction when adjusting each
parameter by 20% of its baseline value. A plus sign indicates that the larger reduction was computed when using the higher parameter value (baseline + 20%),
while a negative sign indicates that the larger reduction was associated with the lower parameter value.

and climate information is relatively insensitive to the exact Variability in N supply was generally a more important
shape of N response. source of uncertainty than variability in N demand. The impact
of N supply uncertainty, however, was strongly dependent on
4. Discussion and conclusions the variability of available indigenous N (sIS). Thus, an
important consideration is whether efforts to reduce N rates by
The goal of this study was to evaluate how the impact of better diagnosis and utilization of indigenous N would result in
uncertainty in crop N demand and soil N supply depends on the reduced variability of soil N. If so, the economic value of soil
various characteristics of cropping systems. While baseline diagnostics would decrease through time as indigenous N was
rates of N varied widely for different hypothetical systems, the slowly depleted. A key issue, then, is the impact of increased
simulated reductions associated with reduced uncertainties NUE on the indigenous soil supply, which is a topic that has not
were quite robust to changes in model parameters. For example, received much attention. The lack of a consistent relationship
changing fertilizer prices by 20% affected simulated reductions between total soil N and available N and large temporal
in N rates by less than 5%. This robustness indicates that field variations in available N (Cassman et al., 1998; Dobermann
studies which have purported to show an important role of soil et al., 2003), make it unclear whether utilization of indigenous
N variability in N management (e.g., Dobermann et al., 2002) N does in fact have an affect on future available N.
are not simply reflecting unique conditions in the chosen Widespread adoption of technologies that reduce uncertainty
experimental regions. Instead, uncertainty evidently represents for N management will occur only if the reduction in costs with
an important contributor to inefficiencies in N use across a wide lower N rates exceeds the cost of the technology itself. The
range of cropping systems. simulations presented here did not account for costs required to
216 D.B. Lobell / Field Crops Research 100 (2007) 210–217

obtain information on YNL or IS, nor did they consider the costs of by University of California, Lawrence Livermore National
site-specific management itself (i.e., purchase of necessary Laboratory under Contract W-7405-Eng-48.
equipment). Several technologies currently exist to measure IS,
including soil tests (Cabrera and Kissel, 1988; Mulvaney et al., References
2006), plant tissue analysis (Fox et al., 1989; Knowles et al.,
1991), and spectral measurements (Raun et al., 2002). While Brown, R.H., 1978. Difference in N use efficiency in C3 and C4 plants and its
current costs appear too high for widespread adoption (Raun implications in adaptation and evolution. Crop Sci. 18, 93–98.
et al., 2005), development of these techniques is an area of active Bullock, D.S., Bullock, D.G., 1994. Calculation of optimal nitrogen fertilizer
research and promise to substantially reduce costs and improve rates. Agron. J. 86, 921–923.
Cabrera, M.L., Kissel, D.E., 1988. Evaluation of a method to predict nitrogen
accuracies in the future. mineralized from soil organic-matter under field conditions. Soil Sci. Soc.
Another important aspect of technologies is the scale at Am. J. 52, 1027–1031.
which they can be implemented. The current simulations were Cassman, K.G., Dobermann, A., Cruz, P.C.S., Gines, G.C., Samson, M.I.,
based on variability of IS measurements for plots 0.05– Descalsota, J.P., Alcantara, J.M., Dizon, M.A., Olk, D.C., 1996a. Soil
0.10 ha in size collected on different fields. If one assumes that organic matter and the indigenous nitrogen supply of intensive irrigated
rice systems in the tropics. Plant Soil 182, 267–278.
variability of plots within fields would be of similar magnitude, Cassman, K.G., Dobermann, A., Walters, D.T., 2002. Agroecosystems, nitro-
then IS measurements would have to be made at this resolution gen-use efficiency, and nitrogen management. Ambio 31, 132–140.
for an entire field to achieve the gains computed here. Gains Cassman, K.G., Gines, G.C., Dizon, M.A., Samson, M.I., Alcantara, J.M.,
would be lower for measurements made at coarser scales, and 1996b. Nitrogen-use efficiency in tropical lowland rice systems: contribu-
tions from indigenous and applied nitrogen. Field Crops Res. 47, 1–12.
presumably higher if measurements were made at finer scales.
Cassman, K.G., Kropff, M.J., Gaunt, J., Peng, S., 1993. Nitrogen use efficiency
For example, Raun et al. (1998) and Solie et al. (1999) report of rice reconsidered—what are the key constraints. Plant Soil 156, 359–362.
significant variability in total soil N within areas of less than Cassman, K.G., Peng, S., Olk, D.C., Ladha, J.K., Reichardt, W., Dobermann,
1 m2. The sensitivity of gains to the resolution and extent of IS A., Singh, U., 1998. Opportunities for increased nitrogen-use efficiency
measurements will, in general, depend on the exact spatial from improved resource management in irrigated rice systems. Field Crops
distribution of soil N within and between fields, and is a topic Res. 56, 7–39.
Dawe, D., Moya, P.F., 1999. Variability of Optimal Nitrogen Applications for
that deserves further study. Rice. Program Report for 1998. International Rice Research Institute,
In addition to ignoring technology costs, these simulations Makati City, Philippines, pp. 15–17.
did not explicitly consider environmental objectives, such as Dillon, J.L., Scandizzo, P.L., 1978. Risk attitudes of subsistence farmers in
the reduction of N leaching into aquatic ecosystems. Such northeast Brazil—sampling approach. Am. J. Agric. Econ. 60, 425–435.
considerations would likely increase the perceived value of Dobermann, A., Cassman, K.G., 2002. Plant nutrient management for enhanced
productivity in intensive grain production systems of the United States and
information since N leaching generally increases with excess Asia. Plant Soil 247, 153–175.
applied N (e.g., Riley et al., 2001; Donner and Kucharik, 2003). Dobermann, A., Witt, C., Abdulrachman, S., Gines, H.C., Nagarajan, R., Son,
In contrast to N rate reductions, environmental benefits may T.T., Tan, P.S., Wang, G.H., Chien, N.V., Thoa, V.T.K., Phung, C.V., Stalin,
even be maintained if indigenous soil N is depleted, since the P., Muthukrishnan, P., Ravi, V., Babu, M., Simbahan, G.C., Adviento,
M.A.A., Bartolome, V., 2003. Estimating indigenous nutrient supplies
amount of excess N in the system will be permanently reduced.
for site-specific nutrient management in irrigated rice. Agron. J. 95,
As a final caveat, the simulations presented here considered 924–935.
uncertainty in YNL and IS but ignored potential uncertainty in Dobermann, A., Witt, C., Dawe, D., Abdulrachman, S., Gines, H.C., Nagarajan,
other model variables, such as the ratio of grain to fertilizer R., Satawathananont, S., Son, T.T., Tan, P.S., Wang, G.H., Chien, N.V.,
prices, or the relationship between crop N uptake and N-limited Thoa, V.T.K., Phung, C.V., Stalin, P., Muthukrishnan, P., Ravi, V., Babu, M.,
yield potential. Given the small influence of changing priceN:G Chatuporn, S., Sookthongsa, J., Sun, Q., Fu, R., Simbahan, G.C., Adviento,
M.A.A., 2002. Site-specific nutrient management for intensive rice crop-
or replacing Eq. (3) with Eq. (10) on simulated N rate ping systems in Asia. Field Crops Res. 74, 37–66.
reductions, it is unlikely that uncertainty in these variables Donner, S.D., Kucharik, C.J., 2003. Evaluating the impacts of land management
would significantly affect the study’s conclusions. and climate variability on crop production and nitrate export across the
Raising crop yields to meet growing food demand while Upper Mississippi Basin. Global Biogeochem. Cycles 17, 1085.
Duvick, D.N., Cassman, K.G., 1999. Post-green revolution trends in yield
simultaneously reducing environmental impacts of agriculture
potential of temperate maize in the north-central United States. Crop Sci.
will require substantial improvements in fertilizer use 39, 1622–1630.
efficiency. Across a wide range of growing conditions for Fox, R.H., Roth, G.W., Iversen, K.V., Piekielek, W.P., 1989. Soil and tissue
the three major cereals, information to reduce uncertainty in N nitrate tests compared for predicting soil–nitrogen availability to corn.
demand and supply can contribute substantially to reducing Agron. J. 81, 971–974.
fertilizer application rates and the associated economic and Galloway, J.N., 1998. The global nitrogen cycle: changes and consequences.
Environ. Pollut. 102, 15–24.
environmental costs. Greenwood, D.J., Lemaire, G., Gosse, G., Cruz, P., Draycott, A., Neeteson, J.J.,
1990. Decline in percentage N of C3 and C4 crops with increasing plant
Acknowledgements mass. Ann. Bot. 66, 425–436.
Haefele, S.M., Wopereis, M.C.S., 2005. Spatial variability of indigenous
I thank Ken Cassman and Ivan Ortiz-Monasterio for many supplies for N, P and K and its impact on fertilizer strategies for irrigated
rice in West Africa. Plant Soil 270, 57–72.
helpful discussions, along with three anonymous reviewers for Hammer, G.L., Holzworth, D.P., Stone, R., 1996. The value of skill in seasonal
comments that greatly improved the manuscript. This work was climate forecasting to wheat crop management in a region with high
performed under the auspices of the U.S. Department of Energy climatic variability. Aust. J. Agric. Res. 47, 717–737.
D.B. Lobell / Field Crops Research 100 (2007) 210–217 217

Hardaker, J., Huirne, R., Anderson, J., Lien, G., 2004. Coping with Risk in Raun, W.R., Solie, J.B., Johnson, G.V., Stone, M.L., Mullen, R.W., Freeman,
Agriculture. CAB International, Cambridge, MA. K.W., Thomason, W.E., Lukina, E.V., 2002. Improving nitrogen use effi-
Hossain, M.F., White, S.K., Elahi, S.F., Sultana, N., Choudhury, M.H.K., ciency in cereal grain production with optical sensing and variable rate
Alam, Q.K., Rother, J.A., Gaunt, J.L., 2005. The efficiency of nitrogen application. Agron. J. 94, 815–820.
fertiliser for rice in Bangladeshi farmers’ fields. Field Crops Res. 93, 94– Raun, W.R., Solie, J.B., Johnson, G.V., Stone, M.L., Whitney, R.W., Lees, H.L.,
107. Sembiring, H., Phillips, S.B., 1998. Microvariability in soil test, plant nutrient,
Janssen, B.H., Guiking, F.C.T., Vandereijk, D., Smaling, E.M.A., Wolf, J., and yield parameters in Bermudagrass. Soil Sci. Soc. Am. J. 62, 683–690.
Vanreuler, H., 1990. A system for quantitative-evaluation of the fertility of Raun, W.R., Solie, J.B., Stone, M.L., Martin, K.L., Freeman, K.W., Mullen,
tropical soils (Quefts). Geoderma 46, 299–318. R.W., Zhang, H., Schepers, J.S., Johnson, G.V., 2005. Optical sensor-based
Knowles, T.C., Doerge, T.A., Ottman, M.J., 1991. Improved nitrogen manage- algorithm for crop nitrogen fertilization. Commun. Soil Sci. Plant Anal. 36,
ment in irrigated durum–wheat using stem nitrate analysis. 1. Nitrate uptake 2759–2781.
dynamics. Agron. J. 83, 346–352. Rauschkolb, R.S., Hornsby, A.G., 1994. Nitrogen Management in Irrigated
Lobell, D.B., Ortiz-Monasterio, J.I., Asner, G.P., 2004. Relative importance of Agriculture. Oxford University Press, New York.
soil and climate variability for nitrogen management in irrigated wheat. Riley, W.J., Ortiz-Monasterio, I., Matson, P.A., 2001. Nitrogen leaching and soil
Field Crops Res. 87, 155–165. nitrate, nitrite, and ammonium levels under irrigated wheat in Northern
Matson, P.A., Naylor, R., Ortiz-Monasterio, I., 1998. Integration of environ- Mexico. Nutr. Cycling Agroecosyst. 61, 223–236.
mental, agronomic, and economic aspects of fertilizer management. Science Saltelli, A., Chan, K., Scott, E.M., 2000. Sensitivity Analysis. John Wiley and
280, 112–114. Sons, New York.
Matson, P.A., Parton, W.J., Power, A.G., Swift, M.J., 1997. Agricultural Sheldrick, W.F., Lingard, J., 2004. The use of nutrient audits to determine
intensification and ecosystem properties. Science 277, 504–509. nutrient balances in Africa. Food Policy 29, 61–98.
Mulvaney, R.L., Khan, S.A., Ellsworth, T.R., 2006. Need for a soil-based Smil, V., 1999. Nitrogen in crop production: an account of global flows. Global
approach in managing nitrogen fertilizers for profitable corn production. Biogeochem. Cycles 13, 647–662.
Soil Sci. Soc. Am. J. 70, 172–182. Solie, J.B., Raun, W.R., Stone, M.L., 1999. Submeter spatial variability of
Raun, W.R., Johnson, G.V., 1999. Improving nitrogen use efficiency for cereal selected soil and bermudagrass production variables. Soil Sci. Soc. Am. J.
production. Agron. J. 91, 357–363. 63, 1724–1733.

Potrebbero piacerti anche