Você está na página 1de 18

Application

of Indirect Measures for Improved Nitrogen Fertilization Algorithms


William Raun, John Solie, Marvin Stone, Ivan Ortiz-Monasterio, and Randy Taylor
Oklahoma State University
International Maize and Wheat Improvement Center, CIMMYT

Abstract
Over the past thirty years, sensing technologies in agriculture have rapidly moved from less
complicated passive sensors to active ones that provide their own light source, thus allowing
for instantaneous calibration regardless of time of day or ambient conditions. Early sensing
technologies available in research included instruments developed by Li-Cor Inc., Lincoln, NE for
measuring carbon dioxide, illumination sensors for determining plant N status, and shortwave
radiation. Affordable spectrometers like those sold by Ocean Optics Inc. (Duneden, FL) capable
of collecting raw radiance values for wavelengths 300-1100 nm were and still are commonly
used in the field. From spectrometer data, the development of indices capable of detecting
nitrogen (N) and water status ensued. The most common index has been the normalized
difference vegetative index or NDVI which was found to be an excellent predictor of forage
biomass, forage N uptake, and final grain yield when combined with climatic data. The
development of active sensors capable of collecting NDVI readings day or night has most
recently been advanced to highly affordable pocket sensors. Combined this technology has
been expanded into complex algorithms that predict fertilizer N needs from mid-season sensor
and climatological data. Today, on-the-go sensor based N management in cereals is
commercially used to maximize farmer profits and minimize environmental risk via the
judicious application of N fertilizers.

Introduction
Electronic sensors are now widely used for a host of different reasons and on a range of
different crops. Early work by Sader et al. (1989) attempted to correlate forest stand and
biomass with NDVI computed using Landsat-Thermic mapper data. In Nebraska, Shanahan et
al. (2003) noted that their remote sensor system measuring GNDVI could be used to identify N
deficient plants. Given the option of using high-clearance applicators configured with a remote
sensor system, they noted that they could likely reduce pre-season N applications and promote
in-season variable N application. Raun et al. (2005a) used a GreenSeeker NDVI (normalized
difference vegetative index) active sensor to determine the growth stage in corn where spatial
variability was most expressed. This was accomplished by collecting daily NDVI readings over
selected rows of corn, and subsequently plotting means and coefficients of variation (CV) as a

function of time. Spatial variability using the CV from sensor readings as an indicator was
greatest at the V6 corn growth stage, just prior to canopy closure. Once leaves began to
overlap, spatial variability decreased markedly (stages greater than V6, Figure 1). Matching the
growth stage with the point where recognizable variability was the greatest was important. It is
unlikely that the CV data generated from NDVI readings would have been useful unless the
sensor had been designed for collecting NDVI readings from a very narrow band capable of
recognizing each plant. Monitoring CVs over time has also been used to identify the growth
stage where maximum variability should be expected in winter wheat (Arnall et al., 2006).


Figure 1. Relationship between the coefficient of variation (CV) for NDVI sensor readings
collected from four rows, August to October, 2002, Texcoco, MX. Taken from Raun et al.
(2005).


Sensors Used to Predict Yield in Season
Yield goals have dominated the N fertilizer landscape for the past 40 to 50 years. This is
because yield level in part determines demand for nutrients, since all cereal grains have
characteristic rates of removal. For example winter wheat grain N generally averages 2.4
percent, while corn grain N averages around 1.2 percent. Knowing these percentages, allows
one to determine how much N would be removed at a given level of yield. Therefore, if yield
level, or yield potential could be determined ahead of time, much like the concept of using
pre-plant yield goals, this information could be very useful when determining how much

fertilizer N should be applied. The problem is that using yield goals essentially employs the law
of averages. Most yield goals are established by taking the average grain yields of the last five
years and adding 20 to 30 percent to that value. Adding a certain percentage to the average
yield of the last five or so years essentially guarantees that the final observed yield will not be
underestimated.
Instead of guessing what the yield will be prior to planting, the concept of using mid-
season indirect measures from actively growing plants was proposed by Stone et al. (1996).
They used NDVI collected at Feekes winter wheat growth stage 5 (Large, 1954) to establish
correlation with final grain N uptake. While this work did not define an actual algorithm for
making in-season fertilizer N rate recommendations, it was a start to using indirect measures
from passive sensor readings. But a problem remained with passive sensors, since white plate
readings were required to convert radiance to reflectance, and unless enough white plate
readings were taken to properly adjust raw radiance data, calibration of the NDVI data
remained difficult. Once a calibrated NDVI value was obtained using active sensors yield
prediction equations became more commonplace since last years data could reliably be
combined with this years data and so on. Prior to this time, combining passive sensor data
collected using the same sensor but different locations at different times was cumbersome.
Not that you couldnt combine the data, but reflectance computed by dividing by a white plate
reading at one location where there may or may not have been clouds, and/or where time of
day influenced raw values restricted the utility of pairing this data together. Furthermore,
passive sensors were often equipped with a variable gain dial in order to manually compensate
for daily changes in lighting conditions, and to ensure that the data collected was enveloped by
the sensors working range. Raun et al. (1999) were first to combine sensor data collected mid-
season with climatological data to predict final grain yield. This was later confirmed in added
work using more data (Raun et al., 2001a). Similar work by Shanahan et al. (2001) noted that
GNDVI ((green-NIR)/(green+NIR)) sensor readings collected during grain filling in corn could be
used to produce relative yield maps on a field scale. However, this work collected readings at
such a late stage, that it prohibited any in-season management decisions.

Sensors Used for Fertilizer Application
Stone et al. (1996), used passive NDVI sensor data to accurately predict wheat forage N uptake,
but that was found to be limited when combining data over sites. Similar work by Taylor et al.
(1998) used NDVI to predict forage N uptake in bermudagrass. Once calibrated NDVI values
became a reality via the use of active sensors, and that could reliably deliver the same value
regardless of time of day, and/or conditions, it exponentially increased the utility of sensor
data. This meant that data collected last year could be used this year, and data collected at one

location could be combined with data obtained at another location. And while this may seem
routine, it was an important breakthrough because passive sensors required that we first have
a calibration equation between sensors, and that we meticulously collect the data under as
similar environmental conditions (time of day, temperature, etc.) as possible. Ensuing work by
Raun et al. (2001) developed constructs of yield prediction equations using solely active sensor
NDVI data, but that also proposed methods that would allow for combined location evaluation.
It was quickly recognized that combining in-season sensor data over sites and years would
exponentially add to the utility and robustness of the data for later use in predictive models.
But, in order to combine NDVI data over sites and years, it would first need to be normalized.
This is because each location was sampled at different times where growth stages were
different (Lukina et al., 2001). To do this, they divided NDVI by the number of days from
planting to sensing and found that this index termed INSEY (in season estimated yield) was
highly correlated with final grain yield, and more so than using NDVI alone. From a totally
utilitarian perspective, this was a huge breakthrough because it allowed for combined location
and year NDVI data to now be used in extrapolative models. So in ensuing years, in-season
INSEY values were plugged into the INSEY-grain yield equation to obtain a predicted yield level,
and combined with the response index (Mullen et al., 2003), used to make fertilizer N rate
recommendations and that improved NUE (Raun et al., 2002). Later work by Raun et al.
(2005b) combined active sensor readings with cumulative growing degree data from planting to
sensing (by site) into a functional algorithm for determining mid-season fertilizer N rates, based
on N responsiveness and predicted yield potential using the GreenSeekerTM NDVI sensor. The
hand held GreenSeeker sensor is now used worldwide, but it remains an expensive
research/extension tool. More recently, the engineering and agronomy team at Oklahoma
State University have produced a much more affordable NDVI pocket sensor that will be
available commercially by the end of 2010. Using commercially available and affordable active
NDVI sensors, crop consultants, and farmers all over the world are positioned to quantitatively
determine accurate mid-season fertilizer N rates. An example web page where Oklahoma State
University is actively extending sensor based N rate recommendations follows.
http://www.soiltesting.okstate.edu/SBNRC/SBNRC.php
Judicious, temporally sensitive N rates are sorely needed in an age when an estimated one
billion dollars of excess N fertilizer flows down the Mississippi River each year (Malakoff, 1998).
This becomes even more alarming when considering that what the USA looses each year
exceeds that amount that all of Sub Saharan Africa with a population of 700 million people
applies annually for cereal production (Edmonds et al., 2009).
The sufficiency approach employed by Blackmer et al. (1994) collected chlorophyll
meter data from the farmer practice, and divided this value by that obtained in reference or
non-N-limiting corn, essentially resulting in a percentage of maximum value. When the
sufficiency index computed using chlorophyll meters was less than 90% at the 8-leaf growth
stage in corn, maximum yields were not achieved with in-season N applications (Varvel et al.,
1997). In this same work, they showed that early N deficiencies could be corrected using

chlorophyll meters and the sufficiency index if the deficiency was not severe. When
deficiencies were severe applied N increased yields, but that resulted in less than the
maximum. Similar work by Scharf and Lory (2009) reported that relative reflectance and
economic optimum N rate (EONR) were correlated. While no actual method was reported, they
indicated that they could convert reflectance measurements into N rate recommendations
during variable rate sidedressing of corn.
Predictive Models and the use of Relative Values
The work by Raun et al. (2005b) is very different from a current study reporting
correlation between an in-season sufficiency index and relative yields (Varvel et al. 2007). It is
important to note that the data used by Varvel et al., (2007) requires that the relative yields or
normalized yield be computed at the end of season. This is problematic because relative
yields are determined after the fact. Their final sufficiency index versus N rate equation was
generated after averaging over years (one site). In-season chlorophyll meter readings used to
compute the sufficiency index (plot under evaluation divided by the N rich plot) are
subsequently plotted against N rate. Their prediction equation of SI versus N rate is thus the
average over all years. So, for ensuing years, the SI value computed would be plugged into
their SI-N rate equation and an average N rate computed accordingly, but that relied on the
average from previously collected historical data. By doing so, they averaged out the temporal
variability.
Alternatively, the methods proposed by Raun et al. (2005b) do not rely on historical
yield data, nor do they employ averaging. Rather based on population data (many sites planted
at different times, sensed at different times, and where yield data was also collected), INSEY
(NDVI/cumulative GDD from planting to sensing) predicts an actual yield possibility by taking
actual mid-season NDVI data, combined with cumulative GDD and predicting the yield potential
using similar environmental conditions encountered up to that point and that are represented
within the INSEY-yield equation. In this fashion, predicted yield potential is temporally
sensitive. Furthermore INSEY is completely predictive in nature since NDVI is an independent
measure, and the cumulative GDDs are site and year specific. Combined, for winter wheat,
NDVI/days from planting to sensing where GDD>0 provides an estimated value that predicts
yield potential across an entire range of yield levels (Figure 2).
With only 5 random sites from any single year where NDVI, cumulative GDD, and yield
were collected, the Oklahoma State University Precision Ag Team showed that a highly accurate
yield prediction equation could be generated, similar to their combined model that includes 48
sites where this data was collected over 9 years (www.nue.okstate.edu/Yield_Potential.htm).
This was important from two perspectives, first that a wide range of environments, even within
one year were sufficient to estimate the overall INSEY-yield equation. This finding showed that

these equations did not depend on nor did they require historical yield data. Second, the outer
boundary of the INSEY-yield equation whether estimated using all sites, or a subset remained
very well defined. In their work, they fit an exponential model to the data set, and then to
determine the outer boundary, they added 1 standard deviation to the original model (dashed
line in Figure 2).


Figure 2. Relationship between INSEY (NDVI at Feekes 5, divided by the number of days from
planting to sensing where GDD>0) and observed winter wheat grain yield.
Formulating N Algorithms
Studies by Ferguson et al. (2002) suggested that improved recommendation algorithms may
often need to be combined with methods (such as remote sensing) to detect crop N status at
early, critical growth stages followed by carefully timed, spatially adjusted supplemental
fertilization to achieve optimum N-use efficiency. Later work by Noh et al. (2005) confirmed
that it was technically feasible to design a machinery-mounted multispectral imaging sensor to
detect crop N stress reliably and accurately.
Zillmann et al. (2006) indicated that sensor-based measurements can be used efficiently for
variable N application in cereal crops when N is the main growth-limiting factor. They further

cautioned that causes for variability must be adequately understood before sensor-based
variable rate fertilization can be properly used to optimize N side-dressing in cereals.
Gautam et al. (2006) showed that leaf nitrogen could be predicted under field conditions using
a genetic algorithm and linear programming.
Work by Tubana et al. (2008) found that using an algorithm that predicted N responsiveness
(RI) and predicted yield potential (YP0), and a modification for plant stand estimated using
sensor coefficients of variation resulted in net returns to N fertilizer that were higher when
spatial variability was treated at 2.32 m2 resolution. This was important, because no other
work has attempted to treat spatial variability in N at this small of a scale.
Ortiz-Monasterio and Raun (2007) showed that by using a combination of an N-rich strip,
together with the use of a GreenSeeker the sensor and an algorithm to interpret the results
from the sensor, allowed farmers to obtain significant savings in N use and thus in farm profits.
Farm income was increased by US$56/ha, when averaged over all trials in all years.

OSU Algorithm Refined
Fundamental to this entire discussion of developing sensor based N algorithms is that yield
potential and N responsiveness are independent of one another (Raun and Solie, 2010).
Because of this, it is critical that they are determined individually. Any practice that combines
the two without considering their independence will result in flawed recommendations. An
example where yield and N responsiveness were not partitioned separately is the use of a
sufficiency concept for recommending fertilizer N employed by Varvel et al. (2007). This work
used normalized chlorophyll meter readings and relative or normalized yields. If you
understand that yield and N responsiveness are independent of one another, and both vary
significantly year to year, using a sufficiency index (SI, or normalized chlorophyll meter readings
to obtain N responsiveness) disregards the variability in N responsiveness and yield level, both
bound by the environment (year). The final N rate recommended was fixed to a percent
sufficiency, and not tied to the yield level that would be achievable that year.
The independence of yield level and N responsiveness was not initially recognized by Oklahoma
State University researchers. Early work by Stone et al. (1996) reported that NDVI alone was an
excellent predictor of wheat forage N uptake. Initially, NDVI data was collected at various
Feekes growth stages, from plots where corresponding forage biomass was also collected. Wet
biomass samples were dried, and ground and analyzed for total N using a dry combustion N
analyzer. Forage N uptake was then determined by taking the product of dry biomass times
total N.









Figure 3. Relationship between winter wheat forage N uptake and NDVI, Hennessey, OK, 1995.
Large differences existed in the NDVI N uptake relationship when evaluated over growth
stages and location, whereby distinct differences in slope and intercept components were
observed (Figure 3). Predicting real time forage N uptake from NDVI readings was exciting, but
our real goal was to predict final grain yield so as to work backwards to formulate an N rate.
Ensuing work targeted the use of NDVI readings collected at or near Feekes growth stage 5 to
predict final grain yield. A typical NDVI-grain yield relationship over three sites is reported in
Figure 4. Although a single curve could easily be fit to the three locations, each site had a
distinctly different NDVI-yield relationship.

Figure 4. Relationship between NDVI collected at Feekes growth stage 5, and final grain yield, 3
locations, 1998-2000.
Because of these differences, it was recognized that some kind of normalization of the data
would be needed in order to simultaneously evaluate NDVI data collected at different stages of
growth and different locations for yield prediction.
As a result, one of the early indices tested for predicting final grain yield was NDVI divided by
the number of days from planting to sensing. While useful, for winter wheat, it was quickly
recognized that the total number of days from planting to sensing would include many days
where there was no growth. Early on, NDVI was recognized as an excellent predictor of plant
biomass, thus dividing by days, to produce an index that was effectively, biomass produced per
day, or growth rate, made sense. In order to compensate for the days where no growth was
possible, especially for winter wheat, those days where growing degree days or GDD were less
than zero were removed.
GDD = ((Tmin + Tmax)/2) -40F
In this formula, Tmin and Tmax are the minimum and maximum observed temperatures on any
given day. In other words, no winter wheat growth is expected when the average temperature
does not exceed 40F. By removing these days, it provided a method of normalizing differences
from one location to the next and one year to the next. For winter wheat, dividing NDVI by the
number of days where GDD>0, was modestly better than dividing by the cumulative GDD from
planting to sensing, but the difference was small. For corn, dividing NDVI by cumulative GDD
(from planting to sensing), over sites, and years, resulted in better in-season estimates of yield
(INSEY).
Initially, N fertilization regimes that were tested employed only NDVI (Figure 3). It was
understood that NDVI was strongly influenced by soil color, and percent cover in the low NDVI
range. As NDVI approached and exceeded 0.70, plant N uptake would peak and level off
(yellow line). Variable rate application of N was tested using feed the rich, starve the poor
and feed the poor, starve the rich (white dashed line), approaches.







Figure 3. Different approaches used for making N recommendations based solely on NDVI
values, in the mid 1990s.
In 1995, a highly homogenous stand of wheat, with visible N deficiencies over short distances
was fertilized in 10x10 foot continuous plots, 210 feet in length. For the variable rate (VRT)
treatment, more N was applied in plots with low NDVI values, and less where NDVI was high
(>0.75) (Figure 4). The fixed rate treatment used the average of all N rates from the variable
rate treatment. Both the fixed rate and variable rate treatments had significantly higher yields
when compared to the check (Figure 4) which basically validated the need for N. However, the
variable rate treatment had a higher average yield, but also the standard deviation about the
mean was smaller when using the variable rate approach. By applying N variably not only were
yields increased, but the variability in actual yields decreased. In this case, the homogenous
stands contributed to this response via the use of an algorithm that would apply more N to
those plots having a low NDVI value. However, at other sites where stands were heterogenous,
this VRT approach did not produce the same results.


Figure 4. Total grain N uptake over 64 m (210) ft for a check strip (no N fertilizer applied),
variable rate strip (N applied based in variable NDVI), and a fixed rate strip
Unpublished data (CIMMYT) showed that N stressed spring wheat in Ciudad Obregon, could
recover completely and produce near maximum yields if fertilized early on in the season (Figure

5). With N stress this severe it was somewhat surprising to see spring wheat recover to
produce optimum yields. So why was this possible? At this location, ideal plant stands were
almost always achieved via irrigation soon after planting. This was further verified by noting
that the CVs from NDVI readings were always low at this site, an indicator that there were
homogenous stands, despite being N stressed (Figure 5).


Figure 5. Severe nitrogen deficiency in spring wheat with ideal plant stands, Ciudad Obregon,
MX.
Alternatively, many field experiments in Oklahoma demonstrated that maximum yields could
not be achieved if N was applied mid-season. For many of these experiments, plant stands
were variable, and CVs from NDVI readings were high (Figure 6).


Figure 6. Highly variable winter wheat plant stands, Hennessey, OK.
Knowing that maximum yields could still be achieved if CVs were low, but not possible when
CVs were high assisted us in recognizing that CVs from NDVI readings could be used to refine
mid-season fertilizer N rates.
The first functional algorithm based on measurable biological properties was reported by
Lukina et al. (2002). They reported that mid-season N fertilization needs were considered to be
equal to the amount of predicted grain N uptake (potential yield times grain N) minus predicted
early-season plant N uptake (at the time of sensing), divided by an efficiency factor (used 0.70).
This method of determining in-season fertilizer need was shown to decrease large area N rates
while increasing wheat grain yields when each 1 m2 area was sensed and fertilized
independently. Work by Raun et al. (2005) used a more practical approach suggesting that mid-
season N fertilizer rates be based on predicted yield potential and a response index (Figure 7,
option A). Their work showed that they could increase NUE by over 15% in winter wheat when
compared to conventional methods, but at a 0.4m2 resolution.

Figure 7. Nitrogen fertilization approaches, A. yield potential (YP0) * RI = yield potential with
added N fertilizer (YPN), B. N fertilizer applied to achieve maximum yield (YPmax) at any given
level of YP0, and C. YP0*RI=YPN but where YPN increases with low CV, and decreases with high
CV.
But this approach (A) conflicted with results in Ciudad Obregon (CIMMYT) reported that severe
early season N stress could be corrected from mid-season N applications (Figure 5, and option
B, Figure 7). In other words, no matter what the YP0 was that was predicted from INSEY, the
final N rate would be the projected grain N uptake at YPMAX minus the projected N uptake at
YPO divided by an expected N use efficiency (between 0.4 to 0.7 depending on the region). At
low YP0s the optimum N rate would be higher and the at high YP0s the optimum N rate would
get smaller and smaller (Figure 7, option B). How could you still achieve maximum yields if N
stress was so severe?


Figure 8. Grain yield of non fertilized (purple diamond) and plots that werent fertilized until
near Feekes 5 (blue square and green diamond).
Spring wheat data from CIMMYT (Ciudad Obregon, MX), for many sites (x axis) from 1989 to
2002 showed that no matter how severe the N stress from planting to first node (Feekes 6),
maximum grain yields could be produced if adequate N is applied mid season. This data was
critical to our thinking since maximum yields were seldom achieved if all N was delayed until
mid-season for winter wheat. The key to understanding the differences between these two
production systems (Figures 5 and 6) was plant stand. As indicated earlier, CVs from NDVI
readings in Figure 5 (winter wheat) were high, while CVs from NDVI readings in Figure 6 (spring
wheat) were low. What we learned was that the homogenous plant stands as in Figure 6,
ensured that the maximum response to fertilizer N could still be achieved. Ideal plant
populations were present, they were just N stressed. Alternatively, the heterogenous plant

stands as in Figure 5, restrict how much of a yield increase one could achieve since plant
populations varied from 1 m2 to the next (Figure 6, option A, Figure 7).
Both algorithms A and B have worked very well in their respective regions (algorithm A, winter
wheat, dryland, Raun et al., 2005; algorithm B, irrigated spring wheat in Mexico, Ortiz-
Monasterio and Raun 2007). Although there are stark differences between the two
approaches, algorithms A and B can be reconciled using CVs. As CVs increase, or plant stands
become more heterogenous, the potential yield with added fertilizer N decreases (Figure 7).
Similarly, as the CVs decrease or plant stands become more homogenous, the potential yield
with added fertilizer N increases. If CVs are low throughout, algorithm C will begin to look
more like algorithm B. If CVs are high throughout as in Figure 6, algorithm C will begin to look
more like algorithm A. Conceptually this makes sense considering how important plant stands
are on final yield. While wheat is capable of tillering, this physiological adjustment is not
capable of buffering enough for bad stands.

Summary
Indirect measures from sensors of all kinds have been used for decades as predictive tools. In
agriculture, the use of active sensor NDVI data has proven to be useful for predicting biomass,
forage N uptake and final grain yield. This knowledge has since been expanded into complex
algorithms that predict fertilizer N needs from mid-season sensor and climatological data.
Predicting yield without accounting for N responsiveness or predicting N responsiveness
without accounting for temporally influenced yield potential can lead to erroneous N fertilizer
recommendations. Each impact N recommendations and each change every year. The
variability in plant stands estimated using continuous NDVI readings is also an important
parameter that needs to be considered when making fertilizer rate recommendations. The CV
from NDVI readings will not affect estimated yield potential (YP0) as yield potential is
determinant. However, the CV from NDVI readings will affect the yield you can achieve if N
fertilizer is applied (YPN), and how much it influences YPN depends largely on what kind of
plant stand was achieved. The potential exists to use indirect measures for other nutrients
other than N, and for basing the use and rate of fungicides or herbicides on accurate estimates
of plant biomass. Finally, with the advent of affordable active NDVI sensors, crop consultants,
and farmers all over the world are now positioned to quantitatively determine accurate mid-
season fertilizer N rates for virtually all cereal crops.

References
Arnall, D.B., W.R. Raun, J.B. Solie, M.L. Stone, G.V. Johnson, K. Desta, K.W. Freeman, R.K. Teal,
and K.L. Martin. 2006. Relationship between coefficient of variation measured by spectral

reflectance and plant density at early growth stages in winter wheat. J. Plant. Nutr. 29:1983-
1997.
Blackmer, T.M., J.S. Schepers. 1994. Techniques for monitoring crop nitrogen status in corn.
Commun. Soil Sci. Plant Anal. 25:1791-1800.
Edmonds, Daniel E., Silvano L. Abreu, Adelheid West, Donna R. Caasi, Travis O. Conley, Michael
C. Daft, Birehane Desta, Brandon B. England, Chelsea D. Farris, Tia J. Nobles, Nehaben K. Patel,
Elliott W. Rounds, Brennan H. Sanders, Samar S. Shawaqfeh, Lakmini, Lokuralalage, Roji
Manandhar, and W. R. Raun. 2009. Cereal nitrogen use efficiency in Sub Saharan Africa. J. Plant
Nutr. 32:2107-2122.
Ferguson, R.B., G.W. Hergert, J.S. Schepers, C.A. Gotway, J.E. Cahoon, and T.A. Peterson. 2002.
Site specific nitrogen management of irrigated maize: yield and soil residual nitrate effects. Soil
Sci. Soc. Am. J. 66:544-553.
Gautam, R., S Panigrahi, and D. Franzen. 2006. Neural network optimization of remotely sensed
maize leaf nitrogen with a genetic algorithm and linear programming using five performance
parameters. Biosystems Eng. 95:359-370.
Large, E.C. 1954. Growth stages in cereals. Plant Pathol. 3:128-129.
Lukina, E.V., K.W. Freeman, K.J. Wynn, W.E. Thomason, R.W. Mullen, A.R. Klatt, G.V. Johnson,
R.L. Elliott, M.L. Stone, J.B. Solie, and W.R. Raun. 2001. Nitrogen fertilization optimization
algorithm based on in-season estimates of yield and plant nitrogen uptake. J. Plant Nutr.
24:885-898.

Malakoff, D. 1998. Death by suffocation in the gulf of Mexico. Science. 281:190-192.

Mullen, R.W., Kyle W. Freeman, William R. Raun, G.V. Johnson, M.L. Stone, and J.B. Solie. 2003.
Identifying an in-season response index and the potential to increase wheat yield with nitrogen.
Agron. J. 95:347-351.

Noh, H., Q. Zhang, S. Han, B. Shin, and D. Reum. 2005. Dynamic calibration and image
segmentation methods for multispectral imaging crop nitrogen deficiency sensors. Trans. ASAE.
48:393-401.
Ortiz-Monasterio, J.I., and W. Raun. 2007. Reduced nitrogen for improved farm income for
irrigated spring wheat in the Yaqui Valley, Mexico, using sensor based nitrogen management. J.
of Agric. Sci. 145:215-222.
Raun, W.R., G.V. Johnson, M.L. Stone, J.B. Solie, W.E. Thomason, and E.V. Lukina. 1999. In-
season prediction of yield potential in winter wheat. Better Crops. 83:(2) 24-25.


Raun, W.R., G.V. Johnson, M.L. Stone, J.B. Solie, E.V. Lukina, W.E. Thomason and J.S. Schepers.
2001a. In-season prediction of potential grain yield in winter wheat using canopy reflectance.
Agron. J. 93:131-138.

Raun, W.R., G.V. Johnson, M.L. Stone, J.B. Solie, E.V. Lukina, W.E. Thomason and J.S. Schepers.
2001b. In-season prediction of potential grain yield in winter wheat using canopy reflectance.
Agron. J. 93:131-138.

Raun, W.R., J.B. Solie, G.V. Johnson, M.L. Stone, R.W. Mullen, K.W. Freeman, W.E. Thomason,
and E.V. Lukina. 2002. Improving nitrogen use efficiency in cereal grain production with optical
sensing and variable rate application. Agron. J. 94:815-820.
Raun, W.R., J.B. Solie, K.L. Martin, K.W. Freeman, M.L. Stone, K.L. Martin, G.V. Johnson, and
R.W. Mullen. 2005a. Growth stage, development, and spatial variability in corn evaluated using
optical sensor readings. J. Plant Nutr. 28:173-182.
Raun, W.R., J.B. Solie, M.L. Stone, K.L. Martin, K.W. Freeman, R.W. Mullen, H. Zhang J.S.
Schepers, and G.V. Johnson. 2005b. Optical sensor based algorithm for crop nitrogen
fertilization. Commun. Soil Sci. Plant Anal. 36:2759-2781.
Raun, William R., and John B. Solie. 2010. Independence of yield potential and N
responsiveness. J. Precision Agric. (in press).
Sader, Stevena, Robert B. Waide, William T. Lwrence, and Armond T. Joyce. 1989. Tropical
forest biomass and successional age class relationships to a vegetation index derived from
Landsat TM data. Remote Sens. Environ. 28:143-156.
Shanahan, John F., James S. Schepers, Dennis D. Francis, Gary E. Varvel, Wallace W. Wilhelm,
James M. Tringe, Mike R. Schlemmer, and David J. Major. Use of remote sensing imagery to
estimate corn grain yield. Agron. J. 93:583-589.

Shanahan, J.F., K. Holland, J.S. Schepers, D.D. Francis, M.R. Schlemmer, and R. Caldwell. 2003.
Use of crop reflectance sensors to assess corn leaf chlorophyll content. In: VanToai, T., Major,
D., McDonald, M., Schepers, J., and Tarpley L., editors. Digital Imaging and Spectral Techniques:
Applications to Precision Agriculture and Crop Physiology. ASA Special Pub. 66. ASA, Madison,
WI. p. 129-144.
Stone, M.L., J.B. Solie, W.R. Raun, R.W. Whitney, S.L. Taylor and J.D. Ringer. 1996. Use of
spectral radiance for correcting in-season fertilizer nitrogen deficiencies in winter wheat. Trans.
ASAE 39(5):1623-1631.

Taylor, S.L., W.R. Raun, J.B. Solie, G.V. Johnson, M.L. Stone, and R.W. Whitney. 1998. Use of

spectral radiance for correcting nitrogen deficiencies and estimating soil test variability in an
established bermudagrass pasture. J. of Plant Nutr. 21:2287-2302.

Tubaa, B.S., D.B. Arnall, O. Walsh, B. Chung, J.B. Solie, K. Girma, and W.R. Raun. 2008.
Adjusting midseason nitrogen rate using a sensor-based optimization algorithm to increase use
efficiency in corn. J. Plant. Nutr. 31:1393-1419.

Varvel, G.E., J.S. Schepers, and D.D. Francis. 1997. Ability for in season correction of nitrogen
deficiency in corn using chlorophyll meters. Soil Sci. Soc. Am. J. 61:12331239.
Varvel, G.E., W.W. Wilhelm, J.F Shanahan, and J.S. Schepers. 2007. An algorithm for corn
nitrogen recommendations using a chlorophyll meter based sufficiency index. Agron. J. 99:701-
706.
Zillmann, E., S. Graeff, J. Link, W.D. Batchelor, and W. Claupein. 2006. Assessment of cereal
nitrogen requirements derived by optical on-the-go sensors on heterogeneous soils. Agron. J.
98: 682-690.

Você também pode gostar