Crop yield estimation model for Iowa using remote sensing

and surface parameters

Anup K. Prasad
a
, Lim Chai
b
, Ramesh P. Singh
a,b,
*
, Menas Kafatos
b
a
Department of Civil Engineering, Indian Institute of Technology, Kanpur 208016, India
b
Center for Earth Observing and Space Research, School of Computational Sciences,
George Mason University, Fairfax, VA 22030, USA
Received 21 October 2004; accepted 8 June 2005
Abstract
Numerous efforts have been made to develop various indices using remote sensing data such as normalized difference
vegetation index (NDVI), vegetation condition index (VCI) and temperature condition index (TCI) for mapping and monitoring
of drought and assessment of vegetation health and productivity. NDVI, soil moisture, surface temperature and rainfall are
valuable sources of information for the estimation and prediction of crop conditions. In the present paper, we have considered
NDVI, soil moisture, surface temperature and rainfall data of Iowa state, US, for 19 years for crop yield assessment and
prediction using piecewise linear regression method with breakpoint. Crop production environment consists of inherent sources
of heterogeneity and their non-linear behavior. A non-linear Quasi-Newton multi-variate optimization method is utilized, which
reasonably minimizes inconsistency and errors in yield prediction.
Minimization of least square loss function has been carried out through iterative convergence using pre-dened empirical
equation that provided acceptable lower residual values with predicted values very close to observed ones (R
2
= 0.78) for Corn
and Soybean crop (R
2
= 0.86) for Iowa state. The crop yield prediction model discussed in the present paper will further improve
in future with the use of long period dataset. Similar model can be developed for different crops of other locations.
# 2005 Elsevier B.V. All rights reserved.
Keywords: Crop yield prediction model; NDVI; Iowa; Soybean; Corn
1. Introduction
Monitoring of crop conditions is important for the
economic development of any nation. The use of
remote sensing has proved to be very important in
monitoring the growth of agricultural crops and in
irrigation scheduling. Efforts have been made to
develop various indices for different crops of different
regions throughout the globe. The production of crop
and prediction of crop yield have direct impact on
year-to-year national and international economies and
play an important role in the food management (Hayes
and Decker, 1996). Using remote sensing data, efforts
have been made to develop various indices such as:
www.elsevier.com/locate/jag
International Journal of Applied Earth Observation
and Geoinformation 8 (2006) 2633
* Corresponding author. Tel.: +91 512 2597295;
fax: +91 512 2597395.
E-mail address: [email protected] (R.P. Singh).
0303-2434/$ see front matter # 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.jag.2005.06.002
normalized difference vegetation index (NDVI), vege-
tation condition index (VCI) and temperature condition
index (TCI). These indices are commonly used for
drought detection, monitoring excessive soil wetness,
assessment of weather impacts on vegetation and
evaluation of vegetation health and productivity
(Unganai and Kogan, 1998; Kogan, 2001, 2002; Kogan
et al., 2003; Singh et al., 2003). The NDVI data have
been used extensively in vegetation monitoring, crop
yield assessment and forecasting (Hayes et al., 1982;
Benedetti and Rossinni, 1993; Quarmby et al., 1993).
The US Corn Belt provides approximately 80% of
the overall maize production for the entire US, and
accounts for 36% of the global maize production
(USDA, 1987). The United States Department of
Agriculture (USDA) forecasts crop supply and
demand estimates including expected crop yields
every month during the crop season beginning in early
January. The ability of the CERES-Maize model to
estimate annual uctuations in maize production for
the US Corn belt was tested for the years 19821985
(Hodges et al., 1987). Spatial interactions in the
CROPGRO-Soybean and CERES-Maize models and
comparison of simulated and measured data are also
studied (Batchelor et al., 2002). Optimizing crop-
growth/yield models for Corn and Soybeans crops in
USA was evaluated using the multi temporal high-
resolution airborne digital imagery.
NDVI has been considered to be a useful way for
crop yield assessment models using various
approaches from simple integration to more compli-
cated transformation. NDVI reects vegetation green-
ness, thus it indicates levels of healthiness in the
vegetation development. Although vegetation devel-
opment of crop elds may differ from those of natural
vegetation because of human inuences involved such
as irrigation, use of fertilizer and pesticides, NDVI is
considered as a valuable source of information for the
crop conditions. Different methods such as neural
network (Stoikos, 1995), autoregressive (AR) state-
space models (Wendroth et al., 2003), least-square
regression (Jones, 1982), exponential-linear (EL) crop
growth algorithm (Oroda, 2001) and numerical crop
yield model (Hayes et al., 1982) have been used to
predict crop yield with moderate success.
The ground and satellite (NOAA, Meteosat, etc.)
measurements are commonly used to deduce various
parameters such as evapotranspiration (Doorenbos
and Kassam, 1979; Oroda, 2001), NDVI, soil type
(Garcia-Paredes et al., 2000), light, carbon dioxide,
temperature, water and the rate of growth and
development (Monteith, 1981) and cropweather
relations (Watson, 1963; Baier, 1977; Frere and
Popov, 1979; WMO, 1982; van Keulen, 1987;
NCMRWF, 1990; Jain and Ranjana, 2000) are
increasingly used to predict crop yield. Present crop
yield estimation is based on various methods and data
sources like eld surveys, expert knowledge, trend
analysis, regression analysis, statistical models and
crop growth simulation models. In this paper, we have
developed a crop yield prediction model based on
Iowa Corn and Soybean yield estimates utilizing
NDVI, surface temperature (ST), precipitation and
soil moisture (SM).
2. Methods
2.1. Selection of a crop region
State of Iowa (Fig. 1a) belongs to the US Corn Belt,
recorded the highest harvested cropland in 1997
(USDA, 1997). We have considered 10 top counties of
Corn and Soybean production in 2000 of Iowa state,
which is total combination of 14 counties to develop a
crop yield assess model for Corn and Soybeans.
Iowa state is lying in the humid temperate zone
which can be approximated as a rectangular area with
corner coordinates (978W, 43.58N), (978W, 40.58N),
(908W, 40.58N) and (908W, 43.58N). The total area of
Iowa is about 35,756,390 and 11,700,000 acres
consists of planted Corn for grain in the year 2001,
which is about 32.7% of the total area. Iowa state
produced 17.9% of all Corn for grain in the US total of
1997, ranking the highest production in the US
(USDA, 1997) and Soybean farming is the second
largest crop after Corn, and about 10,920,000 acres
were harvested in the year 2001.
2.2. Data
Crop yield data (Fig. 1b) of Corn and Soybean for
Iowa state have been used, Corn crop is planted in
early May, growth in biomass occurs from June to
September and is ready for harvesting in September.
Therefore, temporal annual average of NDVI, soil
A.K. Prasad et al. / International Journal of Applied Earth Observation and Geoinformation 8 (2006) 2633 27
moisture, surface temperature and rainfall (RF) data
for period May to September have been used in the
present analysis for 19 years from 1982 to 2001
(excluding 1994). Soybean crop season starts with
active sowing period from mid of May to early June.
Crop growth occurs from June to September and
harvesting mostly in October. For Soybean, temporal
annual average of NDVI, SM, STand rainfall data for
the months June to September for 19 years period
(1982 to 2001, excluding 1994) have been used for
analysis. Monthly composite NDVI data were
spatially averaged over Iowa region (annual mean
of growth season average; Corn (May to September)
and Soybean (June to September)), from1982 to 2001.
Fig. 2ad shows the annual growth season average of
NDVI, SM, ST and RF from 1982 to 2001.
2.2.1. National Agricultural Statistics Service
(NASS) Corn and Soybeans yield estimates
Corn and Soybean yield estimates have been
obtained electronically from NASS/USDA database
site (http://www.usda.gov/nass/). The Corn yield
(Fig. 1b) estimates are available for the years from
1866 to 2004 whereas the Soybeans yield (Fig. 1b)
estimates are available for the years from 1924 to
2004. Note that Soybeans belong to the category of
Oilseeds and Cotton. The datasets provide planted
and harvested acres, yield per acre and production in
bushel.
2.2.2. NDVI
We have used 8 km 8 km monthly composite
continental NDVI datasets from the pathnder
advanced very high-resolution radiometer (AVHRR)
collected by National Oceanic and Atmospheric
Administrations (NOAA) polar orbiters from 1982
to 2001. NDVI is level 3 data derived from Channel 1
(visible band) and Channel 2 (near-infrared band). The
visible wavelength attenuation encountered in
AVHRR observations do not show signicant differ-
ence from the Landsat multi-spectral scanner (MSS)
and Thematic Mapper (TM) sensors or the French
Satellite Probatoire dObservation de la Terre (SPOT)
sensors. The attenuation is considerably more
sensitive to water vapor in the near infrared than
the other land observing sensors (Goward et al., 1991).
The NDVI derived from NOAA 11 data shows values
0.05 higher than earlier NOAA missions for African
desert. An error in the solar zenith angle (SZA) has
also been discovered. However, NDVI in the
pathnder AVHRR land (PAL) dataset received by
the users in HDF format are less affected by errors in
the solar zenith angle (http://daac.gsfc.nasa.gov/
CAMPAIGN_DOCS/LAND_BIO/zenith_angle_me-
mo.html).
Quantitative interpretation of NDVI is compli-
cated by numerous intervening factors such as
instrument calibration, incident solar irradiance,
nominal atmospheric attenuation, variable spatial
A.K. Prasad et al. / International Journal of Applied Earth Observation and Geoinformation 8 (2006) 2633 28
Fig. 1. (a) Map of USAshowing location of Iowa for which Corn and Soybean crop yield prediction model is derived. (b) Corn and Soybean crop
yield from 1982 to 2001.
resolution, anisotropy with off-nadir views, and cloud
occurrence. These factors combine to produce a global
dataset that has, at best, a measurement precision of
E0.1 NDVI units (E10% error) over 1 year at a
temporal resolution somewhere between 10 days and 1
month (Goward et al., 1991).
The dry matter accumulation of the Corn leaves
mature by mid July and almost constant from mid July
to the beginning of September (Iowa State University
of Science and Technology, 1996). The maximum
vegetation greenness of Iowa region generally occurs
in August. We have examined spatially averaged
NDVI for the months of Corn growing season, which
is used for Corn yield estimates. NDVI data are
permanently missing for September to December
1994; therefore, 1994 data is excluded frommodel due
to non-availability of September NDVI.
2.2.3. Precipitation
Rainfall data (http://www1.ncdc.noaa.gov/pub/
data/cirs/) is available as monthly average within a
climatic division calculated using equal weight to
stations reporting both temperature and precipitation
within a division. Monthly average total rainfall
(unit: mm) is taken from above dataset from 1982 to
2001.
2.2.4. Temperature
Monthly surface temperature data is taken from
NOAA NCEP-NCAR CDAS-1 monthly diagnostic
surface temperature database (unit: Kelvin) (http://
iridl.ldeo.columbia.edu/SOURCES/.NOAA/.NCEP-
NCAR/).
2.2.5. Soil moisture
Soil moisture data have been taken from NOAA
NCEP CPC global monthly soil moisture dataset
(unit: mm) http://iridl.ldeo.columbia.edu/SOURCES/
.NOAA/.NCEP/). Soil moisture is based on the water
balance in the soil.
2.3. Methodology
Crop yield is considered as dependent variable that
varies diversely with independent variables like
NDVI, SM, ST and RF. Variations of NDVI, SM,
A.K. Prasad et al. / International Journal of Applied Earth Observation and Geoinformation 8 (2006) 2633 29
Fig. 2. Annual growth season average NDVI (a) and soil moisture (SM) (mm) (b) variation for Iowa (19822001). Annual growth season
average surface temperature (Kelvin) (c) and rainfall (mm) (d) variation for Iowa (19822001).
ST and RF data do not follow any distinct linear
combination and with respect to crop yield. It is
therefore difcult to model such a dynamic relation-
ship using conventional linear methods like multi-
variable multiple regression. Non-linear estimation
approach is used to compute the relationship between
a set of independent variables and a dependent
variable. A two-piece empirical equations is devised
and solved using non-linear Quasi-Newton method.
Crop yield estimation equation with coefcients is
derived by minimizing loss function for Corn and
Soybean crop separately based on the 19 years dataset.
Non-linear piecewise linear regression with break-
point (Quasi-Newton method) (Belegundu and Chan-
drupatla, 1999; Setiono et al., 2002) have been used to
develop model for prediction of crop yield. Various
steps of this model involve (1) identifying an initial
model, (2) iterative convergence using the stepping
criteria, and (3) terminating the search when either
stepping criteria or number of iterations allowed
reached its limit.
Empirical equation is based on piecewise linear
regression method with breakpoint. Quasi-Newton
methods have been used for multi-variate optimization
(Belegundu and Chandrupatla, 1999). It is non-linear
method that has been used to minimize least square
loss function through iterative convergence of pre-
dened empirical equation. In Quasi-Newton Method,
the rst-order derivative of the function at a point is
computed to nd the slope of a function at that point.
Subsequent second-order derivative indicates how fast
the slope is changing at the respective point and its
direction. The Quasi-Newton method evaluates the
function at different points at each step in order to
estimate the rst-order derivatives and second-order
derivatives, which is used to nd out the minimum of
the loss function. Quasi-Newton is an iterative method
that is primarily governed by minimization of chosen
loss function (i.e. achieved global minima point where
observed is closest possible to simulated value which
in principle can be a 100% match). Chosen loss
function can differ based on objective and here it is a
commonly applied least square loss function, i.e.
square of the difference between predicted and
observed value (objective is to achieve lowest possible
difference between observed and predicted value).
The iterative method works for multi-independent
variables and dependent variable crop yield both
above and below the breakpoint. A non-linear
optimization approach achieves acceptable lower
residual values with predicted values very close to
observed values.
2.4. Model
Coefcients of empirical equation have been
obtained using this method. Breakpoint (m) chosen
is the mean of 19 year Corn or Soybean crop yield
of Iowa. In rare case, if breakpoint (m) becomes
equal to crop yield, condition of crop yield >
breakpoint should be applied. The model empirical
equations for Corn and Soybean crops, thus ob-
tained with coefcients (Table 1) is given as:
crop yield c
1
a
1
NDVI a
2
SM
a
3
ST a
4
RF
ffor crop yield <breakpoint mg
or c
2
b
1
NDVI b
2
SM b
3
ST
b
4
RF
ffor crop yield >breakpoint mg
where NDVI, normalized difference vegetation index;
SM, soil moisture (mm); ST, surface temperature
(Kelvin); RF, rainfall (mm); c
1
, c
2
, a
i
, b
i
, for i = 14
are coefcients (Table 1); m = mean Corn or Soybean
crop yield (19822001, excluding 1994) (breakpoint).
A.K. Prasad et al. / International Journal of Applied Earth Observation and Geoinformation 8 (2006) 2633 30
Table 1
Coefcients and breakpoint for Corn and Soybean crop yield model
empirical equation for Iowa
Model variable Coefcients Corn Soybean
Constant c
1
12733.9 445.2006
NDVI a
1
156.4193 1.12634
SM a
2
0.17942 0.02641
ST a
3
51.13045 1.25083
Rainfall a
4
31.7037 0.45429
Constant c
2
4039.71 211.01
NDVI b
1
80.32816 17.57264
SM b
2
0.017716 0.02533
ST b
3
16.03083 0.967642
Rainfall b
4
8.51827 0.45243
Breakpoint 124.4737 40.36842
R 0.88055 0.93053
Variance accounted (%) 77.53 86.58
R
2
0.78 0.86
Loss function used is least square, i.e L
f
= (obser-
ved predicted)
2
The Quasi-Newton method utilizes this loss
function to arrive at a solution closest possible to
observed data. At each iteration loss function is
computed to minimize square of difference between
the observed and predicted crop yield using pre-
dened empirical equation. The method is an
optimization process, which runs as long as initial
values, stepping values, number of iterations and
convergence criteria are favorable. It terminates if
any of these bounding conditions are fullled.
Therefore, loss function can approach theoretically
up to R
2
100%. It depends on degree to which
independent variables considered control depend-
able variable and absence of any other major
governing factor affecting crop yield in a year. This
approach can give results, which are closer to real
value. Small dataset (710 years) gave better result
than large dataset (say 19 years data) due to less
variation in pattern of SM, ST, RF, NDVI and crop
yield.
NDVI, SM, STand RF data are major indicators or
variables controlling the normal crop growth. This
approach can be used to predict any shortfall in crop
yield using empirical equations. Coefcients used in
derived equation largely depend on pool of historical
data. Studying NDVI, SM, STand RF data region wise
before harvesting period and employing above
methodology (with modications) can be used to
predict crop yield for that season.
3. Results and discussion
The tted model (Fig. 3b) agrees well with
Soybean crop. Residual values in individual years
are within acceptable limits with more or less even
distribution of difference from observed crop yield.
Similarly, Corn crop (Fig. 3a) also shows even
distribution of residual values for all years except
for 1993. Still the model agrees well for the year 1993
that has witnessed a steep fall (46%) in observed
crop yield compared with the preceding year (Table 2).
It shows that crop yield is possibly also affected by
some other factors that may dominate in some years
causing steep fall. A moderate to high R
2
-values 0.78
for Corn and 0.86 for Soybean showthat in most years,
crop yield is largely governed by variables considered
in model.
However, other factors like pests, diseases and
human activities can cause local variations in
predicted crop yield. This is a serious limitation to
any forecasting method including this. However,
inclusion of NDVI in model partly takes care of loss
due to diseases or pests that directly affect vegetation.
Crop yield prediction method is expected to yield
good prediction results due to low residual values
comparing with the historical data. The model can be
optimized and evolve to be more rugged with growing
historical data for better prediction. This exercise
includes data for 19 years and yield better prediction
results considering year-to-year variation of control-
ling factors.
A.K. Prasad et al. / International Journal of Applied Earth Observation and Geoinformation 8 (2006) 2633 31
Fig. 3. Observed and predicted crop yield (model) of Corn (a) and Soybean (b) crop for Iowa. R
2
-values are 0.78 for Corn and 0.86 for
Soybean.
This model can be further improved with usage of
high-resolution data with availability of multi-year
data. This method involves simple input dataset and can
be suitably modied to add or drop variables depending
on climate, region and crop type and can be easily
extended for other crops. Modication in approach is
required to use this methodology in prediction mode.
For instance, a weekly or 10 days composite of input
data (before harvesting month) can be compared with
other measures of crop yield like dry matter productiv-
ity (DMP) and net primary productivity (NPP) for
multiple years to arrive at predicted crop yield for
current year.
4. Conclusion
Crop production environment consists of inherent
sources of heterogeneity due to numerous parameters.
The model discussed in the present paper reasonably
minimizes inconsistency and errors in yield prediction
giving high R
2
-values with maximum accounting of
variability in model. The model takes care of most of
the parameters, which control the crop yield. This
method can be used to predict crop yield for other
crops as well as Corn and Soybeans. Based on data
obtained before harvest, crop yield can be predicted
with acceptable accuracy. Piecewise linear regression
equation with breakpoint (Quasi-Newton method) can
be extended to other countries as well, where crop
production is primarily dependent on weather and
climatic conditions. The model developed inthe present
paper shows a promising result, which can be useful for
forecasting crop yields such as Corn and Soybeans and
other crops in regional and global scales.
Acknowledgements
We are grateful to Dr. Katarzyna Dabrowska-
Zielinske and one anonymous Reviewer for their
comments and suggestions, which have helped us to
improve the original version of the manuscript.
References
Baier, W., 1977. Cropweather models and their use in yield
assessments. WMO Technical Note no. 151. World Meteorolo-
gical Organization, Geneva, p. 48.
A.K. Prasad et al. / International Journal of Applied Earth Observation and Geoinformation 8 (2006) 2633 32
Table 2
Predicted and observed crop yield of Corn and Soybean crop for Iowa using non-linear piecewise linear regression with breakpoint (Quasi-
Newton) method
Year Corn crop yield Soybean crop yield
Observed Predicted Residual Observed Predicted Residual
1982 120 106.5419 13.4581 36.5 34.9126 1.5874
1983 87 83.7547 3.2453 35.0 31.2125 3.7875
1984 112 102.6428 9.3572 31.5 33.8647 2.3647
1985 126 138.4765 12.4765 38.0 36.8840 1.1160
1986 135 130.1821 4.8179 41.5 41.9016 0.4016
1987 130 131.7217 1.7217 43.5 43.3660 0.1340
1988 84 98.5832 14.5832 31.0 34.2990 3.2990
1989 118 123.8340 5.8340 39.0 37.7836 1.2164
1990 126 136.5609 10.5609 41.5 41.9585 0.4585
1991 117 110.6744 6.3256 40.5 43.4231 2.9231
1992 147 139.1637 7.8363 44.0 43.6675 0.3325
1993 80 105.4158 25.4158 31.0 33.0437 2.0437
1995 123 109.5531 13.4469 44.0 43.0535 0.9465
1996 138 139.6419 1.6419 44.0 43.6592 0.3408
1997 138 137.0569 0.9431 46.0 46.8009 0.8009
1998 145 139.4707 5.5293 48.0 44.5158 3.4842
1999 149 148.6499 0.3501 44.5 44.7721 0.2721
2000 144 144.5584 0.5584 43.5 44.7258 1.2257
2001 146 138.5173 7.4827 44.0 43.1563 0.8437
Batchelor, W.D., Basso, B., Paz, J.O., 2002. Examples of strategies
to analyze spatial and temporal yield variability using crop
models. Eur. J. Agron. 18 (12), 141158.
Belegundu, A.D., Chandrupatla, T.R., 1999. Optimization Concepts
and Applications in Engineering. Pearson Education Asia.
Benedetti, R., Rossinni, P., 1993. On the use of NDVI proles as a
tool for agricultural statistics: the case study of wheat yield
estimate and forecast in Emilia Romagna. Remote Sens.
Environ. 45, 311326.
Doorenbos, J., Kassam, A.H., 1979. Yield Response to Water. FAO
Irrigation Drainage Paper 33, United Nations, Rome, p. 193.
Frere, M., Popov, G.F., 1979. Agrometeorological crop monitoring
and forecasting. Plant Production and Protection Division, no.
17. Food and Agriculture Organization of the United Nations,
Rome, p. 64.
Garcia-Paredes, J.D., Olson, K.R., Lang, J.M., 2000. Predicting
Corn and Soybean productivity for Illinois soils. Agric. Syst. 64
(3), 151170.
Goward, S.N., Markham, B., Dye, D.G., Dulaney, W., Yang, J.,
1991. Normalized difference vegetation index measurements
from the advanced very high resolution radiometer. Remote
Sens. Environ. 35, 257277.
Hayes, J.T., ORourke, P.A., Terjung, W.E., Todhunter, P.E., 1982.
YIELD: A numerical crop yield model of irrigated and rainfed
agriculture. Publications in Climatology, p. 35.
Hayes, M.J., Decker, W.L., 1996. Using NOAA AVHRR data to
estimate maize production in the United States Corn Belt. Int. J.
Remote Sens. 17, 31893200.
Hodges, T., Botner, D., Sakamoto, C., Haug, J.H., 1987. Using the
CERES-Maize model to estimate production for the US Corn
Belt. Agric. Forest Meteorol. 40, 293303.
Iowa State University of Science and Technology, 1996. Howa Corn
Plant Develops. Special Report No. 48, reprinted February 1996.
Jain, R.C., Ranjana, A., 2000. Statistical models for crop
yield forecasting. In: Singh, S.V., Rathore, L.S., Saseendran,
S.A., Singh, K.K. (Eds.), Proceedings of the National Workshop
on Dynamic Crop Simulation Modeling for Agrometeorological
Advisory, National Centre for Medium Range Weather Fore-
casting, New Delhi, pp. 295304.
Jones, D.R., 1982. A statistical inquiry into cropweather depen-
dence. Agric. Meteorol. 26, 91104.
Kogan, F.N., 2001. Operational space technology for global vegeta-
tion assessment. Bull. Am. Meteorol. Soc. 82 (9), 19491964.
Kogan, F.N., 2002. World Droughts in the new millennium from
AVHRR-based vegetation health indices. Eos Trans. Am. Geo-
phys. Union 48 (557), 562563.
Kogan, F.N., Gitelson, A., Zakarin, E., Spivak, L., Lebed, L., 2003.
AVHRR-based spectral vegetation index for quantitative assess-
ment of vegetation state and productivity: calibration and vali-
dation. Photogrammetric Eng. Remote Sens. 69 (8), 899906.
Monteith, J.L., 1981. Climatic variation and the growth of crops.
Quart. J. R. Meteorol. Soc. 107, 749774.
NCMRWF, 1990. Cropweather modeling. Development of Agro-
meteorology in India. National Center for Medium Range
Weather Forecast, New Delhi, pp. 8491.
Oroda, A., 2001. The International Archives of the Photogrammetry.
Remote Sens. Spatial Inform. Sci. XXXIV, 6672 Part 6/W6.
Quarmby, N.A., Milnes, M., Hindle, T.L., Silicos, N., 1993. The use
of multitemporal NDVI measurements from AVHRR data for
crop yield estimation and prediction. Int. J. Remote Sens. 14,
199210.
Setiono, R., Kheng, L.W., Zurada, J.M., 2002. Extraction of rules
from articial neural networks for nonlinear regression. IEEE
Trans. Neural Netw. 13 (3), 564577.
Singh, R.P., Roy, S., Kogan, F., 2003. Vegetation and temperature
condition indices from NOAA AVHRR data for drought mon-
itoring over India. Int. J. Remote Sens. 24, 43934402.
Stoikos, G., 1995. Sugar beet crop yield prediction using articial
neural networks (in Greek). In: Proceedings of the Modern
Technologies Conference in Automatic Control, Athens,
Greece, pp. 120122.
Unganai, L.S., Kogan, F.N., 1998. Drought monitoring and Corn
yield estimation in Southern Africa from AVHRR data. Remote
Sens. Environ. 63, 219232.
USDA, 1997. 1997 Census of Agriculture, vol. 1. Washington, DC,
US.
USDA, 1987. Major Crop Areas and Climatic Proles. Agriculture
Handbook No. 664. Washington, DC, US.
van Keulen, H., 1987. Forecasting and estimating effects of weather
on yield. In: Wisiol, K., Hesketh, J.D. (Eds.), Plant Growth
Modeling for Resource Management. CRC Press, Boca Raton,
FL, pp. 105124.
Watson, D.J., 1963. Climate, weather and plant yield. In: Evans, L.T.
(Ed.), Environmental Control of Plant Growth. Academic Press,
New York, pp. 337350.
Wendroth, O., Reuter, H.I., Kersebaum, K.C., 2003. Predicting yield
of barley across a landscape: a state-space modeling approach. J.
Hydrol. 272 (14), 250263.
WMO, 1982. The Effect of Meteorological Factors on Crop Yields
and Methods of Forecasting The Yield, Technical Note
no. 174. World Meteorological Organization, Geneva, p. 54,
1982.
A.K. Prasad et al. / International Journal of Applied Earth Observation and Geoinformation 8 (2006) 2633 33