Sie sind auf Seite 1von 9

J Agro Crop Sci (2014) ISSN 0931-2250

DROUGHT STRESS

Predicting Crop Yields with the Agricultural Reference Index


for Drought
P. Woli1, J. W. Jones1, K. T. Ingram1 & G. Hoogenboom2
1 Department of Agricultural and Biological Engineering, University of Florida Gainesville, FL, USA
2 Washington State University Prosser, WA, USA

Keywords Abstract
agricultural reference index for drought;
drought index; stress sensitivity; water stress; A generic agricultural drought index, called Agricultural Reference Index for
yield model; yield prediction Drought (ARID), was designed recently to quantify water stress for use in pre-
dicting crop yield loss from drought. This study evaluated ARID in terms of its
Correspondence ability to predict crop yields. Daily historical weather data and yields of cotton,
P. Woli
maize, peanut and soybean were obtained for several locations and years in the
Department of Agricultural and Biological
south-eastern USA. Daily values of ARID were computed for each location and
Engineering
University of Florida converted to monthly average values. Using regression analyses of crop yields vs.
1741 Museum Road monthly ARID values during the crop growing season, ARID-yield relationships
Gainesville were developed for each crop. The ability of ARID to predict yield loss from
FL 32611 drought was evaluated using the root mean square error (RMSE), the Willmott
USA index and the modelling efficiency (ME). The ARID-based yield models predicted
Tel.: +1 334 844 5636
relative yields with the RMSE values of 0.144, 0.087, 0.089 and 0.142
Fax: +1 334 844 4586
(kg ha1 yield per kg ha1 potential yield); the Willmott index values of 0.70,
Email: premwoli@ufl.edu
0.92, 0.86 and 0.79; and the ME values of 0.33, 0.73, 0.60 and 0.49 for cotton,
Accepted February 14, 2014 maize, peanut and soybean, respectively. These values indicated that the ARID-
based yield models can predict the yield loss from drought for these crops with
doi:10.1111/jac.12055 reasonable accuracy.

not plant physiology based (Woli et al. 2012a). Because


Introduction
photosynthetic assimilation, that is, yield formation, is a
The shortage of water for a specified need is referred to as a plant physiological process, only a physiology-based
drought condition (Dracup et al. 1980). An agricultural drought index can predict the drought-induced yield loss
drought occurs when the amount of plant available water more accurately.
in the root zone due to precipitation falls short of the evap- Considering this aspect and the other essential features of
orative demand of the atmosphere (WMO 1975, Wilhite an agricultural drought index that many current drought
and Glantz 1985, Maracchi 2000). The severity of drought indices lack simplicity, generality, soilplantatmosphere
may be quantified using an indicator, called a drought continuum basis, and daily temporal resolution Woli et al.
index, which integrates all meteorological, hydrological and (2012a) developed a new drought index recently, called the
agricultural information typically into a number and gives Agricultural Reference Index for Drought (ARID). The phys-
a comprehensible big picture on drought conditions (Nara- iological basis to ARID comes from the use of transpiration
simhan and Srinivasan 2005, Hayes 2013). One of the (T) as T is directly related to stomatal opening, which in
intended, and probably the most important, uses of an turn, affects carbon dioxide assimilation. Thus, the amount
agricultural drought index is to predict yield loss from of biomass or marketable yield produced can be determined
drought (Sakamoto 1978, Kumar and Panu 1997, Quiring by the amount of water transpired during the growing sea-
and Papakryiakou 2003). Although a number of drought son. When T falls short of the evaporative demand of the
indices are available for agricultural purposes, the majority atmosphere (potential evapotranspiration), plant water defi-
of them are not able to estimate crop yield loss due to cit occurs. Based on this concept, ARID is computed as the
drought, the ultimate interest of farmers, because they are ratio of plant water deficit to plant water need as

2014 Blackwell Verlag GmbH, 200 (2014) 163171 163


Woli et al.

ARID 1  T=ETo 1 depends on climatic factors that are associated with evapo-
rative demand and is controlled to a large extent by atmo-
where T is transpiration (mm) and ETo is the potential spheric relative humidity (Arkley 1963).
evapotranspiration (mm) of the reference grass (Allen et al. Cole and Mathews (1923) and Mathews and Brown
2005) whose use has provided simplicity and generality to (1938) found similar linear relationship as Eqn 3 between
the index. The index is computed daily, and its values range total evapotranspiration during the crop season and mar-
from 0 to 1, with 0 (when transpiration occurs at potential ketable yield for various crops:
rate) indicating no water stress and 1 (when no transpira-
tion occurs) indicating a full water stress. Between these Y aET 4
two values, ARID decreases linearly with an increase in T.
While T is calculated using a simple root water uptake where Y is the marketable yield or simply yield (kg), and
model (Meinke et al. 1993, Dardanelli et al. 2004), ETo is ET is evapotranspiration (mm). Replacing Y by P (the
estimated using the FAO-56 PenmanMonteith method potential yield, which occurs when soil moisture does not
(Allen et al. 1998). limit growth) and ET by ETp (potential evapotranspiration,
Key features of ARID are simplicity, generality, daily res- which occurs when soil water is not limiting), P can be esti-
olution and soilplantatmosphere basis. The index is mated from ETP as:
physically and physiologically sound and computationally P aET p 5
simple and also can characterize an agricultural drought
better than many other drought indices that are applied to Dividing Eqn 4 by Eqn 5 produces the relative yield rela-
agricultural systems (Woli et al. 2012a). Although ARID, as tionship (Hanks 1974), thus eliminating crop- and loca-
a generic index, uses a fixed set of parameter values, it is tion-specific differences in the actual production of dry
applicable to a wide range of crops, soils, topographies and matter (Doorenbos and Kassam 1979, Tsakiris 1982):
management and has fairly small uncertainties (Woli et al.
ET
2012b). As a measure to quantify and characterize drought R 6
ETp
conditions, ARID can also be used to forecast droughts
(Woli et al. 2013). where R is the relative yield of a crop (the ratio of Y to P),
This study developed ARID-based drought-induced and ET/ETP is the relative evapotranspiration occurred
yield loss prediction models for several field crops during the crop season. Relative yield is the yield that is
grown in the south-eastern United States and evaluated produced under water-stress condition relative to the one
the performance of these models to estimate yield loss that could be produced under non-water-stress condition,
from drought. the other inputs remaining the same. For instance, R = 0.7
means if a crop yields 1 Mg ha1 under irrigated condi-
Materials and Methods tions, it would yield 0.7 Mg ha1 under rainfed conditions.
The fraction of yield loss from drought during the crop sea-
ARID-based yield model son is then computed as (1R). Equation 6 is valid only if
the quantities on both sides of the equality sign belong to
Briggs and Shantz (1913) found a close relationship between the same set of conditions such as of weather, soil, geno-
crop transpiration and crop dry matter production. They type and management. For instance, the relative yield of a
quantified this relationship as the transpiration ratio: crop growing in a nutrient poor soil is directly propor-
tional to the relative evapotranspiration of the crop grow-
Tr T=Ydm 2
ing in the same soil, other variables other than water
1
where Tr is the transpiration ratio (mm kg ), T is transpi- remaining constant.
ration (mm), and Ydm is crop dry matter yield (kg). Because plant physiological processes are different dur-
Crop dry matter yield may be estimated from transpira- ing different crop phenological stages, a differential yield
tion by replacing Tr1 in Eqn 2 by a, a proportionality con- response to water deficiency occurs during each stage of
stant (kg mm1) that depends on crop species and the growing season (Jensen 1968, Hiler and Clark 1971,
growing conditions, as: Hiler et al. 1974, Doorenbos and Kassam 1979). A yield
model that takes into account a series of stages within the
Ydm aT 3
growing season can reflect the effect of water stress on yield
Because a is the reciprocal of transpiration ratio better than the one that considers the whole season as a sin-
(mm kg1), it can also be regarded as water use efficiency gle stage (Jensen 1968, Doorenbos and Kassam 1979).
(Viets 1962), which is the amount of dry matter produced Because crop yield depends on many factors besides
per unit of water transpired. The water use efficiency (a) drought, relative yield may be more reliably estimated than

164 2014 Blackwell Verlag GmbH, 200 (2014) 163171


ARID Predicting Crop Yields

absolute yield based on a series of stage-specific relative cotton, maize and peanut, the SWVT website was the sole
evapotranspiration ratios during the season. Generally, the data source. The maize data were obtained for two loca-
effect of water deficit on crop yield at different growth tions in Georgia, whereas the yields of cotton and peanut
stages can be expressed using either geometric or arithmetic each were obtained from three locations (Table 1). In case
principles (Tsakiris 1982). A geometric method combines of soybean, none of the studies had both irrigated and rain-
effects of deficiencies occurring during different stages in a fed yields for a given year in any location in the region. The
multiplicative form, whereas an arithmetic method com- locations, however, had irrigated yields in other years (that
bines effects in an additive form. To consider that crop did not have rainfed yields). The soybean yield data for
yield is not linearly related to total water use when plants seven locations in Georgia were obtained from SWVT. The
are stressed (Jensen 1968), a geometric method can be used data for 1987 through 1996 were obtained through Mavro-
to reflect the possible drastic reduction in crop yield if a matis et al. (2001), whereas those for 1997 through 2008
severe water stress occurs during a particularly sensitive were obtained from the SWVT website. The soybean data
crop stage. Considering that determinate, flowering crops for one more location Gainesville, Florida were
have several distinct growth stages with differing drought obtained from the DSSAT database (Zur et al. 1982, Wilk-
sensitivities and applying the geometric approach, Eqn 6 erson et al. 1983, Hoogenboom et al. 2004, Kenneth Boote
can be extended to the following equation for estimating personal communication).
relative yield for a determinate crop.
M 
Y 
ET km Study design and computations
R 7
m1
ETP m Although yield data were available, cultivars were not con-
sistent across locations and years. That is, for each crop,
where the symbol indicates a product, m is a there was no common cultivar grown in all locations and
crop stage, M is the number of stages considered in the years, and management was also different across locations
crop growing season, and km is the relative sensitivity of and years. The SWVT website, the principal data source,
the crop to water stress during the m-th stage of growth. consisted of a sheet of data for each crop, location and year,
Because ARID is based on dense and actively growing each containing a range of yield records obtained from sev-
reference grass, it assumes T to be approximately equal to eral trials conducted for a number of cultivars available in a
evapotranspiration (ET). The potential evapotranspiration year (http://www.swvt.uga.edu/2009/cn09/AP101-Coastal.
(ETp) of the reference grass is called the reference grass pdf). To minimize the variability and maximize consistency
evapotranspiration (ETo). Thus, combining Eqn 1 and in the data and to make sure that the data reflect standard
Eqn 7 after replacing ET and ETp in Eqn 7 by T and ETo, management, except water, the varieties that were within
respectively, produces the following model for predicting 95 % of the highest yield each year were picked to produce
relative yields for a determinate crop based on the crop the rainfed yield data used in this study. Like rainfed data,
stage-specific values of ARID. the irrigated crops also had a range of yields for several cul-
Y
M tivars and management. The highest two irrigated yields
R 1  ARIDkmm 8 were averaged to approximate the potential yield for a
m1 given crop, year and location to reduce observation errors.
Along with the yields, the planting and harvesting dates
of each crop in each location and year were obtained, and,
Data
from them, the duration of each crop growing season was
For exploring associations between ARID and crop yields, computed. In general, the growing season of each crop
four major crops grown in the south-eastern USA were across locations and years had the duration of about
studied: maize (Zea mays L.), cotton (Gossypium hirsutum 5 months. To account for the differential responses of
L.), soybean (Glycine max (L.) Merr.) and peanut (Arachis crops to water stress in different periods of time during the
hypogaea L.). Locations in this region were selected based growing seasons, each crop season was divided into five
on the availability of both rainfed and irrigated yield data periods, each with 30 days. The first 30-day period was
from previous studies. The rainfed and irrigated yields were counted from the day of planting. The 5 months were used
needed to compute relative yields. The yield data used in to approximate five growing stages: juvenile, vegetative
this study were obtained from the University of Georgias development, flowering, grain filling and ripening.
Statewide Variety Testing (SWVT) program (www.swvt. To compute ARID values, daily historical weather data
uga.edu/index.html) and from the database of the Decision daily values of ambient minimum, maximum, and dew-
Support System for Agrotechnology Transfer (DSSAT) point temperatures, precipitation, solar radiation and wind
model (Jones et al. 2003, Hoogenboom et al. 2004). For speed for the corresponding locations and years were col-

2014 Blackwell Verlag GmbH, 200 (2014) 163171 165


Woli et al.

Table 1 Various seasons of rainfed cotton, maize, peanut and soybean crops grown in several locations in the south-eastern USA

Crop (i)

Cotton Maize Peanut Soybean

Location (j) Seasons DOP1 Seasons DOP Seasons DOP Seasons DOP

Blairsville, GA 15 (8795) May 26 [3]


Calhoun, GA 17 (9808) Apr 16 [4] 24 (9000) May 20 [3]
Gainesville, FL 4 (7885) Jun 28 [14]
Griffin, GA 28 (8700) May 15 [4]
Midville, GA 202 (9908)3 May 11 [12]4 13 (9708) May 17 [10] 24 (8700) May 22 [8]
Plains, GA 17 (9908) May 02 [2] 27 (9708) May 07 [8] 10 (9498) May 14 [3]
Quincy, FL 5 (9800) Jun 14 [15]
Tifton, GA 25 (9908) Apr 27 [3] 47 (9708) Mar 24 [14] 15 (9708) Apr 22 [15] 20 (8792) May 23 [5]
Total 62 64 55 134
1
Date of planting, mean values.
2
The total number of observation in various seasons.
3
Figures in parentheses are years during which the yield data were available. For instance (99-08), means from 1999 through 2008.
4
Figures in square brackets are standard deviation of planting dates.

lected from various sources. For locations in Florida, the observed yields were estimated for each crop. Finally, to
weather data were obtained from the Florida Automated evaluate the performance of the yield models, estimated rel-
Weather Network (FAWN) and the Florida State University ative yields (Ri,j,k) were compared with computed relative
Center for Ocean-Atmospheric Prediction Studies (FSU- yields (Yi,j,k/Pi,j,k) using the mean absolute error (MAE), the
COAPS). For Georgia, the data were obtained from the modelling efficiency (ME: Nash and Sutcliffe 1970), the
Georgia Automated Environmental Monitoring Network. RMSE and the Willmott Index (d-index: Willmott 1981) as
Using daily historical weather data, the value of ARID the measures of goodness-of-fit. The steps involved in
for a location j, year k and day t (ARIDj,k,t) was computed designing and making computations are also summarized
and later converted to the (1-ARIDj,k,t) value. For each in Table 2. The same steps and activities were applied for
location and year, then, the data set of the (1-ARIDj,k,t) val- each of the four crops cotton, maize, peanut and soybean.
ues during the crop growing season was split into five suc-
cessive 30-day (monthly) subsets and averaged by subset to
Results
create five successive monthly mean values, one for each
designated crop stage m (1-ARIDj,k,m).
Sensitivity coefficients
Because basic plant physiological processes are expected
to be independent of year and location, crop stage-specific The crop stage-specific water-stress sensitivity coefficients
sensitivity to water stress was assumed to be the same (ki,m) estimated for the four crops are presented in Table 3.
across years and locations. The sensitivity was assumed to According to the ki,m values, peanut was most sensitive to
depend only on crop and phenological stage. The sensitiv- water stress during the second month (3160 days after
ity coefficient for crop i at stage m, ki,m, was estimated planting; DAP). In the case of maize, the third month (61
regressing Eqn 9, the linearized form of Eqn 8. The relative 90 DAP) was the most sensitive. The stage-specific sensitiv-
yield for crop i in location j and year k, Ri,j,k, was calculated ity of maize to water deficit occurred in the following
from the observed values of rainfed yield (Yi,j,k) and irri- (descending) order: the third month (flowering and polli-
gated yield (Pi,j,k) as Yi,j,k/Pi,j,k, considering the irrigated nation stage), the fourth month (grain filling stage) and the
yield as the potential yield. second month (vegetative stage). Both cotton and soybean
  X 5    were shown to be most sensitive to water stress during the
ln Ri;j;k ki;m ln 1  ARIDj;k;m 9 fourth month (91120 DAP). In the case of cotton, the ki,m
m1
value was negative for the second month.
Once the values for the parameters of Eqn 8 (km, where
m = 1, 2, . . ., 5) were estimated for an ith crop, an ARID-
Model performance
based empirical yield model was developed for that crop.
Then, using stage-specific monthly values of ARIDi,j,m in Values of the various measures of goodness-of-fit that were
the empirical yield models so developed, relative crop yields used to evaluate the performance of the ARID-based rela-
(Ri,j,k) for the same locations and years belonging to the tive yield models are presented in Table 4. Values of the

166 2014 Blackwell Verlag GmbH, 200 (2014) 163171


ARID Predicting Crop Yields

RMSE for the four crops ranged from 0.09 to 0.14 (kg ing season (peak flowering and pod setting) as shown by
ha1 Y per kg ha1 P). The difference between RMSE and the sensitivity coefficients of the yield models developed in
MAE was the smallest for peanut model, whereas the larg- this study were in good agreement with the findings of sev-
est for soybean model. The per cent error value of the eral researchers. Salter and Goode (1967), Grimes et al.
model, computed as the ratio of RMSE to mean relative (1970), Hiler and Clark (1971), Hiler et al. (1974) and
yield, was <23 % for all crops. The Willmott Index values Doorenbos and Kassam (1979) found that peak flowering
ranged from 0.70 (cotton) to 0.92 (maize). Similarly, the and boll development are the most susceptible stages to
ME value was largest for the maize model (0.73), whereas water stress for cotton, and the stress occurring during this
smallest for the cotton model (0.33). The ARID-based yield period causes severe yield loss in this crop. Peak flowering
models predicted relative yields of the crops with mean and boll development stages fall in the fourth month of the
absolute errors of 0.07 to 0.13 (kg ha1 Y per kg ha1 P). cotton-growing season (Grimes and El-Zik 1990). For
maize, Salter and Goode (1967) and Doorenbos and
Kassam (1979) observed that flowering and early grain
Discussion
formation are the most critical stages of water requirement,
followed by the yield formation stage. Hiler and Clark
Sensitivity coefficients
(1971), Hiler et al. (1974), and Kostandi and Soliman
The occurrences of the most drought-sensitive stage of cot- (1998) also found the reproductive stage silking and tas-
ton in the fourth month (peak flowering and boll develop- selling to soft dough to be the most sensitive of all stages,
ment), of maize in the third month (flowering and followed by the maturity stage after soft dough and ear
pollination), of peanut in the second month (flowering and development. Because water deficit at silking, tasselling and
pegging) and of soybean in the fourth month of the grow- pollination is most detrimental to yield, the sensitivity of

Table 2 Steps involved in designing and making computations in this study

Step Activity

1 For crop i, location j, and year k, the highest yield (Yh) was obtained from the non-irrigated variety performance trial record. This was done
to reflect standard management; that is, to eliminate the effect of management variation across the locations and years
2 For the same crop, location, and year, two largest yields were picked from the irrigated trial record. For soybean, however, these yields were
the two largest yields from among the irrigated years which had no rainfed yields as this crop did not have both irrigated and rainfed yields
in a given year in any location
3 The two yields picked in step 2 were averaged and assumed to be the potential yield (P). This was done to reduce observation errors
4 The 95th percentile value of Yh obtained in step 1 was computed (=0.95Yh). Then, all yield values from the non-irrigated trial record that were
>0.95Yh were picked and considered as rainfed yields (Y). This was done to increase the number of observation
5 The planting and harvesting dates of the crops from the performance trial record were obtained, and from them were computed the length
of crop growing season (L, days)
6 The duration of L was split into five successive 30-day periods, one for each crop stage m: m1 = 130, m2 = 3160, m3 = 6190,
m4 = 91120, and m5 = 121150 days after planting. The five periods were assumed to approximate five growing stages: juvenile,
vegetative development, flowering, grain-filling, and ripening
7 The daily values of ARID for the crop season L were computed and converted into (1-ARID) values. Then, the values for the corresponding
five periods computed in step 6 were split and averaged by stage m
8 All the five stage-specific average (1-ARID) values were assigned to each yield value selected in step 4 as all the yield values had the same
length of growing season
9 The values computed in step 8 were converted into logarithmic values
10 The logarithmic value of relative yield (R) was computed using the P value computed in step 3 and Y values selected in step 4 for each value
of Y as: Ln(R) = Ln(Y)  Ln(P)
11 A 6-column table was prepared with the variables computed above: Ln(R), Ln(1-ARID)m1, Ln(1-ARID)m2, . . ., and Ln(1-ARID)m5
12 Steps 1 through 11 were repeated for each year for which the data were available. Then all yearly tables were combined to produce the
dataset for location j
13 Steps 1 through 12 were repeated for each location in the region for which the data were available. Then all location datasets were
combined to produce the dataset for crop i (i = cotton, maize, peanut, or soybean)
14 To estimate stage-specific sensitivity coefficients (km), Eqn 9 was regressed using the dataset produced in step 13, with Ln(R) as the
dependent and Ln(1-ARID)m1, . . ., Ln(1-ARID)m5 as the independent variables
15 A yield model was developed for crop i by plugging in the values of km estimated in step 14 in Eqn 8
16 Using Eqn 8, relative yields (Rs) were estimated for the same locations and years as the observed yields
17 To evaluate the performance of the yield model for crop i, the estimated Rs (step 16) were compared with the observed Rs computed
as Y (step 4) divided by P (step 3) by using various goodness-of-fit measures

2014 Blackwell Verlag GmbH, 200 (2014) 163171 167


Woli et al.

Table 3 Values of water-stress sensitivity coefficients (ki,m) for various 120 DAP (Reicosky and Heatherly 1990), is the most critical
crops and their phonological stages stage in terms of obtaining maximum yield.
Crop (i) Some researchers have also observed that a mild water
stress stimulates flower production in cotton (Cothren
Stage/month (m) Cotton Maize Peanut Soybean 1999). Water stress during the pre-flowering period of cot-
First (m = 1) 0.12 0.05 0.04 0.16 ton, which falls in the second month of the season (Grimes
Second (m = 2) 0.10 0.09 0.27 0.15 and El-Zik 1990), increases the number of flowers and bolls
Third (m = 3) 0.17 0.36 0.13 0.21 per plant and boll size in some cultivars causing increased
Fourth (m = 4) 0.34 0.11 0.08 0.30 yield (de Bruyn 1964, Singh 1975). With the negative value
Fifth (m = 5) 0.04 0.03 0.07 0.01 of the sensitivity coefficient for the second month for cot-
ton (0.10), the cotton yield model was also able to reflect
the phenomenon of mild water stress increasing the yields
Table 4 Values of the various goodness-of-fit measures used to evalu-
ate the performance of ARID-based relative yield models for various of this crop.
crops Because the information given by the sensitivity coeffi-
cients estimated in this study, which were assumed to be
Crop (i)
regional, was in line with the findings of several other
Measure Cotton Maize Peanut Soybean researchers, the study results implied that all locations
within a region have the same stage-specific sensitivities of
Mean Y/P1 ratio () 0.68 0.71 0.78 0.62
a crop to water stress. The reasonable values of ki,m for each
MAE (kg ha1 Y per kg ha1 P) 0.13 0.07 0.08 0.12
RMSE (kg ha1 Y per kg ha1 P) 0.144 0.087 0.089 0.142
crop studied led to the conclusion that Eqn 9 expressed the
RMSE minus MAE 0.015 0.016 0.014 0.022 relationship between ARID and relative crop yields accu-
Modeling efficiency () 0.33 0.73 0.60 0.49 rately. The results further indicated that ARID can be used
Willmott d-index () 0.70 0.92 0.86 0.79 as a useful tool to detect the stage-specific sensitivity of
1
crops to water stress as well as the yield loss from drought.
Y, rainfed yield; P, potential (irrigated) yield; MAE, mean absolute
error; and RMSE, root mean square error.
Model performance
maize crop growth stages to water deficit occurs in the fol-
lowing order: flowering and pollination, grain filling, and Values of the various measures of goodness-of-fit showed
vegetative development (Musick and Dusek 1980). The that each of the four crop yield models performed reason-
vegetative stage of maize falls in the second month; the sil- ably well at predicting relative yields (Ri,j,k) and thus the
king, tasselling and pollination stages in the third month; yield losses from drought. The mean absolute errors and
and the grain filling stage in the fourth month after planting per cent errors of the ARID-based yield models were low
(Rhoads and Bennett 1990, Nielsen et al. 2002). In the case for all crops (713 % of potential yield). The difference
of peanut, Salter and Goode (1967), Hiler et al. (1974), between RMSE and MAE revealed that peanut had the
Doorenbos and Kassam (1979), and Ramamoorthy and smallest variance in individual errors, whereas soybean had
Basu (1996) found the period between peak flowering and the largest. Values of the Willmott Index, a measure of the
early pegging, which fall during 3557 DAP, to be the most degree to which the observed values are approached by the
susceptible period to drought. The most critical stage of model-estimated values, indicated that the yield model for
peanut crop to water stress is flowering, which starts at maize agreed most closely with the observed data, whereas
about 35 DAP and peaks around 63 DAP, and water deficit the yield model for cotton agreed the least. Values of ME, a
at flowering and pegging stages causes greater yield loss than measure to assess the predictive power of models, also indi-
that at other growth stages (Boote and Ketring 1990). The cated that the agreement between the model-estimated and
greatest effect of water deficit on economic yields has been observed values was best for maize model and relatively
reported to have been during flowering or anthesis because poor for cotton model. Although modelling efficiency
of reduction in grain numbers (Howell 1990). For soybean, varied across yield models, all yield models performed
Salter and Goode (1967), Hiler and Clark (1971), Hiler reasonably well. The positive values of ME indicated that
et al. (1974), Doorenbos and Kassam (1979), Kpoghomou model predictions were more accurate than the means of
et al. (1990), and de Bruyn et al. (1995) observed that the the observed data.
most sensitive stages of this crop to water deficit are peak The predicted relative yield of cotton ranged from 0.46
flowering and pod set. Brown et al. (1985) found this crop to 0.90, whereas the observed relative yield ranged from
to be most susceptible to drought around 100 DAP, with 0.31 to 0.99. For maize, the ranges of predicted and
both weight and number of seed reduced. Doss et al. (1974) observed relative yields were (0.33, 0.97) and (0.22, 1.00),
concluded that the pod-filling stage, which falls around 90 respectively. The predicted relative yield range for peanut

168 2014 Blackwell Verlag GmbH, 200 (2014) 163171


ARID Predicting Crop Yields

was between 0.51 and 0.98, whereas the observed yield Basal, H., N. Dagdelen, A. Unay, and E. Yilmaz, 2009:
range was between 0.40 and 0.99. For soybean, the pre- Effects of deficit drip irrigation ratios on cotton (Gossypi-
dicted and observed relative yield ranges were (0.23, 0.85) um hirsutum L.) yield and fibre quality. J. Agron. Crop
and (0.10, 1.00), respectively. In other words, the width of Sci. 195, 1929.
the range of predicted yields relative to that of the observed Boote, K. J., and D. L. Ketring, 1990: Peanut. In: B. A. Stewart,
yields for cotton, maize, peanut and soybean was 0.64, and D. R. Nielsen, eds. Irrigation of Agricultural Crops, pp.
0.83, 0.80 and 0.68, respectively. Based on these figures 675717. ASA-CSSA-SSSA, Madison, WI, USA.
also, the four crop yield models were ranked in terms of Briggs, L. J., and H. L. Shantz, 1913: The water Requirement of
Plants I: Investigations in the Great Plains in 1910 and 1911
descending performance as: maize, peanut, soybean and
(Bureau of Plant Industry Bulletin No. 284). USDA, Washing-
cotton. The different performances across crops were possi-
ton, DC, USA.
bly due to variations in drought tolerance among the four
Brown, E. A., C. E. Caviness, and D. A. Brown, 1985: Response
crops. Of the four yield models, the performance of maize
of selected soybean cultivars to soil moisture deficit. Agron. J.
model was best because maize is more sensitive to water
77, 274278.
deficit than are the other crops compared (Sangoi and Sal- de Bruyn, L. P., 1964: The Influence of Soil Moisture Deficits on
vador 1998, Heiniger 2001, PIDS 2010). The performance the Growth and Production of Cotton (Technical Communi-
of the cotton model was poorest among the crops, likely cation Vol. 24). South Africa Department of Agricultural
because cotton is the most drought-tolerant crop among Technical Services, Pretoria, South Africa
the four crops compared (Rosenow et al. 1983, Edmisten de Bruyn, L. P., J. P. Pretorius, and J. J. Human, 1995: Water
et al. 1994, Malik et al. 2006, Lv et al. 2007, Basal et al. sensitive periods during the reproductive growth phase of Gly-
2009). The performances of peanut and soybean models cine max L. II Establishing water stress sensitivity. J. Agron.
were intermediate as these crops are less tolerant to Crop Sci. 174, 197203.
drought than is cotton and also less susceptible to drought Cole, J. S., and O. R. Mathews, 1923: Use of Water by Spring
than is maize. These results indicated that ARID-based Wheat on the Great Plains (Bureau of Plant Industry Bulletin
yield models perform better for the crops that are more No. 1004). USDA, Washington, DC, USA.
sensitive to water deficit. Cothren, J. T., 1999: Physiology of the cotton plant. In: W. C.
Despite large variations in observed yield data due to Smith, and J. T. Cothren, eds. Cotton: Origin, History, Tech-
location differences in weather, soil, management, culti- nology, and Production, pp. 207268. John Wiley and Sons
vars and dates of planting and harvesting, the ARID- Inc., New York, NY, USA.
based relative yield models estimated the overall effects Dardanelli, J. L., J. T. Ritchie, M. Calmon, J. M. Andriani, and
of drought on yields of cotton, maize, peanut and soy- D. J. Collino, 2004: An empirical model for root water uptake.
bean reasonably well. The results indicated that ARID Field Crops Res. 87, 5971.
Doorenbos, J., and A. H. Kassam, 1979: Yield Response to Water
has potential to predict the yield loss from drought for
(Irrigation and Drainage Paper No. 33). FAO, Rome, Italy.
several field crops, especially the ones that are more sen-
Doss, B. D., R. W. Pearson, and H. T. Rogers, 1974: Effect of soil
sitive to water stress.
water stress at various growth stages on soybean yield. Agron.
J. 66, 297299.
Acknowledgement Dracup, J. A., K. S. Lee, and E. D. Paulson Jr, 1980: On the defi-
nition of droughts. Water Resour. Res. 16, 297302.
The authors thank the USDA National Institute for Food Edmisten, K., J. Crawford, and M. Bader, 1994: Drought Man-
and Agriculture and the NOAA Regional Integrated Sci- agement for Cotton Production. The University of Georgia,
ences and Assessments program for supporting this work Tifton, GA, USA. http://infohouse.p2ric.org/ref/18/17813.htm
through grants. [last accessed 7 July 2013].
Grimes, D. W., and K. M. El-Zik, 1990: Cotton. In: B. A. Stew-
art, and D. R. Nielsen, eds. Irrigation of Agricultural Crops,
References
pp. 741773. ASA-CSSA-SSSA, Madison, WI, USA.
Allen, R. G., L. S. Pereira, D. Raes, and M. Smith, 1998: Crop Grimes, D. W., R. J. Miller, and L. Dickens, 1970: Water stress
Evapotranspiration: Guidelines for Computing Crop Water during flowering of cotton. Calif. Agric. 24, 46.
Requirements (Irrigation and Drainage Paper no. 56). FAO, Hanks, R. J., 1974: Model for predicting plant yield as influenced
Rome, Italy. by water use. Agron. J. 66, 660665.
Allen, R. G., I. A. Walter, R. Elliott, and T. Howell, 2005: The Hayes, M. J., 2013: Comparison of Major Drought Indices:
ASCE Standardized Reference Evapotranspiration Equation. Introduction. National Drought Mitigation Center, University
ASCE, Reston, VA, USA. of Nebraska, Lincoln, NE, USA. http://drought.unl.edu/
Arkley, R. J., 1963: Relationships between plant growth and planning/monitoring/comparisonofindicesintro.aspx [last
transpiration. Hilgardia 34, 559584. accessed 7 July 2013].

2014 Blackwell Verlag GmbH, 200 (2014) 163171 169


Woli et al.

Heiniger, R. W., 2001: The Impact of Early Drought on Corn Meinke, H., G. L. Hammer, and P. Want, 1993: Potential soil
Yield. North Carolina State University Cooperative Extension, water extraction by sunflower on a range of soils. Field Crops
Raleigh, NC, USA. http://www.ces.ncsu.edu/plymouth/crop- Res. 32, 5981.
sci/docs/early_drought_impact_on_corn.html [last accessed 7 Musick, J. T., and D. A. Dusek, 1980: Irrigated corn yield
July 2013]. response to water. Trans. ASAE 23, 9298.
Hiler, E. A., and R. N. Clark, 1971: Stress day index to character- Narasimhan, B., and R. Srinivasan, 2005: Development and eval-
ize effects of water stress on crop yields. Trans. ASAE 14, 757 uation of soil moisture deficit index and evapotranspiration
761. deficit index for agricultural drought monitoring. Agric. For.
Hiler, E. A., T. A. Howell, R. B. Lewis, and R. P. Boos, 1974: Irri- Meteorol. 133, 6988.
gation timing by the stress day index method. Trans. ASAE Nash, J. E., and J. V. Sutcliffe, 1970: River flow forecasting
65, 393398. through conceptual models part I: a discussion of principles.
Hoogenboom, G., J. W. Jones, P. W. Wilkens, C. H. Porter, W. D. J. Hydrol. 10, 282290.
Batchelor, L. A. Hunt, K. J. Boote, U. Singh, O. Uryasev, W. T. Nielsen, R. L., P. R. Thomison, G. A. Brown, A. L. Halter, J.
Bowen, A. J. Gijsman, A. S. Du Toit, J. W. White, and G. Y. Tsu- Wells, and K. L. Wuethrich, 2002: Corn: delayed planting
ji, 2004: Decision Support System for Agrotechnology Transfer, effects on flowering and grain maturation of dent corn.
Version 4.0. University of Hawaii, Honolulu, HI, USA. Agron. J. 94, 549558.
Howell, T. A., 1990: Relationships between crop production PIDS, 2010: Impact of Drought on Corn Physiology and Yield.
and transpiration, evapotranspiration, and irrigation. In: B. Philippine Institute for Development Studies (PIDS), Makati,
A. Stewart, and D. R. Nielsen, eds. Irrigation of Agricul- Philippines.
tural Crops, pp. 391434. ASA-CSSA-SSSA, Madison, WI, Quiring, S. M., and T. N. Papakryiakou, 2003: An evaluation of
USA. agricultural drought indices for the Canadian prairies. Agric.
Jensen, M. E., 1968: Water consumption by agricultural plants. For. Meteorol. 118, 4962.
In: T. T. Kozlowski, ed. Water Deficits and Plant Growth, Vol. Ramamoorthy, K., and R. N. Basu, 1996: Studies on the effect of
II, pp. 122. Academic Press, New York, NY, USA. moisture stress at different growth phases on seed vigour, via-
Jones, J. W., G. Hoogenboom, C. H. Porter, K. J. Boote, W. D. bility and storability in peanut (Arachis hypogaea L.). J. Agron.
Batchelor, L. A. Hunt, P. W. Wilkens, U. Singh, A. J. Gijsman, Crop Sci. 177, 3337.
and J. T. Ritchie, 2003: The DSSAT cropping system model. Reicosky, D. C., and L. G. Heatherly, 1990: Soybean. In: B.
Eur. J. Agron. 18, 235265. A. Stewart, and D. R. Nielsen, eds. Irrigation of Agricul-
Kostandi, S. F., and M. F. Soliman, 1998: Water stress and N fer- tural Crops, pp. 639674. ASA-CSSA-SSSA, Madison, WI,
tilization effects on yield and smut disease incidence in corn USA.
(Zea mays L.). J. Agron. Crop Sci. 180, 151157. Rhoads, F. M., and J. M. Bennett, 1990: Corn. In: B. A. Stewart,
Kpoghomou, B. K., V. T. Sapra, and C. A. Beyl, 1990: Sensitivity and D. R. Nielsen, eds. Irrigation of Agricultural Crops, pp.
to drought stress of three soybean cultivars during different 569596. ASA-CSSA-SSSA, Madison, WI, USA.
growth stages. J. Agron. Crop Sci. 164, 104109. Rosenow, D. T., J. E. Quisenberry, C. W. Wendt, and L. E. Clark,
Kumar, V., and U. Panu, 1997: Predictive assessment of severity 1983: Drought tolerant sorghum and cotton germplasm.
of agricultural droughts based on agro-climatic factors. J. Am. Agric. Water Manag. 7, 207222.
Water Resour. Assoc. 33, 12551264. Sakamoto, C. M., 1978: The Z-index as a variable for crop yield
Lv, S., A. Yang, K. Zhang, L. Wang, and J. Zhang, 2007: Increase estimation. Agric. Meteorol. 19, 305313.
of glycinebetaine synthesis improves drought tolerance in cot- Salter, P. J., and J. E. Goode, 1967: Crop Responses to Water at
ton. Mol. Breeding 20, 233248. Different Stages of Growth. Commonwealth Agricultural
Malik, T. A., S. Ullah, and S. Malik, 2006: Genetic linkage stud- Bureau, Farnham Royal, Buckinghamshire.
ies of drought tolerant and agronomic traits in cotton. Pak. J. Sangoi, L., and R. J. Salvador, 1998: Maize susceptibility to
Bot. 38, 16131619. drought at flowering: a new approach to overcome the prob-
Maracchi, G., 2000: Agricultural drought a practical approach lem. Ci^encia Rural 28, 699706.
to definition, assessment, and mitigation strategies. In: J. V. Singh, S. P., 1975: Studies on the effects of soil moisture
Vogt, and F. Somma, eds. Drought and Drought Mitigation stress on the yield of cotton. Indian J. Plant Physiol. 18,
in Europe, pp. 6475. Kluwer Academic Publishers, Dordr- 4955.
echt, The Netherlands. Tsakiris, G. P., 1982: A method for applying crop sensitivity
Mathews, O. R., and L. A. Brown, 1938: Winter Wheat and Sor- factors in irrigation scheduling. Agric. Water Manag. 5,
ghum in the Southern Great Plains under Limited Rainfall 335343.
(Circular No. 477). USDA, Washington, DC, USA. Viets, F. G., 1962: Fertilizers and the efficient use of water. Adv.
Mavromatis, M., K. J. Boote, J. W. Jones, A. Irmak, D. Shinde, Agron. 14, 223264.
and G. Hoogenboom, 2001: Developing genetic coefficients Wilhite, D. A., and M. H. Glantz, 1985: Understanding the
for crop simulation models with data from crop performance drought phenomenon: the role of definitions. Water Int. 10,
trials. Crop Sci. 41, 4051. 111120.

170 2014 Blackwell Verlag GmbH, 200 (2014) 163171


ARID Predicting Crop Yields

Wilkerson, G. G., J. W. Jones, K. J. Boote, K. T. Ingram, and J. Woli, P., J. W. Jones, and K. T. Ingram, 2012b: Assessing
W. Mishoe, 1983: Modeling soybean growth for crop manage- the Agricultural Reference Index for Drought (ARID) using
ment. Trans. ASAE 26, 6373. uncertainty and sensitivity analyses. Agron. J. 105, 150
Willmott, C. J., 1981: On the validation of models. Phys. Geogr. 160.
2, 184194. Woli, P., J. W. Jones, K. T. Ingram, and J. O. Paz, 2013: Fore-
WMO, 1975: Drought and Agriculture (Technical Note 138). casting drought using the Agricultural Reference Index for
World Meteorological Organization (WMO), Geneva, Swit- Drought (ARID): a case study. Weather Forecasting 28, 427
zerland. 443.
Woli, P., J. W. Jones, K. T. Ingram, and C. W. Fraisse, 2012a: Zur, B., J. W. Jones, K. J. Boote, and L. C. Hammond, 1982:
Agricultural Reference Index for Drought (ARID). Agron. J. Total resistance to water flow in field soybeans: II. Limiting
104, 287300. soil moisture. Agron. J. 74, 99105.

2014 Blackwell Verlag GmbH, 200 (2014) 163171 171

Das könnte Ihnen auch gefallen