Wheat Yield Prediction Using Remotely Sensed Agromet Trend-Based Models for Hoshiarpur District of Punjab, India
Estimation of crop production in advance of the harvest
has been an intensively researched field in agriculture. The aim of present
study was to predict wheat yield using different agrometeorological indices,
spectral index (NDVI, Normalized Difference Vegetation Index) and Trend
Estimated Yield (TEY) in Hoshiarpur district of Punjab for the years 2001-02
and 2002-03. On the basis of examination of Correlation Coefficients (R),
Standard Error of Estimate (SEOE) and Relative Deviation (RD) values resulted
from different agromet models, the best agromet subset were selected as
Minimum Temperature (Tmin), Maximum Temperature (Tmax)
and Accumulated Heliothermal Units (HTU) for Hoshiarpur district. In order
to improve model accuracy the above mentioned agrometeorological indices
together with NDVI and TEY were used as independent variables for yield
prediction at reproductive stage (2nd week of March) of wheat. It was
found that Agromet-Spectral-Trend-Yield model could explain 96% (SEOE
= 87 kg ha-1) of wheat yield variations for Hoshiarpur district.
Various government agencies and private institutions have provided a
great deal of fundamental information relating spectral reflectance and
thermal omittance properties of soils and crops to their agronomic and
biophysical characteristics. This knowledge has facilitated the development
and use of various remote sensing methods for non-destructive monitoring
of plant growth and development. Estimation of crop production in advance
of the harvest is of great utility in farming. Temporal spectral pattern
of any crop summarizes the phenology as well as growth features of the
crop. The integrated models incorporating satellite based vegetation indices,
agro-meteorological indices, biophysical indices and time trend predicted
well the crop (Pinter et al., 2003; Deosthali and Akmanchi, 2006).
Since crop yield is the culmination of many temporal plant processes and
is affected by various external factors related to soil, weather and technology,
parameterization of these factors and investigation of their relationship
with yield are essential for crop yield modeling. Remotely sensed data
have emerged as a promising tool for yield modeling because of their unique
advantages such as, possibility of obtaining crop specific objective information
with adequate spatial and temporal coverage. The integration of agrometeorological
and spectral indices derived from remotely sensed data may provide more
reliable pre-harvest crop yield estimates (Dadhwal and Ray, 1998). The
wheat yield prediction improved where Trend Estimated Yield (TEY) parameter
(showing the combined effect of technological development on yield) was
incorporated in agromet-spectral-yield relations in Punjab (Kalubarme
et al., 1995; Medhavy et al., 1995) and Haryana (Verma et
al., 2003) states. The present study is an attempt to relate spectral,
agrometeorological indices and TEY for wheat yield prediction in Hoshiarpur
district of Punjab.
MATERIALS AND METHODS
Seventeen years (from 1984-1985 to 2000-2001) historical yield data published
by Bureau of Economics and Statistics (BES) for Hoshiarpur district were
used to develop agromet-spectral-trend-yield models. To integrate various
agromet and spectral indices over different growth phases, wheat growing
season was divided into four phenological stages, starting from the sowing
of crop on 12th of November up to harvesting on 15th of April including:
i) Early Seedling stage (Meteorological Standard Week Numbers (MSWN) 46-50),
ii) Active Vegetative (AV) stage (MSWN 51-7), iii) Reproductive (RP) stage
(MSWN 8-11), iv) Maturity stage (MSWN 12-15). Besides, AV + RP and overall
crop growth were used.
Archived spectral index of Normalized Difference Vegetation Index
NDVI = (NIR-R)/(NIR+R)
||Near infrared reflectance band
||Red reflectance band
corresponding to the maximum vegetative growth stage (maximum leaf area
index, LAI) of wheat were collected from Space Applications Centre (SAC),
Ahmedabad from 1988-89 to 2002-2003 for Hoshiarpur district.
Meteorological parameters: Weekly meteorological data of Ballowal
Saunkhri meteorological observatory located in the study area were used
from 1984-1985 to 2002-2003. Weekly maximum (Tmax) and minimum
(Tmin) temperatures, accumulated rainfall (ARF) and evaporation
data of US Weather Bureau Class A Pan Evaporimeter were used as meteorological
Agrometeorological indices: Growing Degree Days (GDD): The heat
unit or growing degree-days concept was proposed to explain the relationship
between growth duration and temperature. This concept assumes a direct
and linear relationship between growth and temperature (Nuttonson, 1955).
It has been reported that accumulated GDD is the best index to predict
various phenophases in wheat crop under Punjab conditions (Hundal et
A degree-day or a heat unit is the mean temperature (Tmean)
above base temperature. Mathematically, it can be expressed as:
||Growing degree-days (°C day)
||Daily maximum/minimum temperature (°C)
||Base temperature i.e., the lowest temperature below which it is
assumed that there is no growth. A base temperature of 5°C was
selected to determine GDD for different growth stages of wheat (Sharma
et al., 2004; Dubey et al., 1987). If Tmean<
Tb, GDD = 0
||Starting date of phenophase
||Ending date of phenophase
Temperature Difference (TD): Temperature difference was computed
using following expression:
Photothermal Units (PTU) and Heliothermal Units (HTU): Because
of the phasic changes taking place due to the influence of both temperature
and photoperiod, it is better to calculate Photothermal Units (PTU) instead
of heat units for accurate prediction of flowering and maturity. Therefore,
photothermal units are proposed, where in, the degree days are multiplied
by length of the night in case of short- day plants and length of the
day for long-day plants (Reddy and Reddi, 2003).
In general, PTU is product of GDD and day length (maximum possible sunshine
hours, N) and HTU is the product of GDD and bright sunshine hours (actual
sunshine hours, n). Therefore, they can be computed using following expressions:
||Photothermal units (°C day hours)
||Heliothermal units (°C day hours)
||Growing degree days (°C day)
||Maximum possible sunshine hours which collected from Doorenbos and
||Actual sunshine hours
Vapour Pressure Deficit (VPD): Vapour pressure deficit plays a
significant role in crop evapotranspiration. At constant temperature,
changes in atmospheric humidity affect transpiration by changing actual
vapour pressure of the air (ea) and modifying the vapour pressure
gradient from leaf to air (Rao, 2003; Kramer, 1997). The difference between
the saturation vapour pressure (es) and its actual water vapour
pressure is termed as vapour pressure deficit and it can be worked out
using following expressions:
ea = (RHmean es)/100
VPD = es - ea
||Actual water vapour pressure (millibar)
||Mean relative humidity (%)
||Saturated water vapour pressure (millibar) as a function of mean
air temperature which collected from Michael (1978)
||Vapour pressure deficit (millibar)
Potential Evapotranspiration (PET): Baier and Robertson (1967)
demonstrated that the yield of a crop was closely related to the physical
environmental parameters, like evapotranspiration (ET) and soil moisture
than to simple meteorological variables, such as rainfall or temperature.
CropWat for windows package version 4.2 developed by Clarke et al.
(1998) was used to compute PET during different phenological stages. CropWat
for windows is a programme that use the modified Penman-Monteith method
for calculating reference crop evapotranspiration. The method supersedes
the FAO Irrigation and Drainage Paper No. 24 (Doorenbos and Pruitt, 1975).
Monthly maximum and minimum temperature (°C), mean relative humidity
(%), sunshine hours and wind speed (m sec-1) at two meter height
above the ground were used to run the model.
Trend Estimated Yield (TEY): The district-wise data on wheat yield
for Hoshiapur reported by Bureau of Economics and Statistics (BES) from
the year 1984-1985 to 2000-2001 was used for the trend analysis. Scatter
plots of year vs. yield was created for evaluating the time trend used
in fitting a regression line. In this trend analysis time was used as
a dummy variable (Draper and Smith, 2003). The trend-predicted yields
were computed using the following equation:
Y = a + bT
||Yield (kg ha-1)
||Time (years, 1985 = 1)
In order to evaluate the performance of different yield models for prediction
of yields, predicted yield for the years 2001-2002 and 2002-2003 were
compared with corresponding BES estimates using Relative Deviation (RD)
as a measure of accuracy of prediction.
In order to have sufficient information about the order of importance
of the independence variables in predicting the dependent variable y,
a forward stepwise regression method was used. The independent variables
X1, X2,….Xp are entered one-by-one
into the equation according to criterion of a minimum F-to-enter value
equal to 3.5. For the variables, a minimum F-to-remove was also set equal
to 3.2. Once a variable was in the equation, it may be swapped with a
variable not in the equation or it may be removed from the equation altogether.
The meteorological variable used to calculate the different indices (GDD,
TD, HTU, PTU, VPD and PET) were checked for collinearity and it was found
from their low correlation values among these variables that there was
no problem of multicollinearity.
RESULTS AND DISCUSSION
Various possible ways using meteorological parameters/ agrometeorological
indices and NDVI for wheat yield modeling have been attempted. The simple,
multiple-linear and stepwise regression analysis have been developed.
Table 1 shows simple linear correlation coefficient
values between wheat grain yield (kg ha-1) and meteorological
parameters/agrometeorological indices for Hoshiarpur district.
The results obtained from simple linear regression analysis will be discussed
under various wheat growth stages:
Early seedling stage: It was observed that among different meteorological
parameters/agrometeorological indices, minimum temperature (Tmin),
accumulated Temperature Difference (TD), accumulated Evaporation (E) and
Potential Evapotranspiration (PET) were significantly correlated with
wheat grain yield. The model with PET yields the highest correlation coefficient
(r) of 0.831. TD showed positive relationship with yield while Tmin,
E and PET had negative relationship with yield. Baier and Robertson (1967)
reported that high minimum temperatures from emergence to heading may
be responsible for reduction in yields.
Active vegetative stage: In Active Vegetative (AV) stage significant
correlations were obtained between wheat grain yield and Tmin,
accumulated Growing Degree-Days (GDD), Photothermal Units (PTU), accumulated
vapour pressure deficit (VPD), E and PET (Table 1).
In addition, a significant r value was found between wheat grain yield
and accumulated Heliothermal Units (HTU) in Hoshiarpur district. The highest
r values were obtained in the model between yield and PET, which explained
74% of yield variation. These results showed the importance of temperature
based indices and day length in vegetative stage which are in conformity
with the results obtained by Angus et al. (1981). They obtained
a fairly straight line, suggesting that a day-degree system could adequately
account for the different rates of development in wheat crop.
||Simple linear correlation coefficient values between
wheat grain yield (kg ha-1) and meteorological parameters/agrometeorological
indices for Hoshiarpur district
|*: α = 0.05, **: α = 0.01, n =17
Reproductive stage: In Reproductive (RP) stage significant correlations
were found between wheat grain yield and Tmin, TD, E and PET
in Hoshiarpur district and among which the highest correlation coefficient
(r = -0.712) was obtained with Tmin. High value of correlation
coefficient between wheat grain yield and minimum temperature (Tmin)
in reproductive phase (from February 19 to March 18) might be due to the
fact that RP coincides with the transitional period in which change in
Tmin from lower to higher values may influence the rate of
respiration process during night. These results indicated the importance
of high night temperature and confirmed the findings of Peters et al.
Maturity stage: Significant correlations were found between wheat grain
yield and Tmin, TD, E and PET during maturity stage. The highest
r value (-0.700) was obtained between yield and PET for Hoshiarpur district.
These results are in confirmation with the earlier work conducted by Bairagi
and Hassan (2002). The higher correlation coefficient values between yield and
PET indicated that temperature and water use play an important role and influence
grain yield of wheat crop.
Active vegetative to reproductive stage: In combined Active Vegetative
(AV) and Reproductive Period (RP), significant correlations were obtained
between wheat grain yield and Tmin, GDD, PTU, E and PET. Besides,
a negative significant correlation was observed between grain yield and
Overall crop growth: For entire wheat growth season, in Hoshiarpur
district, significant correlations were found between grain yield and
Tmin, TD, GDD, PTU, E and PET among which the highest r value
(r = -0.834) was obtained with PET followed by Tmin (r = -0.785).
Jand (1999) also found a negative correlation between wheat grain yield
and GDD and PTU for Ferozepur district of Punjab.
The best agromet subset were selected on the basis of examination of
correlation coefficients (R), Standard Error of Estimate (SEOE) as well
as RD values resulted from different agromet models including simple/multiple
linear and stepwise regression analysis (Data not given due to brevity)
to develop agromet-spectral-trend-yield models. Accordingly, the suitable
time of prediction was found to be at the end of reproductive stage i.e.,
2nd week of March (11th week after sowing). The best agromet subset to
incorporate in agromet-spectral-trend-yield models were selected as Tmin,
Tmax and HTU for Hoshiarpur district. The final regression
equations of different models are given below:
Model (1), Agromet- Yield
||4300.32-228.54 Tmin-16.76 Tmax + 0.47 HTU
||(R = 0.779, R2 = 0.607, R2adj = 0.516,
SEOE = 293.26, F =6.69**, n = 17)
Model (2), Agromet-Spectral-Yield
||1081.08 -134.68 Tmin + 100.16 Tmax + 0.22 HTU
||(R = 0.944, R2 = 0.891, R2adj = 0.804,
SEOE = 126.80, F= 10.23*, n =10)
Model (3), Agromet-Spectral-Trend-Yield
||1277.36-88.89 Tmin + 25.57 Tmax + 0.25 HTU-520.39
NDVI + 0.51 TEY
||(R = 0.979, R2 = 0.959, R2adj = 0.908,
SEOE = 86.92, F=18.75**, n=10)
Variation explained by models ranged from 61-96% and standard error of
estimate ranged from 87-293 kg ha-1. The results revealed that
in case of agromet-yield model, both minimum and maximum temperature showed
negative relationship with grain yield. It might be due to the fact that
higher day time temperature decreases the period available for photosynthetic
activity before grain maturation and hence will affect the yield adversely
(Marcellos and Single, 1972; Asana and Williams, 1965). Moreover, high
night time temperature associated with accelerated respiration, which
decreases translocation of photosynthates from leaf to grain and hence
reduced the yield.
||Performance evaluation of different agromet-spectral-trend-yield
model at reproductive stage of wheat for the years 2001-2002 and 2002-2003
Heliothermal Units (HTU) showed a positive relationship indicating the
flowering response to bright sunshine hours prevailing during the reproductive
period of the crop growth. The agromet-yield model explained 61% of yield
variability while, the coefficient of determination (R2) value
improved from 61-89% when NDVI was included in agromet-yield model. Similar
results have been reported by Dubey et al. (1994). Agromet-spectral-trend-yield
model accounted for 96% of variations in wheat grain yield over reproductive
stage. It might be due to the fact that trend-incorporated relation gives
better correlation with grain yield than agromet/agromet-spectral-yield
models. Since trend predicted yield is a integrated factors of technological
advancement, improvement in fertilizer insecticide /pesticide/weedicide
use and increased use of high yielding varieties (Verma et al.,
2003; Jand, 1999).
Model validation: In order to evaluate model validity, model predicted
yields were compared with corresponding BES estimates using relative deviation
values for the years 2001-2002 and 2002-2003 (Table 2),
for Hoshiarpur district.
The predicted wheat yield obtained from these models ranged from -23.6
to -0.6% deviation from actual yield in different years. The model developed
using agromet parameters underestimated yield 16.5% during 2002-2003.
With the incorporating of spectral parameters into this model improved
its predictability by reducing the relative deviation between model predicted
and actual observed wheat yield to -9.7% with reduced SEOE. Similarly,
incorporating the trend parameters into these models, further improved
the model with just -0.6. RD values and reduced SEOE. The performance
comparison between wheat yields prediction and its corresponding BES estimates
(Table 2) using different models revealed that the predicted
wheat grain yield computed for the year 2002-2003 were always closer to
actual yields than 2001-2002. It might be attributed to the fact that
the model is not likely to give a very realistic estimate in years of
extreme weather conditions as evident from actual yields in Hoshiarpur
district in the year 2001-2002 which were 3616 kg ha-1 as compared
to its difference from normal yields during 1985 to 2003 (2987 kg ha-1)
whereas, in 2002-2003 yield was 3439 kg ha-1, respectively.
The response of crop meteorological conditions is not always the same
during the entire life cycle of the crop and also during different ranges
of the parameters (Mahey, 1999). It can be concluded that wheat yield
prediction was better when Agromet and spectral indices in the models
were used in combination rather than when used individually. However,
Agromet-Spectral-Trend-Yield models at reproductive stage of wheat crop
give the highest R values of 0.98.
1: Angus, J.F., D.H. Mackenzie, R. Morton and C.A. Schafer, 1981. Phasic development in field crops II. Thermal and photoperiodic responses of spring wheat. Field Crops Res., 4: 269-283.
CrossRef | Direct Link |
2: Asana, R.D. and R.F. Williams, 1965. The effect of temperature stress on grain development in wheat. Aust. J. Agric. Res., 16: 1-13.
Direct Link |
3: Baier, W. and G.W. Robertson, 1967. Estimating yield components of wheat from calculated soil moisture. Can. J. Plant. Sci., 47: 617-630.
CrossRef | Direct Link |
4: Bairagi, G.D. and Z.U. Hassan, 2002. Wheat crop production estimation using satellite data. J. Ind. Soc. Remote Sensing, 30: 213-219.
CrossRef | Direct Link |
5: Clarke, D., M. Smith and K. EL-Askari, 1998. CropWat for windows package, version 4.2. Food and Agriculture Organization of United Nations, Rome.
6: Dadhwal, V.K. and S.S. Ray, 1998. Crop assessment using remote sensing-II: Crop condition and yield assessment. Proceedings of the National Seminar of Remote Sensing and Agricultural Statistics, April 21-22, 1998, Space Applications Centre, Ahmedabad,-pp: 55.
7: Deosthali, V. and A. Akmanchi, 2006. Yield estimation of summer crops grown in a mixed cropping area, a spectral growth curve profile approach. J. Ind. Soc. Remote Sensing, 34: 279-288.
Direct Link |
8: Doorenbos, J. and W.O. Pruitt, 1975. Guidelines for predicting crop water requirements. Tech Note 24. Food and Agriculture Organization of United Nations, Rome.
9: Draper, N.R. and H. Smith, 2003. Applied Regression Analysis. 1st Edn., John Wiley and Sons Inc., New York, pp: 299-368.
10: Dubey, R.P., M.H. Kalubarme, O.P. Jhorar and S.S. Cheema, 1987. Wheat yield models and production estimates for Patiala and Ludhiana districts based on Landsat-MSS and Agrometeorological data. Scientific Note: IRS-UP/SAC/CPF/SN/ 08/87, Space Applications Centre, Ahmedabad.
11: Dubey, R.P., N. Ajwani, M.H. Kalubarme, V.N. Sridhar and R.R. Navalgund et al., 1994. Pre-harvest wheat yield and production estimation for the Punjab, India. Int. J. Remote Sensing, 15: 2137-2144.
CrossRef | Google |
12: Hundal, S.S., R. Singh and L.K. Dhaliwal, 1997. Agro-climatic indices for predicting phenology of wheat (Triticum aestivum) in Punjab. J. Agric. Sci., 67: 265-268.
Direct Link |
13: Jand, S., 1999. Spectral-Agromet yield modeling of cotton and wheat for the South-Western districts of Punjab. M.Sc. Thesis. Punjab Agricultural University, Ludhiana, India.
14: Kalubarme, M.H., R.K. Mahey, S.S. Dhaliwal, S.S. Sidhu and K. Singh et al., 1995. Agromet-spectral wheat yield modeling in Punjab. Proceedings of the National Symposium on Remote Sensing of Environment with Special Emphasis on Green Revolution, November 22-24, 1995, Punjab Agricultural University, Ludhiana, India,-pp: 11.
15: Kramer, P.J., 1969. Plant and Soil Water Relationships: A Modern Synthesis. 1st Edn., Tata McGraw Hill Publishing Co., New Delhi, pp: 296-345.
16: Mahey, R.K., 1999. Acreage estimation and technique development for yield prediction through remote sensing. Final Technical Report: Adhoc Scheme, Department of Agronomy, Punjab Agricultural University, Ludhiana, India, 8-10/95-AFC dt. 5.1.995/ICAR.
17: Marcellos, H. and W.V. Single, 1972. The influence of cultivar, temperature and photoperiod on post-flowering development of wheat. Aust. J. Agric. Res., 23: 533-540.
Direct Link |
18: Medhavy, T.T., T. Sharma, R.P. Dubey, R.K. Mahey and A.K. Sharma, 1995. Development of a wheat yield model for Punjab using remotely sensed data and historical yield trends. J. Ind. Soc. Remote Sensing, 23: 23-30.
CrossRef | Direct Link |
19: Michael, A.M., 1978. Irrigation: Theory and Practice. 1st Edn., Vikas Publishing House Pvt. Ltd., New Delhi, pp: 754-761.
20: Nuttonson, M.Y., 1955. Wheat Climate Relationships and use of Phenology in Ascertaining the Thermal and Photo-Thermal Requirements of Wheat. 1st Edn., American Institute of Crop Ecology, Washing, DC., pp: 54-55.
21: Peters, D.B., J.W. Pendelton, R.H. Hageman and C.M. Brown, 1971. Effect of night air temperature on grain yield of corn, wheat and soybeans. Agron. J., 63: 809-809.
CrossRef | Direct Link |
22: Pinter, Jr. J. Paul, J.L. Hatfield, J.S. Schepers and E.M. Barnes et al., 2003. Remote sensing for crop management. Photogrammetric Eng. Remote Sensing, 69: 647-664.
Direct Link |
23: Rao, G.S.L.H.V.P., 2003. Agricultural Meteorology. 1st Edn., Kerala Agricultural University, Kerala, India, pp: 95-112.
24: Reddy, T.Y. and G.H.S. Reddi, 2003. Principles of Agronomy. 1st Edn., Kalyani Publishers, Ludhiana, India, pp: 48-77.
25: Sharma, A., R.K. Sood and M.H. Kalubarme, 2004. Agrometeorological wheat yield forecast in Himachal Pradesh. J. Agrometeorol., 6: 153-160.
Direct Link |
26: Verma, U., D.S. Ruhal, M. Yadav, A.P. Khera and R.S. Hooda et al., 2003. Wheat production forecast using remote sensing nd agromet variables in haryana state. J. Ind. Soc. Remote Sensing, 31: 141-144.
CrossRef | Direct Link |