Rice yield prediction using Crop Growth Simulation and Agrometeorological models in Gujarat
Loading...
Files
Date
2007
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
AAU, Anand
Abstract
Among the various production commodities of basic importance,
agricultural production is the one which is subjected to wide and irregular
fluctuations of output. The importance of timely and reliable prediction of yield of
principal crops need to be emphasized for the country like India where, the
economy is mainly dependent on rainfed agriculture.
Various yield prediction models are very much useful to the government
agencies, trade and industry for planning about the distribution, storage,
processing, and export/import of crop produce besides taking timely policy
decisions on fixing levy prices as they provide accurate advance estimation of
yields. Besides this, these models have facilitated identification of production
constraints and for assisting in agro-technology transfer.
The present study was undertaken to predict the rice yield using crop
simulation models and to develop an appropriate agrometeorological model for
prediction of rice yield in Gujarat. Crop growth simulation models viz. DSSAT
and WOFOST were calibrated and validated under Gujarat conditions. Suitabl
agrometeorological model for predicting the yield of rice was also developed by
using combined effects of different weather parameters viz. rainfall, bright
sunshine hours, maximum and minimum temperature and morning and afternoon
relative humidity on rice yield. For the present investigation the major rice
growing districts of Gujarat viz., Ahmedabad, Baroda, Bulsar, Kheda,
Panchmahal, Sabarkantha and Surat were considered.
The simulation results of DSSAT model for different districts revealed that
the performance of the model for grain yield for all the districts was found highly
significant with coefficient of determination values of 98.3% and 95.6% during
the years 2004 and 2005 respectively.
Various test criteria were applied to validate the performance of the model.
The simulation performance of grain yield was found better in 2004 than in 2005.
The error per cent during the year 2004 was 9.9% and it was 10.3% in 2005. The
average error as computed by MAE was 139.3 and 135.6 with 173.3 and 186.4
RMSE during 2004 and 2005 respectively. The respective MBE was found to be
89.0 and -94.7 and index of agreement was 0.98 and 0.97.
Sensitivity study of DSSAT revealed that 3°C elevation in mean ambient
temperature decreased the yield by 4 to 41%. The 3°C down scaled ambient
temperature increased the yield by 7 to 76%.
The solar radiation increase from 1 to 3 MJm-2 day-1 resulted in 2 to 26%
yield increase. In case 3 MJm-2 day-1 reduction in solar radiation, 1 to 42%
decrease in yield was observed.
Elevation of 300 ppm in CO2 concentration over its base value increased
yield by 4 to 29%.
The simulation results of WOFOST model simulation in both the years for
different districts revealed that the simulated values of grain yield were found in
perfect agreement with corresponding observed yield. Error per cent of WOFOST
was 6.83% and 7.34% during 2004 and 2005 respectively. The coefficient of
determination was 99.6% and 97.1%; MAE was 93.85 and 91.14, RMSE was
116.98 and 134.60 and MBE was found to be 50.14 and -66.57, respectively
during the year 2004 and 2005.
Three approaches using original observed weather variables (1) weekwise,
(2) stagewise and (3) periodwise average weather vaiables and two approaches
using generated weather variables (1) week number as weight and (2) correlation
coefficient as weight were used for fitting of the agrometeorological models.
Approaches using shorter interval i.e. weekly weather variables either original or
generated were found superior to the approaches using longer period averaged
weather variables i.e. stagewise and periodwise approaches.
The results revealed that the effects of all the weather variables in relation to
their quantum and direction differed over the approaches. However, they were
found important for prediction point of view in rice productivity. The effect of
weather variables also differed within the crop stage and period, indicating that
small interval of crop period results in significantly higher R2 value and thereby
minimizes the error of predicting the rice yield. Different approaches were found
superior over others in different districts.
In Ahmedabad district week number as weight approach provided suitable
yield prediction model, eight week before expected harvest which explained >
75% variation in productivity. The 12 week model fitted with week number as
weight approach is recommended as rice yield prediction model for Ahmedabad
district
Y = 6304.22 + 3.80 RHE1 – 0.69 RHM2 + 0.26 (RF*RHE)0 – 2.37 (MINT*RHE)0
+ 4.36 (MAXT*RHM)2 (Adjusted R2 = 90.6%)
Weekwise approach provided suitable yield prediction model in Baroda
district, 8 weeks before harvest which explained > 75% variability. The 12 week
model fitted with weekwise approach is recommended as rice yield prediction
model for Baroda district
Y = – 5050.94 – 0.33 RF2 + 19.06 BSS6 + 37.45 RHM9 – 42.51 BSS10 + 178.41
BSS11 + 24.46 RHM11 (Adjusted R2 = 99.4%)
The 18 week model fitted with weekwise approach which provided earlier
rice yield prediction (2 weeks before harvest) and explained > 75% variability is
recommended as rice yield prediction model for Bulsar district
Y = 5788.42 – 23.63 BSS3 – 0.15 RF3 – 69.95 MAXT5 + 59.49 MINT5 – 86.85
BSS6 + 106.21 BSS7 + 2.47 RF7 +103.57 MAXT8 – 33.49 BSS9 + 25.88
BSS11 + 3.73 RF11 – 0.84 RF12 – 170.01 MAXT13 – 70.32 MINT13 – 106.46
BSS14 – 41.87 RHE14 + 0.60 RF16 + 41.80 RHM17 (Adjusted R2 = 99.8%)
The 20 week model fitted with weekwise approach which explained > 75%
variability is recommended as rice yield prediction model for Kheda district
Y = 8682.13 + 317.72 MINT10 + 12.87 RHE10 – 202.97 MINT12 – 59.17 MAXT18
– 79.21 BSS19 – 32.92 RHM19 (Adjusted R2 = 99.8%)
The 12 week model fitted with weekwise approach which provided earlier
prediction of the rice yield (8 weeks before harvest) and explained > 75%
variability is recommended as rice yield prediction model for Panchmahal district
Y = 8881.23 – 105.44 BSS2 + 105.02 MAXT2 – 164.14 MINT2 – 8.90 RHE2 –
54.41 RHM7 + 2.78 RF8 – 80.16 MINT10 (Adjusted R2 = 96.0%)
The 14 week model which provided earlier prediction (6 weeks before
harvest) and explained > 75% variability is recommended rice yield prediction
model for Sabarkantha district
Y = – 1075.72 + 37.25 RHM1 + 2.68 RHM3 + 8.99 RHE4 + 28.55 RHM5 – 34.24
RHM6 – 6.40 RHE10 – 59.89 MAXT11 + 68.84 MAXT13
(Adjusted R2 = 99.8%)
The 20 week model fitted with weekwise approach which explained > 75%
variability is recommended as rice yield prediction model for Surat district
Y = 2851.46 + 72.40 MINT1 + 57.63 MAXT4 – 17.24 MAXT5 -0.06 RF8 – 2.83
RHM9 – 1.38 RF11 – 18.61 MINT16 – 2.75 RF18 – 1.66 RF19 – 89.51
MAXT20 – 6.24 RHE20 (Adjusted R2 = 99.8%)
The developed agrometeorological models were equally comparable to the
crop growth simulation models viz. DSSAT and WOFOST in all the districts
except in Baroda district where the performance of WOFOST and DSSAT
respectively was found superior to the developed agrometeorological model.
The average performance of the models was adept in yield prediction for all
the districts. The percent deviation for DSSAT was ±0.9 to ±9.9%, for WOFOST
from ±0.2 to ±6.1 and for agrometeorological model it was ±0.1 to ±16.8%.
The developed agrometeorological models could efficiently predict the rice
yield upto 8 weeks i.e. two months before the actual harvest of the crop with very
high accuracy (>90%). Thus these developed agrometeorological models can help
the government and various other agencies to take appropriate steps is case of
ensuing scarcity or glut situation.
The DSSAT and WOFOST also predicted the rice yield in reasonable limit
with additional advantage to project yield fluctuation with reasonable accuracy in
fluctuating weather particularly temperature, solar radiation and CO2
concentration.
Description
Keywords
agriculture, meteorology, method