Regression a-d Model building - National Food Policy Capacity

advertisement
Crop Forecasting by using Crop-Yield
Weather Regression Model*
by
Dr. M. Rezaul Karim Talukder1
&
Dr. M. Sayedur Rahman2
1Team
2
*
Leader, EWFIS Project, Ministry of Food.
Agricultural Statistics Specialist, EWFIS Project, Ministry of Food.
The views expressed in this report are those of the authors and do not necessarily reflect the official
position of the Government of Bangladesh.
Introduction
Crop production in Bangladesh is largely governed by the
vagaries of nature. Climate is one of the key components influencing
agricultural productivity and has never been stable. The effect of
climate variability on agriculture is very significant (Virmani, 1987).
Several studies (O’Toole and Jones, 1987; Oldeman, 1980; Satake and
Yoshida, 1978) indicate that weather during cropping season strongly
influences the crop growth and account for two-third (67%) of the
variation in productivity while other factors account for only one-third
(33%).
Rice is grown in more diverse environmental condition than any
other major food crop in the world. The different stages of growth and
yields of crops are affected by weather fluctuations that deviate from
optimum. Growth and yield of rice are affected by climate factors such
as temperature, rainfall, hours of bright sunshine, evapotranspiration
and solar radiation (Curry et al., 1990; Yoshida, 1978; 1981). Models
for estimating and predicting crop yields require precise knowledge of
yield weather relationship.
In spite of the adaptation of modern agricultural practices and
plant protection measures, favorable weather condition is essential for
good harvest. Each phase of agricultural activity from preparatory
tillage to plant growth and harvest does have strong dependence for
weather (Rahman, 2000a, b; Ahmed et al. 2000).
Materials and Methods
Modeling Approach
There are several methods of arriving at forecasts of production
of food corps. These are: (a) monitoring crop conditions on the basis of
agro climatic data, (b) making regular survey to assess area, yield and
production of crops; and (c) estimating regression models describing
quantitative relationship between selected climate/input variables and
final yield of the crop. This study used the last approach to estimate
yield and production of boro rice by crop-climate regression model.
2
Study Areas and Data Collection
Estimation of the model required two sets of data
(a) historical crop- yield data;
(b) historical data on a number of agro-climatic variables.
While yield data for this exercise were the yields of different varieties of
Boro rice, agro-climatic data included weekly / periodic data on rainfall,
maximum and minimum temperature and solar radiation, input data
included price and fertilizer.
The agro-climatic parameters that affect crop yield are rainfall,
temperature, solar radiation and evapotranspiration. This study was an
attempt to determine the nature and extent of relationship of these
variables with crop yield at different stages of crop growth.
Data relating to these agro-climatic variables were obtained from the
BMD, yield and input data from the BBS and DAM. However, the
nature of data obtained from the BMD constrained estimation of the
model in both spatial and temporal dimensions. Thus for each
district/region 21 years (1980-2000) data had been considered in this
study.
Finally, data were reduced total rainfall for the week, cumulative
rainfall of the season up to the week, maximum weekly average
temperature, minimum weekly average temperature, average weekly
temperature and diurnal temperature range, net solar radiation and
water balance index in order to fit into the models used.
3
Model Building
Crop-Climate Regression Model
 Regression analysis is a statistical technique for investigating and
modelling the relationship between variables.
Applications of
regression are numerous and occur in almost every field,
including food management, engineering, the physical sciences,
economics, life and biological sciences, and the social sciences. In
fact, regression analysis may be the most widely used statistical
technique.
A regression model that involves more than one
regressor variable is called a multiple regression model. Fitting
and analyzing this model is discussed below.
Regression models are used for several purposes, including




Data Description
Parameter Estimation
Prediction and Estimation
Control
The statistical crop-climate regression model is
m
n
Yi = i+  ij Wj + ik Ik + ui
j=1
k=1
Where Yi is the yield of the ith crop, Wj is the jth agro-climatic variable
in the production of the ith crop, Ik is the kth input variable in the
production of the ith crop, j and k are the coefficient of the relevant
variable, o is the constant and ui is the disturbance term. For a
particular crop, the explicit form of the equation will be determined by
the variables relevant for the crop. In the present exercise, the full
4
model contained ten regressors, which were to explain the yield of Boro
rice for the 2001-02 season.
The explicit formulation of the multiple regression models was:
Y = o + 1 MXT + 2 MNT + 3 AVT + 4 DTR + 5 TRF + 6 CRF
+ 7 NSR +  PR +  FERT +  TIME + e
(1)
Where,
Y= Yield of Boro rice (Mt/ha).
MXT= Maximum temperature (0c)
MNT= Minimum temperature (0c)
AVT= Average temperature (0c)
DTR= Diurnal temperature range (0c)
TRF= Total rainfall of the week (mm)
CRF= Cumulative rainfall for the season up to the week (mm)
NSR= Net solar radiation (cal/cm2/day)
PR = Price value (Tk. per quintal)
FERT = Fertilizer (‘000 Mt.)
TIME= Year
e= Stochastic term/ residual term / error term.
The method of least squares used to estimate the regression coefficients
in equation (1) by MS Excel and SPSS software.
The model was estimated for early reproductive stages of Boro rice crop
growth, corresponding to week 13 (1st week of April) of the year 2002.
For this stage of crop growth, models were estimated for local, HYV
and all Boro rice. The results of initial estimates of the model and those
of the model finally selected through a series of regression diagnostics
are presented in Tables 1 to Table 5 and finally select model using
sensitivity analysis to estimate the yield and production scenarios from
Table 6 to Table 8.
5
Regression Diagnostics
Model Adequacy and Sensitivity Test
Since most regression problems involving time series data (Box and
Jenkins, 1976) exhibit positive autocorrelation, the hypothesis usually
considered in the Durbin-Watson test (Durbin and Watson, 1971). The
use and interpretation of a multiple regression model often depends
explicitly or implicitly on the estimates of the individual regression
coefficients. Several techniques have been proposed for detecting
multicollinearity (Farrar and Glauber, 1967; Montgomery and Peck,
1982). These are as follows:
 Examination of the Correlation Matrix;
 Variance Inflation Factors;
 Eigensystem Analysis of X1X.
In multiple regression problems certain tests of hypothesis about
the model parameters are useful in measuring model adequacy. The
test of significance of regression is a test to determine if there is a linear
relationship between response variable and any of the regressor
variables. Using F-test, we conclude that yield is related to climatic
variables.
However, this does not necessarily imply that the
relationship found is an appropriate one for predicting yield as a
function of climatic variables. Further tests of model adequacy are
required.
Evaluating model adequacy is an important part of a multiple
regression problem. R2 is a measure of the reduction in the variability of
dependent variable obtained by using the regressor variables.
However, a large value of R2 does not necessarily imply that the
6
regression model is a good one. Adding a regressor to the model will
always increase R2, regardless of whether or not the additional
regressor contributes to the model. Thus it is possible for models with
large values of R2 to perform poorly in prediction or estimation.
Our focus was on techniques to ensure that the functional form of
the model was correct and that the underlying assumptions were not
violated.
In some applications, theoretical considerations or prior
experience can be helpful in selecting the regressors to be used in the
model.
Building a regression model that includes only a subset of the
available regressors involves two conflicting objectives:
 we would like the model to include as many regressors as
possible so that the “information content” in the factors can
influence the predicted value of response variable;
 we want the model to include as few regressors as possible
because the variance of the estimated response variable
increases as the number of regressors increase. Also, the more
regressors there are in a model, the greater the costs of data
collection and model maintenance.
The process of finding a model that is a compromise between
these two objectives is called selecting the “best” regression equation.
Experience, professional judgment in the subject matter field and
subjective considerations all enter into the variable selection problem
(Hocking, 1976).
Backward elimination attempts to find good model by working in
the opposite direction. That is we begin with a model that includes all K
as regressors.
Then the partial F-statistic is computed for each
regressor as if it were the last variable to enter the model. Backward
7
elimination is often a very good variable selection procedure (Allen,
1974). Analysts, who like to see the effect of including all the candidate
regresses, just so that nothing will be missed, particularly favor it.
In general if a model is to extrapolate well, good estimates of the
individual coefficients are required.
When multicollinearity is
suspected, the least squares estimates of the regression coefficients may
be poor. This may seriously limit the usefulness of the regression model
for inference and prediction (Marquardt and Snee, 1975; Mason et al.,
1975).
The method of least squares used to estimate the regression
coefficients in equation (1) by MS Excel and SPSS software. The model
was estimated for early reproductive stages of boro rice crop growth,
corresponding to weeks 13 (Ist week of April) of the year 2001-02. For
this stages of crop growth models were estimated for local, HYV and all
Boro rice. The ridge regression technique was applied for sensitivity
analysis but not found suitable stable estimate. That’s why finally we
had to consider O.L.S model for estimation of yield and production of
Boro rice crop.
8
Ridge Regression Simulation Study for Sensitivity
Analysis
When the method of least squares is applied to non-orthogonal
data, poor estimates of the regression coefficients are usually obtained.
The variance of the least squares estimates of the regression coefficients
may be considerably inflated and the length of the vector of least
squared parameter estimates are too long on the average. This implies
that the absolute value of the least squares estimates is too large. The
Gauss-Markov property referred that the least squares estimator has
minimum variance but there is no guarantee that this variance will be
small (Montgomery and Peck, 1982).
The ridge regression estimates may be computed by using an
ordinary least squares computer program and augmenting the data as
follows:
 X
XA  
 KI p



Y 
YA   
 OP 
Where KI P is a pxp diagonal matrix with diagonal elements equal to
the square root of the basing parameter and Op is a Px1 vector of zeros.
The ridge estimates are then computed from

 R  (X 1A X A ) 1 X 1A YA  (X 1 X  K1P ) 1 X 1 Y
Where K0 is a constant selected by the analyst. Note that when K= o
the ridge estimator is the least square estimator.
9
Methods for Choosing Biasing Parameter K
Hocrl et al., (1976) have suggested that an appropriate choice for K is
2
K
P
1 

Where,

=
standardized regression coefficient and
2

= mean sum
square error are found from the least squares estimator of the
standardized regression coefficient, and analysis of variance table.
The following sequences of estimates of  and K Via Simulation;


p
k0 

2
^ / 
 
 2

 R (k 0 )
p
k1 
 /


R ( K1 )
 R(K )
1
2

 R (k 1 )
k2

p
/

R ( k1 )

/
R ( k1 )
.
.
The procedure is terminated when
k i 1  k i
The ridge regression technique was applied but was not found suitable
for this crop-climate regression model because after a series of iteration
the stable estimates were no found.
That’s why finally we had to
consider O.L.S model for estimation of yield and production of Boro
rice crop.
10
Results and Discussions
Equations
were
estimated
with
respect
representing the 23 old districts of Bangladesh.
to
23
locations
However, for the
purpose of demonstrations of the details about the parameter estimates
and the regression diagnostics, equation for one district was examined.
For this purpose, Mymensingh was chosen on the basis of the fact that
total area under Boro in Mymensingh was one of the highest in the
country.
Table 1, Table 2, and Table 3 show the coefficients and related
statistics of the model estimated at the early maturity stage falling in the
13th week of the calendar year and 1st week of April 2002. The local
Boro equation for 13 week (Table 1) showed that R2 = .67, implying that
67% of the variability in yield was explained by the variables included
in the model.
Overall
(F-test)
regression
coefficients
were
statistically
significant and some regression coefficients were statistically significant
by using single t-test. Also Durbin-Watson statistic showed that the
calculated d value d=2.739 (where tabulated d value dL=1.20, du=1.41)
the null hypothesis was not rejected at the 5% level of significance.
Thus there was no positive autocorrelation at all. Fig.1& 2 showed that
histogram and normal probability plot did not indicate any serious
departure from the assumptions. Similar result found by Draper and
Smith (1981). These statistical tests on residuals would make relatively
confident that including them would not seriously limit the use of the
model.
11
The HYV Boro equation for 13 week (Table 2) showed that R2
was .862, implying that 86.2% of the variability in yield was explained
by the variables included in the model. Overall (F-test) indicated that
regression coefficients were statistically significant and some regression
coefficients were statistically significant by using single t-test.
Also
Durbin-Watson statistic showed that the calculated d value d=1.734
(where tabulated d value dL=1.20, du=1.41). Thus the null hypothesis
was not rejected at 5% level of significance, implying that there was no
positive autocorrelation. Fig.3& 4 showed that histogram and normal
probability plot did not indicate any serious departure from
assumptions. Similar result found by Draper and Smith (1981). This
statistical test on residual would make one relatively confident that
including them would not seriously limit the use of the model.
The results of all Boro equation for 13 week (Table 3) showed that
R2 = .904, implying that 90.4% of the variability in yield was explained
by the variables included in the model. Overall (F-test) indicated that
regression coefficients were statistically significant and some regression
coefficients were statistically significant by using single t-test.
Also
Durbin-Watson statistic showed that the calculated d value d=2.358
(where tabulated d value dL=1.20, du=1.41), the null hypothesis was not
rejected at 5% level of significance. Thus there was no positive
autocorrelation at all. Fig.7 & Fig. 8 shows that histogram and normal
probability plot did not indicate any serious departure from the
assumptions. Similar result found by Draper and Smith (1981). This
statistical test on residual would make one relatively confident that
including them would not seriously limit the use of the model.
For week 13 the correlation coefficient between MXT & AVT
(r=.6527), MXT & DTR (r=.6427), MXT & TRF(r=-.748), MXT &
12
CRF(r=-.618),
MNT
&
AVT(r=.761),
MNT
7DTR(r=-.761),
TRF&CRF(r=.696), FERT& TIME(r=.73) and PRICE &TIME (r=.85)
were highly significant. Collinearity diagnostic test for multicollinearity
problems of local, HYV and all Boro for week 13 (Table 4) showed that
multicollinearity was present in the data. Also simple correlation of
independent variables showed that some variables were highly
correlated which indicated that multicollinearity was suspected. The
ridge regression simulation study could play a vital role to solve this
multicollinearity problems; but in this case, stable estimator could not
be found through a long simulation study.
The fitted regression model of local Boro for 13 (Table 5) week
showed that maximum temperature and time had positive impact on
local Boro yield whereas average temperature, diurnal temperature
range, total rainfall, net solar radiation, fertilizer and price had
negative impact on local Boro yield. Also R2=.67 and F= 3.045 values
were significantly improved through variable selection and model
adequacy test. Overall (F-test) regression coefficients were statistically
highly significant. Among the individual coefficients, the coefficient of
the maximum temperature, average temperature, diurnal temperature
range, net solar radiation, price and time variable was statistically
highly significant.
The equation for HYV Boro for 13 week (Table 5) showed that
maximum temperature and time had positive impact on HYV Boro
yield whereas average temperature, diurnal temperature range, total
rainfall, cumulative rainfall, net solar radiation and fertilizer had
negative impact on HYV Boro yield. Also R2=.864 and F=9.555 values
were significantly improved through variable selection and model
adequacy test. Overall (F-test) regression coefficients were statistically
13
highly significant. Among the individual coefficients, the coefficient of
the maximum temperature, average temperature, diurnal temperature
range and time variable was statistically highly significant.
Finally, the equation for all Boro for 13 week (Table 5) showed
that maximum temperature and time had positive impact on all Boro
yield whereas average temperature, diurnal temperature range, total
rainfall, net solar radiation, fertilizer and price had negative impact on
all Boro yield. Also R2=.902 and F=13.766 values were significantly
improved through variable selection and model adequacy test. Overall
(F-test) regression coefficients were statistically highly significant.
Among the individual coefficients, the coefficient of the maximum
temperature, average temperature, diurnal temperature range and time
variable was statistically highly significant.
The estimated yields and production of local, HYV and all Boro
rice (Table 6, 7 & 8) in 23 districts for 2001-02 from the summary
statistics of the equations were calculated by using fitted regression
model. It can be seen from Table 6, 7 and 8 that the estimated yields of
local, HYV and all Boro rice for the Mymensingh district were 2.0961,
3.283588 and 3.546412 metric tones/ha respectively. When multiplied
by the respective area coverage, total production of local, HYV and all
Boro rice for the district 13.41567, 616.0011 and 688.0039 thousand
metric tones respectively. The predicted production of Boro of other
districts is also presented in Tables 6, 7 and 8. The estimated national
production of Boro rice, as obtained from the equations, were 335.4826,
11592.01 and 12501.35 thousand metric tones for of local, HYV and all
Boro rice respectively. The detail modelling exercises were reported for
different rice crop like Boro, Aus and Aman in the report number
“EWFIS-15 & 35, 23 and 29 respectively.
14
Concluding Remarks
Conclusions
The purpose of this study is to set forth the key issues that must be
resolved in formulating a national food policy that can help achieve
food security. The overall targeted output of the research program is to
formulate locally based public policies and strategies for the sustainable
development of agricultural markets. The crop rotation system in
Bangladesh is very complex. The output of the model showed the
potential it had in predicting yield.
The results obtained from the
analysis provide a picture of predicted production of Boro rice for the
2001-2002 seasons both for individual districts and for the country as a
whole. These predictions are made at the early maturity stage of the
crop growth cycle. An important consideration would be to judge the
reliability of the estimation, particularly in the content of the real world
situation. If the findings are judged to be in conformity with the real
world environment, they should be providing important clues for
advance food planning both at national and regional levels.
It may be mentioned that the actual Boro rice production in the
country in the immediate past two years were 11027 thousand metric
tons in 1999-2000 and 11920 thousand metric tons in 2000-2001. Thus
there may be substantial yield losses of Boro in some locations of the
country.
A comparison of the predicted Boro rice production in week 13 of
2002 with the actual production of the two preceding years shows that
while the predicted production increased to the actual production of
2000-2001 by 4.67% respectively.
Thus the predicted Boro rice
production in 2001-2002 may be taken as a closer approximation of
15
reality, particularly in view of the agroclimatic environments that have
prevailed during the crop growth stages.
On the basis of the modeling output the following conclusions
could be drawn: (1) Delay beyond optimum planting time will reduce
Boro production of the country. The model output opened a new
horizon in the course of the research work and eventually it was felt
necessary to check thoroughly all the driving variables contributing to
the production system so as to find out the reasons behind such a low
yield.
Recommendations
Further improvements will make the model a more useful tool for
the agriculture sector of Bangladesh. The model could be used to:
 estimate rainfed yield;
 investigate the reasons for low yields;
 assist as a tool to complement the soil and water management
research;
 train agriculture extension personnel in the growth and
management of the crop.
In future, the model integrate with GIS enables a quick interpretation
of the results by the user, as well as a quick definition of alternative
scenarios (Fig. 7). The agronomic community, including farmers, land
managers, fellow scientists, policy makers and the general public should
benefit from this evolving and expanding field.
The integrated GIS-Crop model plays a vital role in alleviation of
poverty of Bangladesh.
The model also helps the government and
relevant organizations to identify methods for reducing poverty and
16
achieving sustainable development in agriculture and land use. This
integrated GIS-Crop model had four main uses:
 as a national, regional and local level conservation policy
studies;
 in programme planning and evaluation;
 in project planning and design;
 as a research and teaching tool.
 Stabilizing the price food in the market and continued supply
of food grains in the rural markets are the best options for
increasing household food security.
 The second priority option for mitigation household food
insecurity was to improving PFDS by making it efficient and
effective.
 Finally suggested to improving rural communication for better
transportation of goods and that would certainly benefit the
locality as well as community.
This study will help identify the types of public policies needed for the
development of competitive and efficient agricultural markets that can
contribute in reducing rural poverty and promoting agricultural
economic growth in the country.
17
Table 1 :Summary statistics of multiple regression model of local Boro
for Mymensingh district (Week 13, 2002).
Model
Constant
MXT
AVT
DTR
TRF
CRF
NSR
FERT
PRICE
TIME
Regression
t
Significant
coefficients
value
value
2.96
2.567
.026
5.544
3.08
.01
-5.584
-3.108
.01
-2.749
-3.074
.011
-.00258
-.971
.353
.0001457
.126
.902
-.00174
-2.322
.040
-.000012
-1.649
.127
-.000063
-2.115
.058
.06121
2.863
.015
R2
F
DW
.67
2.487**
2.739
** 1% level of significance.
Histog ram
Norm al P-P Plot of Re gression Standa rdized Re sidu
Depe ndent Variable: YLB O
Depe ndent Variable: YLBO
10
1.00
8
.75
6
2
.50
Expected Cum Prob
Frequen cy
4
.25
Std. Dev = .74
Mean = 0.0 0
N = 21.00
0
-2.00
-1.50
-1.00
-.50
0.00
.50
1.00
0.00
0.00
1.50
Regress ion Standardi zed Residual
.25
.50
.75
1.00
Observed Cum Prob
Fig. 1: Histogram for regression standarised
local Boro rice.
Fig. 2: Normal probability plot of residul for
regression standardize residual for local
Boro rice.
18
Table 2 Summary statistics of multiple regression model of HYV Boro
for Mymensingh district (Week 13, 2002).
Model
Regression
t
Significant
coefficients
value
value
1.581
1.628
.132
3.980
2.637
.023
-3.95
-2.624
.024
-1.98
-2.64
.025
-.000533
-.239
.815
-.000529
-.549
.594
-.000717
-1.142
.278
-.00000537
-.881
.397
.000006237
.025
.981
.05066
2.805
.017
.862
7.786**
1.734
Constant
MXT
AVT
DTR
TRF
CRF
NSR
FERT
PRICE
TIME
R2
F
DW
** 1% level of significance.
Histogram
Dependent Variable: YHBO
3.5
Norm al P-P Plo t of Regression Stan dardized Residual
3.0
Depe ndent Vari able: YHB O
1.00
2.5
2.0
.75
Expecte d Cum Prob
Frequen cy
1.5
1.0
Std. Dev = .74
.5
Mean = 0. 00
N = 21.00
0.0
-1.50
-1.00
-1.25
-.50
-.75
0.00
-.25
.50
.25
1.00
.75
1.25
.50
.25
0.00
0.00
Regress ion Standardize d Residual
.25
.50
.75
1.00
Observe d Cum Prob
Fig. 3: Histogram of regression standarised
residul for HYV Boro rice.
Fig. 4: Normal probability plot of
regression standardized
residual for HYV Boro rice.
19
Table 3: Summary statistics of multiple regression model of All Boro
for Mymensingh district (Week 13, 2002).
Model
Regression
t
Significant
coefficients
value
value
1.781
1.602
.137
4.373
2.532
.028
-4.362
-2.523
.028
-2.166
-2.573
.028
-.00186
-.731
.48
.000597
.541
.599
-.00109
-1.519
.157
-.0000103
-1.473
.169
-.000031
-1.085
.301
.09282
4.521
.001
.904
11.599**
2.358
Constant
MXT
AVT
DTR
TRF
CRF
NSR
FERT
PRICE
TIME
R2
F
DW
** 1% level of significance
Norm al P-P Pl o t of Regre ssi on Stan dardi zed R esi dua
Histog ram
Depen dent Varia ble: Y ABO
Depe ndent Vari abl e: YAB O
7
1.00
6
.75
5
Frequen cy
3
2
Expecte d Cum Prob
4
.50
Std. Dev = .74
1
Mean = 0.0 0
0
N = 21.00
-1.25
-.75
-1.00
-.25
-.50
.25
0.00
.75
.50
1.25
.25
0.00
1.00
0.00
Regres sion Standard ized Residual
.25
.50
.75
1.00
Obs erve d Cum Prob
Fig. 5: Histogram for regression standarised
residul for all Boro rice.
Fig. 6: Normal probability plot of
regression standardized residual
for all Boro rice.
20
Table 4: Collinearity diagnostic test for multicollnearity of multiple
regression model of Boro rice for Mymensingh district (week 13, 2002).
Model
Constant
MXT
AVT
DTR
TRF
CRF
NSR
FERT
PRICE
TIME
Eigen value
8.352
.901
.426
.151
.08669
.04402
.02888
.008953
.0008789
.0000003058
Condition
index
1.0000
3.044
4.427
7.427
9.815
13.774
17.086
30.541
97.478
5225.912
21
VIF
8313.705
5000.018
4877.093
3.371
2.573
1.998
2.673
6.008
10.046
Table 5: Summary statistics and estimated values of the O.L.S
regression coefficients of the fitted Crop Yield-Weather Regression
Model for Boro rice, 2001-2002 season, (Week 13, 2002;early maturity
stage) of Mymensingh district.
Variables/statistic
Regression coefficients/(t-values)
Local Boro
Const.
MXT
HYV Boro
All Boro
3.01
1.581
1.958
5.583
(3.287)
3.978
(2.758)
4.535
(2.747)
-5.625
(-3.321)
-2.77
(-3.286)
-.00241
(-1.094)
-3.947
(-2.742)
-1.979
(-2.764)
-.00055
(-.271)
-.000521
(-.599)
-.000722
(-1.261)
-.0000054
(-.955)
-4.529
(-2.751)
-2.251
(-2.747)
-.00117
(-.547)
MNT
AVT
DTR
TRF
CRF
NSR
FERT
PRICE
TIME
R2
-.001736
(-2.432)
-.0000121
(-1.771)
-.0000617
(-2.294)
.06066
(3.025)
.67
.05104
(5.65)
.864
-.00103
(-1.498)
-.0000109
(-1.641)
-.0000259
(-.989)
.09054
(4.645)
.902
F
3.045
9.555
13.766
d
2.739
1.794
2.358
22
Table 6: Estimated yields and production of local Boro rice in 23
district for 2001-2002 season, as obtained from the Crop YieldsWeather Regression Model (Week 13, 2002; early maturity stage).
District
Yield per
hectare
Area of Local
Boro
(m. tons)
Bandarban
Barisal
Bogra
Chittagong
Comilla
Dhaka
Dinajpur
Faridpur
Jamalpur
Jessore
Khagrachari
Khulna
Kishoreganj
Kushtia
Mymensingh
Noakhali
Pabna
Patuakhali
Rajshahi
Rangamati
Rangpur
Sylhet
Tangail
Total
(.000 hectare)
0.413014
1.750407
2.76611
1.850436
1.789772
1.43796
1.050629
1.424805
1.69851
1.991233
1.530155
1.898927
2.011517
1.829548
2.096199
1.44835
1.124639
1.436442
1.78015
2.568856
1.491291
1.212157
1.39182
23
0.089
13.122
1.76
0.2
5.92
8.714
0
12.293
4.43
1.385
0
9.157
34.808
0.242
6.4
1.15
5.21
7.955
3.47
0
4.12
100.145
1.715
222.285
Production of
Local Boro (.000
metric tons)
0.036758
22.96884
4.868354
0.370087
10.59545
12.53038
0
17.51513
7.524399
2.757858
0
17.38847
70.01688
0.442751
13.41567
1.665603
5.859369
11.4269
6.177121
0
6.144119
121.3915
2.386971
335.4826
Table 7: Estimated yields and production of HYV Boro rice in 23
district for 2001-2002 season, as obtained from the Crop YieldsWeather Regression Model (Week 13, 2002; early maturity stage).
District
Yield per
hectare
(metric tons)
Bandarban
Barisal
Bogra
Chittagong
Comilla
Dhaka
Dinajpur
Faridpur
Jamalpur
Jessore
Khagrachari
Khulna
Kishoreganj
Kushtia
Mymensingh
Noakhali
Pabna
Patuakhali
Rajshahi
Rangamati
Rangpur
Sylhet
Tangail
Total
2.067025
3.258924
2.794693
2.701032
3.285356
3.20296
2.882748
3.629764
3.13164
3.334255
2.38361
2.83159
3.714392
3.086332
3.283588
3.077122
3.017652
1.866792
2.31936
2.840352
3.065608
3.19782
3.1674
24
Area of HYV
Boro
(.000 hectare)
4.381
103.281
247.57
111.9
280.148
238.151
184.14
174.016
161.635
261.363
6.571
84.393
259.556
77.068
187.6
99.269
165.79
0.6
321.32
6.7
358.745
256.151
147.225
3737.573
Production of
HYV Boro (.000
metric tons)
9.055637
336.5849
691.8821
302.2455
920.3859
762.7881
530.8292
631.637
506.1826
871.4509
15.6627
238.9664
964.0927
237.8574
616.0011
305.4628
500.2965
1.120075
745.2568
19.03036
1099.772
819.1248
466.3205
11592.01
Table 8: Estimated yields and production of all Boro rice in 23 district
for 2001-2002 Season, as obtained from the Crop Yields-Weather
Regression Model (Week 13, 2002;early maturity stage).
District
Yield per
hectare
Area of all Boro
(.000 hectare)
(metric tons)
Bandarban
Barisal
Bogra
Chittagong
Comilla
Dhaka
Dinajpur
Faridpur
Jamalpur
Jessore
Khagrachari
Khulna
Kishoreganj
Kushtia
Mymensingh
Noakhali
Pabna
Patuakhali
Rajshahi
Rangamati
Rangpur
Sylhet
Tangail
Total
2.067025
3.258924
2.800243
2.701032
2.989424
3.164383
2.92866
3.03409
3.016505
3.525775
2.38361
2.723985
3.534677
3.158878
3.546412
3.478304
3.048235
2.232723
3.491757
2.514574
3.251704
2.855149
3.105781
25
4.47
116.403
249.33
112.1
286.068
246.865
184.14
186.309
166.065
262.748
6.571
93.55
294.364
77.31
194
100.419
171
8.555
324.79
6.7
362.865
356.296
148.94
3959.858
Production
of all Boro
(.000 metric
tons)
9.239602
379.3485
698.1846
302.7857
855.1785
781.1754
539.2835
565.2783
500.9359
926.3903
15.6627
254.8288
1040.482
244.2129
688.0039
349.2878
521.2482
19.10095
1134.088
16.84765
1179.93
1017.278
462.575
12501.35
CABINET
FPMC
MOF
EWTC
FPMU
EWC
Assessment
Information
Memo,
Brief
EWFIS
Source of Information
BBS
DAM
DGF
SPARRSO
BWDB
BMD
BARC
DAE
Others
Forecasting Model
Statistical Inference & Sensitivity Analysis
Modelling Output
Integrate Model with GIS
GIS Map & Report
Fig.7: An Integrated GIS-Crop Production Forecasting Model of Early
Warning System for Food Management in Bangladesh.
26
Download