Journal of the Indian Society of Remote Sensing (October 2018) 46(10):1701–1711

https://doi.org/10.1007/s12524-018-0825-8
(0123456789().,-volV)(0123456789().,-volV)

RESEARCH ARTICLE

Yield Forecasting of Spring Maize Using Remote Sensing and Crop

Modeling in Faisalabad-Punjab Pakistan
Ishfaq Ahmad1 • Umer Saeed1 • Muhammad Fahad1 • Asmat Ullah2 • M. Habib ur Rahman3 •

Ashfaq Ahmad1 • Jasmeet Judge4

Received: 13 December 2017 / Accepted: 25 July 2018 / Published online: 7 August 2018
Ó Indian Society of Remote Sensing 2018

Abstract
Real time, accurate and reliable estimation of maize yield is valuable to policy makers in decision making. The current
study was planned for yield estimation of spring maize using remote sensing and crop modeling. In crop modeling, the
CERES-Maize model was calibrated and evaluated with the field experiment data and after calibration and evaluation, this
model was used to forecast maize yield. A Field survey of 64 farm was also conducted in Faisalabad to collect data on
initial field conditions and crop management data. These data were used to forecast maize yield using crop model at
farmers’ field. While in remote sensing, peak season Landsat 8 images were classified for landcover classification using
machine learning algorithm. After classification, time series normalized difference vegetation index (NDVI) and land
surface temperature (LST) of the surveyed 64 farms were calculated. Principle component analysis were run to correlate
the indicators with maize yield. The selected LSTs and NDVIs were used to develop yield forecasting equations using least
absolute shrinkage and selection operator (LASSO) regression. Calibrated and evaluated results of CERES-Maize showed
the mean absolute % error (MAPE) of 0.35–6.71% for all recorded variables. In remote sensing all machine learning
algorithms showed the accuracy greater the 90%, however support vector machine (SVM-radial basis) showed the higher
accuracy of 97%, that was used for classification of maize area. The accuracy of area estimated through SVM-radial basis
was 91%, when validated with crop reporting service. Yield forecasting results of crop model were precise with RMSE of
255 kg ha-1, while remote sensing showed the RMSE of 397 kg ha-1. Overall strength of relationship between estimated
and actual grain yields were good with R2 of 0.94 in both techniques. For regional yield forecasting remote sensing could
be used due greater advantages of less input dataset and if focus is to assess specific stress, and interaction of plant genetics
to soil and environmental conditions than crop model is very useful tool.

Keywords Crop model  Remote sensing  Landcover classification  Regional yield forecasting

& Ishfaq Ahmad 1

Agro-climatology Lab, Department of Agronomy, University
ishfaq.ahmad@uaf.edu.pk of Agriculture Faisalabad, Faisalabad, Pakistan
Umer Saeed 2
Agronomic Research Station, Karor-Layyah, Punjab,
usagrarian@yahoo.comay Pakistan
Muhammad Fahad 3
Department of Agronomy, MNS-University of Agriculture,
fahadagr@outlook.com Multan, Pakistan
Asmat Ullah 4
Center for Remote Sensing, Agricultural and Biological
aumalik.aari@gmail.com Engineering Department, University of Florida, Gainesville,
M. Habib ur Rahman USA
habib.rahman@mnsuam.edu.pk
Ashfaq Ahmad
ashfaqchattha@uaf.edu.pk
Jasmeet Judge
jasmeet@ufl.edu

123
1702 Journal of the Indian Society of Remote Sensing (October 2018) 46(10):1701–1711

Introduction the crop season can help in yield forecasting. Crop growth
profile can be formed by calculating statistical average of
Accurate and real time crop yield forecasting is very NDVI in certain regions (Zhang et al. 2004). Use of LST
important for policy makers and grain agencies in decision and NDVI in regional yield forecasting gave good index of
making (Mkhabela and Mkhabela 2000). Population of agreement with observed yield (Johnson 2014).
Pakistan is increasing, and future scenarios showed that it Crop simulation models and statistical approaches like
would be 271 million by 2050, but food production is not regression models are the two main approaches used to
increasing at same rate (Schafer and Victor 2000). Severe forecast yield using weather data (Lee et al. 2013; Dumont
drought and floods due to climate change also affect the et al. 2015). Johnson (2014) developed regression tree-based
availability of food. Access of policy makers and planners models to study the relationship of remotely sensed pre- and
to real time crop condition and yield estimates is very within-season NDVI, land surface temperature, and rainfall
important to sustain food availability for ever growing with crop yield to forecast the yield of corn and soybeans.
population of Pakistan (Briscoe and Qamar 2006). Crop simulation models need a lot of input data such as crop
Maize is the third most important crop by area after phenology, soil characteristics, cultivar coefficients, weather
wheat and rice. It is the highest yielding crop of world and data, and cultural practices data for calibration and simula-
has significant importance in many countries including tions (Van Wart et al. 2013). Linear multiple regression
Pakistan. Early and accurate crop yield estimation help the models use less input data, but higher numbers of indepen-
decision maker in import and export policy (Dorosh and dent variables and the results might not be as good because it
Salam 2008). For this, remote sensing and crop models are assumes that all independent variables are linearly correlated
becoming important tools that help in yield forecasting. with dependent variables (Osborne and Waters 2002), which
Remote sensing is also used to monitor the vegetation is not always the case. Different forecasting models account
stress though different satellite observations like normal- for single source of variability in crop yield, which could be
ized difference vegetation index (NDVI) and Land surface explained by combined effects of remotely sensed indices
temperature (LST) (Friedl et al. 2002). Crop models are and climatic conditions. Crop models that use crop man-
used to understand the interaction of plant with soil, water agement practices, soil and weather data fail to include
and environmental factors. Crop models are crop specific information from remote sensing and also do not take into
and generally simulate water and nitrogen balance in soil account the spatial variability (Batchelor et al. 2002). The
and plant. Most of the crop models cannot simulate forecasting models that are based on remote sensing indices
micronutrients; pest and diseases. On these basis, one can alone cannot explain yield variability due to climatic factors
say that currently there is no comprehensive crop model (Singh et al. 2003).
(Boote et al. 1996). Combined use of remote sensing and crop models for
Decision support system for agro-technology transfer accurate and real-time yield forecasting of maize could be
(DSSAT) has been used in yield forecasting of many crops very useful. Yield forecasting models, based on satellite
at different growth stages (Yun 2003). CERES-maize derived indices, use less input data as compared to crop
model, under the shell of (DSSAT), is comprehensive and models. Crop models need large amount of input data (soil,
process based model that has been evaluated to study the weather, yield and yield components) for calibration and
effect of different management practices in maize (Jones evaluation and simulate crop growth on daily basis. Crop
et al. 1986). Many researchers used this to evaluate the models are also very helpful to explain different mecha-
potential response of maize to farmers’ management nism in crop growth but yield forecasting models (regres-
practices (Gaiser et al. 2010). Erda et al. (2005) used sion) based on remotely sensed data cannot help to explain
CERES-Maize for yield forecasting of maize in china. growth mechanisms. The study was planned with specific
Fontana et al. (2000) reported that yield forecasting in objective to forecast crop yield by using both remote
Brazil was done after crop harvest. CERES-Maize model sensing and crop modeling. Instead of relying on one
simulate the crop growth and development on daily basis technique, researchers could use different yield forecasting
from planting to maturity and describe the interaction of tools to get reliable production estimates.
soil and environmental conditions (Fraisse et al. 2001).
This shows that CERES-Maize can be used to forecast
maize yield for different farmers’ management practices. Materials and Methods
Vegetation indices calculated from remotely sensed data
during crop peak growth season has strong relationship Site of Study Area
with crop yield. Crop growth can be monitored continu-
ously using time series indices like NDVI. NDVI during Study was conducted at Faisalabad district in Central
Punjab Pakistan. The latitude 31.25 N, Longitude 73.06 E

123
Journal of the Indian Society of Remote Sensing (October 2018) 46(10):1701–1711 1703

and the elevation from the seas level is 184.5 m It has and amount, tillage operation, pesticide applications and
semi-arid climate characteristic. The annual temperature harvest operations etc.) and initial conditions (previous
and rainfall is 24.2 °C and 346 mm. Mainly summer sea- crop etc.).
son has more rainfall than the winter. The mean maximum
and minimum temperature gradually start to decrease from Application of Crop Model at Regional Scale
November to January and then rise. The soil of Faisalabad
is silt loam or very fine sandy loam. Faisalabad has mixed Before application of crop model at regional scale, first
cropping zone, in which wheat, rice, maize and sugarcane calibration and evaluation of model is done with experi-
etc. are cultivated. mental data set. A complete set of experiment on spring
maize was conducted during the year 2015 and 2016.
Survey Data Collection Experiment include four Maize hybrids (Pioneer-1543,
Mosanto-DK6103, Syngenta-NK8711) of different com-
Comprehensive survey was conducted in Faisalabad dis- panies and four sowing dates (27 January, 16 February, 08
trict to collect the crop management data from the farmers. March, and 28 March). Calibration (adjustment of genetic
Stratified random sampling technique were used for the coefficients) of CERES-Maize model V4.6.1 was done
selection of farms. A total of 64 farm were surveyed during with the best sowing date of 27 January 2015 and evalu-
2015 as shown in Fig. 1. Mobile Agricultural Geotagging ation (testing with independent data set) was done with
Information System (MAGIS) system were used to collect remaining sowing date of 2015 and with data of 2016. The
data. It is mobile based application which allows to collect performance of model was checked using Mean Absolute
the digital and georeferenced field data. It is based on Error (MAE) and Mean Absolute Percent error.
GeoODK platform. MAGIS form was created in the
1X n
Microsoft excel sheet and uploaded on database (https:// MAE ¼ jYSi  YOi j ð1Þ
n i¼1
ona.io/). The developed form was excess using GeoODK
Collect app by cell phone. The farmers interview were 1X n
jYSi  YOi j
conducted at their farms. Survey data includes crop man- MAPE ¼  100 ð2Þ
n i¼1 YOi
agement data (sowing time, irrigation and fertilizer date

Fig. 1 Farm surveyed in Faisalabad district during 2015

123
1704 Journal of the Indian Society of Remote Sensing (October 2018) 46(10):1701–1711

where Si and Oi are the simulated and observed value, n is (PCA) were run to see which NDVI and LST were closely
the number of parameter to be compared and i each value related to yield. Then closely related NDVI and LST were
of comparison. MAE measure the manganite of error in used to get co-efficient by using Least Absolute Shrinkage
observed and simulated values, while MAPE showed the and Selection Operator (LASSO) Analysis. The developed
relative term of mistake made in predicted values. coefficients were used to predict the yield of each farm.
After calibration model was applied at regional scale. Detail methodology is given in flow diagram of Fig. 2.
For this translation tools (ADA, QuadUI, and ACMO UI)
were used to create model file and to get arranged output.
The collected survey data of 64 farms was put in the .xlsx Results and Discussion
sheet, which have some variables for the conversion into
model files. The weather data of 2015 was collected from Model Calibration and Evaluation Results
the weather observatory, which includes maximum and of CERES-Maize Model
minimum temperature, precipitation, wind speed. Solar
radiation were calculated from the temperatures using the Data presented in Table 1 showed the calibration of three
formula describe by (Allen et al. 1998). Soil series data was maize hybrid with 27 January planting date and evaluation
collected form the Soil Survey of Pakistan (SSOP). with remaining planting dates and 2016 data. Results
Lyallpur soil series of Faisalabad was used. Soil layers indicate that the performance of model to the recorded
(0–190 cm) wise data were used which include silt%, parameters during experiment was good. In calibration (3
Clay%, organic carbon%, total nitrogen and cation number of pairs) model showed a close match between
exchange capacity (meq/100gm). The drainage upper limit, observed and simulated values of phenology (days to
lower limit, saturation%, bulk density and saturation anthesis and maturity), with MAE of 1.66. The value of
hydraulic conductivity were calculated from the model. MAE was little high of about 718 in case of grain yield.
Crop management, soil and weather data was feed into The LAI and biomass has values of MAPE are 0.35 and
survey sheet and data regarding cultivar information like 4.39.
row spacing, number of plant m-2 etc. were used in field Model evaluation results showed the accuracy of cali-
overlay sheet. The linkage file was created to link these brated model. The anthesis and maturity days were eval-
files. These three files of .xlsx format were converted into uated well, with MAPE values 2.22 and 3.92. Similar
.CSV format using AgMIP data assistant (ADA), then results were recorded for grain yield and LAI maximum.
QuadUI was used to convert the .CSV file into model file The high values of MAE were recorded in simulation of
having one batch file. By using this batch file model was final grain yield (Table 1).
run and output was taken using agricultural crop model Our results are similar to Saseendran et al. (2005) who
output (ACMO-UI) found that CERES-maize indicate a good accuracy in
simulation of planting on grain yield of three maize
Satellite Remote Sensing for Regional Yield hybrids. The CERES-Maize was calibrated and evaluated
Forecasting by Lin et al. (2015) who reported that model showed a
good root mean square error (RMSE) of 8.44 kg ha-1 in
For regional yield forecasting, first landcover classification simulating the grain yield. CERES-maize model was cali-
for spatial distribution maize were done using machine brated and evaluated using planting and nitrogen fertilizer.
learning algorithm. Three Landsat-8 images taken in sec- Model prediction in days to anthesis and maturity showed
ond week of May which correspond spatially and tempo- 2–3 days difference between observed and simulated.
rally to the local, peak growing season for maize were Model is useful tool and can accurately simulate the phe-
gathered. By sampling the second, third, and fourth bands nology, growth, grain yield and biomass (Chisanga et al.
of the Landsat-8 imagery at these reference locations, 2015). Calibrated the CERES-model by adjusting the
training data was constructed for a variety of machine genetic co-efficients and evaluated with independent data
learning algorithms. Cross validation was used for param- set. Results showed a good agreement with observed values
eter tuning as well as estimating the generalized and reported that CERES-maize model is useful tool in
performances. yield prediction of maize (Liu et al. 2012). (Mubeen et al.
After classification, the time series NDVI and LST for 2016) reported that CERES-maize model has ability to
growing season of spring maize were calculated for 64 predict the biomass and grain yield accurately. During
farms that were surveyed. NDVI and LST were calculated model calibration and evaluation, the simulated LAI, grain
with 16 days interval, which starts from end week of Jan- yield and biomass are within 10% error. Model is useful
uary to second week of June. Principle component analysis tool in decision making and can help in yield estimation at
farm level (Table 1).

123
Journal of the Indian Society of Remote Sensing (October 2018) 46(10):1701–1711 1705

Method 1 Method 2

Remote Sensing Model Calibration

Experimental Data

Landsat 8 (images)
Farmer survey data

Classification
(Machine Learning Algorithm)
Crop Management data

NDVI and LST Climate Data

(Pakistan Meteorological Department)

Soil Data
(Soil Survey of Pakistan)
PCA Analysis

Crop Model
LASSO Regression

Model Development Yield Forecasting

Fig. 2 Flow diagram methodology for yield forecasting using remote sensing and crop modeling

Table 1 Calibration and

Paramters No. of paired Observed range MAE MAPE
evaluation of CERES-maize
model with experimental data Anthesis day (DAP) 3 71–76 1.66 2.26
set
Physiological maturity day (DAP) 3 107–114 1.66 1.52
LAI maximum 3 7990–9380 33 0.35
Yield at harvest maturity (kg ha-1) 3 19,051–22,850 718 3.51
-1
Tops weight at maturity (kg ha ) 3 5.75–6.27 0.26 4.39
Anthesis day (DAP) 21 45–72 2.42 3.92
Physiological maturity day (DAP) 21 88–111 2.33 2.22
LAI maximum 21 5118–9203 454.9 6.71
Yield at harvest maturity (kg ha-1) 21 16,471–22,265 878.6 4.74
Tops weight at maturity (kg ha-1) 21 3.86–6.24 0.22 4.16

Yield Estimation from Crop Model 2015. Simulated yield was compared with the observed
yield of each farm as shown in Fig. 3. Results indicated
Farm survey data were used to create crop management that there is good agreement between the simulated and
file, initial condition, soil characteristic and observed observed 64 farmer yields of maize with value of coeffi-
weather data were used as input into model. After cali- cient of regression and RMSE were 0.99 and 255 (Fig. 4).
bration and evaluation model was run with for the year The performance of the model differed for good and poor

123
1706 Journal of the Indian Society of Remote Sensing (October 2018) 46(10):1701–1711

Fig. 3 Comparison of observed

and simulated yield of 64 farms

management practices of the farmers. The difference (2007) reported that model is very sensitive to management
between simulated and observed yield was less for those practices and environmental variable. If there is any stress
farmers whose management practices were according to or excess of management practices model under or over-
recommendations. Planting time, plant population, number stimulated the results (Jagtap and Jones 2002). Study was
of irrigation, irrigation at critical stages, fertilizer applica- conducted for regional yield estimation of maize. Different
tion dates, application at crop critical stages, weed man- scenarios of planting, soil combination were incorporating
agement and disease control were better in case of into the model. Model showed the very high accuracy with
progressive farmers field and model also simulated almost the actual. Model estimated the yield by interaction of plant
same yield as observed (Fig. 3). to soil, climate and other management factors (Moen et al.
CERES-wheat model was applied at 150 farms in rice– 1994).
wheat cropping system of Punjab Pakistan. The perfor-
mance of model was well in simulation of grain yield of Yield Estimation Using Remote Sensing
wheat with observed having RMSE of 749 kg ha-1 (Ah-
mad et al. 2015). CERES maize model has been used at Image Classification
regional scale for yield estimation under different climate
scenarios. The performance of model was asses through The tuned models of Machine Learning algorithms were
closeness of observed and simulated grain yield, Root used to determine the spatial distribution of maize fields for
mean square difference was 1098 kg ha-1. Xiong et al. growing seasons in the Faisalabad district using parallel

Fig. 4 Relationship of observed

and simulated yield of 64 farm

123
Journal of the Indian Society of Remote Sensing (October 2018) 46(10):1701–1711 1707

Table 2 Accuracy of machine learning algorithm used for spatial 85.9%. LST and NDVI has been for the monitoring of
distribution of maize drought in Turkey, results showed a good agreement
Classifier Accuracy % between real time ground and satellite data with R2 of 0.90.
(Orhan et al. 2014). An empirical relationship of NDVI and
Support vector machine (SVM)-radial basis 0.97
LST was developed by Anbazhagan and Paramasivam
SVM-linear 0.92 (2016), which indicated the crop health with respect to land
Quadratic discriminant analysis (QDA) 0.95 surface temperature.
Linear discriminant analysis (LDA) 0.90
Random forests 0.94 Analysis of Time Series NDVI and LST
Decision trees 0.93
k-nearest neighbor (KNN) 0.92 Figure 6 represents the principle component analysis of
Boosting 0.90 time series NDVI (Fig. 6a) and LST (Fig. 6b). The results
of NDVI (a) showed that that NDVI before peak season
(X5) * 60–62 days after planting (DAP), peak season
processing to improve computation time as shown in NDVI (X6) * 76–78 DAP and after peak season NDVI
Table 2. All classifiers of machine learning showed the (X7) * 90–92 DAP, are highly correlate to the grain yield.
accuracy greater then 90%, however support vector Results of LST showed that LST (X3) which is surface
machine with radial basis showed higher accuracy of 97%. temperature * 10–13 days after recommended planting
The best classifier (SVM-radial basis) were used for the date and LST (X4) which is * 28–30 DAP are highly
classification of maize. The accuracy of estimated area correlated to yield. PCA is multivariate techniques which
using machine learning algorithm was 91%, compared with can be used with time series observations and correlate
Crop Reporting Service (CRS) of Punjab Pakistan (Fig. 5). dependent variables (Algur et al. 2002). Time series NDVI
CRS reported the spring maize area of 6400 acres, while and LST correlation was carried out by PCA to develop the
our model predicted the area of 5824 acres. yield estimation model (Ramachandra et al. 2016).
Johnson et al. (2012) used the machine learning algo- The correlated variable were used to develop the yield
rithms for the classification of forest with accuracy of prediction model. LASSO regression was applied to get the

Fig. 5 Classified map of maize for Faisalabad district during 2015

123
1708 Journal of the Indian Society of Remote Sensing (October 2018) 46(10):1701–1711

Fig. 6 Principle component analysis of NDVI (a) and LST (b) of 64 farms during 2015

coefficient for model. The final developed with coefficients NDVI value is high over high biomass due to saturation of
are given below signal, but its relationship with yield may not good.
Yield ¼ 5020:26 þ 200:80  LST1 þ 25:65  LST2
 518:04  NDVI1 þ 12561:22  NDVI2 Comparison of Crop Model and Remote Sensing
 1154:30  NDVI3
The inter-comparison of crop model and remote sensing
LST1 = Land surface temperature (10–13 days after rec- are mainly based on the forecasting error like RMSE. This
ommended sowing date); LST2 = Land surface tempera- error was calculated from the predicted yield of 64 farm
ture (28–30 days after recommended sowing date); with their observed yield. RMSE results showed that model
NDVI1 = Values of NDVI before peak season (60–62 days results are more precise then the remote sensing. While
after sowing); NDVI2 = Values of NDVI at peak crop overall strength of relationship between predicted and
growth season (76–78 days after sowing); NDVI3- observed was good. The indicator like NDVI performed
= Values of NDVI after peak crop growth season well during early season of crop but remote sensing indi-
(90–92 days after sowing) cator some time did not if there is any stress in later stage
Results of this model are presented in Figs. 7 and 8. of crop. Based on the performance the remote sensing
There was a close relation between observed and predicted indicator perform batter in some situation as shown in
yield with R2 of 0.94 (Fig. 8). However, few of the farms Fig. 7. If long term time series observation are available for
have large variation in from the observed yield which regional yield estimation then remote sensing has greater
higher RMSE of 397.4 kg ha-1. The 95% prediction and advantages (Genovese et al. 2006) then model. Crop model
confidence interval showed that there is variation in predict the yield by interacting soil, water, plant and
observe and simulated values. The variation in yield of few environmental conditions and it gave us simulation on
farms might be due to high temperature at later productive daily basis. But for crop model we need lot of data set of
stage of maize causes the reduction in yield. In our case we soil parameters, weather conditions, crop management
have taken the LST at vegetative growth stages and NDVI data, which is labor intensive. So, in regional yield fore-
at peak season. High temperature at reproductive stage casting we can get good results with remote sensing,
reduces the time for grain filling duration that lead to without collection of intensive survey, but if we want to
decrease the yield. Panda et al. (2010) reported that NDVI assess the specific stress like temperature, water, nutrient at
has been widely used for vegetation health monitoring, specific stage then crop model are useful tools.

123
Journal of the Indian Society of Remote Sensing (October 2018) 46(10):1701–1711 1709

Fig. 7 Comparison of observed

and simulated yield of 64 farms

Fig. 8 Relationship of observed

and simulated yield of 64 farms

Conclusion for land cover classification of maize that showed the

accuracy of 92%, as compared with the area estimated by
Accurate and timely yield forecasting is becoming more crop reporting service (CRS) Punjab-Pakistan. Yield fore-
important in decision making. In this study remote sensing casting results of crop model were precise with RMSE of
and crop model were used for yield forecasting of maize. 255 kg ha-1 then the remote sensing, having RMSE of
CERES-Maize model was calibrated and evaluated using 397 kg ha-1. But overall strength of relationship between
field experimental data, which showed the mean absolute predicted and actual grain yield was good with R2 of 0.94
percent error ranged from 0.35 to 6.71 for recorded in both techniques. So, it can be concluded for yield
parameters. While in Remote sensing, machine learning forecasting remote sensing could be used due to greater
algorithm were used for image classification that showed advantages of less input dataset and if focus is to assess
the accuracy greater than 97%. The best classifier was used specific stress and interaction of plant genetics to soil and

123
1710 Journal of the Indian Society of Remote Sensing (October 2018) 46(10):1701–1711

environmental conditions than crop model is very useful farming systems in tropical sub-humid West Africa and semi-
tool. arid Brazil. Agriculture, Ecosystems and Environment, 135,
318–327. https://doi.org/10.1016/j.agee.2009.10.014.
Genovese, G. P., Fritz, S., & Bettio, M. (2006). A comparison and
evaluation of performances among crop yield forecasting models
based on remote sensing: Results from the geoland observatory
References of food monitoring. International Archives of the Photogram-
metry, Remote Sensing and Spatial Information Sciences, 36,
Ahmad, A., Ashfaq, M., Rasul, G., Wajid, S. A., Khaliq, T., & Rasul, 71–77.
F., et al. (2015). Impact of climate change on the rice–wheat Jagtap, S. S., & Jones, J. W. (2002). Adaptation and evaluation of the
cropping system of Pakistan. In C. Rosenzweig, & D. Hillel CROPGRO-soybean model to predict regional yield and pro-
(Eds.), Handbook of climate change and agroecosystems: The duction. Agriculture, Ecosystems and Environment, 93(1),
agricultural model intercomparison and improvement project 73–85.
(AgMIP). ICP Series on climate change impacts, adaptation, and Johnson, D. M. (2014). An assessment of pre-and within-season
mitigation, (Vol 3(Part 2), pp. 219–258). remotely sensed variables for forecasting corn and soybean
Algur, S. P., Goudannavar, B. A., & Bhat, P. (2002). Classification yields in the United States. Remote Sensing of Environment, 141,
and analysis of web multimedia data using. La Crau: Principal 116–128.
Component Analysis. Johnson, B., Tateishi, R., & Xie, Z. (2012). Using geographically
Allen, R. G., Pereira, L. S., Raes, D., & Smith, M. (1998). Crop weighted variables for image classification. Remote Sensing
evapotranspiration-Guidelines for computing crop water require- Letters, 3(6), 491–499.
ments-FAO Irrigation and drainage paper 56. FAO, Rome, Jones, C. A., Kiniry, J. R., & Dyke, P. T. (1986). CERES-Maize: A
300(9), D05109. simulation model of maize growth and development. College
Anbazhagan, S., & Paramasivam, C. R. (2016). Statistical correlation Station: Texas A&M University Press.
between land surface temperature (LST) and Vegetation Index Lee, B.-H., Kenkel, P., & Brorsen, B. W. (2013). Pre-harvest
(NDVI) using multi-temporal landsat TM data. International forecasting of county wheat yield and wheat quality using
Journal of Advanced Earth Science and Engineering, 5(1), 333. weather information. Agricultural and Forest Meteorology, 168,
Batchelor, W. D., Basso, B., & Paz, J. O. (2002). Examples of 26–35.
strategies to analyze spatial and temporal yield variability using Lin, Y., Wu, W., & Ge, Q. (2015). CERES-Maize model-based
crop models. European Journal of Agronomy, 18(1–2), 141–158. simulation of climate change impacts on maize yields and
Boote, K. J., Jones, J. W., & Pickering, N. B. (1996). Potential uses potential adaptive measures in Heilongjiang Province, China.
and limitations of crop models. Agronomy Journal, 88(5), Journal of the Science of Food and Agriculture, 95(14),
704–716. 2838–2849. https://doi.org/10.1002/jsfa.7024.
Briscoe, J., & Qamar, U. (2006). Pakistan’s water economy: Running Liu, H., Yang, J., Ping, H., Bai, Y., Jin, J., Drury, C. F., et al. (2012).
dry. Oxford: University Press Karachi. Optimizing parameters of CSM-CERES-Maize model to
Chisanga, C. B., Phiri, E., Shepande, C., & Sichingabula, H. (2015). improve simulation performance of maize growth and nitrogen
Evaluating CERES-Maize model using planting dates and uptake in northeast China. Journal of Integrative Agriculture,
nitrogen fertilizer in Zambia. Journal of Agricultural Science, 11(11), 1898–1913.
7(3), 79. Mkhabela, M. S., & Mkhabela, M. S. (2000). Exploring the
Dorosh, P., & Salam, A. (2008). Wheat markets and price stabilisa- possibilities of using noaa, vhrr data to forecast cotton yield in
tion in Pakistan: An analysis of policy options. The Pakistan swaziland. UNISWA Journal of Agriculture, 9(1), 13–21.
Development Review, 47, 71–87. Moen, T. N., Kaiser, H. M., & Riha, S. J. (1994). Regional yield
Dumont, B., Basso, B., Leemans, V., Bodson, B., Destain, J.-P., & estimation using a crop simulation model: Concepts, methods,
Destain, M.-F. (2015). A comparison of within-season yield and validation. Agricultural Systems, 46(1), 79–92.
prediction algorithms based on crop model behaviour analysis. Mubeen, M., Ahmad, A., Wajid, A., Khaliq, T., Hammad, H. M.,
Agricultural and Forest Meteorology, 204, 10–21. Sultana, S. R., et al. (2016). Application of CSM-CERES-Maize
Erda, L., Wei, X., Hui, J., Yinlong, X., Yue, L., Liping, B., et al. model in optimizing irrigated conditions. Outlook on Agricul-
(2005). Climate change impacts on crop yield and quality with ture, 45(3), 173–184.
CO2 fertilization in China. Philosophical Transactions of the Orhan, O., Ekercin, S., & Dadaser-Celik, F. (2014). Use of landsat
Royal Society B: Biological Sciences, 360(1463), 2149–2154. land surface temperature and vegetation indices for monitoring
Fontana, D. C., Weber, E., Ducati, J., Figueiredo, D. C., Berlato, M. drought in the Salt Lake Basin Area, Turkey. The Scientific
A., & Bergamaschi, H. (2000). Monitoramento e previsão de World Journal. https://doi.org/10.1155/2014/142939.
safras no Brasil. In Simposio lationamericano de percepcion Osborne, J. W., & Waters, E. (2002). Multiple regression assump-
remota sistemas de informacion espacial (Selper) (Vol. 9, tions. Washington, DC: ERIC Digest.
pp. 87–96). SELPER Puerto Iguazú. Panda, S. S., Ames, D. P., & Panigrahi, S. (2010). Application of
Fraisse, C. W., Sudduth, K. A., & Kitchen, N. R. (2001). Calibration vegetation indices for agricultural crop yield prediction using
of the CERES–MAIZE model for simulating site–specific crop neural network techniques. Remote Sensing, 2(3), 673–696.
development and yield on claypan soils. Applied Engineering in Ramachandra, T. V., Kumar, U., & Dasgupta, A. (2016). Time-series
Agriculture, 17(4), 547. MODIS NDVI based vegetation change analysis with land
Friedl, M. A., McIver, D. K., Hodges, J. C. F., Zhang, X. Y., surface temperature and rainfall in Western Ghats, India (Vol.
Muchoney, D., Strahler, A. H., et al. (2002). Global land cover 100, pp. 50–100). ENVIS Technical Report.
mapping from MODIS: Algorithms and early results. Remote Saseendran, S. A., Ma, L., Nielsen, D. C., Vigil, M. F., & Ahuja, L. R.
Sensing of Environment, 83(1), 287–302. (2005). Simulating planting date effects on corn production
Gaiser, T., Barros, I. De, Sereke, F., & Lange, F. (2010). Agriculture, using RZWQM and CERES-Maize models. Agronomy Journal,
Ecosystems and Environment Validation and reliability of the 97(1), 58–71.
EPIC model to simulate maize production in small-holder

123
Journal of the Indian Society of Remote Sensing (October 2018) 46(10):1701–1711 1711

Schafer, A., & Victor, D. G. (2000). The future mobility of the world Xiong, W., Matthews, R., Holman, I., Lin, E., & Xu, Y. (2007).
population. Transportation Research Part A: Policy and Prac- Modelling China’s potential maize production at regional scale
tice, 34(3), 171–205. under climate change. Climatic Change, 85(3), 433–451.
Singh, R. P., Roy, S., & Kogan, F. (2003). Vegetation and Yun, J. I. (2003). Predicting regional rice production in South Korea
temperature condition indices from NOAA AVHRR data for using spatial data and crop-growth modeling. Agricultural
drought monitoring over India. International Journal of Remote Systems, 77(1), 23–38.
Sensing, 24(22), 4393–4402. Zhang, F., Wu, B. F., Liu, C. L., Luo, Z. M., Zhang, S. W., & Zhang,
Van Wart, J., Kersebaum, K. C., Peng, S., Milner, M., & Cassman, K. G. L. (2004). A method for extract regional crop growth profile
G. (2013). Estimating crop yield potential at regional to national with time series of NDVI Data. Journal of Remote Sensing, 8(6),
scales. Field Crops Research, 143, 34–43. 515–528.

123