Journal of Geomatics Vol. 17, No.

2, October 2023

Development of Machine Learning based Models for Multivariate Prediction of Wheat Crop
Yield in Uttar Pradesh, India
Sukirti1. Kamal Pandey2*, Abhishek Danodia3, . Harish Chandra Karnatak2
1Indian Institute of Remote Sensing (IIRS), Dehradun
2Department of Geoweb Services, IT and Distance Learning, IIRS, ISRO, Dehradun
3 Agriculture and Soils Department, IIRS, Dehradun

*[email protected]

(Received: 21 April 2023; in final form 27 November 2023)

DOI: https://doi.org/10.58825/jog.2023.17.2.70

Abstract: The consequences of climate change have a substantial impact on agricultural crop production and
management. Predicting or forecasting crop yields well in advance would help farmers, agriculture corporations and
government agencies manage risk and design suitable crop insurance plans. Ground survey is the traditional way of
determining yield, which is subjective, time-consuming, and expensive. While Machine Learning (ML) techniques make
yield prediction less expensive, less time taking and more efficient. In this study, thirteen years of meteorological
parameters and wheat yield data (2001-2013) of Uttar Pradesh were used to train and analyze three Machine Learning
Regression models viz. Support Vector Regression, Ordinary Least Squares, and Random Forest. Each model's
performance was assessed using Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error
(RMSE). Results revealed that the Random Forest model with a MAE of 0.258 t/ha, MSE of 0.096 t/ha and RMSE of
0.311 t/ha proved to be the best model in the yield prediction of wheat when results are statistically compared with others.
Researchers and decision-makers can use the findings to estimate pre-harvest yields and to ensure food security.

Keywords: Meteorological parameters, Wheat, Multivariate, Yield prediction, Machine Learning, Random Forest.

1. Introduction (Arora 2019). Rising temperature is also leading to gradual

decline in annual crop yield. Crop failures are more likely
India has a long history of agriculture (Madhusudan, 2015) to occur in the short term as a result of altered precipitation
dating back to the Indus Valley Civilization. Agriculture patterns, and diminished crop productivity \(Poudel and
sector is very important for India's socioeconomic Shaw 2016).
structure as it is one of the largest economic sectors in
terms of population diversity. The inability to integrate Until recently, yield prediction was being done by
technology to provide the intended results is one of the considering the farmer's knowledge of a particular area and
biggest challenges Indian agriculture is facing (Nigam et. crop. Empirical and crop growth models (Jørgensen, 1994)
al., 2019). Due to the uneven climatic trends brought on were created using meteorological data to for the yield
due to the negative impacts of global warming, which has prediction, but these techniques also have a lot of problems
affected crop production as well, farmers find it owing to the spatial distribution of the weather stations.
challenging to precisely predict temperature and rainfall The outputs from these operations are accessible only after
patterns for their agricultural practices. Farmers play the crops have been harvested,. Therefore, it is essential to
important role to ensure food security for the human decrease expenses and longer waits involved in the
civilization. Food security is provided through agriculture, traditional practices while also improving yield prediction
and is crucial for a nation's economic growth (Barman, accuracy. For this purpose, machine learning techniques
2020). Since the launch of Earth Observation (EO) provide an alternate option with improved and proven
satellites and the explorer missions that followed, satellite performance matrix. (Nigam et. al., 2019; Jaafar and
remote sensing has enabled us gather remote sensing data Mourad 2021; Cunha et. al., 2018; Pantazi et. al., 2016).
at improving scales (Karthikeyan et. al., 2020; Shetty et.
al., 2021; Virnodkar et. al., 2020; Feizizadeh et. al., 2023; Researchers across the globe have attempted and
Jamali 2019; Chatziantoniou et. al., 2017; Ienco et. al., developed machine learning techniques for crop yield
2019. forecasting. Preseason forecasting was a technique
developed that used a machine learning-based system to
Global temperature is rising as a result of the continuing predict soybean yields before the start of the crop season
rise in greenhouse gas emissions. Some effects of climate (Cunha et. al., 2018). The system was based on a recurrent
change include shifting seasons, melting glaciers, neural network (RNN) that was trained using historical
increasing precipitation, rise in extreme weather data on municipal-level soybean and/or maize yields as
occurrences, and so on. World population and well as parameters such as precipitation, temperature, and
socioeconomic growth, along with climate change, soil conditions. A reanalysis-based seasonal forecast
represent a danger to food security. An extremely high rate product of temperature and precipitation, which enabled
of land degradation brought on by climate change is predicting up to seven months in advance, providing the
creating accelerated desertification and nutrient-deficient meteorological data operationally (Cunha et. al., 2018).
soils. The problem of land degradation is said to be a The findings were on par with, and in some cases better
serious global threat and is getting worse day by day than, similar models that are limited to early season

© Indian Society of Geomatics

211
Journal of Geomatics Vol. 17, No. 2, October 2023

forecasting due to the need for remote sensing data across best one for multivariate analysis of yield prediction. The
the farm. This study concentrated on the quantification of study aims to promote wide spread use of ML models in
machine learning algorithms and their practical decision making in a farming sectors for the countries like
application. The method mentioned also considered the India where agriculture has a majoe share in economy.
unpredictable rainfall and temperature in order to obtain a
steady trend. On the basis of mean absolute error, the 2. Materials and Methods
results of different algorithms were compared. The most
accurate regressor for predicting yield was found to be the 2.1 Study Area
Random Forest Regressor. A sequential model called Uttar Pradesh (UP) is the fourth largest state of India with
Simple Recurrent Neural Network performed better at an area of 240,928 km2. UP is the most populous state of
forecasting rainfall than the LSTM for forecasting the India. It is located between latitude 24º to 31º North and
temperature (Nigam et. al., 2019). longitude 77º to 84º East (Figure 1).

A technique to anticipate agricultural yield based on

historical data was designed and implemented by Bondre
et al (2019) Using agricultural data, machine learning
methods like Support Vector Machine and Random Forest
were used to determine the recommended fertiliser for a
number of crops (Bondre and Mahagaonkar 2019). It
concentrated on developing a prediction model that may
be applied to crop yield forecasting in the future. Different
sources were used to collect a variety of datasets, including
those for crops, crop yield, geography, soil and crop
nutrients, and fertiliser.

Haque et al (2020) suggested two distinct Machine

Learning (ML) algorithms in their study to examine the
crop yield (Haque et.al., 2020). With 140 data points
collected, the Support Vector Regression (SVR) and
Linear Regression (LR) algorithms proved to be quite
effective for evaluating the performance of the parameters
involved in the forecast. The Mean Square Error (MSE)
and Coefficient of Determination were used to calculate
the error rate (R2) and the same dataset was used to directly
compare the results obtained from the ML models. Pantazi
et al (2016) attempted a study wherein online multi-layer
soil data and crop growth parameters from satellite Figure 1. Map of Study Area (Uttar Pradesh)
imagery were used to forecast within-field variation in
wheat yield. An unsupervised learning system was There are 18 divisions and 75 districts in Uttar Pradesh.
combined with supervised self-organizing maps that could Due to the predominately agrarian economy, Uttar Pradesh
handle data from numerous soil and crop sensors. The is prominently dependent on the performance of industries
effectiveness of XY-fused Networks (XY-Fs), Supervised like agriculture, horticulture, animal husbandry, dairy,
Kohonen Networks (SKNs), and Counter-Propagation fishery, etc. With four distinct seasons, Uttar Pradesh has
Artificial Neural Networks (CP-ANNs) for predicting a humid subtropical climate. The state receives an average
wheat production in a field in Bedfordshire, UK, was of 650 mm of rainfall per year in the southwest and 1000
compared during a single cropping season (Pantazi et. al., mm in the eastern and south eastern regions (Nihar et.al.,
2016). 2022).

In the field of crop yield analysis, machine learning (ML) Approximately 47% of the population is directly
is a new topic of research. By providing inputs for growing dependent on agriculture for their livelihood, and climate
the best possible crop and anticipating the yields, ML has is the primary factor affecting the production. Given the
the potential to transform agriculture (Nigam et. al., 2019) size of the state's geographic area and the access to the
to new dimension. In order to help farmers for selecting a fertile Indo-Gangetic plains, UP makes a considerable
suitable crop for cultivation to obtain maximum yield, ML contribution to the country's food security. The state
model takes into consideration parameters like produces about 12% of India's rice and 28% of it is wheat.
temperature, rainfall, area, etc. The ML based approaches Additionally, a significant amount of sugarcane is
has the potential to improve the expanding agricultural produced, making up 44% of the nation's total output
industry of countries like India and, taken together, raise (Gulati et.al., 2021). The major crops of the state are rice,
the living standard of farmers. The objectives of the wheat, maize, sugarcane, chickpea and pigeon pea. About
presented study are in line with the above argument viz. (i) 24% of the state's agricultural area is used to grow wheat
To develop machine learning models to predict yield of throughout the state.. On an average the total geographical
wheat for the state of Uttar Pradesh, and (ii) To compare area under wheat crop is 9730.60 ha, total production is
the performance of different models and to find out the 32799.71 tons and total yield is 3371 kg/ha.

212
Journal of Geomatics Vol. 17, No. 2, October 2023

2.2 Data Used

Wheat yield was predicted using net sown area, and In cases where the dataset is less complicated and tuning
meteorological variables such as mean rainfall, mean of hyperparameters is not required , only the training and
temperature, mean relative humidity, mean wind speed, test sets can be used. Accordingly, the data array was
mean wind direction, and surface net solar radiation (for divided into two subsets, one for training and the other for
Rabi Season). The Indian government website testing, using the Sklearn library's data splitting function
(data.gov.in) provided historical district-level wheat yield known as train test split. By default, it divides the two
data (2001-2013). The Climate Change Service subsets into random divisions. In this study, 80% of the
(https://climate.copernicus.eu/), a website run by the data were used for training, with the remaining data being
ECMWF as part of The Copernicus Program, was used to utilised to test the algorithm's performance..
collect all the meteorological data for the time period of
2001-2013. The datasets were gridded with hourly 2.4 Machine Learning Algorithms
temporal resolution and a horizontal resolution of 0.5° x The models were built using long-term datasets for the
0.5°. The ECMWF Copernicus services provide ever- years from 2001 to 2013. The presented research work is
changing datasets by continuously monitoring the climate focused on developing yield estimation models for wheat
and atmospheric composition and using climate and using the three ML models viz. Random Forest (RF),
atmosphere reanalysis.. Ordinary Least Squares (OLS), and Support Vector
Regression (SVR),.. The performance of these models
2.3 Data Pre-processing were further compared to select the best performing model.
The crop production data that were downloaded from
data.gov.in were district- and season-specific, and they 2.4.1 Random Forest
date back to 1997. Required yield data of wheat crop of Random Forest is an ensemble based ML model wherein
Uttar Pradesh from 2001-2013 was extracted from the multiple decision trees are combined for producing
downloaded dataset. The collected data was organized in a aggregated output.. Ensemble learning is the process of
Comma Delimited (CSV) format before being used with combining the predictions or classifications of different
other meteorological parameters to form the dataset used models that have all been trained on the same data. The
for this study. underlying principle of ensemble learning is that each
model's errors—in this case, the errors of a decision tree—
2.3.1 Principal Component Analysis are separate and unrelated to one another. A Random
In order to study, model, and explain complex multivariate Forest model's forecast is produced by averaging the
real-world systems, various statistical methods are predictions of various decision trees and hence there isn't
available and Principal Component Analysis (PCA) is one a single equation that summarizes the RF model.
of the widely used method to study the collinearity among Similar study has been done in which Random Forests and
the parameters . It is a useful tool for conducting data Multiple Linear Regressions were compared, for crop
analysis on huge data sets and for identifying the main production prediction, for wheat, maize, and potato at
trends and the variables that influence them. It can quickly global and regional scales in response to meteorological
determine which primary components contribute the most and biophysical variables. They trained the RF models to
to data variability. Instead of performing the whole process predict crop yield using a variety of biophysical predictors
of attempting to estimate which of the variables might be (Jeong et. al., 2016).
of the most relevance, effort might instead be spent on
trying to find and comprehend these few most critical 2.4.2 Ordinary Least Squares
components (Bloomer and Rehm 2014). In this research Ordinary Least Square is one of the popular regression
study, PCA was used to identify and address model in ML. Building a model that lowers the overall
multicollinearity issues. squared deviations between the expected and actual data is
an approach used to estimate unknown parameters. OLS
2.3.2 Data Splitting tries to reduce this sum (Brinkhoff and Robson 2021). It is
The primary requirement of any ML model is data a linear regression method that aims to minimize the sum
splitting, and the available dataset must be divided into of squared residuals between the predicted values and the
three parts: actual values. The equation (1) for OLS is:
a. Using the training set, the model is trained or fitted. y = w0 + w1x1 + w2x2 + ... + wnxn + ε (1)
The optimal weights or coefficients for ML models Here, y is the predicted output, x1, x2,..., xn are the input
can be found using the training set. variables, w0, w1, w2,..., wn are the coefficients (weights)
b. The validation set is utilised to improve the model to be learned, and ε represents the error term. The study
performance by adjusting the the hyperparameters carried out by Sharma and team (2013) utilized OLS for
used in the model. For each potential setting of the the yield prediction of rainfed maize and soyabeans crops
hyperparameters, the model is fitted using the and reported the outputs with high accuracy. (Sharma et.
training set, and the performance is assessed using al., 2013).
the validation set.
c. The test set is necessary for a fair assessment of the
final model..

213
Journal of Geomatics Vol. 17, No. 2, October 2023

Figure 2. Flowchart of Methodology

Figure 3. Correlation Matrix of Independent Variables

214
Journal of Geomatics Vol. 17, No. 2, October 2023

2.4.3 Support Vector Regression R2 value associated with the SVR model indicated a
A well-known machine learning approach called Support weaker correlation and less accurate predictions. These
Vector Machine (SVM) is widely utilized in both findings highlight the superior performance of the RF
classification and regression. The Support Vector model in predicting crop yield, while underscoring the
Regression (SVR) aims to fit the best line within a limitations of SVR in this particular context.
threshold value, in contrast to other regression models that
aim to minimize the difference between the actual and
predicted value. Equation (2) is the generic representation
for the line in support vector regression:
y= wx+b (2)

SVR has been used in similar studies, a study in Andhra

Pradesh, India which concentrated on the forecasting of
important Kharif crops employed modular artificial neural
networks to first anticipate the quantity of monsoon
rainfall before applying support vector regression and
rainfall data to calculate the potential yield of the main
kharif crops. They concluded that the proposed strategy
outperformed earlier machine learning algorithms in
estimating kharif crop production after comparing their
study with other ML methods (Khosla et.al., 2020).

3. Results and Discussion

Figure 4. Importance of Predictor Variables
3.1 Correlation Analysis
Mean Squared Error (MSE), and Root Mean Squared Error
Collinearity check among the participating parameter is
(RMSE) of yield forecasts were estimated for the ML
carried out before developing a predictive model in order
models and out of all three models, RF showed the best
to reduce the redundancy. If the correlation among
predictive capability for wheat yield (R2 = 0.72).
different variables is high, it gives rise to multicollinearity
problems and estimates are unstable. Therefore, to
examine the co-linearity between the independent
3.3 Discussion
variables and prevent the multicollinearity issue, principal
component analysis (PCA) is used in the present study. It
The ability of these ML models (RF, OLS, and SVR) to
was found that there was no significant correlation
predict the yield of wheat in Uttar Pradesh was tested using
between the variables as shown in Figure 3..
Mean Absolute Error (MAE), Mean Squared Error (MSE),
and Root Mean Squared Error (RMSE). Results for the ML
3.2 Importance of Predictor Variables
models are summarized in Table 1. Random forest
The significance of these predictor variables was
successfully predicted the wheat yield and was compared
examined using the Random Forest model's
against the test data, which was not used in training the
characteristics. These factors were ranked according to
model. Comparison between the observed and predicted
their importance, and it was observed that each variable
yield (Figure 5a) produced an R2 of 0.72. The model was
has its importance for yield estimation differently (Figure
able to explain with MAE value of 0.258 t/ha, MSE value
4). Surface net solar radiation is observed to be the most
of 0.096 t/ha and RMSE value of 0.311 t/ha.
crucial factor in the yield estimation, followed by mean
temperature, mean wind speed, and area. Since wheat is a
The method we followed in this study was found similar to
rabi (winter) crop, temperature and net solar radiation both
the method adopted by Schwalbert et al 2020. They found
have a significant impact on the crop's ability to grow.
that combining meteorological information with multi-
Strong winds can lead to lodging of the crop and hence
temporal satellite imagery could produce useful data, thus
mean wind speed also plays very important role in for the
enabling the creation of more precise yield forecast
overall crop yield.
models, which is consistent with our work. Using a
regression model between historical meteorological data
3.3 Model Comparison
and yield data for food crops, the effects of climate change
The density scatter plots for each algorithm are shown in
were investigated in the mountainous regions of Nepal.
Figures 5. Among the models, RF demonstrated the
They came to the conclusion that different crops' yield is
highest predictive accuracy, as evidenced by its scatter plot
affected differently by climate variables. However, the
exhibiting the strongest alignment between observed and
current climate trends have a negative impact on winter
predicted yield. The data points are closely clustered
crops (Poudel and Shaw 2016).
around a central line, with only a few outliers indicating a
robust relationship between the two variables. This was
further supported by the highest coefficient of
determination (R2) achieved by the RF model. On the other
hand, SVR performed relatively poorly as compared to the
other models and as evident in its scatter plot. The lower

215
Journal of Geomatics Vol. 17, No. 2, October 2023

machine learning techniques like Support Vector Machine

and Random Forest and obtained the accuracy of 99.47%
and 97.48% for . SVM and RF respectively. Therefore,
SVM method was found to be good for agricultural yield
prediction as opposed to our result. This might be due to
the difference in predictor variables.

4. Conclusion

. With the help of machine learning algorithms and

publicly available data, this study established a
methodology for estimating winter wheat production on a
regional scale that might possibly be used to estimate crop
yield globally and in regions with low frequency of
observed data. By merging crop models with additional
data for crop yield estimation, forecasting, and catastrophe
monitoring in vast areas, the framework can be further
enhanced. In order to estimate winter wheat yield from
2001 to 2013 along with 07 parameter of winter season viz.
mean temperature, mean relative humidity, mean pressure,
mean wind speed, mean wind direction, mean precipitation
and mean solar radiation, three ML models (Support
Vector Regression, Ordinary Least Squares, and Random
Forest) were used in this study. Using three error metrics—
Mean Absolute Error (MAE), Mean Squared Error (MSE),
and Root Mean Squared Error (RMSE)—accuracy was
compared to assess each model's performance. Overall, all
three models performed well but RF model performed the
best. With MAE values of 0.258,

MSE values of 0.096, and RMSE values of 0.311, it was

found that RF outperformed SVR and OLS. The poorest
result was obtained by SVR. This study has the potential
to open up new possibilities for productive research on
agricultural production prediction utilizing a variety of
meteorological factors. It will encourage and promote
research into the use of machine learning techniques for
yield forecasting.

Acknowledgements:
Authors express their sincere thanks to Copernicus
Figure 5. Scatter plot of Predicted yield against (https://scihub.copernicus.eu/) for providing satellite data
Observed Yield for (top) Random Forest, (middle) at no cost. The authors are also thankful to the Indian
Ordinary Least Squares. And (bottom) Support Vector Institute of Remote Sensing for providing the laboratory
Regression facilities for the completion this work.
Table 1. Prediction Performance of Different Competing Interests: We, the authors declare that we
Algorithms have no competing interests.
Algorithm MAE MSE RMSE
(t/ha) (t/ha) (t/ha) References
Random Forest 0.258 0.096 0.311 Arora N.K. (2019), “Impact of climate change on
Support Vector 0.439 0.304 0.552 agriculture production and its sustainable solutions,”
Regression Environ. Sustain., vol. 2, no. 2, pp. 95–96, doi:
Ordinary Least 0.312 0.156 0.395 10.1007/s42398-019-00078-w.
Squares
Barman S. (2020), “The Political Economy of Food
Security in India: Evolution and Performance,
Cabas et al 2010 concluded in their study that the non-
International Journal of management, 11(12), pp 1156-
climatic variables had a relatively minor impact on the
1162”
yield distribution, indicating that climatic factors should
dominate this relationship (Cabas et. al., 2010). A system Bloomer C. and G. Rehm (2014), “Using principal
to forecast agricultural yield based on historical data has component analysis to find correlations and patterns at
also been proposed by Bondre and Mahagaonkar (Bondre diamond light source,” IPAC 2014 Proc. 5th Int. Part.
and Mahagaonkar 2019). They used agriculture data and Accel. Conf., pp. 3719–3721

216
Journal of Geomatics Vol. 17, No. 2, October 2023

Bondre D. and S. Mahagaonkar (2019), “Prediction of crop Jørgensen S.E. (1994). “Models as instruments for
yield and fertilizer recommendation using machine combination of ecological theory and environmental
learning algorithms,” International Journal of Engineering, practice,” Ecol. Modell., vol. 75–76, no. C, pp. 5–20, doi:
Applied Science and technology vol. 4(5), pp. 371–376 10.1016/0304-3800(94)90003-5.
Brinkhoff J. and A. J. Robson (2021). “Block-level Karthikeyan L., I. Chawla and A.K. Mishra (2020). “A
macadamia yield forecasting using spatio-temporal review of remote sensing applications in agriculture for
datasets,” Agric. For. Meteorol., vol. 303, doi: food security: Crop growth and yield, irrigation, and crop
10.1016/j.agrformet.2021.108369. losses,” J. Hydrol., vol. 586, 2020, doi:
10.1016/j.jhydrol.124905.
Cabas J., A. Weersink and E. Olale (2010). “Crop yield
response to economic, site and climatic variables,” Clim. Khosla E., R. Dharavath and R. Priya (2020). “Crop yield
Change, vol. 101, no. 3, pp. 599–616, 2010, doi: prediction using aggregated rainfall-based modular
10.1007/s10584-009-9754-4. artificial neural networks and support vector regression,”
Environ. Dev. Sustain., vol. 22, no. 6, pp. 5687–5708, doi:
Chatziantoniou A., G. P. Petropoulos and E. Psomiadis
10.1007/s10668-019-00445-x.
(2017). “Co-Orbital Sentinel 1 and 2 for LULC mapping
with emphasis on wetlands in a mediterranean setting Madhusudan L. (2015). “Agriculture role on
based on machine learning,” Remote Sens., vol. 9, no. 12, Indian..Economy. - Business and economics journal, vol
6, no. 4
Cunha R.L.F, B. Silva and M. A. S. Netto (2018). “A
scalable machine learning system for pre-season Nigam A., S. Garg, A. Agrawal and P. Agrawal (2019)
agriculture yield forecast,” Proc. - IEEE 14th Int. Conf. “Crop yield prediction using machine learning
eScience, e-Science pp. 423–430, doi: algorithms,” Fifth International Conference of Image
10.1109/eScience.2018.00131. Information Processing. Shimla, 125-130
Feizizadeh B., D. Omarzadeh, M. Kazemi Garajeh, T. Nihar A., N. R. Patel and A. Danodia, (2022). “Machine-
Lakes and T. Blaschke (2023). “Machine learning data- Learning-Based Regional Yield Forecasting for Sugarcane
driven approaches for land use/cover mapping and trend Crop in Uttar Pradesh, India,” J. Indian Soc. Remote Sens.,
analysis using Google Earth Engine,” J. Environ. Plan. vol. 50, no. 8, pp. 1519–1530, doi: 10.1007/S12524-022-
Manag., vol. 66, no. 3, pp. 665–697, doi: 01549-0.
10.1080/09640568.2021.2001317.
Pantazi X.E., D. Moshou, T. Alexandridis, R. L. Whetton
Gulati A., P. Terway and S. Hussain (2021) “Performance and A. M. Mouazen (2016). “Wheat yield prediction using
of agriculture in Uttar Pradesh”, In: A. Gulati, R. Roy and machine learning and advanced sensing techniques,”
S. Saini (eds) Revitalizing Indian Agriculture and Comput. Electron. Agric., vol. 121, pp. 57–65, doi:
Boosting Farmer Income. Indian Studies in Business and 10.1016/j.compag.2015.11.018.
Economics, Springer
Poudel S. and R. Shaw (2016). “The relationships between
Haque F.F., A. Abdelgawad, V. P. Yanambaka and K. climate variability and crop yield in a mountainous
Yelamarthi (2020). “Crop Yield Analysis Using Machine environment: A case study in Lamjung District, Nepal,”
Learning Algorithms,” IEEE World Forum Internet Climate, vol. 4, no. 1, doi: 10.3390/cli4010013.
Things, WF-IoT 2020- Symp. Proc., doi: 10.1109/WF-
Schwalbert R.A., T. Amado, G. Corassa, L.P. Pott, P.V.V.
IoT48130.2020.9221459.
Prasad and I.A. Ciampitti (2020). “Satellite-based soybean
Ienco D., R. Interdonato, R. Gaetano and D. Ho Tong yield forecast: Integrating machine learning and weather
Minh (2019). “Combining Sentinel-1 and Sentinel-2 data for improving crop yield prediction in southern
Satellite Image Time Series for land cover mapping via a Brazil,” Agric. For. Meteorol., vol. 284, doi:
multi-source deep learning architecture,” ISPRS J. 10.1016/j.agrformet.2019.107886.
Photogramm. Remote Sens., vol. 158, pp. 11–22, doi:
Sharma V., D. R. Rudnick and S. Irmak, (2013).
10.1016/j.isprsjprs.2019.09.016.
“Development and evaluation of ordinary least squares
Jaafar H. and R. Mourad (2021). “Gymee: A global field- regression models for predicting irrigated and rainfed
scale crop yield and et mapper in google earth engine maize and soybean yields,” Trans. ASABE, vol. 56, no. 4,
based on landsat, weather, and soil data,” Remote Sens., pp. 1361–1378, doi: 10.13031/trans.56.9973.
vol. 13, no. 4, pp. 1–30, doi: 10.3390/rs13040773.
Shetty S., P.K. Gupta, M. Belgiu and S. K. Srivastav
Jamali A. (2019). “Evaluation and comparison of eight (2021). “Assessing the effect of training sampling design
machine learning models in land use/land cover mapping on the performance of machine learning classifiers for land
using Landsat 8 OLI: a case study of the northern region cover mapping using multi-temporal remote sensing data
of Iran,” SN Appl. Sci., vol. 1, no. 11, doi: 10.1007/s42452- and google earth engine,” Remote Sens., vol. 13, no. 8,
019-1527-8. doi: 10.3390/rs13081433.
Jeong J.H, J.P. Resop, N.D. Mueller, D.H. Eleisher, K. Virnodkar S.S., V. K. Pachghare, V. C. Patil and S. K. Jha
Yun, E.E. Butler, D.J. Timlin, K. Shim, J.S. Gerber, V.R. (2020). “Remote sensing and machine learning for crop
Reddy and S. Kim (2016). “Random forests for global and water stress determination in various crops: a critical
regional crop yield predictions,” PLoS One, vol. 11(6), review,” Precis. Agric., vol. 21, no. 5, pp. 1121–1155, doi:
doi: 10.1371/journal.pone.0156571. 10.1007/s11119-020-09711-9.

217