Emerging Trends in Machine Learning to Predict Crop Yield and Study Its Influential Factors: A Survey

Abstract

Agriculture is one of the most crucial field contributing to the development of any nation. It not only affects the economy of nation but also has impact on the world food grain statistics. For agriculturist obtaining sustainable production of crop is always a challenge. Achieving optimum crop yield has always been a challenge for the farmer due to ever changing environmental conditions. The major reasons for unpredictability of crop yield are: land types, availability of resources, and changing nature of weather. Thus, the scientists all over the world are trying to discover techniques which can efficiently and accurately estimate the crop yield in much advance so that the farmers can take suitable actions to meet the future challenges. The main objectives of the study include: (a) Exploration of various machine learning techniques used in crop yield prediction; (b) Assessment of advanced techniques like deep learning in yield estimations; and (c) To explore the efficiency of hybridized models formed by the combination of more than one technique. The reviews done have shown good inclination towards hybrid models and deep learning techniques as means of crop yield prediction. The study also reviewed the works done by researchers in assessing the influence of various factors on crop yields and temperature and precipitation have been found to have maximum influence on the yields of different crops. Apart from climatic factors, agronomic practices adopted by farmers at various stages of growth of a plant also have considerable influence of the final yield of crop.

Full text

Vol.:(0123456789)
1 3
Archives of Computational Methods in Engineering
https://doi.org/10.1007/s11831-021-09569-8
SURVEY ARTICLE
Emerging Trends inMachine Learning toPredict Crop Yield andStudy
Its Influential Factors: ASurvey
NishuBali1· AnshuSingla1
Received: 20 August 2020 / Accepted: 9 March 2021
© CIMNE, Barcelona, Spain 2021
Abstract
Agriculture is one of the most crucial field contributing to the development of any nation. It not only affects the economy
of nation but also has impact on the world food grain statistics. For agriculturist obtaining sustainable production of crop
is always a challenge. Achieving optimum crop yield has always been a challenge for the farmer due to ever changing envi-
ronmental conditions. The major reasons for unpredictability of crop yield are: land types, availability of resources, and
changing nature of weather. Thus, the scientists all over the world are trying to discover techniques which can efficiently
and accurately estimate the crop yield in much advance so that the farmers can take suitable actions to meet the future chal-
lenges. The main objectives of the study include: (a) Exploration of various machine learning techniques used in crop yield
prediction; (b) Assessment of advanced techniques like deep learning in yield estimations; and (c) To explore the efficiency
of hybridized models formed by the combination of more than one technique. The reviews done have shown good inclination
towards hybrid models and deep learning techniques as means of crop yield prediction. The study also reviewed the works
done by researchers in assessing the influence of various factors on crop yields and temperature and precipitation have been
found to have maximum influence on the yields of different crops. Apart from climatic factors, agronomic practices adopted
by farmers at various stages of growth of a plant also have considerable influence of the final yield of crop.
1 Introduction
Agricultural scientists all over the world are struggling with
the problem of agriculture sustainability owing to the threats
posed by various factors like increase in the price of food
and energy, changing climates, continuous use and degrada-
tion of natural resources, alarming decrease in water avail-
ability and at the same time expected increase in popula-
tion in the coming centuries. Crop yield is one amongst the
important fields of agriculture that has attracted the attention
of scientists owing to its strong impact on the national and
international economy and to solve the problem of food scar-
city. An accurate and timely forecast of crop yield can not
only help the government of any country in taking various
strategical decisions like planning import/export, formulat-
ing future policies like cost/selling price of crops and timely
gauging the future threats but can also be a great help to a
farmer whose livelihood is totally based on the expected
yield of the crop [1].
1.1 Crop Yield Prediction: AGlobal Need
Agriculture is an important field that contributes to the
overall development of any nation. Growing population of
the world and the unexpected changes in climatic and soil
conditions is forcing global researchers to uncover meas-
ures that can increase crop yield without adversely affecting
our natural resources, called sustainable Agricultural prac-
tices. An accurate and timely crop yield prediction of cur-
rent growing season can be an important step in this direc-
tion and can contribute in framing any Agricultural policy.
Global bodies like Joint Research Centre (JRC), an initiative
by European Commission, was established for Monitoring
Agriculture Resources (MARS) in 1988. This body performs
regular crop yield forecasting and provides monthly bulle-
tins on expected yields to implement a Common Agriculture
Policy (CAP) for the entire world. Early warnings on crop
shortage or failure provides valuable information to food
insecure countries and help in global food security. The sys-
tem that generates this information uses real-time data like
* Anshu Singla
anshu.singla@chitkara.edu.in
1 Chitkara University Institute ofEngineering andTechnology,
Chitkara University, Punjab, India

N.Bali, A.Singla
1 3
weather related observations and forecasts, data obtained
from remote sensing like soil maps, crop characteristics and
administrative regions in the form the inputs to the system.
With these inputs, crop conditions are simulated. On regular
intervals new yield statistics are added [2].
1.2 Global Demand andSupply ofCrops
According to FAO (Food and Agriculture Organization), the
demand and consumption of cereals has witnessed a steep
growth in comparison with production in developing coun-
tries. There has been a continuous growth in the demand
of rice, wheat and other coarse grains from 1964 to 2030
[3]. To meet the growing demands, there had been a con-
siderable rise in imports of cereals in developing countries
from 39 million tonnes a year (1970) to 130 million tonnes
a year by 1997–1999. This rise in imports is expected to
continue and may aggravate in coming years. By year 2030,
these developing countries are expected to import 265 mil-
lion tonnes of cereals, which amounts to almost 14 percent
of their annual consumption, annually. Thus, the conditions
of global market are quite volatile and are also on a falling
trend in regard to real prices. These market conditions can
be devastating for the progress of nations that are not think-
ing of taking steps to decrease their overall dependence on
imports for the traditional crops. Thus it’s a world challenge
to change the present scenario in future and make coun-
tries more and more self-sufficient in fulfilling their food
demands [4] which in turn requires a timely and an accurate
estimate of yield of a crop.
1.3 Crop Yield Estimation Approaches
Crop yield estimation is essential but the involvement of
complex interrelated environmental factors makes its precise
measurement a very difficult and challenging task. Weather
changes influence the plant growth at various stages leading
to large intra-seasonal yield variations. Also the spatial vari-
ability of soil properties, farmer choices such as frequency of
irrigations, pest and fertilizer application, crop rotation and
land preparation practices adopted add to the complexity of
accurately measuring the yield of crops. Thus, development
of accurate and efficient crop yield forecast method requires
correct assessment of weather and soil parameters through
implementation of crop and soil management experiments.
There are two diverse approaches, as shown in Fig.1, fol-
lowed to predict the preharvest yield of crops even though
they are not mutually exclusive.
1.3.1 Crop Growth Model
Crop yield is always affected by the environmental factors.
The impact of these factors varies at different stages of crop
growth. Mathematical models can be used to represent these
diverse interactions of plant physiological processes with
environment to predict the final production or yield of a crop.
Such mathematical models are called crop process models or
crop growth models. These mathematical models use daily
crop growth simulator to estimate biomass production poten-
tial and provide an abstract view of the implementation of
dynamic behaviour of the plant’s physiological phases. The
actual data and various speculations about soil types, solar
radiations, various management practices adopted, rainfall
and temperature changes serve as an input fed through the
models of seed formation and plant growth. Mostly, the
mechanistic models are crop-specific [5] and some of the
models for specific crops are shown in Table1. In recent
studies also, quite good results are obtained by using these
semi empirical crop models [10]. Although efficient but
these kind of models usually turn out to be expensive in
terms of time and money, and are proven to be impractical
for massive applications and agricultural planning.
1.3.2 Data Driven Model
Another approach, called empirical approach, is compar-
atively more practical and easy to use than crop growth
model. In this approach, crop yield data for several years
is considered and the set of parameters most effective or
contributing to the yield variations are determined. Accept-
ing these efficacious parameters as independent and harvest
yield as dependent variables, empirical equations are formu-
lated to compute the coefficients of these parameters. These
coefficients are used to estimate the final crop yields. Every
statistical model determines one set of parameters. Such
Fig. 1 Crop yield estimation approaches
Table 1 Crop specific mechanistic models [5]
References Crop Model
Wilkerson etal. [6] Legumes SOYOGRO
Jones and Kiniry [7] Corn CERES-Maize
Porter [8] Wheat AFRCWHEAT2
Jamieson etal. [9] Wheat Sirius

Emerging Trends inMachine Learning toPredict Crop Yield andStudy Its Influential Factors:…
1 3
techniques are relatively less expensive and easy in applica-
tion and also they do not need any prior information on the
various physiological processes involved in the growth of the
plant or predefined structure of the model [5].
Thus, both the approaches have their own pros and cons
and the need of the hour is a united framework that has the
capability of modelling nonlinear relationship between soil
factors, weather conditions and biomass and the yield of
crop [5].
2 Methodology
This section presents the methodology adopted for an exten-
sive literature review on the topic. A thorough and com-
plete analysis of the domain required two steps to be taken
(a) Collection of related literature, (b) Analysis of the final
selected work. For accomplishing the first step, appropri-
ate keywords were selected for searching related conference
& journal papers from scientific databases and scientific
indexing services like IEEE explore and Google scholar.
The search keywords included words like, machine learning,
crop yield estimation, neural networks, influential factors in
crop yield. The collection consisted of more than 100 papers
which were studied and the papers having close relevance
with the domain were selected. It was found that collected
papers were having related information but can be further
categorized into separate sub domains relevant to the topic
under study. The steps taken for planned literature study are
as shown in the Fig.2.
Thus, the filtering of papers as per the categories men-
tioned above led to identification of 17 papers studying the
role of machine learning in quantifying the effect of various
environmental factors and other agronomic practices on the
final yield of various crops, 43 papers examining the role of
machine learning in the field of crop yield prediction and
15 papers studying the role of deep learning in the field.
Among the final selected papers, 28 papers belonged to Sco-
pus indexed journals and rest of the papers also belonged to
reputed journals. The final identified papers were thoroughly
studied to find answers to following questions:
1. What type of machine learning techniques have been
used in quantifying the effect of various environmental
factors on final yield for various crops?
2. How machine learning techniques have contributed to
the study of crop yield estimation?
3. What is the efficiency of neural networks in the field as
compared to other machine learning techniques?
4. Which deep learning techniques have been explored in
the field and how efficient are they in predictions?
5. What type of data sets have been used by the authors?
6. How remote sensing data has contributed to the study?
The detailed study of various sections has been summa-
rized in Tables2 and 3 in Sect.3 and Sect.4 respectively
and final results have been discussed in conclusion section.
3 Emergence ofMachine Learning
Techniques inYield Prediction
Machine learning, an established field of computer science,
has shown a promising future in different research areas. It
is largely used by the data scientists and researchers who
Fig. 2 Methodology for Systematic Review

N.Bali, A.Singla
1 3
Table 2 Existing machine learning techniques for crop yield prediction on different crop varieties
References Anshal
Savla etal.
[11]
Ahamed
etal. [32]
Lamba and
Dhaka [33]
Zhang
etal.
[16]
Gonzalez-
Sanchez etal.
[31]
Nari etal.
[37]
Bose etal.
[38]
Pantazi
etal.
[39]
Kaul etal.
[40]
Chlinga-
ryan etal.
[41]
Crop type
MT ✓
CB ✓ ✓
P✓
W✓ ✓ ✓
C✓ ✓ ✓
R✓
T✓
SB ✓ ✓ ✓ ✓
M
S
A
CP ✓
Machine learning technique used
SVM ✓ ✓
RT ✓
REPT ✓ ✓
BG ✓ ✓
BA ✓
LR ✓
KNN ✓ ✓ ✓
ANN ✓ ✓ ✓ ✓
SVR ✓
RF ✓ ✓
DL ✓
SNN ✓
CP-ANN ✓
XYF ✓
SKN ✓
MR ✓ ✓
MP ✓
ERT ✓
References J.H. etal.[42] Dai etal. [43] Ji etal. [45] Gandhi
etal. [47]
Uno etal. [48] Cheng
etal. [50]
Ghodsi
etal. [51]
Singh [52] Shastry
etal.
[56]
Crop type
MT
CB
P✓
W✓ ✓ ✓
C✓
R✓ ✓
T
SB
M✓ ✓
S✓
A✓
CP

Emerging Trends inMachine Learning toPredict Crop Yield andStudy Its Influential Factors:…
1 3
SVM: Support Vector Machine; RT: Random Tree; REPT: REP Tree; BG: Bagging; BA: Bayes; LR: Linear regression; KNN: K Nearest Neigh-
bour; ANN: Artificial Neural Network; ARIMA: Auto regressive Moving Average Model; ANFIS: Adaptive neuro-fuzzy Inference System; SVR:
Support Vector Regression; RF: Random Forest; DL: Deep Learning; SNN: Spiking Neural Network; CP-ANN: Counter Propagation-ANN;
XYF: XY Fused Networks; SKN: Supervised Kohonen Networks; MR: Multiple regression; MP: M5-Prime Regression Tree; ERT: Extremely
randomized Tree; MT: Mexicon Tomato ;CB: Common Bean; P: Potato; W: Wheat; C: Corn; R: Rice; T: Tomato; SB: Soya Bean; M: Maize;
CP: Chick Pea; A: Apple; S: Sunflower
Table 2 (continued)
References J.H. etal.[42] Dai etal. [43] Ji etal. [45] Gandhi
etal. [47]
Uno etal. [48] Cheng
etal. [50]
Ghodsi
etal. [51]
Singh [52] Shastry
etal.
[56]
Machine learning technique used
SVM
RT
REPT
BG
BA
LR
KNN
ANN ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
SVR
RF ✓
DL
SNN
CP-ANN
XYF
SKN
MR ✓ ✓ ✓ ✓
MP
ERT
want to predict or find trends in the raw data. As data per-
taining to agriculture is quite vast and is ever growing day
by day, so machine learning techniques can be of great help
in analysis of such huge data. The machine learning tech-
niques are characterized into two broad categories: super-
vised and unsupervised learning techniques. In Supervised
learning, machine is made to learn from the data provided
and is trained to make decisions on new or unseen data.
ANN, Bayesian network, decision tree, support vector
machines, ID3, k-nearest neighbour, hidden Markov model
are some examples of supervised learning. Unsupervised
machine learning is a technique in which machine is made
to infer based on the patterns identified in a dataset without
prior knowledge of any referenced or labelled outcomes.
Self organizing map, and partial based clustering, hierarchi-
cal clustering, k-means clustering are examples of unsuper-
vised machine learning techniques. Recent progressions in
the field of machine learning has inspired researchers all
over the world to explore the potential of emerging tech-
niques in different areas related to crops like yield predic-
tion and quantification of factors affecting the crop yield.
Following section describes the findings of prominent
research works done in mentioned areas.
3.1 Effectiveness ofFactors Affecting Crop Yield
Crop yield is a complex phenomenon involving contribution
of various weather and soil parameters. Apart from uncon-
trolled factors related to climate and soil, there are many
controlled factors contributing to the variations in the yield
of a crop like farm practices employed by the farmers, type
and quantity of fertilizers applied, frequency of irrigations
applied on field etc. In this scenario, it becomes essential to
quantify the contribution of various factors responsible for
crop yield. Following section provides a review of works
done in the said area.
Vashisht etal. piloted experiments to assess the contribu-
tion of various agriculture practices like planting time and
date, crop varieties and crop irrigation schedules on varia-
tions in bread wheat yield under different climatic variations.
Field experiments were conducted using six seasons for
time slice (PTS; 2008–2013). Simulation studies revealed a
reduction in crop yield with rise in temperatures (maximum
and minimum) [11].
In another study, increase in growing season average tem-
perature was found to have negative impact on the winter
wheat yield whereas growing season precipitation (GSP)

N.Bali, A.Singla
1 3
showed encouraging impact on the yield. Cobb–Douglas
production function was used to quantify the effect of vari-
ous factors involved in the study [14].
Linear Regression model was examined for quantify-
ing the effect of different meteorological parameters on the
rice yield in district Raipur, India. Particular stages of plant
growth such as seedling, tillering, 50% flowering and matu-
rity, were selected and effect of parameters on plant during
these stages was analysed. Different correlations (positive
and negative) were exhibited between various parameters
and growth stages [15].
Zhang etal. explored the utility of remote sensing tech-
nique in collecting and analyzing data obtained at different
stages of plant growth to find the contribution of growth
stages on final yield of winter wheat crop. Hyperspectral
information was gathered at three different stages (jointing,
heading and grain filling) and its effect on final crop yield
was analysed. The study proposed an enhanced 2D corre-
lation spectral analysis method to identify the perceptive
wavebands. The contributions of different phases of crop
growth to the estimated yield was determined by the models
based on coefficients of partial least square method using
complete spectral information. Support Vector machine
model was found to perform well with satisfactory accuracy
and robustness [16].
The effect of different land use practices on the overall
variations in surface temperature and its adverse effect on
the yield of rice and wheat crop was studied. The study was
limited to three different geoclimatic regions of Punjab. The
satellite data under study was categorized into four major
LULC classes: water, vegetative, built-up and plain soil.
There was found to be increase in temperature in areas where
the land use was transformed from agriculture, plain soil
and forest to urban. Normalized difference vegetation index
(NDVI) of the area was found to have positive correlation
with rice and wheat yield, but significantly negative correla-
tion with LST [17].
In another study, authors explored the impact of weather
parameters and technological advancements like improved
pesticide use and adoption of high yielding varieties on crop
produce in various regions of Haryana. Principle component
analysis (PCA) was carried out for preharvest yield estima-
tion in Haryana on various agro-climatic zones. Four cli-
matic zones were used for the study include different regions
of Haryana, India. The estimated yield(s) as per the designed
models were in good match with DOA wheat yield estimates
in most of the districts [18].
Mukherjee etal. reviewed the influence of various cli-
matic factors on yield of wheat in the states of Northwest
India. Various important factors were included in study such
as daily air temperature, standard precipitation and evapo-
transpiration index and ground water variability. The rise in
count of days having temperature above 35o C during matu-
rity period led to loss in yield of wheat crop. Also there was
Table 3 Deep learning techniques for crop yield prediction on different crop varieties
CNN: Convolutional Neural network; DNN: Deep Neural Network; LSTM: Long Short term memory; DCNN: Deep Convolutional neural Net-
work; RNN: Recurrent Neural Network; SOM: Self Organizing Map; CNN: Convolutional Neural network; BM: Bitter Melon; SB: Soya Bean;
CSB: County Soya Bean; C: Corn; OR: Orchard; A: Apple; W: Wheat; R: Rice; M: Maize; CC: County Corn
References Villanueva and
Salenga [82]
Oliveira
etal.[12]
Wang
etal.
[81]
You etal. [80] Kuwata and
Shibasaki [85]
Fourie et.
al. [83]
Bargoti and
Underwood
[6]
Mohan and
Patil [8]
Jiang et. al. [9]
Crop Type
BM ✓
CSB ✓
C✓
OR ✓
A✓
W✓
R✓
M✓
CC ✓
Machine Learning Technique used
CNN ✓ ✓ ✓
DNN ✓ ✓ ✓ ✓
LSTM ✓
DCNN
RNN ✓
SOM ✓

Emerging Trends inMachine Learning toPredict Crop Yield andStudy Its Influential Factors:…
1 3
depletion in ground water and surface water for irrigation
owing to less rainfall in the wheat growing season (Novem-
ber–March). Thus, high temperatures along with acute scar-
city of water and less irrigations led to an overall increase in
yield reduction [19].
In another study, the authors reviewed the effect of date of
sowing along with climate features on distinctive stages of
growth of wheat crop in Uttar Pradesh, India. Effect of the
two factors, date of sowing and climatic factors, was studied
primarily on time of germination, plant height and number
of tillers per spike. Correlation study revealed relationship
that was directly proportional with temperature, rainfall and
indirectly proportional with humidity for plant height and
number of tillers [20].
Jiayu etal. conducted a study on effect of meteorologi-
cal factors on different growing stages of rice and wheat
crops. The study used counties of China from 1980 to 2012
and 2481 weather stations for the day-degree data. The AIC
method was used to optimize the combination of various
meteorological factors and to select the most influencing
parameters. This method, outweighs the previous researches
in identifying the factors with less influence on the yield and
in exploring the relationship between meteorology and the
yield [21].
Epule etal. compared the impact of climatic and non-
climatic factors on the yield of 31 food and cash crops in
Uganda. Multiple linear regression model was used for the
analysis. Non climatic factors were found to have more effect
on the yields of the crop as compared to climatic factors.
Among climatic factors, temperature played major role in
affecting the yield of crop which was followed by precipita-
tion and CO2 emanations due to deforestation. Among non-
climatic factors, forest area dynamics (t value: − 11.11; p
value: 0.012 (1.20%); R: − 0.5), wood fuel (t value: − 9.40; p
value: 0.032 (3.16%); R: 0.3) and tractors used (t value: 8.46;
p value: 0.041 (4.09%); R: 0.2) showed their importance in
the order given [22].
Another study explored some of the key climatic and
agronomic factors affecting production of quality bread
wheat seed production. It being difficult to identify all pro-
duction factors at once, some important factors were initially
selected for the study. The selected factors included rainfall
and temperature among climatic factors while seed rate and
nitrogen fertilization were the agronomic factors. The con-
clusions from the study stated that temperature, amount of
rainfall and nitrogen fertilization of the soil were some of
the most important influential factors affecting the state of
the physiological processes in seeds and finally affected the
overall yield and quality of seeds [23].
Authors in a study quantified the effect of different
weather parameters on yield of wheat phenophase and grain
yield. The variations in climate was studied from 1981 to
2014. The data collected from 10 regions of Mongolia was
analysed using Agricultural Production Systems Simulator
(APSIM) model. Owing to significant climate warming from
year 1981 to 2014, there was a considerable reduction in
spring wheat yield, with an average of 3564kg ha−1. The
air and surface temperature variations were found to be the
major weather parameters affecting the phenophase of spring
wheat in Inner Mongolia. Between maximum and minimum
average temperatures, the former had more pronounced
effect than the latter. This was trailed by the relative damp-
ness and sun powered radiation. Precipitation, wind speed
and reference crop evapotranspiration were found to be least
affecting among various climatic factors. Regarding spring
wheat yield, temperature, solar radiation and air relative
humidity were the major contributing climatological factors
affecting in the eastern and western Inner Mongolia [24].
Meng etal. conducted a study for 20 districts over the
time slice of 1987–2010 to explore the effect of climatic fac-
tors like precipitation and temperature on canola and spring
wheat yield. The moment-based methods were used to
analyse asymmetric associations between climate and crop
yields. There was a rise in crop yield with the rise in grow-
ing season degree-days and pre-growing season precipitation
whereas extreme temperatures during the growing season
adversely affected the crop. Also the effect of variations in
temperature was found to be more impactful than precipita-
tion on yield distribution [25].
Safa etal. explored the efficiency of ANN in wheat crop
yield prediction in New Zealand. Out of 140 factors, 6 were
found to be showing considerable effect on the yield and
were used as input to the model. The study explored indirect
factors affecting the crop yield like conditions of farm land,
area of wheat and irrigation frequency, machinery conditions
and farm inputs in the form of N and fungicides consump-
tions. The final ANN model could predict wheat production
with very low error margin of ± 9% (± 0.89t ha−1) [26].
An assessment of climatic variables as factors affect-
ing the yield of corn and soybean crop was done by John-
son, D.M. Regression analysis was used to find correlation
between the factors and the final yield of crop. The study
concluded that NDVI and daytime land surface temperature
(LST) were positively and negatively correlated with the
yield of both the crops whereas precipitation and night LST
did not show prominent impact on the yield [27].
Parekh and Suryanarayana studied the effect of dif-
ferent combinations of weather parameters on crop yield
using ANN technique in Vallabh Vidyanagar for the period
1981–1999. The model was further used to validate the data-
set from 2000 to 2006. It was concluded that the combined
effect of all the three parameters, is essential for accurate
prediction of the crop yield [28].
Ruß etal. explored the role of IT in precision agricul-
ture using Neural Networks in the field of wheat crop yield
prediction. Among different factors, topology of network is

N.Bali, A.Singla
1 3
found to be the most important factor that affects the effi-
ciency of the network model, It was found that by increasing
the amount of available data, the prediction accuracy of the
model increases but size of network is not found to be having
much effect on the effectiveness of the model [29].
The reviews done have shown that temperature and pre-
cipitation have been important influential factors affecting
the yield of most of the crops. Agronomic practices adopted
by farmers like amount of irrigation, time of sowing and
pesticides used have also been found to be prominent in
the studies. Also, machine learning techniques like linear
regression, and neural networks have shown good potential
in rating the factors by exploring different combinations of
factors.
3.2 Machine Learning Techniques forCrop Yield
Prediction
An accurate and timely forecasting of crop yield before har-
vest is extremely important. Due the involvement of diverse
natured factors in the study, it has always been a challenge
for researchers to predict the yield accurately. The advances
in the field of machine learning has shown promising future
in the field. Following section reviews the application of
various machine learning techniques for crop yield predic-
tion for different crop varieties.
Wheat crop yield estimation was done using support
vector regression model. Various models were tested which
included nine base learner models and two ensemble mod-
els. The results showed that out of the nine models, SVR
showed the best learning efficiency and also ensemble mod-
els, in spite of considerable increase in cost, did not report
much improvement in accuracy. Also, an increase in training
data led to better results for all the models [30].
An extensive comparison of various machine learning
approaches in crop yield prediction was done for multiple
crops. The authors in this paper compared the machine
learning techniques for predictive accuracies. Four accuracy
rubrics [root relative square error (RRSE), correlation factor
(R), normalized mean absolute error (MAE), and root mean
square error (RMS)] were used to check the accuracy of the
models. Results favoured M5-Prime and k-NN techniques
with lowest RMSE errors (5.14 and 4.91), the smallest
RRSE errors (79.46% and 79.78%), very low average MAE
errors (18.12% and 19.42%), and at the same time showing
highest correlation factors (0.41 and 0.42) [31].
In another study, Ahamed etal. explored data min-
ing techniques to see the impact of various environmental
(weather) factors including biotic factors and production
area on the crop produce in different districts of Bangladesh.
Clustering technique was applied to divide regions on the
basis of attributes to be studied and suitable classification
techniques (Linear Regression, KNN and neural network)
were applied for predicting crop yield. Results of RMSE
proved ANN as being better in predicting for some of the
crops with missing values i.e. wheat, potato and aus. Linear
Regression gave good results for boro and amo [32].Another
study done by Lamba and Dhaka also found superiority of
neural network model over other models (Statistical, Met-
rological, Simulation, Agronomic, Remote Satellite Sensed,
Synthetic and Mathematical) [33].
Nath etal. explored the efficiency of Box Jenkin’s Autore-
gressive Integrated Moving Average model, ARIMA (1, 1,
0), a time series modelling approach to predict wheat pro-
duction for India. The forecast was done using previous yield
data from 1949–1950 to 2016–2017 (68years) and an effort
was made to do the predictions for ten leading years. As
model can be used only on stationary data, the data was
converted to stationary data by differencing the time series.
It was concluded that ARIMA (1, 1, 0) provided a reason-
able predictive model and proposed a raise in production for
duration of 10years (2017–2018 to 2026–2027) [34].
Another study was done in Ukraine by Kogan etal. to
explore the application of remote sensing for forecasting
yield of winter wheat crop. Three approaches of forecasting,
empirical regression based model, with Moderate Resolution
Imaging Spectroradiometer (MODIS), empirical regression
model based on weather-related parameters and CGMS were
compared. The most reliable and accurate predictions for
2010 were obtained using the CGMS system while perfor-
mance of all three approaches was same for 2011 (0.6t ha−1
in April) [35].
In another study, the accuracy of calendar-day based
approach in forecasting the phonological growth of soybean
was explored. Prediction models were built using various
machine learning techniques as artificial neural network
(ANN), k-nearest neighbour (kNN) and regression. Results
proved the approach under study to be feasible in predictions
as all three methods achieved acceptable levels of accuracy
for vegetative and reproductive stage using ANN, kNN and
Regression models [36].
Kim and Lee studied remote sensing data for predicting
corn yield in Iowa State using four machine learning tech-
niques (SVM, RF, ERT and DL). Results proved in favour of
machine learning techniques as a means for estimating the
yield especially for DL which showed more stable results
[37].
Bose etal. performed remote sensing spatiotemporal
analysis using spiking neural network for crop yield valua-
tion. The study made preharvest yield prediction six weeks
prior to harvest with an accuracy of 95.4% and average error
of prediction of 0.236 t/ha and correlation coefficient of
0.801 using a nine-feature model [38].
Pantazi etal. studied the variations in the wheat yield
based on online multilayer soil data and satellite imagery
crop growth characteristics. Three machine learning

Emerging Trends inMachine Learning toPredict Crop Yield andStudy Its Influential Factors:…
1 3
techniques, counter-propagation artificial neural networks
(CP-ANNs), XY-fused Networks (XY-Fs) and Supervised
Kohonen Networks (SKNs), were compared in performance.
Results indicated that yield prediction in case of cross vali-
dation generated yield for low yield class is more than 91%
which is highly significant in reference to intricate relation-
ship between controlling aspects and the yield. For average
and high-pitched yield class, accuracies obtained were 70%
and 83% respectively. Among the three models, SKN, CP-
ANN and XYF, SKN showed highest accuracy of 81.65%
proving it to be the best model [39].
Kaul etal. compared ANN models in terms of their pre-
diction capabilities and their performance with multiple
linear regression models. Results indicated that ANN con-
sistently produced more accurate predictions as compared
to linear regression model [40].
Chlingaryan etal. reviewed various machine learning
methods for predicting yields in remote sensed data. Review-
ers discussed researches done in last 15years on machine
learning techniques for yield prediction. The study finds a
promising future of remote sensing and machine learning
techniques in providing good solutions for better crop esti-
mation and decision making [41].
A comparative study of Random Forest models with Mul-
tiple linear regression models for yield estimation of varied
crops (wheat, maize, and potato) was done by Jeong etal.
Better results were shown by Random Forest models in all
performance metrics. For RF models, the root mean square
errors (RMSE) ranged from 6 to 14% of the mean obtained
yield for all test cases whereas the values varied from 14 to
49% for MLR models. Thus, the study proved RF to be an
efficient and useful technique for yield prediction not only
at regional but also on global scale [42].
Apart from climate, soil is another important and con-
tributing factor to the variation in yields of a crop. Dai etal.
studied the effect of two soil properties, soil moisture and
salt content of soil, on the yield of sunflower crop. Two
machine learning techniques ANN and MLR were compared
in the study for efficiency in analysing the effect of differ-
ent input variables related to soil as soil moisture and soil
salinity during different phases of crop growth. Connection
weight method was adopted to measure crop sensitivity to
soil moisture and salt stress at different growth stages. In this
method, the connection weights of input-hidden and hidden-
output were taken into consideration. Compared with MLRs,
both ANN models (ANN-10 and ANN-6) showed better pre-
cision according to RMSE, RE and R2 values. Overall, the
models based on neural network showed good performance
in exploring diverse relationships between yield of the crop
and and soil features during different stages of crop growth
[43].
Usually the model proposed for crop yield predictions
in an area is applicable for only that specific area but in
a study, a generalized regression model was proposed for
doing predictions in multiple areas. The model was first used
for doing predictions in Kansas and the same model was
tested in Ukraine. The results gave good accuracy of the
proposed model by reporting 7% error in Kansas and 10%
in Ukraine [44].
In another study, Ji etal. investigated the capability of
artificial neural network on rice crop yield prediction in
mountainous regions of The Fujian province of China where
continuous weather aberrations such as typhoons, floods and
droughts constantly impend the rice production. The objec-
tive of the study was to test the feasibility of ANN model to
predict rice yield in typical climatic conditions and to com-
pare it with multiple linear regression models. ANN model
parameters such as learning rate and number of hidden layers
were found to have significant effect on the accuracy of the
model in rice yield predictions. Smaller data sets were found
to require less hidden nodes and lower learning rates for
model optimization. Comparative analysis of ANN models
and MLR for accuracy of predictions favoured ANN models
over MLR models [45].
Four BPN models of ANN were studied for corn yield
prediction based on topographic features, vegetation indices
and textural indices. Results confirmed that the use of topo-
graphic data along with vegetation and textural indices can
greatly improve the prediction accuracies. Also, efficiency
of ANN as a prediction model was highlighted in study [46].
Another study explored the potential of ANN in rice yield
prediction in various districts of Maharashtra state in India.
The study was based on different environmental predictor
variables including temperature and precipitation on the
yield of the Kharif season for the years 1998 to 2002. A
multilayer perceptron neural network model was used on the
current dataset. The results gave high accuracies of 97.5%
with a sensitivity of 96.3 and specificity of 98.1. WEKA tool
was used for the analysis purpose [47].
In recent times, the remote sensing has shown great
potential in vegetation growth analysis and has good pros-
pects in helping in the field of crop yield prediction. Uno
etal. studied the importance of accurate crop yield maps in
precision farming. The ability of remote sensing systems to
acquire information for a large area in a brief span of time
makes it more beneficial over harvester-mounted crop yield
monitoring units for yield map creation. Both statistical and
ANN models were used to develop yield estimation models.
In order to reduce large amount of unnecessary information,
usually generated in hyper spectral imagery, and to deal with
the problem of over fitting, PCA was used. Results showed
greater accuracy in prediction with ANN models than with
either of three traditional empirical models [48].
Balaghi etal. studied the efficiency of least square regres-
sion models to estimate the yields of wheat crop in Morocco
at province and national level. The predictions used NDVI/

N.Bali, A.Singla
1 3
AVHRR, rainfall sums and average monthly air temperatures
as input parameters. The study tried to explore the future of
NDVI in the field of crop yield prediction. Yields associ-
ated with provinces were assessed with errors between 80 to
762kg ha−1, depending on the province. whereas at national
level, yield prediction showed 73kg ha−1 error. The study
recommended that proposed model can be useful for early
forecast of wheat yield in Morocco [49].
A new artificial neural network approach was proposed
for early yield prediction in fruit crops by means of image
analysis. Two BPNN models were established for two phases
of the season: opening period and the ripening period.
Results were analysed in terms of various measuring fac-
tors such as correlation coefficients (R2), mean forecast error
(MFE), mean absolute percentage error (MAPE), and root
mean square error (RMSE). For early periods, the values
came out to be 0.81, − 0.05, 10.7%, 2.34kg/tree, respec-
tively whereas for the ripening period, these measures were
0.83, − 0.03, 8.9%, 2.3kg/tree, respectively [50].
Among various factors affecting the yields of a crop, cli-
matic factors play an important role. Ghodsi etal. examined
the effect of different climatic parameters on the yield of
wheat crop in Iran. ANN models were used for analysing
the data. ANNs are an encouraging substitute to economet-
ric models. Eight important factors were considered in the
study. In order to select the best ANN model, 11 varied ANN
models with different number of neurons in hidden layers
were tried and the optimum model was selected. ANN-
MLP model based on conjugate gradient back propagation
algorithm reported lowest MAPE making it the preferred or
optimum model. The results reported have also supported
the proposed ANN model as the suitable way for wheat yield
prediction [51].
Another study was conducted to explore Multi-layered
feed forward model of artificial neural network. Two learn-
ing algorithms namely gradient descent algorithm (GDA)
and conjugate gradient descent algorithm (CGDA) were
explored to train ANN for maize crop yield forecasting.
Three layered MLFANN with two hidden layers containing
(11, 16) units was found to be the best for training, validation
and test sets [52].
Alvarez studied the effect of soil and climatic parameters
on wheat yield in Argentina pampas to propose an accurate
and efficient model for crop yield estimation. The region
under study was split into 10 geographical areas. Two tech-
niques, surface regression and ANN were studied and com-
pared for efficiency. ANN has been found to be much better
and accurate in predictions (RMSE = 0.05) as compared to
surface regression method [53].
Park etal. compared three adaptive techniques, artificial
neural networks (ANNs), general linear models (GLMs),
and regression trees (RTs) in forecasting maize crop yield
in eleven dissimilar land management tests in southern
Uganda. GLM showed poorest results whereas RT showed
the best results. ANN also showed promising results [54].
In another study, authors explored the potential of regres-
sion model in making the preharvest estimate of what crop
yield in Ludhiana district Punjab. The efficiency of regres-
sion models was found to be highly significant at 5% signifi-
cance level. The regression models developed on weather
parameters justified 69% variations in crop yield [55].
A default ANN consist of an input layer, a hidden layer
and one output layer. Shastry etal. compared the efficiency
of default ANN with customized ANN (C-ANN) and MLR
techniques for wheat crop yield prediction using some of
the important influential factors for the study like amount
of rainfall, crop biomass, soil evaporation, transpiration,
Extractable Soil Water (ESW) and amount of fertilizer
applied (NO3). Results showed significant improvement
in yield prediction in C-ANN with higher R2 statistics and
lower percentage errors as compared to MLR and D-ANN
techniques [56].
Bhangale etal. proposed a methodology for crop yield
estimations. The study was pertaining to various states of
India. Each state was divided into different agriculture zones
according to the amount of rainfall received and geographi-
cal location. The proposed model used SVM for analysing
weather changes, K means approach to classify the soil and
plants and ANN for crop yield prediction on the basis of
data collected from previous techniques. The proposed agri-
cultural DSS framework provided a means to predict the
cropping information in advance from a set of inputs [57].
Bejo etal. explored the efficiency of ANN in various
aspects related to crop yield estimation. These aspects
included ANN in study of Environmental factors, soil and
soil–plant Hydrology, Sensing Techniques, Biomass factor
prediction, and controlled environment studies. The authors
emphasized the need to explore the current trends in crop
yield prediction for precision agriculture. ANN has been
found to be showing great future in all the studied areas and
can definitely prove to be an asset for agriculture research-
ers [58].
In this paper, Dahikar and Rode studied the effect of vari-
ous predictor variables such as type of soil, PH value, pres-
ence of various nutrients in soil, depth temperature, rainfall
and humidity on yields of various crops using ANN models.
Study varied the model architectures by wavering the hidden
layers used. The results of study showed a promising future
for ANN in crop yield estimation even in rural areas [59].
Application of ANN for prediction of crop yield for
rice, sugarcane and wheat crops was studied. The proposed
model used weather variables as input whereas crop yield
was the output produced. Different MLP algorithms have
been explored. Among various algorithms and architectures
explored, forecasts done using MLP architecture based on
Conjugate gradient descent algorithm was found to be

Emerging Trends inMachine Learning toPredict Crop Yield andStudy Its Influential Factors:…
1 3
closest to the actual observations in almost all the cases.
The study authenticates ANN as a promising technique for
crop yield prediction [60].
In the area of green house operations, crop yield predic-
tion is still a manual affair. Qaddoum, Hines and Iliescu
proposed an efficient technique to estimate yield of tomato
crop to help human operators in anticipating and accordingly
take care of problems of both over demand and over pro-
duction accurately. The influential parameters selected for
the study comprised of different ecological variables inside
greenhouse, such as, temperature, vapour pressure deficit
(VPD), CO2 and radiation, as well as yields from previous
years. The proposed system was an intelligent system called
EFuNN (evolving fuzzy neural network). Results reported
gave weekly predictions with an accuracy of 90% [61].
Khoshnevisan etal. considered the impact of different
energy input contributions on wheat harvest yield utilizing
a consolidated model of ANN and fuzzy framework, ANFIS.
Several ANFIS models were structured and trained for the
study each utilizing the learning ability of ANN to formulate
if-then rules of fuzzy system and to develop suitable func-
tions designed from training pairs to generate inferences.
The models were compared for accuracy with ANN for pre-
diction. The most promising ANFIS structure found in the
study reported R, RMSE and MAPE as 0.976, 0.046 and 0.4,
respectively proving that ANFIS model has better prediction
capability than ANN [62].
Naderloo etal. proposed adaptive neuro fuzzy system for
crop yield prediction using different energy inputs. Owing to
such a large number of inputs, the input vector was clustered
into two groups and two networks were trained. The RMSE
and R2 values were found 0.013 and 0.996 for ANFIS 1
and 0.018 and 0.992 for ANFIS 2, respectively. The values
predicted by ANFIs1 and ANFIS 2 models were utilized as
input for third ANFIS model. Results showed that param-
eters used in first ANFIS model had more profound effect
on yield than other energy inputs. Also, the ANFIS 3 model
using outputs of ANFIS 1 and ANFIS 2 models showed
RMSE and R2 values as 0.013 and 0.996, respectively [63].
Kouchakzadeh and Ghahraman explored the effect of var-
ying weather conditions on the wheat crop yield using ANN
and ANFIS models. Various weather related parameters such
as evapotranspiration, precipitation, daily temperature (max,
min, and dew temperature), daily average relative humidity
for twenty-two years at nine stations and net radiation, were
considered as part of the study. The results of ANFIS model
were found to be consistently more precise in terms of vari-
ous statistical indices (R2 = 0.67, RMSE = 151.9 kg ha−1,
MAE = 130.7kg ha−1), when temperature (max, min, and
dew temperature) data was used as an independent variable
[64].
Pandey etal. compared the efficiency of ANN in pre-
dicting wheat crop yield in comparison to fuzzy time series
method. Authors found fuzzy methods quite subjective as
their output has an element of user’s interpretation which
leads to different results interpreted by different analysts.
This drawback is not there in ANN which is objective
in nature as in ANN the prediction is done solely by the
designed network and there are multiple interpretations of
the results [65].
Balakrishnan and Muthukumarasamy compared effi-
ciency of various machine learning techniques for various
crops yield predictions. The significant classification meas-
ures used in the study were Support Vector Machine (SVM)
and Naive Bayes. Two proposed ensemble models, AdaSVM
and AdaNaive were explored to estimate the crop produc-
tion over a particular selected time slice. The results showed
great perfection by reporting very few errors in prediction
and also there was appreciable amount of fall in the classifi-
cation error for both the proposed techniques [66].
Priya etal. studied the efficiency of random forest tech-
nique for crop yield prediction for kharif and rabi seasons
for rice crop. Data analysis was done using R-Tool and the
results showed a promising future for random forest tech-
nique for massive crop yield predictions [67].
In another study, authors proposed a model for crop yield
prediction employing data mining with association rules as
a tool for prediction in different districts of Tamil Nadu in
India. The study revealed that the projected model is efficient
in predicting the yield [68].
Shree studied the impact of parameters related to soil and
atmosphere on variations in crop yield. This paper predicts
the crop yield and also suggests the best crop that should be
sown for improving the quality and profits incurred from the
agricultural sector. The soil and climatic factors were taken
as inputs for the study including type of soil, temperature
variations, relative humidity, ground water level, spacing,
depth, pH of the soil, seasonal variations, fertilizers used and
months of cultivation. This prediction was aimed at help-
ing the farmers in determining whether the specific crop is
suitable for a given soil type. The Bayesian algorithm was
employed for predictions for achieving high accuracy and
speed [69].
Ingole etal. explored the efficiency of Fuzzy logic in crop
yield prediction and in selecting suitable crop under particu-
lar conditions. Authors generated a decision support system
that gets values of inputs and provides the name of the crop
that can be sown for the given inputs. Two sensing mecha-
nisms were used light sensors and temperature sensors [70].
Use of Fuzzy logic for crop yield prediction was explored
in which different Fuzzy models based on different partitions
of Universe of Discourse and their effect on wheat crop yield
prediction was studied. This paper proposed a method for
wheat crop forecasting by using actual production as the
universe of discourse and intervals based partitioning. The
proposed method was found to be optimal and gave high

N.Bali, A.Singla
1 3
precision with insignificant mean square and average fore-
casting error rate. The proposed fuzzy approach has proved
to be an inerrant and efficient way to estimate wheat produc-
tion [71].
Efficiency of Neuro fuzzy model (ANFIS) was explored
for rice crop yield prediction. Various meteorological param-
eters were included in study and Gamma test was done to
find the factors closely related with the yield of crop. The
prediction was done on a time series data of 27years and
results have shown good efficiency of ANFIS model for rice
crop yield prediction [72].
Table2 provides a summarized view of the works done
by researchers on crop yield prediction in terms of different
crop type and machine learning techniques used. The col-
umns in table containing less markings shows the need of
further exploration in those crop types and ML techniques.
Also maximum markings are in the column of ANN show-
ing that neural networks have been a preferred choice of
many researchers for the study.
Recent advancements in the field of ANN in the form
of Deep learning has opened new avenues in the field. Fol-
lowing section has reviewed works done in the field of deep
learning for crop yield prediction.
4 Deep Learning: AnEmerging Trend
intheField ofAgriculture
Among various machine learning techniques, neural net-
works have shown good results owing to their efficiency in
dealing with both linear and non linear aspects of the data.
Deep learning is an extended version of neural networks.
Deep Learning extends classical machine learning by con-
tributing more “depth” into the model and transforms the
data by using several functions providing data a hierarchical
representation, with several levels of abstraction [73]. The
ability of feature learning, extracting information from raw
data, inherent in deep learning architecture makes it par-
ticularly advantageous for solving complex problems. DL
models provide classification accuracy even on very large
datasets which are difficult to be dealt with by other machine
learning techniques. DL architecture comprises of various
different components, based on the overall network architec-
ture adopted like Unsupervised Pre-trained Networks, Con-
volutional Neural Networks, Recurrent Neural Networks,
Recursive Neural Networks. Deep learning has achieved
considerable popularity in analyzing raster based data like
videos and images although it can be used on any form of
data, such as audio, speech, and natural language, and more
commonly to continuous or point data like weather data,
soil chemistry and population data [74]. Apart from yield
estimation, deep learning based approaches have also shown
good results in other agriculture based fields like early plant
disease detection in different crops including fruit crops like
banana [75, 76]. Following section discusses the contribu-
tion of deep learning in the field of crop yield estimation.
4.1 Deep Learning inCrop Yield Prediction
Deep learning models consist of a highly complex hierarchi-
cal structure with a large learning capacity. These features
make these models particularly suitable for dealing with
classification and prediction challenges. Various architec-
tures of deep learning have been used depending upon the
type of data involved. Newlands etal. studied the role of
deep learning in assessing the potential risk in agricultural
insurance or management. In this paper, the authors empha-
sized the need of accurate crop yield prediction for a proper
index insurance of agricultural products. In the study, the
authors compared the forecasting power of deep learning
by gauging its functioning with other established predic-
tive techniques. The study revealed a good potential of deep
learning in the field by showing highest predictive accu-
racies. Also the authors inferred that use of deep learning
avoids underestimating the inefficiencies and costs of insur-
ance coverage occurring due to the use of vague metrics of
real risk exposure [77]. In another study, deep learning was
used for estimation of corn yield and was compared with
SVR technique. The results favored DL as an efficient and
accurate technique for yield estimation [78].
4.1.1 Deep Learning Approach toCater Enormous Data
Requirements
Application of any machine learning technique requires
enormous amount of data. More the amount of data, more
is the accuracy in predictions. In case of deep learning, this
requirement becomes one of the bottlenecks for some areas
where large data acquisition is quite difficult. Yield predic-
tion in developed countries like USA is largely facilitated by
easy access to large scale survey data and common variables
related to crop growth. The developing countries have to
struggle to get right amount of information for accurate yield
predictions. Researchers are trying to propose techniques
through which this data requirement can be fulfilled.
Cunha etal. proposed a new method of yield prediction for
a crop before the start of the season called pre season predic-
tion using data from multiple sources. Normalized Difference
Vegetation Index (NDVI) has commonly been used as an indi-
cator of vegetation activity of an area and is obtained through
remote sensing of the farm. Although these indices provide a
good insight of the conditions but they usually come with a
cost i.e. cost of data acquisition and data processing to gener-
ate required analytics. This study proposes a new ML based
method using data from multiple sources to perform crop yield
estimation of soybean crop before the start of the season. The

Emerging Trends inMachine Learning toPredict Crop Yield andStudy Its Influential Factors:…
1 3
system comprises of RNN, Recurrent Neural Network model
trained using precipitation, temperature, soil properties and
previous years observed soybean yield as data for 1500 + cit-
ies in Brazil and USA. The two main highlights of the study
were that system could perform yield predictions with much
lesser amount of data as compared to existing yield forecast-
ing systems and also the predictions could be done before the
beginning of the crop season. Results have shown that it is pos-
sible to obtain dependable yield forecasts with very less data
requirements as the proposed neural network model has the
ability to identify and exploit unnecessary information inher-
ent in soil and weather related data. Also the results showed
that the prediction accuracy also varies with the type of crop
into consideration as they vary in their physiological proper-
ties [79].
You et al. proposed a new approach based on modern
representation learning ideas to predict county-level soy-
bean yield in U.S. The lack of sufficient training data was
accomplished through a new dimensionality reduction tech-
nique. In this, the raw images were treated as histograms
on which deep learning architectures, including CNNs and
LSTMs were trained to predict the crop yield. The account
for spatio-temporal dependencies between data points was
compensated by incorporating a Gaussian process layer
ahead of Neural Network model. Experimental results have
shown that proposed model has outperformed the custom-
ary remote sensing centred techniques by 30% in terms of
RMSE and USDA national estimates by 15% in terms of
Absolute Percentage Error (MAPE) [80].
Wang etal. introduced the concept of transfer learn-
ing approach to predict the yields with less available data.
The author emphasized the need of abundant ground truth
training data for success of any deep learning model. This
approach is particularly useful when deep learning needs
to be applied in region with little training data. In such
cases, fine tuning the pre-trained models can be of consid-
erable help. In this study, model proposed by You etal. for
predicting crop yield on remote sensed data has been fine
tuned to predict the yield of soya bean crop in Brazil. Unlike
the approach used by You etal. who tested and trained the
model in same region, the authors tested the ability to trans-
fer a model trained in one region to another. The study has
shown a monotonic increase in the accuracy of prediction
with the increase in the amount of data in most of the cases.
Also the results have favoured the approach of transfer learn-
ing as an exciting new approach specifically helpful in the
regions with less available data [81].
4.1.2 Convolutional Neural Network forImage Data
Analysis
Advancements in the field of computers has made images a
good source of input for researches in the field of agriculture.
Nowadays, various remote sensing devices can capture
images from areas which were considered to be unapproach-
able and difficult to study. The need of the hour is to explore
techniques which can efficiently extract information from
these images and use this information for accurate future
predictions. Convolutional neural network, a class of deep
learning neural network, has been a breakthrough in the area
of image recognition and analysis. Following section reviews
some of the works done by researchers in the said area.
Villanueva and Salenga studied the fruit bearing ability
of bitter melon or bitter gourd crop using CNN method. The
study was based on the scrutiny of the health of leaves of
plant gathered from Ampalaya farms. The fruit bearing capa-
bility of the plant was judged on the basis of color and shape
of leaves as small size, deformed shape and dark green, yel-
low or brown color of leaves signified bad class with no
fruit bearing capability whereas normal size, green or light
green color of leaves signified good fruit bearing capabil-
ity of plant. Training of data was done through Keras, ten-
sor Flow and Python worked together. The study concluded
that testing of at least 293 images was required to train the
CNN model to correctly predict the fruit bearing capability
of the plant. On increasing the count of training images, the
neurons of the model will also increase which will in turn
increase the prediction capability of the model [82].
Accurate crop yield prediction in orchards is particularly
useful for efficient load management. Traditionally, this is
done by manually taking a measure of important features of
the fruit trees (wood, buds, flowers, fruitlets, and fruit) dur-
ing various stages of growth that can affect the crop yield.
This is quite laborious and expensive process and may lead
to inaccuracies. Fourie, Hsiao and Werner in this paper pro-
posed an automated yield prediction system that optically
estimates crop yield during various stages of growth. Deep
convolutional neural networks (DCNN) was used to build
object detectors that extract regions from the image that
represents the fruit. InceptionV3 model, pretrained on Ima-
geNet database, was used as an image feature extractor and
was customized using own classifier for classifying parts of
an image containing fruit or background features. The same
framework can be applied to detect leaves, branches or other
parts of orchard canopy [83].
Another study for fruit detection and counting on orchard
image dataset was done by Bargoti and Underwood. Two
feature learning algorithms, CNN and MLP were used along
with an image segmentation approach. Metadata pertaining
to the methods of capturing image data was added to the
networks. Results showed improvement in the fruit image
segmentation with the inclusion of metadata. The count esti-
mate results produced using CNN and WS gave a squared
correlation coefficient of r2 = 0.826 [6].
Kuwata and Shibasaki explored deep learning techniques
and machine learning technique SVR for estimation of

N.Bali, A.Singla
1 3
Illinois corn yield. In this study, the researchers proposed
a methodology for crop yield estimation using deep learn-
ing to unearth features prominently affecting crop growth
and yet to be quantified. Various environmental factors
selected for the study included NDVI (Normalized Vegeta-
tion Index), APAR (Absorbed Photosynthetically Active
Radiation), canopy surface temperature and water stress
index. Convolutional Architecture for Fast Feature Embed-
ding (Caffe) was used to implement the estimation model of
deep learning. Results showed highest accuracy in predic-
tion with two InnerProductLayer model. Correlation coef-
ficient was reported as 0.810 and RMSE was 6.298. For
single InnerProductLayer, the values were 0.727 and 7.427
respectively. For SVR, correlation coefficient was found to
be 0.644 and RMSE was 8.204. Results indicated that two
InnerproductLayer model with trained model of Caffe can
estimate crop yield index more accurately as compared to
SVR model which overestimates once the crop yield index
goes below 0.8 [85].
Accurate weather forecasting is one of the key features
in success of any yield prediction in agriculture. Mohan
and Patil proposed a dimensionality decreasing strategy,
Self organizing Map (SOM), along with Latent Dirichlet
Allocation (LDA) for predicting appropriate season and
crop for agriculture purpose. Suitable season for an appro-
priate crop was decided with the help of deep neural network
classification system. The Results of the study claims that
proposed approach when compared to the other approaches
for weather and crop prediction, proved to be more effec-
tive(7–23%) according to accuracy, sensitivity, specificity,
precision and recall, than the previous methods [8].
Jiang etal. proposed deep LSTM, a special form of RNN
model, to predict County level Corn yields. A large data
comprising of county level corn yield and hourly weather
data necessitated the need for using deep learning model.
This paper claims to be the first to employ LSTM for corn
yield prediction. The model gave quite good predictions
which shows a promising future for LSTM in the field of
crop yield prediction [9].
Table3 shows the summarized view of deep learning
techniques studied on different crops emphasizing the areas
yet to be delved into. The success rate of deep learning tech-
niques in the area of crop yield estimation can be an inspira-
tion for researchers to do more exploration in the area.
5 Conclusion
Crop yield estimation is one of the important areas of agri-
culture which is essential for agriculture planning involving
proper crop selection, giving farmers the correct estimates of
their gains for the crop they are planning to sow and also for
deciding the import and export decisions of the government
of any nation. Many studies have already been done on dif-
ferent types of crops and different techniques for the estima-
tions. Traditional ways are quickly becoming obsolete with
the intervention and success of machine learning as a tool
in various practical fields.
Machine learning techniques have the ability to extract
information and identify patterns from structured as well
as unstructured data and that too without the intervention
of any human intelligence. These properties make it well
suited for studies requiring future predictions from raw
data. Also its ability to cater enormous amount of varied
nature data further help in areas like agriculture where cli-
matic and soil data is involved having spatial and temporal
variations.
The accuracy of a machine learning technique is meas-
ured in terms of different evaluation metrics like Root Mean
Square Error (RMSE), Correlation factor (R2), Root relative
Squared Error (RRSE), Mean Absolute Error (MAE) and
Mean Absolute Percentage Error (MAPE). Figure3 shows
values of R and RMSE reported for multiple techniques by
different authors in their study.
Some inferences drawn from the study:
(a) The variations of same technique in different stud-
ies is owing to the use of different crops and diverse
parameters of study for the models. Apart from weather
and soil parameters, agronomic practices adopted by
farmers have also shown considerable effect on the
final yield of crop. The evidence shows that there is no
standardization of parameters and how the parameters
are being tuned. The focus is required to study the opti-
mized deduction of parameters on which the model is
based.
(b) As is evident from the graph, best results are shown by
Neural Network technique, ANN and ANFIS. Hybrid-
ized model based on fuzzy and ANN has shown good
accuracies proving to be efficient in the area. We can
explore more such hybridized techniques in future.
(c) Advanced techniques of machine learning like deep
learning has shown good potential in dealing with
huge amounts of data. Also their efficiency in learn-
ing through pattern recognition in the data without any
outside training has made them particularly suitable for
the area of crop yield predictions.
6 Future Work
As is evident, among machine learning techniques, ANN
based techniques hold a very promising future in the field
of crop prediction. The enormous amount of data and
that too from varied sources force the need of more effi-
cient models than ANNs. Deep learning is a very recent

Emerging Trends inMachine Learning toPredict Crop Yield andStudy Its Influential Factors:…
1 3
advancement in the field of ANN. The technique has
shown promising results in various research areas like
speech recognition, medical diagnosis, drug design and
many other important areas. Recently, its contribution in
solving the food problems of the world has shown tre-
mendous success. The advanced techniques like LSTM
and RNN have shown good results paving way for more
exploration for deep learning techniques in the field. Also,
Fuzzy techniques have shown good results both in crop
yield prediction and in parameter evaluation. Fuzzy theory
is an important concept in decision making theory and
science. However fuzzy logic is characterized by its mem-
bership function lying between 0 and 1 but not a non mem-
bership function. To overcome this, the concept of intui-
tionistic FS (IFS) was introduced for both membership
and non membership functions between 0 and 1 which was
extended to a more generalized form as Interval-valued
intuitionistic FS (IVFS). As IFS and IVFS cannot cover
indeterminate and inconsistent information, Neutrosophic
sets have been a new trend to express indeterminate and
inconsistent information which can be widely explored in
the area of crop yield prediction as it involves factors of
indeterminant nature.
Author contributions All authors contributed to the study as follows:
NB: conceptualization, original draft preparation, literature search,
editing the drafts. AS: supervision and guidance, suggestions for
improvements, directions, critically reviewed and revised the work.
All authors read and approved the final manuscript.
Funding This study was not funded by any organization, institution
or research centre.
Data availability Not applicable.
Code availability Not applicable.
Declarations
Conflict of interest Authors Nishu Bali, Anshu Singla have received
no research grants or honorarium from any Company, or are members
of any committee. The authors declare that they have no conflict of
interest.
References
1. Lipper L etal (2014) Climate-smart agriculture for food security.
Nat Clim Change 4(12):1068–1072. https:// doi. org/ 10. 1038/ nclim
ate24 37
2. Atzberger C (2013) Advances in remote sensing of agriculture:
context description, existing operational monitoring systems and
major information needs. Remote Sens 5(2):949–981. https:// doi.
org/ 10. 3390/ rs502 0949
3. Prospects by Major Sector (2020, April 10). http:// www. fao. org/3/
Y3557E/ y3557 e08. htm
4. Wright BD (2012) International grain reserves and other instru-
ments to address volatility in grain markets. World Bank Res Obs
27(2):222–260. https:// doi. org/ 10. 1093/ wbro/ lkr016
5. Basso B, Cammarano D, Carfagna E (2013) Review of crop yield
forecasting methods and early warning systems. In: The first meet-
ing of the scientific advisory committee of the global strategy to
Fig. 3 Values of various evaluation metrics for various machine learning and hybrid techniques

N.Bali, A.Singla
1 3
improve agricultural and rural statistics, pp 1–56. https:// doi. org/
10. 1017/ CBO97 81107 415324. 004
6. Wilkerson GG, Jones JW, Boote KJ, Ingram KT, Mishoe JW
(1983) Modeling soybean growth for crop management. Trans
ASAE 26(1):63–73
7. Jones CA, Kiniry JR (1986) CERES-Maize: A simulation Model
of Maize Growth and Development. Texas A&M Press, College
station
8. Porter JR, (1993) AFRCWHEAT2: A model of the growth and
development of wheat incorporating responses to water and
nitrogen. European J Agronomy 2(2):69–82
9. Jamieson PD, Semenov MA, Brooking IR, Francis GS (1998)
Sirius: a mechanistic model of wheat response to environmental
variation. European J Agronomy 8(3–4):161–179
10. Chen Y, Donohue RJ, McVicar TR, Waldner F, Mata G, Ota N,
Houshmandfar A, Mata G, Lawes RA (2020) Nationwide crop
yield estimation based on photosynthesis and meteorological
stress indices. Agric For Meteorol 284:107872
11. Savla A, Israni N, Dhawan P, Mandholia A, Bhadada H, Bhard-
waj S (2015, March) Survey of classification algorithms for
formulating yield prediction accuracy in precision agriculture.
In 2015 International Conference on Innovations in Information,
Embedded and Communication Systems (ICIIECS) (pp. 1-7).
IEEE
12. Oliveira I, Cunha, RL, Silva B, Netto MA (2018) A scalable
machine learning system for pre-season agriculture yield forecast.
arXiv preprint arXiv: 1806. 09244
13. Vashisht BB, Maharjan B, Jalota SK (2019) Management prac-
tice to optimize wheat yield and water use in changing climate.
Arch Agron Soil Sci 65(13):1802–1819. 10. 1080/ 03650 340. 2019.
15789 57
14. Geng X etal (2019) Climate change impacts on winter wheat yield
in Northern China. Adv Meteorol. https:// doi. org/ 10. 1155/ 2019/
27670 18
15. Jain A et al (2019) Developing regression model to forecast
the rice yield at Raipur condition. J Pharmacogn Phytochem
8(1):72–76
16. Zhang L etal (2010) Simulation and prediction of soybean growth
and development under field conditions. Am-Euras J Agric Envi-
ron Sci 7(4):374–385
17. Majumder A etal (2020) Influence of land use/land cover changes
on surface temperature and its effect on crop yield in different
agro-climatic regions of Indian Punjab. Geocarto Int 35(6):663–
686. https:// doi. org/ 10. 1080/ 10106 049. 2018. 15209 27
18. Jeev S, Verma P, Verma U (2018) Development of weather based
wheat yield forecast models in Haryana. Int J Curr Microbiol App
Sci 7(12):2973–2978. https:// doi. org/ 10. 20546/ ijcmas. 2018. 712.
340
19. Mukherjee A, Wang SYS, Promchote P (2019) Examination of
the climate factors that reduced wheat yield in northwest India
during the 2000s. Water (Switzerland) 11(2):1–13. https:// doi. org/
10. 3390/ w1102 0343
20. Agrawal DK, Nath S (2018) Effect of climatic factor and date of
sowing on wheat Crop in Allahabad condition, Uttar Pradesh. Int
J Curr Microbiol App Sci 7(09):1776–1782. https:// doi. org/ 10.
20546/ ijcmas. 2018. 709. 214
21. Jiayu Z etal (2018) The influence of meteorological factors on
wheat and rice yields in China. Crop Sci 58(2):837–852. https://
doi. org/ 10. 2135/ crops ci2017. 01. 0048
22. Epule TE etal (2018) The determinants of crop yields in Uganda:
what is the role of climatic and non-climatic factors? Agric Food
Secur 7(1):1–17. https:// doi. org/ 10. 1186/ s40066- 018- 0159-3
23. Nadew BB (2018) Effects of climatic and agronomic factors on
yield and quality of bread wheat (Triticum aestivum L.) seed: a
review on selected factors. Adv Crop Sci Technol 06(02):356.
https:// doi. org/ 10. 4172/ 2329- 8863. 10003 56
24. Zhao J etal (2017) Assessing the combined effects of climatic
factors on spring wheat phenophase and grain yield in Inner Mon-
golia, China. PLoS ONE 12(11):1–17. https:// doi. org/ 10. 1371/
journ al. pone. 01856 90
25. Meng T etal (2017) Analyzing temperature and precipitation
influences on yield distributions of canola and spring wheat in
Saskatchewan. J Appl Meteorol Climatol 56(4):897–913. https://
doi. org/ 10. 1175/ JAMC-D- 16- 0258.1
26. Safa M, Samarasinghe S, Nejat M (2015) Prediction of wheat
production using artificial neural networks and investigating indi-
rect factors affecting it: case study in Canterbury Province, New
Zealand. J Agric Sci Technol 17(4):791–803
27. Johnson DM (2014) An assessment of pre-and within-season
remotely sensed variables for forecasting corn and soybean yields
in the United States. Remote Sens Environ 141:116–128
28. Parekh FP, Suryanarayana TMV (2012) Impact of climatological
parameters on yield of wheat using neural network fitting. Int J
Mod Eng Res 2(5):3534–3537
29. Ruß G etal (2008) Data mining with neural networks for wheat
yield prediction. In: Lecture notes in computer science (including
subseries lecture notes in artificial intelligence and lecture notes in
bioinformatics), 5077 LNAI, pp 47–56. https:// doi. org/ 10. 1007/
978-3- 540- 70720-2_4.
30. Kamir E, Waldner F, Hochman Z (2020) Estimating wheat yields
in Australia using climate records, satellite image time series and
machine learning methods. ISPRS J Photogramm Remote Sens
160:124–135
31. Gonzalez-Sanchez A, Frausto-Solis J, Ojeda-Bustamante W
(2014) Predictive ability of machine learning methods for massive
crop yield prediction. Span J Agric Res 12(2):313–328. https://
doi. org/ 10. 5424/ sjar/ 20141 22- 4439
32. Ahamed ATMS etal (2015) Applying data mining techniques to
predict annual yield of major crops and recommend planting dif-
ferent crops in different districts in Bangladesh. In: 2015 IEEE/
ACIS 16th international conference on software engineering, arti-
ficial intelligence, networking and parallel/distributed computing,
SNPD 2015—proceedings. https:// doi. org/ 10. 1109/ SNPD. 2015.
71761 85
33. Lamba V, Dhaka VS (2014) Wheat yield prediction using artificial
neural network and crop prediction techniques (A Survey). Int J
Res Appl Sci Eng Technol 2:330–341
34. Nath B, Dhakre D, Bhattacharya D (2019) Forecasting wheat pro-
duction in India: An ARIMA modelling approach. J Pharmacogn
Phytochem 8(1):2158–2165
35. Kogan F etal (2013) Winter wheat yield forecasting in Ukraine
based on Earth observation, meteorological data and biophysical
models. Int J Appl Earth Obs Geoinf 23(1):192–203. https:// doi.
org/ 10. 1016/j. jag. 2013. 01. 002
36. Zhang Y etal (2018) Optimal hyperspectral characteristics
determination for winter wheat yield prediction. Remote Sens
10(12):1–18. https:// doi. org/ 10. 3390/ rs101 22015
37. Kim N, Lee YW (2016) Machine learning approaches to corn
yield estimation using satellite images and climate data: a case
of Iowa State. J Korean Soc Surv Geod Photogramm Cartogr
34(4):383–390. https:// doi. org/ 10. 7848/ ksgpc. 2016. 34.4. 383
38. Bose P etal (2016) Spiking neural networks for crop yield estima-
tion based on spatiotemporal analysis of image time series. IEEE
Trans Geosci Remote Sens 54(11):6563–6573. https:// doi. org/ 10.
1109/ TGRS. 2016. 25866 02
39. Pantazi XE etal (2016) Wheat yield prediction using machine
learning and advanced sensing techniques. Comput Electron Agric
121:57–65. https:// doi. org/ 10. 1016/j. compag. 2015. 11. 018
40. Kaul M, Hill RL, Walthall C (2005) Artificial neural networks for
corn and soybean yield prediction. Agric Syst 85(1):1–18. https://
doi. org/ 10. 1016/j. agsy. 2004. 07. 009

Emerging Trends inMachine Learning toPredict Crop Yield andStudy Its Influential Factors:…
1 3
41. Chlingaryan A, Sukkarieh S, Whelan B (2018) Machine learn-
ing approaches for crop yield prediction and nitrogen status
estimation in precision agriculture: a review. Comput Electron
Agric 151(November 2017):61–69. https:// doi. org/ 10. 1016/j.
compag. 2018. 05. 012
42. Jeong JH etal (2016) Random forests for global and regional
crop yield predictions. PLoS ONE 11(6):1–15. https:// doi. org/
10. 1371/ journ al. pone. 01565 71
43. Dai X, Huo Z, Wang H (2011) Simulation for response of crop
yield to soil moisture and salinity with artificial neural network.
Field Crops Res 121(3):441–449. https:// doi. org/ 10. 1016/j. fcr.
2011. 01. 016
44. Becker-Reshef I, Vermote E, Lindeman M, Justice C (2010)
A generalized regression-based model for forecasting winter
wheat yields in Kansas and Ukraine using MODIS data. Remote
Sens Environ 114(6):1312–1323
45. Ji B etal (2007) Artificial neural networks for rice yield pre-
diction in mountainous regions. J Agric Sci 145(3):249–261.
https:// doi. org/ 10. 1017/ S0021 85960 60066 91
46. Serele CZ, Gwyn QHJ, Boisvert JB, Pattey E, McLaughlin N,
Daoust G (2000) Corn yield prediction with artificial neural
network trained using airborne remote sensing and topographic
data. In: IGARSS 2000. IEEE 2000 international geoscience
and remote sensing symposium. Taking the Pulse of the Planet:
the role of remote sensing in managing the environment. Pro-
ceedings (Cat. No. 00CH37120), vol 1. IEEE, pp 384–386
47. Gandhi N, Petkar O, Armstrong LJ (2016) Rice crop yield pre-
diction using artificial neural networks. In: Proceedings—2016
IEEE international conference on technological innovations in
ICT for agriculture and rural development, TIAR 2016 (Tiar),
pp 105–110. https:// doi. org/ 10. 1109/ TIAR. 2016. 78012 22
48. Uno Y etal (2005) Artificial neural networks to predict corn
yield from Compact Airborne Spectrographic Imager data.
Comput Electron Agric 47(2):149–161. https:// doi. org/ 10.
1016/j. compag. 2004. 11. 014
49. Balaghi R etal (2008) Empirical regression models using
NDVI, rainfall and temperature data for the early prediction
of wheat grain yields in Morocco. Int J Appl Earth Obs Geoinf
10(4):438–452. https:// doi. org/ 10. 1016/j. jag. 2006. 12. 001
50. Cheng H etal (2017) Early yield prediction using image analysis
of apple fruit and tree canopy features with neural networks. J
Imaging 3(1):6. https:// doi. org/ 10. 3390/ jimag ing30 10006
51. Ghodsi R, Yani RM, Jalali R, Ruzbahman M (2012) Predict-
ing wheat production in Iran using an artificial neural networks
approach. Int J Acad Res Bus Soc Sci 2(2):34
52. Singh RK (2008) Artificial neural network methodology for
modelling and forecasting maize crop yield. Agric Econ Res
Rev 21(347-2016–16813):5–10
53. Alvarez R (2009) Predicting average regional yield and produc-
tion of wheat in the Argentine Pampas by an artificial neural
network approach. Eur J Agron 30(2):70–77. https:// doi. or g/ 10.
1016/j. eja. 2008. 07. 005
54. Park SJ, Hwang CS, Vlek PLG (2005) Comparison of adaptive
techniques to predict crop yield response under varying soil and
land management conditions. Agric Syst 85(1):59–81. https://
doi. org/ 10. 1016/j. agsy. 2004. 06. 021
55. Bal SK etal (2004) Wheat yield forecasting models for Ludhi-
ana district of Punjab state. J Agromet 6(January):161–165
56. Shastry KA, Sanjay HA, Deshmukh A (2016) A parameter
based customized artificial neural network model for crop yield
prediction. J Artif Intell 9(1–3):23–32. https:// doi. org/ 10. 3923/
jai. 2016. 23. 32
57. Bhangale PP, Patil PYS, Patil PDD (2017) Improved crop yield
prediction using neural network. IJARIIE 3(2):3094–3101
58. Bejo S, Mustaffha S, Wan Ismail W (2014) Application of arti-
ficial neural network in predicting crop yield: a review. J Food
Sci Eng 4(1):1–9
59. Dahikar SS, Rode SV (2014) Agricultural Crop Yield Prediction
Using Artificial Neural Network Approach. Int J Innov Res Electr
Electron Instrum Control Eng 2(1):2321–2004
60. Laxmi RR, Kumar A (2011) Weather based forecasting model for
crops yield using neural network approach. Stat Appl 9(1):55–69
61. Qaddoum K, Hines EL, Iliescu DD (2013) Yield prediction for
tomato greenhouse using EFuNN. ISRN Artif Intell 2013:1–9.
https:// doi. org/ 10. 1155/ 2013/ 430986
62. Khoshnevisan B etal (2014) Development of an intelligent system
based on ANFIS for predicting wheat grain yield on the basis of
energy inputs. Inf Process Agric 1(1):14–22. https:// doi. org/ 10.
1016/j. inpa. 2014. 04. 001
63. Naderloo L etal (2012) Application of ANFIS to predict crop
yield based on different energy inputs. Meas J Int Meas Confed
45(6):1406–1413. https:// doi. org/ 10. 1016/j. measu rement. 2012. 03.
025
64. Kouchakzadeh M, Ghahraman B (2011) ‘Ar’, 13, pp 627–640
65. Pandey AK, Sinha AK, Srivastava VK (2008) A comparative
study of neural-network & fuzzy time series forecasting tech-
niques-case study: wheat production forecasting. Int J Comput
Sci Netw Secur 8(9):382–387
66. Balakrishnan N, Muthukumarasamy G (2016) Crop production—
ensemble machine learning model for prediction. Int J Comput Sci
Softw Eng 5(7):148–153
67. Priya P, Muthaiah U, Balamurugan M (2018) Predicting yield
of the crop using machine learning algorithm. Int J Eng Sci Res
Technol 7(1):1–7
68. Manjula E, Djodiltachoumy S (2017) A model for prediction of
crop yield. Int J Comput Intell Inform 6(4):298–305
69. Preethaa KS, Nishanthini S, Santhiya D, Shree KV (2016) Crop
yield prediction. Int J Eng Technol Sci III:111–116
70. Ingole K, Katole K, Shinde A, Domke M (2013) Crop predic-
tion and detection using fuzzy logic in MATLAB. Int J Adv Eng
Technol 6(5):2006
71. Garg B, Aggarwal S, Sokhal J (2018) Crop yield forecasting using
fuzzy logic and regression model. Comput Electr Eng 67:383–
403. https:// doi. org/ 10. 1016/j. compe leceng. 2017. 11. 015
72. Kumar P (2011) Crop yield forecasting by adaptive neuro fuzzy
inference system. Math Theory Model 1(3):1–7
73. Schmidhuber J (2015) Deep Learning in neural networks: an over-
view. Neural Netw 61:85–117. https:// doi. org/ 10. 1016/j. neunet.
2014. 09. 003
74. Kamilaris A, Prenafeta-Boldú FX (2018) Deep learning in agri-
culture: a survey. Comput Electron Agric 147(February):70–90.
https:// doi. org/ 10. 1016/j. compag. 2018. 02. 016
75. Amara J, Bouaziz B, Algergawy A (2017) A deep learning-
based approach for banana leaf diseases classification. Daten-
banksysteme für Business, Technologie und Web (BTW
2017)-Workshopband
76. Francis M, Deisy C (2020) Mathematical and visual understand-
ing of a deep learning model towards m-agriculture for disease
diagnosis. Arch Comput Methods Eng 1–17
77. Newlands N, Ghahari A, Gel YR, Lyubchich V, Mahdi T (2019)
Deep learning for improved agricultural risk management. In: Pro-
ceedings of the 52nd Hawaii international conference on system
sciences
78. Kuwata K, Shibasaki R (2015) Estimating crop yields with deep
learning and remotely sensed data. In: 2015 IEEE international
geoscience and remote sensing symposium (IGARSS). IEEE, pp
858–861
79. Cunha RLF, Silva B, Netto MAS (2018) A scalable machine
learning system for pre-season agriculture yield forecast. In:

N.Bali, A.Singla
1 3
Proceedings—IEEE 14th international conference on eScience,
e-Science 2018, pp 423–430. https:// doi. org/ 10. 1109/ eScie nce.
2018. 00131.
80. You J etal. (2014) Deep Gaussian process for crop yield predic-
tion based on remote sensing data, pp 4559–4565
81. Wang AX, Lobell D, Ermon S (2015) Deep transfer learning for
crop yield prediction with remote sensing data
82. Villanueva MB, Salenga MLM (2018) Bitter melon crop yield
prediction using Machine Learning Algorithm. Int J Adv Comput
Sci Appl 9(3):1–6. https:// doi. org/ 10. 14569/ IJACSA. 2018. 090301
83. Fourie J, Hsiao J, Werner A (2017) Crop yield estimation using
deep learning. In: 7th Asian-Australasian conference on precision
agriculture, pp 1–10
84. Bargoti S, Underwood JP (2017) Image segmentation for fruit
detection and yield estimation in apple orchards. J Field Robot
34(6):1039–1060. https:// doi. org/ 10. 1002/ rob. 21699
85. Kuwata K, Shibasaki R (2016) Estimating Corn Yield in the
United States With Modis Evi and Machine Learning Methods.
ISPRS Ann Photogramm Remote Sens Spat Inf Sci 3(8):131–136.
https:// doi. org/ 10. 5194/ isprs annals- iii-8- 131- 2016
86. Mohan P, Patil KK (2018) Deep learning based weighted SOM to
forecast weather and crop prediction for agriculture application.
Int J Intell Eng Syst 11(4):167–176. https:// doi. org/ 10. 22266/ ijies
2018. 0831. 17
87. Jiang Z etal (2018) Predicting county level corn yields using deep
long short term memory models. http:// arxiv. org/ abs/ 1805. 12044
Publisher’s Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.