2021 2nd International Conference for Emerging Technology (INCET)

Belgaum, India. May 21-23, 2021

Machine Learning Approaches Used for Weather

Attributes Forecasting
Rudrappa B Gujanatti Nataraj Vijapur Sushant S Jadhav
Department of ECE Department of ECE Department of ECE
KLE Dr. M S Sheshgiri College of RV College of Technology and KLE Dr. M S Sheshgiri College of
Engineeringand Technology Management Engineeringand Technology
(VTU) Bangalore, India (VTU)
Belagavi, India [email protected] Belagavi, India
[email protected] [email protected]

Prabhakar Manage Anand Konnur

Department of ECE Department of ECE
KLE Dr. M S Sheshgiri College of KLE Dr. M S Sheshgiri College of
Engineeringand Technology Engineeringand Technology
(VTU) (VTU)
Belagavi, India Belagavi, India
[email protected] [email protected]

Abstract—In the not so distant future, weather is the state desire for a system which enhances the efficiency of
of the environment; and it forecasts the state of the forecasting; these models can then later be used in above
atmosphere. Weather prediction is a difficult challenge since mentioned applications. Weather forecasting is significant in
the environment is an unbroken, multi-dimensional process. composition and checking out our day by day programs.
Satellites, Doppler radar, radio signals, aircraft and earth Based on the forecasting information we can take call for
senses gain explicitly environment conditions. Different factual specific tasks on a given day. There's a desire for a system
and AI methods are essential for gathering knowledge. These that improves the prognostication results which can be used
strategies can blend an exceptionally complex mechanized in the above-mentioned applications. Weather prediction
numerical model with the moderately simple perception of the
techniques are based on sky perception, victimization of
sky. Weather determination is also a difficult problem, as the
weather is unpredictable and turbulent. Indeed, even though
microwave signals, use of satellite imageries or ground-
the system for weather guidance is in the present techniques, based imageries, and Machine Learning (ML) techniques [6 -
weather consistency can in any case be ignored and these 9]. The research by the specialist is not exclusively for
frameworks are therefore still to be improved. The goal of the current weather patterns but also foresees the weather for a
study is to explore how AI can help us improve the estimation longer term [10]. All the sky pictures have become a fact
of climatic boundaries. In this paper we discuss the different among different scientific communities. It finds it’s
strategies used by driving experts, enforce challenges and application is many domains such as farming, surveillance,
present our perspectives on improving the successful system etc [9]. Studies are carried out for several aspects of weather
for weather control. forecasts [8]. The regional data analysis reveals that the
environment tends to be complex at a specific site and that
Keywords— Machine Learning, Artificial Intelligence, time shifts provide extravagant data and will impact wider
Weather forecasting, Image processing areas [7]. The findings are more comprehensive. The
predictions are divorced into 6 different categories: [3, 5]
I. INTRODUCTION ranging from Now casting (NC)to long range forecasting.
In any technical and scientifically disruptive issue the The accuracy related to forecasting of the above-mentioned
meteorological forecast is one of the most relevant in the methods moves from very high accuracy to very low
world [1]. Complicated mathematical models were accuracy respectively.
accelerated due to the advent of the technologies to get
precise predictions. The introduction of mainly dependent II. LITERATURE SURVEY
models of machine learning results in obtaining higher and The authors of [6] presented prediction of temperature
better results [2 - 4]. Recent meteorological characteristics using Convolutional Neural Network (CNN). If single image
are not only extremely complicated, but also extremely is to be used for temperature prediction then CNN is
quantitative. Different forms the weather area unit used for employed and if the temperature prediction uses series of
forecasting: synoptic, numerical and applied mathematical images the RNN (Recurrent Neural Network) is used.
methods [5]. Dataset form [11] and from [12] were used for training and
The composition and testing of our regular programs is testing the NN (Neural Networks). In [7] the authors applied
critical in weather forecasts. Weather forecast perfect for a RF (Random Forest) classifiers to estimate weather data
given day, i.e., weather prediction, will affect our day-to-day from single image. In [8] authors presented cooperative
activities. An honest forecast model will be used in timely learning approach for classification of weather, and estimate
and sufficient meteorology to minimize damage and loss [2]. whether the weather is sunny or cloudy in nature. This was
In the majority of applications such as agriculture, severe achieved with the help of CNN and “Sun Dataset” [13],
weather alert systems, air traffic controlling, etc., there is a “Labelme Dataset” [14] and “Flickr” were used. In [9]

978-1-7281-7029-9/21/$31.00 ©2021 IEEE 1

authors used “multi-feature texture analysis” and built a IV. TYPES OF WEATHER ATTRIBUTES FORECASTING
standard NNs system. This was later used for weather Different methods of weather forecasting have been
analysis with the help of the machine-controlled incorporated from the ancient times which relied in
interpretation of satellite cloud pictures for classification. In observing the sky to present day systems which are making
[15] authors proposed cluster learning primarily based for use of artificial intelligence. Below are few methods used for
weather estimation model to assist forecast road hazards weather attributes forecasting.
because of completely different climatic conditions. This was
achieved with the help of CCTVs placed at different A. Persistence Weaher Forecasting
locations. In [16] authors presented a technique for automatic This is the simplest method of weather attribute
cloud detection with the help of superpixel segmentation forecasting. It assumes that the weather will be the same
(SPS). tomorrow as it will be today. Such method is suitable for
very short projections of weather attributes forecasting.
III. PROBLEMS FACED IN WEATHER FORECASTING
A significant objective of any research endeavor is to B. Trend Weather Forecasting
predict correctly. In the hope of identifying some In this method the weather forecasting relies on finding
fundamental principles which can be used for predicting the the information related to wind speed, wind direction,
outcomes of certain experiments based on these principles. In atmospheric pressure, etc., and determining the where these
addition, most science laws only provide very detailed features will be identical in near future. Such kind of weather
forecasts of the outcomes of certain types of experiments. forecasting method is generally suitable for nowcasting. This
However, few physicians face more nuanced or complicated weather forecasting method relies on numerical information.
forecast issues than the meteorologist. First and foremost, the
meteorological laboratory encompasses the entire world, C. Analogue Weather Forecasting
making it incredibly difficult to calculate the actual condition This method of weather attributes forecasting involves
of the atmosphere. Moreover, the Earth's surface is a mixture relating the information regarding weather today and
of soil and water that react differently to the source of correlating it to the preciously occurred weather incident.
energy. The atmosphere itself is, then too, a mixture of The results of this method would give an information as to
gasses, liquids and solid components, all of which have an weather will be behaving the same way as it had behaved in
effect on the energy balance of the planet. The circulations in the past incident.
the atmosphere are often incredibly broad, with a life span of
D. Statistical Weather Forecasting
just a few seconds, and can last for weeks or months or
minute whirls. This method of statistical weather attributes forecasting
involves use of historical information and applying statistical
Apart from the above-mentioned issues we also face analysis in the information and then predict what would
different challenges when we want to model weather happen in future.
attributes forecasting. Challenges involve tasks such as
E. Numerical Weather Prediction (NWP) Method
 Identifying the relation between various atmospheric
NWP method uses models which are developed with the
parameters. This task becomes challenging as this
help of programming languages are used to achieve weather
consists of relating non-linear aspects for weather
attributes forecasting. Such method can help us overcome the
attributes forecasting. To present day many forecasts
non-linearity present between the weather parameters to
still fail because of the non-linearity present in the
some extent.
information required for weather attribute forecasting.
The methods mentioned above are presently combined
 Data obtained for weather attributes forecasting can with modern day Artificial Intelligence (AI) to obtain better
turn out to be imperfect at sometimes. This may be weather attributes forecasting results. The AI projections are
the due to the data being mis read, or data being mainly intended to use techniques such as Artificial Neural
corrupted, or sensor problem imposing data to be Networks (ANN), Machine Learning (ML), MultiLayer
imperfect. Perceptron (MLP), etc. This helps us to process complicated
 When image processing related to ground based information in a minimal amount of time using different
images and satellite images is involved in the weather concepts. This process helps us achieve better precision. ML
attributes forecasting then it involve challenges for can handle non-linear information very well hence many
pre-processing the images for extracting the features forecasting methods look forward to use ML techniques.
from the image. Here we may face challenges in
processing the images. For example, if we want to V. METHODOLOGY TO BE IMPLEMENTED
process the ground based image for cloud The major motivation of the presented methodology here
classification then the intensity present in the image is to implement a method for weather attribute forecasting.
may pose challenges in processing and classifying the Later using the forecasted attributes can be used in various
clouds into low level, middle level, and high level fields such as aviation, military, agriculture, etc. For the
clouds. Resolution of the image acquired also plays a weather attribute forecasting purpose, the meteorological
vital role in such types of image classification. data along with images related to sky images and/ satellite
images can be used. The block diagram of the planned
methodology is shown in Fig. 1.

2
 pressure for these clouds is in the range from 680 – 1000 Pa.
The cloud optical thickness for Cumulus Clouds ranges from
0 – 3.6 tau, for Stratocumulus Clouds it ranges from 3.6 – 23
tau and for Stratus Clouds it ranges from 23 – 379.

Fig. 1. Block diagram of proposed method

A. Meteorological Data
This includes ground based image of clouds, satellite
images and the statistical parameters related to weather
attribute (rainfall) forecasting. Datasets such as presented in
[11], [17], [12], etc., could also be utilized for training and
testing functions.
B. Preprocessing
Pre-processing involves data cleaning of the statistical
parameters, pre-processing the images (satellite and ground
based images) to remove noise if any and process the image
for further cloud classification approach that will be used for
rainfall forecasting [3-4]. Image Pre-processing step can Fig. 2. Cloud classification
involve measures to improve the image quality by filtering
out unwanted distortions and making the image ready for C. Database Creation (Traing Data and Test Data)
extracting its feature in future steps. In [18] and [19] authors The meteorological data obtained from to image pre-
used Satellite imagery. In [20] authors used aerial processing and the statistical parameters need to be train and
photographs, especially, for large-scale surveys. Images test the model needs to be stored which will act as database.
captured can be of very high quality because of the This data in the database will be used for training and testing
advancements in imaging technologies and the devise used the model developed for rainfall forecasting.
for image acquisition [21]. The image processing outputs
along with other meteorological will be used for weather D. Training and Testing the Prediction System Using
attribute forecasting. Fig. 2 [28] shows the types of clouds Machine Learning
that need to be categorised. We need to pre-process the Once the database is created with all the meteorological
ground based images in order to classify the clouds as shown information stored in it, the same information has to be used
in Fig. 2. The categories of clouds mentioned fall into for training the machine learning model which will be used
following broad classes for prediction of rainfall. To verify the accuracy of the model
developed we need to test the model. For these two purposes
 High Level Clouds the meteorological data stored in database has to be divided
These cloud types are basically above 6KM (20,000 ft) into training and testing data. In this process prior to testing
from ground level. These include subtypes of clouds which we may need to run optimization algorithms such as Jaya
are Cirrus Clouds, Cirrocumulus Clouds and Cirrostratus improvement rule (JOA) [26], Rider improvement rule
Clouds (Deep Convective Clouds). From Fig. 2 we observe (ROA) [27], etc., on the data stored in database.
that the cloud top pressure for these clouds is in the range E. Forecasting Rainfall Precipitation
from 50 – 440 Pa. The cloud optical thickness for Cirrus
Clouds ranges from 0 – 3.6 tau, for Cirrocumulus Clouds it The accuracy of the model developed needs to be tested
ranges from 3.6 – 23 tau and for Cirrostratus Clouds (Deep with ground truth results. This can give us useful information
Convective Clouds) it ranges from 23 – 379. regarding the steps to be included to enhance the prediction
results of rainfall. Why we choose for rainfall forecasting?
 Middle Level Clouds Rainfall affects multiple domains or fields such as
agriculture, aviation section, military applications, etc., since
These clouds are fall in the range of 2KM (6,500 ft) to
the impact of rainfall is high and of significance importance,
6KM (20,000 ft) from ground level. These include subtypes
we attempt to forecast rainfall once the model is developed
of clouds which are Altocumulus Clouds, Altostratus Clouds
using machine learning techniques. One of the major
and Nimbostratus Clouds. From Fig. 2 we observe that the
dependencies on rainfall we observe is in the field of
cloud top pressure for these clouds is in the range from 440
agriculture. In agriculture if we provide precise information
– 680 Pa. The cloud optical thickness for Altocumulus
regarding the rainfall then farmers can take advantage of
Clouds ranges from 0 – 3.6 tau, for Altostratus Clouds it
such systems. The overall crop productivity can be increased,
ranges from 3.6 – 23 tau and for Nimbostratus Clouds it
the loss due to unprecedented rainfall can also be avoided or
ranges from 23 – 379.
minimised, crop health can also improved by cultivating the
 Low Level Clouds crop based in the rainfall prediction, etc.
These clouds are fall in the range of 0KM to 2KM (6500 VI. CONCLUSION
ft) from ground level. These include subtypes of clouds
The review study found that it was very difficult for
which are Cumulus Clouds, Stratocumulus Clouds and
linear systems to forecast rainfall, hence non-linear approach
Stratus Clouds. From Fig. 2 we observe that the cloud top
which involves the use of machine learning techniques will

3
be helpful in prediction weather attributes. The bulk of the [12] Wei-Ta C, K Ho, and Ali B, "Visual Weather Temperature
form of Mathematics applied ends up in linear model Prediction," Proceedings of IEEE Winter Conference on Applications
of Computer Vision, pp. 234-241, 2018.
modification and machine learning operates under nonlinear
[13] Jianxiong X, James H, Krista A E, Aude O, and Antonio T, “SUN
models. It is difficult to understand the non-linearity or one- Database: Large-scale Scene Recognition from Abbey to Zoo," IEEE
dimensionality of mathematical knowledge if the drag Conference on Computer Vision and Pattern Recognition, 2010.
includes universal knowledge. Interval could not be a grave Available:
concern in declaration models, but accuracy can be a big https://www.cc.gatech.edu/~hays/papers/sun.pdf
challenge. Hence, huge number of advanced models are [14] Bryan C R, Antonio T, Kevin P M, and William T F, “LabelMe: a
being built to obtain higher precision. The different database and web-based tool for image annotation,” Int. Jour. of
prediction techniques / methods were studied from the Computer Vision, 2007.
papers. Together, it appears to assume that the form of [15] Jiwan L; Bonghee H; Sunghoon J; Victor C, “Clustering learning
application of mathematics ends up adjusting to linear model of CCTV image pattern for producing road hazard
meteorological information,” Future Generation Computer Systems,
models in most situations, with machine learning techniques ISSN: 0167-739X, Volume 86, pp. 1338-1350, 2018.
suitable for under nonlinear approaches. Later the results of [16] Shuang L, Linbo Z, Zhong Z, Chunheng W, and Baihua X,
the rainfall prediction system can be used in various fields "Automatic Cloud Detection for All-Sky Images Using Superpixel
such as aviation, agriculture, etc. In this paper we also Segmentation," in IEEE Geoscience and Remote Sensing Letters,
propose to use image processing to classify the clouds as low volume 12, issue 2, pp. 354-358, February 2015.
level clouds, medium level clouds and high level clouds with [17] Nathan J, Walker B, Nick F, Austin A, Kylia M, Bobby H B, Andrew
the help of ground based images; this information along with D R, and Robert P, “The Global Network of Outdoor Webcams:
Properties and Applications,” in ACM International Conference on
other weather parameters we propose to develop a weather Advances in Geographic Information Systems, pp.111–120.
attribute forecasting model. The attribute of prime https://dl.acm.org/citation.cfm?id=1653789
importance to be considered will be rainfall or precipitation.
[18] Tomohiro S, Takeshi M, Rajesh B T, Manabu W, and Masanobu S,
This attribute is considered as it has major impact various April 2014, "Comparative Assessment of Supervised Classifiers for
activities related to the fields or domains such as agriculture, Land Use–Land Cover Classification in a Tropical Region Using
aviation section, military applications, etc. Time-Series PALSAR Mosaic Data," in IEEE Jour. of Selected
Topics in Applied Earth Observations and Remote Sensing, volume 7,
REFERENCES issue 4, pp. 1186-1199.
[19] Ik-Hyun L, and Muhammad T M, "Robust Registration of Cloudy
[1] Garima J and Bhawna M, “A Review on Weather Forecasting
Satellite Images Using Two-Step Segmentation," in IEEE Geoscience
Techniques,” in International Journal of Advanced Research in
and Remote Sensing Letters, volume 12, issue 5, pp. 1121-1125, May
Computer and Communication Engineering, Vol. 5, no. 12, pp. 177–
2015.
180, December, 2016.
[20] Qing Z, and Chunxia X, "Cloud Detection of RGB Color Aerial
[2] H. Saima, J. Jaafar, S. Belhaouari and T. A. Jillani, "Intelligent
Photographs by Progressive Refinement Scheme," in IEEE
methods for weather forecasting: A review," 2011 National
Transactions on Geoscience and Remote Sensing, volume 52, issue
Postgraduate Conference, Kuala Lumpur, pp. 1-6, 2011.
11, pp. 7264-7275, November 2014.
[3] Gujanatti R, Nataraj V, “Intelligent Methods Used for Obtaining
[21] Soumyabrata D, Yee H L and Stefan W, "Color-Based Segmentation
Weather Derivatives: A Review,” Engineering and Applied Sciences.
of Sky/Cloud Images From Ground-Based Cameras," in IEEE Jour. of
Vol. 4, No. 6, 2019, pp. 144-148. doi: 10.11648/j.eas.20190406.12
Selected Topics in Applied Earth Observations and Remote Sensing,
[4] Gujanatti Rudrappa and Nataraj Vijapur, “Cloud Classification using volume 10, issue 1, pp. 231-242, January 2017.
K-Means Clustering and Content Based Image Retrieval Technique,”
[22] C. Papin, P. Bouthemy and G. Rochard, "Unsupervised segmentation
International Conference on Communication and Signal Processing –
of low clouds from infrared METEOSAT images based on a
2020 (iccsp - 20), July 28 - 30, 2020, India
contextual spatio-temporal labeling approach," in IEEE Transactions
[5] V K Didal, Brijbhooshan, Anita T and K Choudhary, "Weather on Geoscience and Remote Sensing, vol. 40, no. 1, pp. 104-114, Jan.
Forecasting in India: A Review", Intertnational Journal of Current 2002.
Microbiology and Applied Science, Vol. 6, no. 11, pp-577-590, 2017.
[23] Majid M, Nicolas H Y, Valentine G A, James A, and Shantia Y, "On
[6] Wei-Ta C, Kai-Chia H and Ali B, "Visual Weather Temperature the Use of a Cluster Ensemble Cloud Classification Technique in
Prediction," 2018 IEEE Winter Conference on Applications of Satellite Precipitation Estimation," in IEEE Journal of Selected
Computer Vision (WACV), Lake Tahoe, NV, pp. 234-241, Topics in Applied Earth Observations and Remote Sensing, volume 5,
2018.Available: issue 5, pp.1356-1363, October, 2012.
https://arxiv.org/abs/1801.08267 [24] J. M. Weiss, S. A. Christopher and R. M. Welch, "Automatic contrail
[7] Wei-Ta C, Xiang-You Z, Ding-Shiuan D, “Camera as weather sensor detection and segmentation," in IEEE Transactions on Geoscience
Estimating weather information from single images,” Journal of and Remote Sensing, volume 36, issue 5, pp. 1609-1619, September
Visual Communication and Image Representation, Vol. 46, pp-233- 1998.
249, July 2017. [25] Andrei B, Elena B, "Meteorological Data Analysis and Prediction by
[8] Cewu L, , Di L, J Jiaya J and Chi-Keung T, "Two-Class Weather Means of Genetic Programming," Proceedings of the Fifth Workshop
Classification," IEEE Transactions on Pattern Analysis and Machine on Mathematical Modeling of Environmental and Life Sciences
Intelligence, Volume 39, Issue 12, pp. 2510-2524, 1 December 2017. Problems Constant¸a, Romania, pp. 35–42, 2006.
[9] Christodoulos I. C, Silas C. M, and Constantinos S. P, "Multifeature [26] R. Venkata Rao, "Jaya: A simple and new optimization algorithm for
texture analysis for the classification of clouds in satellite imagery," solving constrained and unconstrained optimization problems," in
IEEE Transactions on Geoscience and Remote Sensing, volume 41, International Journal of Industrial Engineering Computations, volume
Issue 11, pp. 2662-2668, November 2003. 7, pp.19–34, 2016.
[10] Sandeep K M, Anamika U and Channabasva G, "Rainfall prediction [27] D. Binu and B. S. Kariyappa, "RideNN: A New Rider Optimization
based on 100 years of meteorological data," 2017 Int. Conf. on Algorithm-Based Neural Network for Fault Diagnosis in Analog
Computing and Communication Technologies for Smart Nation, Circuits," in IEEE Transactions on Instrumentation and Measurement.
Gurgaon, pp. 162-166, 2017. doi: 10.1109/TIM.2018.2836058.
[11] Radu P M, Scott W, Zack B, Nathan J, “Sky Segmentation in the [28] Website URL:
Wild: An Empirical Study,” in IEEE Winter Conference on https://www.google.com/search?q=cloud+classification&rlz=1C1CH
Applications of Computer Vision, pp-1–6. Available: BD_enIN871IN871&sxsrf=ALeKk00O75mkiD-YoHoISj96-
http://cs.uky.edu/~jacobs/papers/mihail2016skyfinder.pdf OCydndG0A:1598520640450&tbm=isch&source=iu&ictx=1&fir=iw
D9Hu2s7MN2cM%252CSuiPrLHw7WSY0M%252C_&vet=1&usg=
AI4_-kQ30e2-AMFTSBZa_-

4
 k5DTn8czV2ZA&sa=X&ved=2ahUKEwjOqcK3ibvrAhVxIbcAHR9 D9Hu2s7MN2cM%252CSuiPrLHw7WSY0M%252C_&vet=1&usg=
_AbcQ_h0wAnoECAwQBg&biw=1229&bih=578#imgrc=P7kbhB5q AI4_-kQ30e2-AMFTSBZa_-
cq1J-M (Accessed on 20-08-2020) k5DTn8czV2ZA&sa=X&ved=2ahUKEwjOqcK3ibvrAhVxIbcAHR9
[29] Website _AbcQ_h0wAnoECAwQBg&biw=1229&bih=578#imgrc=YyMej3x
https://www.google.com/search?q=cloud+classification&rlz=1C1CH SuH24gM (Accessed on 22-08-2020)
BD_enIN871IN871&sxsrf=ALeKk00O75mkiD-YoHoISj96-
OCydndG0A:1598520640450&tbm=isch&source=iu&ictx=1&fir=iw