Download Deep Learning technique in Weather Forecasting and more Study Guides, Projects, Research Software Engineering in PDF only on Docsity!
Cardiff Metropolitan University
Assignment Cover Sheet
Student Details ( Student should fill the content) Name Aravinda Dharmalingam Student ID CL/BSCSD/18/38^ st Scheduled unit details Unit code Unit title Unit enrolment details Year 3 Study period 2019 Lecturer Mode of delivery Full Time Assignment Details Nature of the Assessment Topic of the Case Study (^) Deep Learning in Weather forecasting – CW 2 Learning Outcomes covered Word count Due date / Time 12/ 14 / Extension granted? Yes No Extension Date Is this a resubmission? Yes No Resubmission Date Declaration I certify that the attached material is my original work. No other person’s work or ideas have been used without acknowledgement. Except where I have clearly stated that I have used some of this material elsewhere, I have not presented it for examination / assessment in any other course or unit at this or any other institution Name/Signature Date Submission Return to: Result Marks by 1st Assessor
Signature of the 1st^ Assessor Agreed Mark
Marks by2nd Assessor
Signature of the 2nd^ Assessor
Comments on the Agreed Mark.
For Office use only (hard copy assignments) Receipt date Received by
Assignment 2 This assignment is worth 50% of the overall assessment for this module. STUDENT NAME: STUDENT NUMBER:
Module Number & Title : Semester:
Assignment Type & Title: For student use: Critical feedback on the individual progression towards achieving the assignment outcomes
For the Assessors’ feedback Indicate the Task number strength and Weaknesses and the marks for each task Task No/Questi on No
Strengths (1st Assessor) Strengths (2nd^ Assessor)
CMU B.Sc. (HONS) SE- ASSIGNMENT FEEDBACK SHEET – ICBT CAMPUS
Contents
- Acknowledgement
- Introduction
- Introduction to the concept of deep leaning
- The deep learning techniques and methods used in weather forecast
- Climate Models
- Grid enabled integrated earth system model (GENIE)
- Complexity.............................................................................................................................
- Artificial Neural Network
- Convolutional Neural Network
- Recurrent Neural Network and LSTM
- The summary of selected research papers
- Conclusion
- References
Acknowledgement
For the outcome of my assignment to be a success it required a great deal of guidance and tips from many expertise people whom without their assistance I wouldn’t have been able to complete the assignment with success.
First and foremost, I would like to express my gratitude towards my dear lecturer Ms. Ravindi De Silva for expertly guiding me and providing me tips on my studies, I would like to thank her whole heartedly for her support and assistance throughout the completion of my assignment. My appreciation and heartfelt thanks and gratitude also extends to my parents for supporting and been by my side while providing me all the resources required to complete my assignment. Lastly, not forgetting my dear friends for their insights which meant a lot in making my assignment a success.
Without their help this assignment would have been impossible to complete. So once again I would like to make this the opportunity to thank everyone, Thank You!
Introduction
Deep learning allows computational models to learn by gathering knowledge from experience. Complex concepts can be learnt by deep learning approach due to its hierarchical conceptualization. Deep learning has significantly benefitted for the state-of-the-art in many recurring domains in the modern world. Deep Learning has an immense influence on fields such as computer vision, object detection, object recognition and speech recognition.
Deep leaning is one of the best fitted methods of implementing face recognition systems. There are various architectures used in the face recognition domain. In this report, a detailed description and introduction to the concept of deep learning is provided. Also a summary of two different deep learning methods used in the weather forecasting domain is evaluated. The report also consists of a detailed analyze of reputed research papers done on the same field.
Weather forecasting
Weather determining is a use of science and modernization to understand the status of the air for given range of space and time period. Climate estimates are produced by collecting significant and measurable data about the current circumstances of the environment at a estimated spot and handling climate forecast to expand how air will transform. (Shchur, 2019)
Climate information for learning shared portrayals utilizing verifiable information and anticipating climate components for various client characterized climate stations at the same time in a start to finish style. The inserted highlight learning part of the models just as coupling the educated highlights of various info layers have appeared to significantly affect the forecast errand. (SiamakMehrkanoon, 2019)
The deep learning techniques and methods used in weather forecast
Climate model
To make lengthy atmosphere model will run of various complexity, they utilized the Planet Simulator transitional complexity GCM, and the dynamical Centre. Each model was run for a long time with two diverse flat goals and 10 steep heights. Initial 3 decade of every run was disposed of as turn up, leaving 8 centuries of every day information for every model. Runs will starting now and into the foreseeable future be alluded. The PUMA runs do exclude sea, utilize the Newtonian for warming or cooling and orography. PLASIM runs incorporate orography, yet no sea model. Primary distinction among PLASIM and genuine GCMs is a substitute-frameworks of the world framework other than an atmosphere. (Scher,
1. Grid Enabled Integrated Earth (GENIE)
This model has been created as a secluded system that takes into consideration a range of transitional intricacy. Earth framework models to be made by choosing various alternatives for the different atmosphere and carbon cycle segments. Earth framework models made within GENIE have been arranged for distributed examinations spreading over a wide scope of land ages crosswise over Paleozoic, Mesozoic and Cenozoic Eras. GENIE system models are regularly equipped for mix over multi-millennial timescales and a few of the distributed investigations have included a great many long periods of re- enactment time, consolidating long runs and huge outfits. The system has been intended to be secluded to encourage the coupling of increasingly complex part modules as accessible figuring power increments. Invariably, utilizations of GENIE have utilized arrangements that speak to the climate with a computationally quick vitality dampness balance model. (Holden, 2015)
subtleties into more significant level structures. The transient measurement can be added to these systems by adding a third hub to the convolutional bits. This work shows how CNNs can be used to decipher the yield of Numerical Weather Prediction (NWP) consequently to create neighborhood conjectures. CNNs can give a model to decipher numerical climate model fields legitimately and to create nearby climate conjectures. (Dou, 2018)
LSTM and Recurrent Neural Network (RNN)
A repetitive neural framework (RNN) is a sub of ANN where associations between units structure a planned diagram with an arrangement. This empowers to present dynamic fleeting behavior for a period game plan. This is far-fetched with feed forward method system, intermittent neural system can utilize its mind from inside stockpiling to proceeding grouping of information sources. Climate estimating model, utilized the intermittent neural system with LSTM calculation basically expects to assemble information that is climate parameters, similar to temperature, mugginess, force, dew point, wind pace, drizzle and perceivability. Those are taken as neurons of contribution to repetitive neural system. Climate determining is finished by gathering data related to present day climate with respect to the past and the current state of the climate and also using these data to prepare LSTM model. (Subashin, 2019)
The summary of selected research papers
Climate forecast utilizing deep learning methods has livened up the consideration of numerous people in the IT people group, in this way numerous meeting papers, inquire about papers and exchanges were expounded on the issue. Among them I have picked three research papers to help my report.
“Advance climate forecasting prediction using deep learning”, a research paper written by namely A.Subashini, S.M.Thamarai and Dr.T.Meyyappan from Department of computer application in IJRAS and Engineering Technology. They discussed how weather forecast prediction can be done with the usage of deep learning techniques like recurrent neural system, LST memory connections are the supporting units to the RNN. As per the researches the repetitive neural system with LSTM calculation basically expects to assemble information that is climate parameters, similar to temperature, mugginess, force, dew point, wind pace, drizzle and perceivability. Those factors are considered as neurons of contribution to intermittent neural system. Climate anticipating is finished by gathering data related to present day climate with respect to the past and the current state of the climate and also using this data to prepare LSTM model.
In their experiment the chronicled climate information is provided by Dark Sky (Weather information Provider) from November 2008 to November 2018 decade of information utilized for preparing. 2019 information utilized for examination. This dataset include more climate qualities like climate condition, mugginess, dew, weight, perceivability and weather condition. Those qualities are offered data to the neural system and prepared utilizing LSTM calculation .The test result presents that a suitable precision could be accomplished utilizing LSTM method. (A. Subashini, 2019)
The second paper is “ANN effect for weather forecasting using back propagation” by a bunch of authors namely, B.Syam, N.Raja Nayak, K.Vagdhan Kumar, Ch.Jyosthna Devi and Prasad Reddy from computer science department in (IJETT) International journal of engineering trends. In this paper a normal neural system comprises of tiers. In an individual layered system there is an information tier of source hubs also, a yield tier of neurons. A multi-tier organize in Expansion at least one concealed tier. A multi-tier neural arrange is shown in additional shrouded neurons provide the system's capacity to extricate highest request measurements from information. Besides a system is commanded to be completely associated if each hub in every
Conclusion
Over the years weather forecast has gained great acceptance and importance. The climate expectation area has consistently been exposed to many research and exchange matters in PC vision. With the development of deep learning strategies climate forecast has accomplished a colossal progression. This is because of the unfathomable learning limit of deep learning techniques. As of now climate forecast frameworks executed with deep learning techniques are utilized worldwide because of its strength and precision. In my perspective the explanation for this tremendous intrigue for climate expectation frameworks actualized with deep learning methods is non-meddling.
In my point of view utilizing deep learning procedures in the climate expectation area has upgraded a monstrous productivity and precision level while additionally making frameworks that are hearty on discovery of rain, storm, windy and sunny with the guide of deep learning methods the climate forecast space has arrived at its top with critical execution enhancements.
References
A. Subashini, S. D., 2019. Advanced Weather Forecasting Prediction, Karaikudi: International Journal for Research in Applied Science & Engineering Technology (IJRASET). Ch.Jyosthna Devi, B. P. R. K. K. B. R. R. n., 2012. ANN Approach for Weather Prediction, Andhra: International Journal of Engineering Trends and Technology. Devi, C., 2012. ANN Approach for Weather Prediction, Andhra: International Journal of Engineering Trends and Technology. Dou, J., 2018. Short-term Wind Power Forecasting, Beijing: IOP Conference Series. Holden, P. B., 2015. PLASIM–GENIE v1.0: a new intermediate complexity AOGCM, Hamburg: Geosci. Model Dev. Magnimind, 2019. Magnimind academy. [Online] Available at: https://magnimindacademy.com/deep-learning-and-its-5-advantages/ [Accessed 5 December 2019]. Mihajlovic, I., 2019. Hackernoon. [Online] Available at: https://hackernoon.com/introduction-to-deep-learning-9064d6b87a [Accessed 5 December 2019]. Mohamed Elhoseiny, S. h. H. A. E., 2015. Weather classification with deep convolutional neural networks, New Jersey: International Conference on Image Processing. Scher, S., 2019. Weather and climate forecasting with neural networks, Uppasala: s.n. Scher, S., 2019. Weather and climate forecasting with neural networks, Uppsala: Geosci. Model Dev. Shchur, A., 2019. Towards data science. [Online] Available at: https://towardsdatascience.com/weather-forecasting-with-data-science- approaches-cb8f2afd3f [Accessed 7 December 2019]. SiamakMehrkanoon, 2019. science direct. [Online] Available at: https://www.sciencedirect.com/science/article/abs/pii/S [Accessed 7 december 2019]. Subashin, A., 2019. Advanced Weather Forecasting Prediction, Karaikudi: International Journal for Research in Applied Science & Engineering Technology.
