[Audio] Weather Forecasting.. Weather Forecasting.
[Audio] Weather forecasting is imperative to various aspects of economics and human livelihood. In this presentation, I will analyze the current and future prediction models, identify limiting factors, and see if they can be improved upon. My goal was to better understand the current weather prediction methods and establish a performance comparison relative to their accuracy and limitations in day-ahead forecasts. My review will cover a brief history of Numerical Weather Prediction and its computational limitations in solving the equations of motion. I will then introduce the application of Artificial Intelligence and machine learning methods, as stand-alone models , and follow up with the incorporation of hybrid modeling for fine tuning forecasts. The primary consensus of the review indicate higher accuracy in hybrid prediction models, particularly in models with a base in numerical weather prediction. Potential methods for future weather forecast models are also discussed. I will also review the methods and results of my own model constructed via MATLAB. The weather data from several airports across the United States from the years 2000 to 2022 was obtained from NOAA and was split into training and testing data of ten-year intervals, The training data, 2000 to 2010, was fit into a regression model and the predictions were evaluated via the test data, 2010 to 2022, with a higher RSME than expected. The results of my model indicate inaccuracy in standalone machine learning models via regression. The application of numerical weather prediction and machine learning would provide a lower RSME as shown in the articles reviewed..
[Audio] I will begin with the elements of weather prediction with a brief review of its history. I will then cover different models and methods for forecasting. I will conclude with the results of my own model via MATLAB..
[Audio] Atmospheric Data includes Temperature, Pressure, Wind Speed and Direction, Water Vapor, and Air Pollution Concentration. Atmospheric anomalies include, sudden stratospheric warming..
[Audio] Data collection on land includes the tthermometer for Temperature the Barometer for Pressure, the Anemometer and Wind Vine for Wind Speed and Direction, the Hygrometer for Humidity, the Rain Gauge for Precipitation and Wave Patterns detection via Doppler Radar, Microwave, and Infrared. Geospatial data includes low orbital and geostationary satellites..
[Audio] In the 1940s, the development of the electronic computer helped advance the application of numerical weather prediction. In the 1950s, the 2-level prediction model was introduced, but was not comparable to a human forecaster. By the 1960s, the 3-level model was relatively equivalent to the human forecaster. From the 60s to the 80s the further research and development improved the results and provided objective guidance to 72h ahead predictions ( Young MV et al. 2022)..
[Audio] The types of weather forecasting include Climatology, Persistence, Analog, and Numerical. The climatology method is a regional calendar forecast and involves reviewing several years of statistical weather data to make a forecast. The problem: it requires pattern consistency. The persistence method only looks at current conditions. The analog method involves filtering through past weather data of a specific day with similar conditions. The problem minute differences can affect the outcome. Numerical Weather Prediction includes statistical and graphical weather prediction..
[Audio] The two primary modes of machine learning is classification and regression. Classification is meant for categorical variable outputs with discrete values. In a Review on weather prediction using machine learning, the Naïve Bayes Algorithm was found to yield the best results in weather prediction with an accuracy of 100% and had the highest values in Recall, compared to other classification algorithms ( Khan S et al. 2022). Regression is meant for continuous outputs. In an article on the Comparison of machine learning methods for photovoltaic power forecasting based on numerical weather prediction, out of 24 machine learning models, the most accurate were Kernel Ridge and Multilayer Perceptron with approximately 13.9% RMSE improvement over a baseline linear regression model ( Markovics D et al. 2022)..
[Audio] Due to the chaotic nature of the atmosphere, it is difficult to build a model that can handle the uncertainty. Ensemble approach is only practically affordable way to solve the problem of model uncertainty. ( Astakhova E et al. 2021). Ensemble is like hybrid models, but it does not typically combine the models. Each model serves as series of references for the actual forecaster. Accurate results require high computation power and are generally time consuming. Another method for handling chaos is with Neural Networks. Using Machine Learning approaches, chaotic time series Neural Network prediction methods were developed, which enhanced their efficiency and accuracy ( Ramadevi B et al. 2022). Neural Networks have shown reasonable computational accommodation and often can be implemented in smaller systems..
[Audio] Artificial Neural Networks have shown positive results in recent years, particularly with the Multilayered perceptron model ( MLP). A recent scientific report implemented Neural Networks for prediction. The results showed an overall mean square error for the entire forecast was 3.11 C [Degrees Celsius]. ( Del IV et al. 2021). In an article on Developing machine learning algorithms for meteorological temperature and humidity forecasting at Terengganu state in Malaysia, several Artificial Neural Networks were implemented, and the results showed the Multilayered Perceptron Model as the most accurate model for predicting daily Temperature ( T) and relative humidity ( Rh) ( Hanoon MS et al. 2021). In both cases, the MLP has the best accuracy in weather prediction verses Numerical Weather Prediction and other Neural Networks..
[Audio] The data in these tables were from the article on Deep Learning-Based Effective Fine-Grained Weather Forecasting Model. Results indicate the DeepSTF is the best model. The DeepSTTF model has a greater advantage in the extraction of spatial and temporal features of meteorological predictors while maintaining high accuracy even under seasonal changes and terrain differences ( Hewage P et al. 2022)..
[Audio] The results of the survey on Deep learning-based weather prediction showed the prediction accuracy of specially designated DLWP may be higher than Numerical Weather Prediction for specific variables or events ( Ren X et al. 2021). The inaccuracy of Deep Learning occurs in anomalies. This will cause some of the prediction to be interpretable as outliers. A combination of Deep learning and machine learning for better classification could improve forecasts by detecting anomaly that skew the data. Deep learning is good at predictor extraction and accommodating for seasonal changes ( Kong W et al. 2022). Deep Learning has been shown to accurately predict weather forecasts, but it does not handle weather anomalies. Deep learning is unable to account for extreme weather (Kong W et al. 2022)..
[Audio] For my own model, I chose the MATLAB platform. I utilized NOAA's weather database to acquire airport data from several states spanning the United States. The average temperature of Missouri was my prediction goal, and I added the temperature averages, maximums, and minimums of seven other states as features to capture a forecast perimeter. I used MATLAB's regression learner to train and test multiple Regression models. Exponential and rational quadratic Gaussian Process Regression provided the best results. My initial RSME validation was incredibly small at 8.5583e-05. After reanalyzing the data, I found that I included the maximum and minimum temperature values for Missouri. Since the average temperature is determined via the maximum and minimum values, I believe this invalidates my small RSME. I removed the variables and re-trained and tested the models. My average RSME was between 4 to 6. The prediction values captured general momentum of the actual values but were not accurate in forecasting. My greatest challenge was formatting the data. I was working with a 10-year span and there was an apparent number of empty cells. After filling in the empty cells, my training and testing speeds significantly reduced. From my research and experience with my model, a future model would provide better accuracy and require less computational resources if the data had better classification..
Bibliography. Young MV, Grahame NS. The history of UK weather forecasting: the changing role of the central guidance forecaster. Part 2: the birth of operational numerical weather prediction. Weather. 2022 [accessed 2022 June 18] https ://rmets.onlinelibrary.wiley.com/doi/full/10.1002/wea.4216 . doi:10.1002/wea.4216 Davis NA, Richter J H, Glanville AA, Edwards J, LaJoie E. Limited surface impacts of the January 2021 sudden stratospheric warming. Nature communications. 2022 [accessed 2022 June 19];13(1):1-11. https://www.nature.com/articles/s41467-022-28836-1 . doi:10.1038/s41467-022-28836-1 Eyre JR, English SJ, Forsythe M. Assimilation of satellite data in numerical weather prediction. Part I: The early years. Quarterly Journal of the Royal Meteorological Society. 2019 [updated 2020 May 2020; accessed 2022 June 16];146(726):146(726):49-68. https://rmets.onlinelibrary.wiley.com/doi/10.1002/qj.3654 . doi:10.1002/qj.3654. Zhang G, Yang D, Galanis G, Androulakis E. Solar forecasting with hourly updated numerical weather prediction. Renewable and Sustainable Energy Reviews. 2022 [accessed 2022 June 16];154:111768. https://www.sciencedirect.com/science/article/pii/S1364032121010364 . doi:10.1016/j.rser.2021.111768 He B, Ye L, Pei M, Lu P, Dai B, Li Z, Wang K. A combined model for short-term wind power forecasting based on the analysis of numerical weather prediction data. Energy Reports. 2022 [accessed 2022 June 16]; 8:929-939. https://www.sciencedirect.com/science/article/pii/S2352484721011215?via%3Dihub . doi:10.1016/j.egyr.2021.10.102 Markovics D, Mayer MJ. Comparison of machine learning methods for photovoltaic power forecasting based on numerical weather prediction. Renewable and Sustainable Energy Reviews. 2022 [accessed 2022 June 16];161:112364. https://www.sciencedirect.com/science/article/pii/S136403212200274X?via%3Dihub . doi:10.1016/j.rser.2022.112364 Trenchevski A, Kalendar M, Gjoreski H, Efnusheva D. Prediction of air pollution concentration using weather data and regression models. In Proceedings of International Conference on Applied Innovation in IT. Anhalt University of Applied Sciences. 2020 [accessed 2022 June 16];8(1):55-61. https://icaiit.org/paper.php?paper=8th_ICAIIT_1/2_2 . doi:10.25673/32749 Khan S, Mishra MP, Khatoon R, Singh R. A Review on weather prediction using machine learning. 2022 [accessed 2022 June 16];3(2):137-138. https ://theijire.com/assets/pdf/archives/1649784712_9ddfb9c0c3ede260edf4.pdf . Hanoon MS, Ahmed AN, Zaini NA, Razzaq A, Kumar P, Sherif M, Sefelnasr A, El- Shafie A. Developing machine learning algorithms for meteorological temperature and humidity forecasting at Terengganu state in Malaysia. Scientific Reports. 2021 [accessed 2022 June 18];11(1):1-9. https://www.nature.com/articles/s41598-021-96872-w . doi:10.1038/s41598-021-96872-w.
Bibliography Cont …. Jang P A, Matteson DS. Spatial correlation in weather forecast accuracy: A functional time series approach. arXiv preprint. 2021 [accessed 2022 June 16]; 2111(11381). https ://arxiv.org/abs/2111.11381 . doi:10.48550/arXiv.2111.11381 Astakhova E, Alferov D, Alferov Y, Bundel A. Ensemble approach to weather forecasting. In Journal of Physics: Conference Series. 2021 [accessed 2022 June 19];1740(1):012070. https ://iopscience.iop.org/article/10.1088/1742-6596/1740/1/012070 . doi:10.1088/1742-6596/1740/1/012070 Kong W, Li H, Yu C, Xia J, Kang Y, Zhang P. A Deep Spatio -Temporal Forecasting Model for Multi-Site Weather Prediction Post-Processing. Commun . Comput . Phys. 2022 [accessed 2022 June 16];31(1): 131-153. v31_131.dvi (global-sci.org) . doi:10.4208/cicp.OA-2020-0158 Hewage P, Trovati M, Pereira E, Behera A. Deep Learning-Based Effective Fine-Grained Weather Forecasting Model. Pattern Analysis and Applications. 2022 [accessed 2022 June 16];24(1):343-366. https://link.springer.com/article/10.1007/s10044-020-00898-1 . doi :10.1007/s10044-020-00898-1 Sasaki H, Urano S. Daily peak load demand forecast considering weather conditions. In 2022 12th International Conference on Power, Energy and Electrical Engineering (CPEEE). 2022 [accessed 2022 June 16];12(1):195-200. https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9738671 . doi:10.1109/CPEEE54404.2022.9738671 Ren X, Li X, Ren K, Song J, Xu Z, Deng K, Wang X. Deep learning-based weather prediction: a survey. Big Data Research. 2021 [accessed 2022 June 16];23(1):100178. https ://www.sciencedirect.com/science/article/pii/S2214579620300460?via%3Dihub . doi:10.1016/j.bdr.2020.100178 Ramadevi B, Bingi K. Chaotic Time Series Forecasting Approaches Using Machine Learning Techniques: A Review. Symmetry. 2022 [accessed 2022 June 16];14(5):955. https ://www.mdpi.com/2073-8994/14/5/955/htm . doi:10.3390/sym14050955 Kim H, Ham Y G, Joo Y S, Son S W. Deep learning for bias correction of MJO prediction. Nature communications. 2021 [accessed 2022 June 19];12(1):1-7. https://www.nature.com/articles/s41467-021-23406-3#Sec6 . doi:10.1038/s41467-021-23406-3 Del IV, Starchenko AV. Forecast of the near ground air temperature based on the multilayer perceptron model. In Journal of Physics: Conference Series. 2021 [accessed 2022 June 16];1989(1):012025. IOP Publishing. https://iopscience.iop.org/article/10.1088/1742-6596/1989/1/012025 . doi:10.1088/1742-6596/1989/1/012025 Singh G, Diamantopoulos D, Hagleitner C, Gómez-Luna J, Stuijk S, Mutlu O, Corporaal H. NERO: A near high-bandwidth memory stencil accelerator for weather prediction modeling. In 2020 30th International Conference on Field-Programmable Logic and Applications (FPL). 2020 [accessed 2022 June 16];9-17. https ://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9221526 . doi :10.1109/FPL50879.2020.00014.