Weather Prediction Literature Review

Transcription

1 Robert Clark Clark 1 Stem I Weather Prediction Literature Review Introduction The weather has destroyed cities, damage has cost trillions of dollars, and tragic events have killed thousands of people, globally (Gould & Bryan, 2017). Thankfully, these terrible aspects of weather do not affect most people, but the weather still plays a role in the lives of every person on Earth. Trying to understand the weather is not new. Many different cultures had, and still have, gods and shrines dedicated to the very topic. Today, as humanity turns to the future and pushes boundaries in science and technology, weather prediction could be much better. Table 1: Billion-dollar Weather and Climate Disaster, Disaster Number of Frequency Losses (USD Avg. Cost Deaths Events Billions) (USD Billions) Drought % $232.5 $9.7 2,933 Flooding % $118.7 $ Freeze 8 3.8% $27.3 $ Severe Storm 89 42% $200 $2.2 1,578 Tropical % $580.7 $16.6 3,210 Cyclone Wildfire % $35.6 $ Winter Storm % $42.7 $3.1 1,013 All Disaster % $1,237.5 $5.8 9,680 (Gould & Bryan, 2017) The ability to predict the weather accurately is important. Understanding the weather can help people to plan for the future, which can improve planning for things from commutes, to livelihoods. Weather prediction is not an easy task. There is an array of different factors, methods, and data need to be understood before accurate weather prediction can be achieved (Samenow & Fritz, 2015).

2 Clark 2 Issues with Weather Prediction Weather prediction has many different problems. Even the simplest weather predictions are not perfect. The one-day forecast typically falls within two degrees of the actual temperature. Although this in accuracy is not bad, as predictions are made for further in time. Also, when predicting precipitation, accuracy can be even worse. Furthermore, weather prediction, in areas where the climate lacks consistency, is off by even more. For example, in a place like New England where temperatures have great variance the temperature predictions are more inaccurate than a place like the tropics (Samenow & Fritz, 2015). Inaccuracies in weather prediction have different effects on different people. To some people, these inaccuracies may mean getting wet on the walk to the car or having to take off an extra layer of clothes, but to others, these inaccuracies can lead to unnecessary evacuations, home damage, and possibly death. Better weather prediction would allow for better knowledge of when evacuation is needed, when people need to board up their homes, and when and where people are most going to need help. Inaccurate weather prediction systems must be improved when so much is at stake. However, in order to understand what needs to be improved, it is important to understand how weather is currently predicted (Gould & Bryan, 2017). Current Weather Prediction Numerical Model Typically, professional weather prediction is made using mathematical simulations. Numerical models are used to simulate what is most likely going to happen based on the known state of the atmosphere. Forecasters use an array of models and then, based on the outputs of these simulations, produce their predictions. For example, if a forecaster is looking at three different numerical models, and two models predict that a storm is going to hit a certain place, the forecaster would most likely predict that the storm is going to hit the area. These numerical

3 Clark 3 models work well and are being tweaked all the time, but they still have errors because some of the equations used by the models are not precise. Also, if all the different initial aspects of a weather system are not known, it is difficult to predict the future (University of Illinois, 2010). Trends The trends method is less focused on multiple different outside factors and generally focuses on just one. The method focuses on looking at how weather is currently moving. For example, if a weather event is moving east and crosses half the United States in twenty-four hours, it would generally be predicted to cross the next half over the next twenty-four hours. However, this method has problems. If the weather changes direction or begins to move at a different speed, the prediction method failed. The trends method can work well for analyzing big systems that have a likelihood of staying on the same course, but if weather event changes direction the trends system begins to fail (University of Illinois, 2010). Persistence Method The persistence method of prediction assumes that weather today is going to be the same as weather tomorrow. The method uses generalized information about how the weather for an area is supposed to be, and then based on what the weather was that day, it predicts the next day to be very similar. For example, if the weather in Puerto Rico is generally 95 degrees in the summer, and the weather was 95 degrees that day, the persistence method would predict that tomorrow to be the same. This prediction method is generally successful in areas where the weather does not normally change or experience that many deviations. However, if the area of weather is in an area where it changes constantly and is in frequent flux, the prediction method becomes inaccurate (University of Illinois, 2010). Climatology Method The climatology method is very similar to the persistence method. However, this method takes a more in depth look at the past when predicting the future. The climatology method

4 Clark 4 generally works by taking data from a certain time of the year and then averaging what has happened in the past to predict the future. For example, if the weather for a certain area in 2015, 2016, and 2017 was 10, 12, and 8 degrees Fahrenheit, a person using the climatology method would predict that day in 2018 to be 10 degrees Fahrenheit. This method works very well in areas where the weather has little variance and follows similar patterns over many years. However, if an area does not meet these conditions it will, generally have faulty predictions (University of Illinois, 2010). Analog Method The analog method of prediction combines different aspects of other prediction methods. This method functions by looking in the past and finding a time period when the weather was similar to how it is right now. For example, say the weather of the past three days has been 10, 12, and 16 degrees Fahrenheit, the past would then be analyzed to find a few days where the weather lined up similarly, like 11, 13, and 16, degrees Fahrenheit. Then, based on what happened on the next day in the past, a similar prediction would be made for the future. This method has potential, but currently, this method does not have enough data or the capability to make accurate predictions. (University of Illinois, 2010). Weather Background The weather has a tendency to follow several different patterns. One well known pattern is that of the seasons. Because of the tilt of the axis, the Northern Hemisphere cools down in the winter as the sun light becomes less direct, and warms up in the summer as the opposite happens. As well as the seasons, the uneven heating of the Earth also causes the global circulation of air, creating wind. The sun s energy creates low pressure areas near the equator. As these areas move away from the equator these winds are met by winds coming south from the poles to create another low pressure zone. Eventually the wind returns back to the equator. These wind patterns

5 Clark 5 are able to push storms and other systems around the globe. The place where two masses meet is called a front (Parkins & Mcmurry, 1909). The jet stream push weather systems across the United States and all throughout the globe. The jet stream is theorized to be caused by the energy created by the westerly winds which are winds that blow up from the Equator. These winds meet up with others and release much of their energy at the point of the jet stream. The jet stream is narrow and have a strong concentration of isotherms, areas of the same temperature. The jet stream can create fluctuations in temperature, and weather (Namias & Clapp, 1949). Another major part of weather is the El Nino Southern Oscillation (ENSO), which is a powerful weather phenomenon on Earth. El Nino is when ENSO tends to be lower pressure than normal, and La Nina is the opposite. El Nino causes weather on the surface of the Pacific to become warmer than normal and the winds blow less powerfully to the East than normal. El Nino is important because it affects global conditions and weather patterns. ENSO can lead to large thunderstorms, droughts, or mild weather for the areas that it affects. These conditions can have large impacts on people s lives and weather predictions (National Oceanic and Atmospheric Association, n.d.a). Climate change is another part of weather. As more carbon is released into the atmosphere, it creates a greenhouse on the planet. The carbon holds the heat from the sun in the atmosphere and does not allow for the extra energy to be released into space. The energy that remains on the planet continues to warm up icebergs and raise sea levels. Also, the extra energy causes more water to be evaporated and moved into the atmosphere. The extra energy and vapor in the air create more extreme storms and weather systems than would have previously existed. These more powerful weather events can be dangerous and cause incredible amounts of damage

6 Clark 6 and destruction. As the globe continues to warm, it is necessary for weather prediction to include these aspects of climate in its forecasting (National Oceanic and Atmospheric Association, n.d.b). These different global weather aspects combine with more local ones to produce the weather on Earth. Other aspects of weather that affect what is experienced in a certain location are high and low pressure area. Low pressure areas are more likely to be hit by storms and larger weather systems than high pressure areas (National Snow and Ice Data Center, n.d.). Massive wind currents far up in the atmosphere continue to push around different parts of weather. For example, the Labrador current pushes cold weather to the South toward Massachusetts. On the coast, the land cooling and warming in the winter and summer respectively can create winds that blow onto or off of the land from the sea. These different factors and more combine to affect the weather (Parkins & McMurry, 1980). Data Extrapolation Methods In order to make predictions, it is important to be able to extrapolate from given data. The basis of extrapolation is analyzing numbers and trying to identify a pattern. The ability to extrapolate from data is crucial to making predictions as it allows for numbers beyond the given data set to be predicted. Forecasting weather is based around the principal of data extrapolation because it allows for a prediction based on a given data set (Armstrong, 1985). Markov Chain Markov chains are one method of prediction. These chains use a recent pattern of behavior and continuous data to make predictions. The computer would create a matrix, which would be used for analysis and predictions when there is a stable event. Markov chains also require an analysis of how an event changes and evolves from one period to the next. These can allow for slight changes in a dynamic system. However, if the prediction and extrapolation of

7 Clark 7 data require a frequently changing situation, the Markov chain process typically fails (Armstrong, 1985). Exponential Smoothing Exponential smoothing is creating a graph that extrapolates using averages. When using exponential smoothing, one would use past averages of what has happened to predict what is going to happen in the future. However, exponential smoothing also requires for the more recent data to have more weight. This means that a system can be more dynamic and not need to worry too much about the rapid changes. By having more weight on the more recent results, exponential smoothing produces curvier data extrapolations than traditional linear ones. The steps to accomplishing exponential smoothing are cleaning data, calculating new averages and new trends, and then predicting future data (Armstrong, 1985). Moving Averages Moving averages are a similar way to extrapolate from data. These have the functionality of finding certain upcoming points. Similar to exponential smoothing, moving averages tend to give more power to the most recent points. The most recent data points have more weight which makes sense when the most recent points have the most relevance. However, if the data is changing and evolving or there is no relevance based on the closeness of the points to the data extrapolation, it becomes less reliable. These work similarly to exponential smoothing, but are a bit more linear (Armstrong, 1985). Genetic Algorithms Genetic algorithms are a different sort of method for extrapolation. The algorithms are a way of finding an exact answer to a problem by testing a variety of solutions. For example, genetic algorithms would be very useful for finding the most optimal length of trips to a certain destination. These algorithms can be useful for trying to work out difficult problems because

8 Clark 8 they are capable of finding precise known solutions. However, genetic algorithms tend to require a lot of known information that is not always available when making prediction (Strauss, 2009). Genetic Algorithms function in a very unique way. Initially, two different sorts of methods for finding an exact solution are created. After creating two different algorithms that are very similar but slightly different, the better algorithm is then selected for, and the other one dies. Then a next round is done with the successful network duplicated, but one is slightly adjusted and the other is not. This repeats until the most optimal algorithm has been reached. Once the best solution has been developed the algorithm has been determined. The reason it is called a genetic algorithm is that it functions similar to how adaptations are created in genetics through natural selection and killed off traits (Rogers, 1990). Artificial Neural Networks Artificial Neural Networks (ANNs) are also excellent for analyzing data. ANNs function by studying a certain data set and the outcomes from such a data set, and then, from the studying, create an algorithm to use for extrapolation. These networks function by studying and then learning from the amount of error they have. The error is used to tweak the different parts of an algorithm. Once the algorithm goes through many different training iterations, it develops a way to predict data from a given data set (Strauss, 2009). ANNs are created using matrices of data. The network will have the responsibility to do something such as predict the sixth unit or try to analyze a photo. The algorithm goes through thousands of training iterations where the weight, the value of each node, is adjusted. As the node values are adjusted, the network becomes more fine tuned for the data that is being analyzed and becomes better equipped to make predictions from a given data set (Black, 1999).

9 Clark 9 The networks have the capability of working in areas where there is fuzzy logic or outcomes are not defined in a binary way. ANNs can be used with things such as the stock market, and even in an understanding of the qualitative factors of investing. The capability of understanding the issues with stock markets, represents the capabilities and applications of neural networks (Khou, 1998). ANNs have made weather predictions before, for they have been used to help predict the temperature in Georgia to further improve the ability of farmers to protect their crops. This use of ANNs was both a situation that involved uncertainty and an application that involved weather (Smith, 2006). Other Possibilities Other possibilities for data set extrapolation is by using root mean square error (RMSE) and other ways of calculating residuals. Based off residuals and RMSE it is possible to find data sets that best align with given data. A large amount of data can be searched through to find the certain values which line up most closely. One possible way of weather analysis can be finding the temperatures in the past that most closely relate to the present. Then the next value with other calculations can be predicted as the next temperature. RMSE can be very useful when analyzing data sets and trying to find the points at which a data set most closely aligns with the littlest difference (Armstrong, 1985). Conclusion The weather is a complex topic. There is an array of different factors that go into influencing how the weather is going to change and evolve over a certain period of time. As the weather changes, it is important to take different factors into account when trying to forecast the weather.

10 Clark 10 Data extrapolation is another crucial part of weather prediction. As the weather changes, it is important to be able to predict beyond a given timeline to try and understand what the future holds. Data extrapolation can be done in a variety of different ways and using even more complex methods to analyze data such as an ANN or a genetic algorithm. By using these tools, it is possible to better understand the future state of weather.

11 Clark 11 References Armstrong, J. S. (1978). Long-range forecasting: from crystal ball to computer. Retrieved from: Black, Christopher Lee. System Training of an ANN Patent. 31 Mar Gould, S., & Bryan, B. (2017, September 02). Hurricane Harvey is the 10th natural disaster in 2017 to cause more than $1 billion in damage. Business Insider. Retrieved from: Khou, R. J., Chen, C. H., Hwang, Y.C. (1988). An Intelligent Stock Trading Decision Support System through Integration of Genetic Algorithm Based Fuzzy Neural Network and Artificial Neural Network. Fuzzy Sets and Systems Retrieved from: pdfs.semanticscholar.org/40c2/38be863d8e1a38638d4b365911e9e38537a5.pdf Namias, J., & Clapp, P. F. (1949). Confluence Theory Of The High Tropospheric Jet Stream. Journal of Meteorology, 6(5), doi: / (1949)006<0330:ctotht>2.0.co;2 National Oceanic and Atmospheric Administration. (n.d.a). El Nino. Retrieved from: National Oceanic and Atmospheric Administration. (n.d.b). Global Warming. Retrieved from: National Snow and Ice Data Center. (n.d.)..patterns in Arctic Weather and Climate. Retrieved from Parkins, A. E. & McMurry, F. M. (1909). Advanced Geography. Retrieved from: vanced+geography+winds&source=bl&ots=d_a2lkgq8m&sig=h7iir- TIX92h4lfAAnnCQPoeoFE&hl=en&sa=X&ved=0ahUKEwj_xov- 2_DXAhVQSt8KHSywArMQ6AEINDAC#v=onepage&q&f=false Rogers, D. Predicting Weather Using a Genetic Memory: a Combination of Kanerva's Sparse Distributed Memory with Holland's Genetic Algorithms. Retrieved from: pdfs.semanticscholar.org/6531/1f4bec3e5c4a5b2d311a865fdbfc670993e4.pdf Samenow, J. & Fritz, A. (2015, January 02). Five myths about weather forecasting. The Washington Post. Retrieved from: myths-about-weather-forecasting/2015/01/02/e49e8950-8b86-11e4-a085-34e9b9f09a58_story.html?utm_term=.6067ae6442dc Smith, B A. (2006). Air Temperature Prediction Using Artificial Neural Networks. Retrieved from Penn. State University website: Strauss, D. (2009, September 9). When to use Genetic Algorithms vs. when to use Neural Networks? [closed][comment]. Retrieved from: University of Illinois Department of Atmospheric Science. (1999). The Weather World Retrieved from: