Wednesday, 15 January 2020: 3:15 PM
260 (Boston Convention and Exhibition Center)
Weather operations at Kennedy Space Center and Cape Canaveral Air Force Station (KSC/CCAFS) are complicated by unique requirements for near-real time determination of risk from lightning. KSC/CCAFS experiences one of the world's highest incidence of lightning, impacting both the launch of space vehicles and daily support activity. Accurate prediction of lightning risk is complicated by sensor data that is both inherently noisy and collected in time series. This study applies discrete wavelet transformations (DWT) and semi-parametric modeling to improve the timeliness and accuracy of lightning prediction for KSC/CCAFS.
Predictive models such as linear regression are easy to interpret due to their prevalence, however they require certain assumptions to be made concerning underlying relationships within the data. These assumptions may cause a model to over smooth a predicted response, resulting in a failure to capture a significant event that is of most interest to research. Furthermore, these models are not well suited to time series data and can not mitigate noise in the data. The proposed approach employs DWT as a computationally efficient method to transform a meteorological time series for accurate modeling while simultaneously reducing observed noise. Additionally, semi-parametric models are used to capture complex phenomena observed in meteorological events without assumptions of the underlying data.
Wavelet transforms are a relatively new method that allows re-expression of data from the time domain into a frequency domain in an very computational efficient manner. These methods facilitate accurate modeling of a complex response in time, producing analysis of both frequency and time content simultaneously. Motivated by the Fourier transform, the DWT consists of a linear transformation that reduces a complex response to a single vector of coefficients. An inverse DWT (IDWT) can then be applied to perfectly reproduce the original data. This method allows manipulation to remove noise or extraneous data, with common applications in signal analysis, data compression, and image analysis. This method is especially useful in evaluation of time series as allows analysis without auto-correlation that would otherwise cause overestimation of a response.
Semi-parametric models, such as a single index model or generalized liner model, are methods that bridge the capabilities of parametric and non-parametric models. Non-parametric methods describe a data strictly using the data and without any required assumptions. This technique allows for the modeling of abrupt changes in data, however suffers due to a lack of interpretability in data sets of large dimension. Semi-parametric techniques combine elements of both parametric and non-parametric techniques to exploit the benefits of each method.
The following case study for use will be replaced with findings if available at the time of presentation. A wide array of sensors are employed at KSC/CCAS that provide discretized readings over time, primarily returns from the Lightning Detection and Ranging (LDAR) network and conventional weather data. LDAR data consists of returns from a system of electric field mills located throughout KSC/CCAFS. A wavelet transform is applied to these readings, which are then fed into a semi-parametric model to produce a measure of assessed lightning risk. This model is designed to provide both the accuracy and timeliness required to effectively support operations at KSC/CCAS.
Predictive models such as linear regression are easy to interpret due to their prevalence, however they require certain assumptions to be made concerning underlying relationships within the data. These assumptions may cause a model to over smooth a predicted response, resulting in a failure to capture a significant event that is of most interest to research. Furthermore, these models are not well suited to time series data and can not mitigate noise in the data. The proposed approach employs DWT as a computationally efficient method to transform a meteorological time series for accurate modeling while simultaneously reducing observed noise. Additionally, semi-parametric models are used to capture complex phenomena observed in meteorological events without assumptions of the underlying data.
Wavelet transforms are a relatively new method that allows re-expression of data from the time domain into a frequency domain in an very computational efficient manner. These methods facilitate accurate modeling of a complex response in time, producing analysis of both frequency and time content simultaneously. Motivated by the Fourier transform, the DWT consists of a linear transformation that reduces a complex response to a single vector of coefficients. An inverse DWT (IDWT) can then be applied to perfectly reproduce the original data. This method allows manipulation to remove noise or extraneous data, with common applications in signal analysis, data compression, and image analysis. This method is especially useful in evaluation of time series as allows analysis without auto-correlation that would otherwise cause overestimation of a response.
Semi-parametric models, such as a single index model or generalized liner model, are methods that bridge the capabilities of parametric and non-parametric models. Non-parametric methods describe a data strictly using the data and without any required assumptions. This technique allows for the modeling of abrupt changes in data, however suffers due to a lack of interpretability in data sets of large dimension. Semi-parametric techniques combine elements of both parametric and non-parametric techniques to exploit the benefits of each method.
The following case study for use will be replaced with findings if available at the time of presentation. A wide array of sensors are employed at KSC/CCAS that provide discretized readings over time, primarily returns from the Lightning Detection and Ranging (LDAR) network and conventional weather data. LDAR data consists of returns from a system of electric field mills located throughout KSC/CCAFS. A wavelet transform is applied to these readings, which are then fed into a semi-parametric model to produce a measure of assessed lightning risk. This model is designed to provide both the accuracy and timeliness required to effectively support operations at KSC/CCAS.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner