Scholarlink

<p>Reliable wind speed and direction forecasting is crucial to ensuring operational safety and efficiency in aircraft landings, takeoffs, and the reliable preparation of Terminal Aerodrome Forecasts (TAFs). This study investigates and assesses how consistent and predictable wind speeds are at Patna Airport (ICAO: VEPT), an important location in the Indo-Gangetic Plain (IGP), by analyzing hourly Meteorological Aerodrome Report (METAR) data from April 2014 to June 2024. Persistence analysis using unit root tests—Augmented Dickey-Fuller (ADF), Phillips-Perron (PP), and Kwiatkowski-Phillips-Schmidt-Shin (KPSS)—along with Levene’s variance test, confirms data stationarity and reveals significant monthly variability. The average wind speed is 3.86 knots, with gusts occasionally reaching 30.0 knots. Wind Speed Duration Curve (WSDC) analysis shows the dominance of low wind conditions, which is crucial for aviation safety, wind energy assessment, and infrastructure planning. Unlike previous studies that relied primarily on statistical or shallow learning models, this study evaluates four advanced deep learning (DL) architectures—conventional Long Short-Term Memory (LSTM), hybrid Convolutional Neural Network-LSTM (CNN-LSTM), attention-based LSTM, and a novel self-supervised LSTM—using multivariate time series data. The self-supervised LSTM achieves the most accurate wind speed predictions, while the CNN-LSTM outperforms in forecasting wind direction. Both models attain the lowest Mean Absolute Error (MAE), meeting World Meteorological Organization (WMO) standards for TAF reliability. This study highlights the effectiveness of Machinel Learning(ML)/DL-based approaches—particularly the self-supervised LSTM and hybrid CNN-LSTM models—in enhancing short-term forecasting of wind speed and direction. These models demonstrate superior performance compared to conventional statistical Auto Regressive Integrated Moving Average (ARIMA), especially in complex, low-speed wind environments such as that of Patna Airport. Their ability to learn temporal patterns and adapt to non-linear dynamics also makes them promising tools for automating TAFs.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Evaluating wind persistence and nowcasting of wind in aviation services: Harnessing deep learning for enhanced terminal aerodrome forecast

  • 期刊论文

  • 发表日期：

    2025年8月18日

查看全文
  • Anand Shankar,
  • Bikash Chandra Sahana,
  • Ashish Kumar,
  • Abhishek Anand

摘要

Reliable wind speed and direction forecasting is crucial to ensuring operational safety and efficiency in aircraft landings, takeoffs, and the reliable preparation of Terminal Aerodrome Forecasts (TAFs). This study investigates and assesses how consistent and predictable wind speeds are at Patna Airport (ICAO: VEPT), an important location in the Indo-Gangetic Plain (IGP), by analyzing hourly Meteorological Aerodrome Report (METAR) data from April 2014 to June 2024. Persistence analysis using unit root tests—Augmented Dickey-Fuller (ADF), Phillips-Perron (PP), and Kwiatkowski-Phillips-Schmidt-Shin (KPSS)—along with Levene’s variance test, confirms data stationarity and reveals significant monthly variability. The average wind speed is 3.86 knots, with gusts occasionally reaching 30.0 knots. Wind Speed Duration Curve (WSDC) analysis shows the dominance of low wind conditions, which is crucial for aviation safety, wind energy assessment, and infrastructure planning. Unlike previous studies that relied primarily on statistical or shallow learning models, this study evaluates four advanced deep learning (DL) architectures—conventional Long Short-Term Memory (LSTM), hybrid Convolutional Neural Network-LSTM (CNN-LSTM), attention-based LSTM, and a novel self-supervised LSTM—using multivariate time series data. The self-supervised LSTM achieves the most accurate wind speed predictions, while the CNN-LSTM outperforms in forecasting wind direction. Both models attain the lowest Mean Absolute Error (MAE), meeting World Meteorological Organization (WMO) standards for TAF reliability. This study highlights the effectiveness of Machinel Learning(ML)/DL-based approaches—particularly the self-supervised LSTM and hybrid CNN-LSTM models—in enhancing short-term forecasting of wind speed and direction. These models demonstrate superior performance compared to conventional statistical Auto Regressive Integrated Moving Average (ARIMA), especially in complex, low-speed wind environments such as that of Patna Airport. Their ability to learn temporal patterns and adapt to non-linear dynamics also makes them promising tools for automating TAFs.