Temporal and Spatial Characteristics of Meteorological Elements in the Vertical Direction at Airports and Hourly Airport Visibility Prediction by Artificial Intelligence Methods

Jin Ding, Guoping Zhang, Jing Yang, Shudong Wang, Bing Xue, Xiangyu Du, Ye Tian, Kuoyin Wang, Ruijiao Jiang, Jinbing GaoSustainability 2022 · DOI

TL;DR

This paper investigates the vertical distribution of meteorological elements and their relationship with airport visibility using L-band sounding data. It then employs nine AI models to predict hourly airport visibility, finding that Huber Regression (HBR) and Random Sample Consensus Regression (RANSAC) offer the most consistent predictions with observations, outperforming other methods like PAR and IST.

What Problem Does This Paper Address?

Airport visibility is crucial for flight safety and economic operations. Predicting it accurately, especially in the short term, is a significant challenge. While previous studies have explored visibility prediction using various meteorological factors, there's a gap in understanding the detailed vertical characteristics of meteorological elements and their specific relationships with airport visibility at different altitudes. This paper aims to bridge this gap by analyzing L-band sounding data to understand these vertical patterns and then leveraging this knowledge to build more accurate AI-driven hourly visibility prediction models.

Methods

The study utilizes two main methodological approaches:

  1. Trend Analysis: The Mann-Kendall (MK) test and Sen's slope estimator are employed to analyze the temporal and spatial trends of meteorological elements (temperature, air pressure, relative humidity, and wind speed) in the vertical direction. This helps identify how these elements change with altitude and across different geographical regions.
  2. AI-based Visibility Prediction: Nine different artificial intelligence algorithms are used to predict hourly airport visibility. These include:
  3. Isotonic Regression (IST)

    • Bayesian Ridge Regression (BRR)
    • Least Absolute Shrinkage and Selection Operator (LASSO)
    • Passive Aggressive Regression (PAR)
    • Random Sample Consensus Regression (RANSAC)
    • Huber Regression (HBR)
    • Elastic-Net Regression (ENR)
    • Automatic Relevance Determination Regression (ARD)
    • Tweedie Regression (TWD)

    The models are trained on historical hourly meteorological data and then tested to evaluate their prediction performance. The study focuses on comparing the effectiveness of these diverse AI models for this specific task.

Datasets

  • L-band Sounding Data: Second-level L-band sounding data from 2010-2020, collected twice daily (8:00 a.m. and 8:00 p.m. BJT), was used to analyze vertical meteorological element characteristics. This data underwent strict quality control.
  • Hourly Meteorological Data: Hourly measurements of 26 meteorological elements from 2018-2020 were used for training and testing the visibility prediction models. This data was sourced from national and regional weather stations, with visibility being an average of the five closest stations to the airport.

Key Insights and Ideas

  • Geographical Trends in Vertical Elements: The study reveals distinct geographical patterns in the vertical trends of meteorological elements. Temperature and relative humidity show a clear negative trend that intensifies from northwestern to southeastern China. Air pressure also exhibits a significant negative vertical trend, particularly pronounced in eastern China.
  • Altitude-Specific Visibility Correlations: Airport visibility's relationship with meteorological elements is not uniform across altitudes. Visibility is strongly correlated with air pressure within approximately 2000 geopotential meters (gpm) from the ground (mostly negative correlation). Within 400 gpm, visibility is negatively correlated with relative humidity. The correlation with wind speed shows diurnal variations, being positive within 2000 gpm at 8:00 a.m. but primarily within 400 gpm at 8:00 p.m.
  • AI Model Performance Varies: Not all AI models are equally effective for airport visibility prediction. The Passive Aggressive Regression (PAR) and Isotonic Regression (IST) models performed poorly, indicating limitations in their ability to capture the complex relationships in this domain.
  • Robust Models Emerge: Huber Regression (HBR) and Random Sample Consensus Regression (RANSAC) demonstrated the most consistent prediction results, with their simulation dispersions closely matching observed data. This suggests their robustness in handling potential outliers or complex data distributions.

Key Results

  • Vertical Trends: Negative vertical trends for temperature and relative humidity were observed to increase from northwest to southeast China. Air pressure showed a greater negative trend in eastern China.
  • Visibility Correlations: Within ~2000 gpm, visibility strongly correlated negatively with air pressure. Within ~400 gpm, visibility negatively correlated with relative humidity. At 8:00 a.m., visibility positively correlated with wind speed within ~2000 gpm, while at 8:00 p.m., this positive correlation was mainly observed within ~400 gpm.
  • Model Performance: PAR and IST models showed the largest gap between predicted and observed visibility. HBR and RANSAC models exhibited the most consistent dispersion between predictions and observations. ARD, BRR, ENR, LASSO, and TWD models showed similar, good performance.
  • Airport-Specific Performance: The study identified specific airports where visibility prediction was more accurate (e.g., Nanjing, Hangzhou) and others where it was poorer (e.g., Lhasa, Lanzhou). This suggests that airport location and local conditions play a significant role.

Limitations and Future Work

  • Data Limitations: The study acknowledges that human activities (e.g., incense burning, fossil fuel combustion) significantly impact visibility but were not directly included as input features. Future work should incorporate these environmental factors (like PM2.5, PM10, NO2, SO2, O3, CO) to improve prediction accuracy.
  • Model Limitations: While AI methods are advanced, they still have limitations. The study suggests that incorporating interpretable factors like aerodynamics could enhance prediction and early warning capabilities.
  • Generalizability: While the study analyzed 47 airports, further research could explore the applicability of these models to a wider range of airports and climate zones.
  • Model Robustness: The performance of PAR and IST models highlights the need for careful model selection, as their online learning or nonparametric nature may not be ideal for complex, large-scale meteorological data.

Summary updated

Abstract

Based on second-level L-band sounding data, the vertical distribution and variation of meteorological elements at airports in 2010–2020 are investigated. At the same time, the relationships between airport visibility and meteorological elements at different potential heights are also investigated. Then, based on hourly measurements of 26 meteorological elements in 2018–2020, the hourly visibility of airports is predicted by 9 artificial intelligence algorithm models. The analyses show: (1) For the vertical changes in four meteorological elements of the airports, the negative vertical trends of temperature and relative humidity increase clearly from northwestern to southeastern China. The significant negative trend of air pressure in the vertical direction in the eastern China is greater. (2) Within about 2000 geopotential metres (gpm) from the ground, the visibility has a strong correlation with the air pressure, and most of them are negative. Within 400 gpm from the ground, airport visibility is negatively correlated with the relative humidity. At 8:00 a.m., airport visibility is positively correlated with the wind speed within 2000 gpm from the ground at most airports, while at 20:00 p.m., the positive correlation mainly appears within 400 gpm from the ground. (3) The passive aggressive regression-(PAR) and isotonic regression-(IST) based models have the worst effect on airport visibility prediction. The dispersion degree of the visibility simulation results obtained by Huber regression-(HBR) and random sample consensus regression-(RANSAC) based models is relatively consistent with the observations.
BibTeX
@inproceedings{ding2022temporal,
  author = {Jin Ding and Guoping Zhang and Jing Yang and Shudong Wang and Bing Xue and Xiangyu Du and Ye Tian and Kuoyin Wang and Ruijiao Jiang and Jinbing Gao},
  title = {{Temporal and Spatial Characteristics of Meteorological Elements in the Vertical Direction at Airports and Hourly Airport Visibility Prediction by Artificial Intelligence Methods}},
  booktitle = {Sustainability},
  year = {2022},
  doi = {10.3390/su141912213},
}
Figure