高海拔机场危险天气感知预报预警模型算法数据

主要是雷暴、低空风切变、沙尘、低云低能见度天气的预报预警模型算法数据。具体包括：拉萨机场雷暴、风切变、沙尘的感知预报预警模型算法数据；林芝机场低云低能见度天气的感知预报预警模型算法数据。低云低能见度预报预警模型算法：通过机场局地地理气候要素分析、数值模式对机场气象要素的基本预报能力分析、数值预报机场低云低能见度分析、数值预报机场雷暴分析，在此基础上构建了多源线性回归的传统MOS方法以及包括长短期记忆神经网络LTSM、支持向量机SVM、决策树TREE的深度学习方法相结合的机场低云低能见度预报模型算法。结合12小时的预报作为背景场，结合深度学习方法YOLO识别出的低云低能见度区域，以及快速更新的机场探测资料（包括机场自观AWOS和垂直温湿廓线雷达探测），进行2小时内的低云低能见度快读更新预报，输出频率为：每6分钟输出一次。
雷暴预报预警模型算法：主要功能是对雷达观测数据进行读取和基础处理，然后利用光流方法实现雷达回波的外推预测。首先对原始雷达数据进行格式解析和必要的预处理；接着，通过光流算法来捕捉和计算不同时间帧之间回波的移动特征，从而推算未来一段时间内回波的大致演变趋势。最终，这些预测结果会被转换成灰度图像输出，便于后续处理。在光流预测结果的基础上进行强度修正。它会读取前一步生成的灰度图像，结合计算得到的回波特征参量，对预测中可能存在的强度偏差进行调整，使结果在保持原有运动趋势的同时，更加贴近真实观测。
风切变预报预警模型算法：主要功能是利用是数值模式预报数据和风切变预警结果，使用卷积神经网络和循环神经网络构建了拉萨机场低空风切变短时临近预报模型。先使用卷积残差网络提取数字模式的背景特征，然后将临近两小时的风切变预警结果拼接天气背景特征，作为循环神经网络的输入。最终输出拉萨机场未来两小时低空风切变的预报；对测风激光雷达观测的三维风场进行数据质控等预处理，然后利用组合切变算法进行水平风水平切变识别和线性外推，最后输出以雷达为中心方圆5km范围的风切变大小，根据不同风切变大小对飞行的影响不同，对风切变影响等级进行分类，输出对应的风切变级别强度。
沙尘预报预警模型算法：使用计算流体动力学仿真计算，采用基于拉格朗日粒子法对机场沙尘天气进行数字建模，模拟沙源地在不同风向风速条件下沙子的移动路径，从而计算得到机场沙尘浓度，根据沙尘浓度和光学特性得到沙尘天气能见度结果。结合数值模式对机场风场、温湿场的基本预报结果，进行沙尘能见度快速更新预报。

This dataset primarily contains forecast and early warning model algorithm data for thunderstorms, low-level wind shear, dust, and low cloud and low visibility weather conditions. Specifically, it includes: perceptual forecast and early warning model algorithm data for thunderstorms, wind shear, and dust at Lhasa Airport; and perceptual forecast and early warning model algorithm data for low cloud and low visibility weather at Nyingchi Airport.

Low cloud and low visibility forecast and early warning model algorithm: Based on the analysis of local geographic and climatic factors of the airport, the basic forecast capability of numerical models for airport meteorological elements, numerical forecast analysis of airport low cloud and low visibility, and numerical forecast analysis of airport thunderstorms, a low cloud and low visibility forecast model algorithm for airports is constructed by combining the traditional Model Output Statistics (MOS) method with multi-source linear regression, and deep learning methods including Long Short-Term Memory Network (LSTM), Support Vector Machine (SVM), and Decision Tree (TREE). Combined with the 12-hour forecast as the background field, the low cloud and low visibility area identified by the YOLO deep learning method, and the rapidly updated airport detection data (including airport Automatic Weather Observing System (AWOS) and vertical temperature and humidity profile radar detections), a rapid update forecast of low cloud and low visibility within 2 hours is performed, with an output frequency of once every 6 minutes.

Thunderstorm forecast and early warning model algorithm: Its main functions include reading and basic processing of radar observation data, and then using the optical flow method to realize the extrapolation prediction of radar echoes. First, format analysis and necessary preprocessing are performed on the original radar data; then, the optical flow algorithm is used to capture and calculate the movement characteristics of echoes between different time frames, so as to infer the approximate evolution trend of echoes over a future period of time. Finally, these prediction results are converted into grayscale images for output to facilitate subsequent processing. Intensity correction is performed based on the optical flow prediction results. It reads the grayscale images generated in the previous step, combines the calculated echo characteristic parameters, and adjusts possible intensity deviations in the prediction, so that the results maintain the original movement trend while being closer to real observations.

Low-level wind shear forecast and early warning model algorithm: Its main functions include using numerical model forecast data and wind shear early warning results to construct a short-term nowcasting model for low-level wind shear at Lhasa Airport using Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). First, the convolutional residual network is used to extract the background features of the numerical model, then the wind shear early warning results of the adjacent 2 hours are spliced with the weather background features as the input of the recurrent neural network. Finally, the forecast of low-level wind shear at Lhasa Airport in the next 2 hours is output; preprocessing such as data quality control is performed on the three-dimensional wind field observed by the wind-measuring lidar, then the combined shear algorithm is used to identify horizontal wind shear and perform linear extrapolation, and finally the wind shear magnitude within a 5km radius centered on the radar is output. According to the different impacts of different wind shear magnitudes on flight, the wind shear impact levels are classified, and the corresponding wind shear level intensity is output.

Dust forecast and early warning model algorithm: Computational fluid dynamics simulation is used for calculation, and the Lagrangian particle method is adopted to digitally model the dust weather at the airport, simulating the movement paths of sand at sand source areas under different wind directions and wind speeds, so as to calculate the airport dust concentration. The visibility result of dust weather is obtained based on the dust concentration and optical characteristics. Combined with the basic forecast results of the airport wind field, temperature and humidity field from numerical models, a rapid update forecast of dust visibility is performed.

搜集汇总

背景与挑战

背景概述

该数据集聚焦于高海拔机场的危险天气预报预警，涵盖雷暴、低空风切变、沙尘和低云低能见度等天气类型的模型算法数据。它结合了深度学习、数值预报和传统方法，为拉萨和林芝机场提供具体的感知与预报解决方案。