Airport Arrival Rate (AAR) is a parameter specifying the number of arrival aircrafts that an airport can accept within an hour and it is commonly used for air traffic flow control. The Hong Kong International Airport (HKIA) is one of the busiest airports in the world for passenger and the world’s busiest for cargo in 2024.  Situated in a sub-tropical climate, HKIA often experiences significant convective weather and squall lines associated with cold fronts, low-pressure troughs and tropical cyclones during spring and summer. These intense convective activities sometimes disrupt the air traffic flow and cause significant travel delays in HKIA. An assessment of the impact of convective activities (related to its intensity and coverage) to AAR would facilitate early planning and necessary air traffic flow control. This study attempts to predict the Airport Arrival Rate via machine learning techniques using aircraft data, airport operational data and meteorological data. The aircraft data specifies the aircraft positions extracted from ADS-B data which can be used to evaluate the degree of congestion in the air space. The airport operational data shows the scheduled AAR for the day, which varies depending on the seasons and holidays. For meteorological data, the weather radar data indicates the location and intensity of convection. Besides, radar nowcast products for the next two hours can provide information on the short-term evolution of the significant convection. This paper will present the formulation of a convolutional neural network (CNN) model to predict the AAR. Due to imbalanced data and the changes in air traffic throughout the years, techniques of coupled CNN model and incremental learning are employed and evaluated. The results and limitations for these techniques will be discussed in the paper.  It is worth noting that there are many other factors affecting AAR, not just the weather, but this study may serve as an estimation of the weather’s impact on air traffic.  Following validation and evaluation, the Hong Kong Observatory may utilize the model to work with the air traffic flow management team in Hong Kong to improve air traffic flow planning.

How to cite: Leung, C. Y., Tam, H. Y., and To, K. C.: Predicting airport arrival rate using ADS-B data and weather radar data via machine learning techniques, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-165, https://doi.org/10.5194/ems2025-165, 2025.

In operational meteorology, ceilometers have been widely used for decades. The initial purpose of detecting cloud bases has greatly expanded. As a ground-based LIDAR instrument, the ceilometer provides high temporal resolution vertical profiles of backscatter signals up to 15000 m, depending on the model and configuration. Ceilometers are very important at airports where information on the heights of low clouds is essential for aviation operations, especially in so-called low visibility procedures.

The main advantage of ceilometers is their ability to provide continuous high-frequency measurements every 10-60 seconds. At Croatian airports, Vaisala ceilometers are in use, and together with Lufft instruments operated by the National Meteorological Institute (DHMZ), they form a national ceilometer data network. This spatial representation is important because weather conditions in Croatia can vary significantly between continental, mountain and coastal regions. In Croatia control, we developed visualisation for the forecasters, which includes raw backscatter data profiles (up 15km, PBL zoom up to 1200 - 3000m and up to 150m) together with derived cloudiness, cloud base height over time and direct algorithm cloud data output which is used operationally. From a novel Vaisala CL61 ceilometers at some airports, the linear depolarisation ratio (LDR) is also visualised and used to distinguish lithometeors and hydrometeors. Dozens of examples of typical and particular cases will be shown in the presentation (e.g. convection, growth of PBL, fog, stratus, drizzle, aerosol layers, snow and hail). In order to understand and recognise patterns, a targeted education for forecasters (and observers) is needed, which we started in 2019. 

Using the ceilometer network, which complements the observing system, forecasters can better diagnose weather processes in the atmosphere. This particularly refers to the boundary layer and area around the airports where crucial landing and take-off operations are performed.

How to cite: Jurković, J., Babić, N., and Babić, K.: Ceilometers as a diagnostic tool for forecasters, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-348, https://doi.org/10.5194/ems2025-348, 2025.

To avoid hazards associated with deep convective clouds, airplanes must fly above or around their tops. Therefore, accurate diagnosis of deep convective cloud top height is crucial in aviation meteorology.

This work employs an established operational method that involves comparing infrared satellite brightness temperature (BT) with a calculated parcel curve temperature. The intersection of BT and the parcel curve corresponds to a theoretical cloud top pressure (CTP) level, which is directly related to altitude (flight level) in the standard atmosphere.

Typically, the parcel curve is calculated iteratively from surface temperature and dewpoint, but this process can be computationally intensive for large datasets. In the first phase of this study, inspired by previous work on non-iterative calculations of moist adiabats, we found that the best approximation for moist adiabats is 5th-degree polynomial, with variable coefficients which are all functions of the wet bulb potential temperature. These coefficients can also be approximated with 4th-degree polynomials. In this approximation, a total of 6 polynomials (comprising 30 coefficients) must be evaluated, rendering it computationally very efficient. This represents a novel approach, as previous non-iterative approximations of moist adiabats employed a total of 200 coefficients and a different methodology to model the changing shape of moist adiabats.

In the second phase of the study the developed approximation was implemented in an operational environment (Visual Weather visualization software). For the method to perform effectively for elevated convection, the temperature and dewpoint of the most unstable layer should be used for calculating the moist adiabat. For this purpose, the layer with the maximum equivalent potential temperature is assumed to be the most unstable layer. Additionally, it is important to note that this method's validity is limited to convective clouds, as other cloud types lack the updraft required for temperatures to follow moist adiabats.

The aim of the third phase of this study is the validation of the automated cloud top height (CTH) diagnostic method. This validation can be challenging due to the lack of ground truth CTH data. This presentation will demonstrate the performance of our method across various convective situations and convective cloud top ranges (e.g., 6000-15000 m) by comparing the calculated CTH with radar vertical cross-sections, which are taken as ground truth. We also compared it with other similar products, such as the NWC SAF CTTH and radar ECHO TOPS products. Based on the analysis of all considered cases, we can conclude that our new method demonstrates very good performance.

How to cite: Šoljan, V. and Jurković, J.: Validation of Fast Approximation for Cb Top Height Diagnosis, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-497, https://doi.org/10.5194/ems2025-497, 2025.

Reliable road weather forecasts are crucial for ensuring road safety and optimizing road maintenance. They help to time road saltings effectively and to keep roads ice-free while avoiding unnecessary saltings. The Finnish Meteorological Institute produces road weather forecasts using a specified road weather model. This model is a one-dimensional heat balance model that utilizes atmospheric forecast from large-scale numerical weather prediction (NWP) model as input. One of the most important outputs of the road weather model is the road surface temperature, as it is essential to know whether it is below or above freezing.

Neither NWP forecasts nor the physics of the road weather model are perfect, which leads to errors in the forecasts. However, it is possible to reduce these errors by using machine learning model to learn from the past behavior of the forecast errors under different weather conditions. In this study, we develop a machine learning model using XGBoost (Extreme Gradient Boosting) to predict the difference between observed and forecasted road surface temperatures. This model can then be used to correct forecast errors. XGBoost is a type of gradient boosting algorithm that builds an ensemble of decision trees sequentially, where each new tree focuses on correcting the errors made by the previous ones. Unlike random forests, which aggregate the predictions of many independent trees, gradient boosting improves prediction accuracy by iteratively minimizing error.

The road weather model was run to road weather station points in Finland using atmospheric forecasts from the MEPS (MetCoOp Ensemble Prediction System) control member as forcing. Road weather station observations were used in the model initialization. The forecasts were done for each control member run, which means 8 forecasts for each day. The XGBoost model was trained with data from four winter periods (September-May) between 2020 and 2024. The winter period of 2024-2025 will be used for the final evaluation of the model's performance. The variables used as predictors contained forecasted atmospheric values from MEPS, forecasted road surface temperatures as well as time- and location-specific variables. Before training or evaluation, data quality control was performed on the data to remove outliers and missing values. Hyperparameter tuning is being performed to optimize the model’s performance. Although the tuning and training process is ongoing, initial results show that the machine learning model can reduce forecast error. The mean absolute error (MAE) of road surface temperature predictions across all lead times (1–62 hours) was 1.7 °C on the training set. After optimizing the number and depth of decision trees, the MAE decreased to 1.5 °C. This error estimate is based on four-fold cross-validation, where in each fold, one winter period served as the validation set while the remaining periods were used for training.

How to cite: Karsisto, V.: Post-Processing Surface Temperature Forecasts with XGBoost, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-188, https://doi.org/10.5194/ems2025-188, 2025.

The World Meteorological Organization’s Early Warnings for All (EW4All) initiative aims to ensure that every person, everywhere, is protected by life-saving early warning systems by 2027. Yet, the maritime navigation—particularly in offshore and remote oceanic regions— remains underserved in this global effort. The challenges arise mainly due to sparse meteorological data coverage, unreliable network connection, and a lack of tailored, context-aware alerting services designed specifically for sailors and seafarers.

In response, we present a novel AI- and ML-based maritime early warning system that provides real-time, personalized alerts for hazardous weather conditions, including dangerous sea states, strong winds, frontal systems, and convective thunderstorms.

The system integrates conventional numerical weather prediction (NWP) outputs and satellite-based remote sensing with machine learning algorithms and computer vision techniques to detect, track, and nowcast evolving hazardous features in the marine environment. Specifically, we apply front-detection algorithms to identify synoptic-scale boundaries using visual patterns in NWP outputs, while convective activity is monitored and nowcasted based on satellite imagery.

A key feature of the system is its user-centric design: alerts are dynamically adapted to individual vessel types, planned routes, and predefined risk thresholds, allowing for operationally relevant and personalized decision support. This personalization is crucial in fostering trust among end users, particularly in offshore sailing and commercial maritime operations.

Early results are promising, demonstrating the potential of AI to support marine situational awareness even in data-sparse regions. Looking forward, we aim to integrate deep learning models and ensemble forecasting techniques to enhance alert precision and better represent uncertainty in meteorological predictions.

By extending the reach of early warning systems to the open ocean, our solution contributes to the EW4All initiative’s goal of truly global coverage—bringing personalised, intelligent tools to improve safety at sea. It also exemplifies the role of private-sector innovation in complementing public meteorological infrastructure, reinforcing the value of collaborative frameworks.

How to cite: Zygarlowska, E., Dumard, C., and Rochut, B.: Integrating Machine Learning and Computer Vision for Personalized Maritime Weather Hazard Warnings in Data-Sparse and High-Uncertainty Environments, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-399, https://doi.org/10.5194/ems2025-399, 2025.