Forecasting refugee migration is challenging, exacerbated by the high dimensional and dynamic nature of its drivers, such as climatic, economic, and political stressors. This article introduces a novel forecasting framework based on the Dynamic Elastic Net (DynENet) algorithm, which incorporates a time-varying regularization and a new model selection criterion: the penalized deviance ratio (PDR). Unlike conventional metrics such as the deviance ratio (DR), which emphasize in-sample fit, PDR explicitly penalizes model complexity, enhancing generalization in high-dimensional covariate setting. We apply this framework to forecast asylum-seeker rates (ASR) from Somalia to EU member states, leveraging a comprehensive set of district-level predictors. Extensive validation demonstrates that PDR-tuned models consistently outperform DR-based benchmarks in out-of-sample accuracy, reducing average point prediction errors by 40% and improving interval forecasts by 79%. Furthermore, we demonstrate how the DynENet framework supports explanatory insights at multiple levels-origin district, destination, and temporal-revealing both persistent and transient nature of migration drivers. The proposed methodology not only advances forecasting accuracy under high-dimensional covariate conditions but also enhances the interpretability of complex and evolving migration systems.
Relative navigation is a fundamental task in space proximity operations and autonomous rendezvous. The nonlinear relative orbital dynamics are equivariant and admit a semidirect product Lie group symmetry. This property is used to design an Equivariant Filter (EqF). The filter estimates the relative position and velocity between a chaser spacecraft and a target by lifting the filter dynamics to the group, while respecting the underlying geometry of the problem. Simulations demonstrate the filter's performance and effectiveness.
Digital television signals received from a geostationary satellite can be used as signals of opportunity to calculate and predict the satellite orbit. This is an inexpensive and practical way to evaluate satellite collision risks, a key aspect of space surveillance and space safety. The technique is based on measuring the Time Difference of Arrival (TDoA) between pairs of tracking stations located across the satellite footprint, using a high-gain parabolic antenna and a signal digitiser. To calculate the TDoA accurately, the stations must timestamp the measurements according to a common time reference, which is achieved by using a GNSS receiver. Since the GNSS equipment setup can be different at each station (e.g., different cable lengths), it is necessary to calibrate the total GNSS chain delay and compensate for it. Also, the signal from the geostationary satellite itself undergoes a time delay as it travels across the tracking station hardware, which should also be accounted for. This paper describes GMV's Focusear passive tracking system, with an emphasis on signal delay calibration aspects.
The availability of realistic covariance information for the orbit of every Resident Space Object (RSO) contained in a catalogue is crucial for Space Situational Awareness activities, e.g., collision avoidance services. The most comprehensive of these catalogues is the Special Perturbations Catalogue (SPCAT), maintained by the U.S. 18th Space Defense Squadron. The SPCAT is the high-precision ephemeris version of the Two Line Elements RSOs catalogue, publicly available on databases such as Space Track and Celestrak. However, covariance information is not provided with the mean state of the SPCAT ephemerides. So-called observed covariance values can be obtained via a comparison procedure between consecutive orbit information updates referring to the same SPCAT RSO. This paper proposes new methodologies for calculating covariance values for catalogues deprived of such information, including the application and adaptation of existing data-fusion methods from literature. The main final goal is to compute covariance matrices that are more realistic and reliable than those obtained with the currently available methods. Another key objective is the integration of the new methodology in an operational environment. Computational efficiency is then a relevant factor, and the baseline method to be developed is selected and improved taking into account such efficiency criterion. A new routine that considers the Orbit Determination epoch of each RSO ephemeris arc to coherently combine covariances based on their propagation time is developed and implemented. Two fusion methods are deployed, Covariance Intersection and Covariance Union, and the realism of the results is tested with a well-established metric, the Mahalanobis distance and its fitting of the Chi-square distribution according to appropriate Empirical Distribution Function tests such as Cramer-von Mises. The realism of the combined covariances is validated against precise ephemeris of LEO Sentinel satellites. While Covariance Intersection is proved inadequate as a stand-alone fusion method due to the characteristics of the SPCAT observed covariances, Covariance Union provides covariance values that are consistently more realistic than the ones obtained with the baseline method.
This paper presents the development (including data gathering, transformation, training and validation) and performance evaluation of a time-series Transformer neural-network model for GPS satellite clock-bias correction prediction for up to two-hour ahead of time horizon. The motivation is to provide clock-bias correction for a stand-alone receiver. The training and validation leverages IGS final clock products as the ground truth to compare predictions. The developed model provides fast and reliable forecasting of clock-bias corrections for stand-alone (without network connections) single-frequency GNSS receivers without changing the infrastructure of the receivers, such as adding additional sensors. From the results, the clock-bias correction prediction can achieve less than 2 ns. The application of predicted clock-bias corrections can improve clock-bias prediction up to 50% more accurate than IGS rapid products and CODE-MGEX products. The practicality of this forecasting capability is to improve the accuracy of a remote and low-cost receiver, such as single frequency receiver, by correcting the clock-bias error component offline without the need of an internet connection.
