From weather radars to bird migration fluxes - Process-guided machine learning for spatio-temporal forecasting and inference - Chapter 6 - Reconstructing movements from imperfect radar data

Aug 4, 2026

Upload by Aurelia

Process-guided machine learning methods reconstruct fine-scale animal movement patterns from partial, noisy low-level weather radar data. The approach combines classical data assimilation with physics-informed deep learning by mapping high-resolution radar measurements to a low-dimensional latent space where Kalman filtering or smoothing performs inference. Temporal relationships are encoded through a locally linear Gaussian state-space model, while a physics-informed loss enforces mass conservation for physical consistency. The chapter motivates the inverse problem, coverage limits, and partial observability of density and radial velocity, highlighting challenges in ecological dynamics.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Reconstructing movements from imperfect radar data

Introduction

Physics-informed process-guided inference framework

UvA-DARE (Digital Academic Repository) From weather radars to bird migration fluxes Process-guided machine learning for spatio-temporal forecasting and inference Lippert, F. Publication date 2025 Link to publication Citation for published version (APA): Lippert, F. (2025) . From weather radars to bird migration fluxes: Process-guided machine learning for spatio-temporal forecasting and inference. [Thesis, fully internal, Universiteit van Amsterdam] . General rights It is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons) . Disclaimer/Complaints regulations If you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, stating your reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Ask the Library: [https://uba.uva.nl/en/contact](https://uba.uva.nl/en/contact), or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam, The Netherlands. You will be contacted as soon as possible. UvA-DARE is a service provided by the library of the University of Amsterdam ( [http](https://dare. uva. nl)[s](https://dare. uva. nl)[://dare. uva. nl](https://dare. uva. nl)) Download date:06 Jan 2026 6 RECONSTRUCTING MOVEMENTS FROM IMPERFECT RADAR DATA After focusing on the development, application and interpretation of continental-scale migration forecast models in Part I of this thesis, we now shift our attention to finescale local movement patterns. Specifically, we work towards new methods to infer these patterns from partial and noisy low-level weather radar data. In this chapter, we take a first step in this direction by combining classical data assimilation techniques with physics-informed deep learning. The key idea is to use neural networks to map high-resolution radar measurements into a low-dimensional latent space, in which we can perform inference using the Kalman filter or smoother. This approach encodes prior knowledge about the strong temporal relationships between bird densities and velocities in consecutive time steps via a linear state space model. To encourage physical consistency, we additionally introduce a physics-informed loss term that leverages known mass conservation constraints. 6.1 introduction Doppler weather radars provide high-resolution information about the distribution and movement of objects in the atmosphere. Although designed to monitor weather, radar beams are reflected not only by precipitation but also by animals passing the airspace around the radar. This offers invaluable opportunities for ecologists to study mass movements of birds, bats and insects that otherwise remain hidden due to low light conditions and high flight altitudes (Bauer et al., 2017) . However, since the measurement range of radars is limited, the spatial coverage of operational weather radar networks is typically incomplete (Kranstauber et al., 2020). Moreover, while the amount of energy reflected back to the radar antenna can be translated directly into animal density estimates, movements can be captured only partially by measuring the radial velocity, i.e. the component of movement along the direction of the radar beam, based on the Doppler shift (Doviak et al., 2006) . Inferring the complete underlying density and velocity fields from partial observations of This chapter is based on the following publication: F. Lippert, B. Kranstauber, E. E. van Loon, and P. D. Forré (2022b) .“Physics-informed inference of aerial animal movements from weather radar data.” In: NeurIPS 2022 Workshop on AI for Science reconstructing movements from imperfect radar data weather radar networks remains a challen

FAQs

What problem does the chapter address?

It addresses reconstructing fine-scale bird movement patterns from partial and noisy low-level weather radar measurements, which form a challenging inverse problem.

How does the proposed method perform inference?

It maps radar measurements into a low-dimensional latent space using neural networks, then runs inference with a Kalman filter or smoother under a linear Gaussian transition model.

Why is physics-informed learning used?

A physics-informed loss term leverages known mass conservation constraints, encouraging physically consistent reconstructions of densities and velocities across time steps.

Aug 4, 2026

Upload by Aurelia