arXiv subjects

Marius Appel

Publications and source records attributed to Marius Appel.

4 recordsLinked to original sources

arXiv2026

A Machine Learning API for Earth Observation Data Cubes Based on openEO

Earth Observation (EO) data are increasingly organized as spatio-temporal data cubes, while machine learning (ML) methods operate on tabular feature matrices or structured tensor inputs. This mismatch forces platform-specific transformations that are difficult to reproduce or transfer across cloud infrastructures. The openEO specification provides a unified interface for EO data access and processing across heterogeneous backends, but lacks a standardized approach for ML integration. We propose a process-level ML specification for openEO structured into three stages: model initialization, model actions (training, tuning, inference, validation), and model management. It supports classical algorithms such as Random Forest and SVM, as well as deep learning architectures for time-series and spatial patch-based modeling, including TempCNN, Temporal Attention Encoders, and foundation model inference. Three prototype implementations in R and Python demonstrate feasibility across diverse technology stacks. A crop type mapping use case demonstrates cross-backend interoperability by submitting an identical process graph to independent R and Python backends and comparing predictions and evaluation metrics. Two further use cases demonstrate deep learning on time series and foundation model inference, each executed on a dedicated backend. The prototypes reveal, however, that full cross-backend portability requires deeper harmonization of serialization formats and execution semantics than the process level alone can enforce; backend library versions and preprocessing conventions outside the specification's boundary also affect reproducibility. Addressing both through explicit backend conformance profiles represents the most important near-term direction. The specification advances the reproducibility, portability, and accessibility of ML workflows on EO data cubes across cloud platforms.

cs.LG↗

arXiv2025

Spatial Data Science Languages: commonalities and needs

Recent workshops brought together several developers, educators and users of software packages extending popular languages for spatial data handling, with a primary focus on R, Python and Julia. Common challenges discussed included handling of spatial or spatio-temporal support, geodetic coordinates, in-memory vector data formats, data cubes, inter-package dependencies, packaging upstream libraries, differences in habits or conventions between the GIS and physical modelling communities, and statistical models. The following set of insights have been formulated: (i) considering software problems across data science language silos helps to understand and standardise analysis approaches, also outside the domain of formal standardisation bodies; (ii) whether attribute variables have block or point support, and whether they are spatially intensive or extensive has consequences for permitted operations, and hence for software implementing those; (iii) handling geometries on the sphere rather than on the flat plane requires modifications to the logic of {\em simple features}, (iv) managing communities and fostering diversity is a necessary, on-going effort, and (v) tools for cross-language development need more attention and support.

stat.CO↗

arXiv2022

Efficient data-driven gap filling of satellite image time series using deep neural networks with partial convolutions

The abundance of gaps in satellite image time series often complicates the application of deep learning models such as convolutional neural networks for spatiotemporal modeling. Based on previous work in computer vision on image inpainting, this paper shows how three-dimensional spatiotemporal partial convolutions can be used as layers in neural networks to fill gaps in satellite image time series. To evaluate the approach, we apply a U-Net-like model on incomplete image time series of quasi-global carbon monoxide observations from the Sentinel-5P satellite. Prediction errors were comparable to two considered statistical approaches while computation times for predictions were up to three orders of magnitude faster, making the approach applicable to process large amounts of satellite data. Partial convolutions can be added as layers to other types of neural networks, making it relatively easy to integrate with existing deep learning models. However, the approach does not quantify prediction errors and further research is needed to understand and improve model transferability. The implementation of spatiotemporal partial convolutions and the U-Net-like model is available as open-source software.

cs.LG↗

arXiv2020

Spatiotemporal Multi-Resolution Approximations for Analyzing Global Environmental Data

Technological developments and open data policies have made large, global environmental datasets accessible to everyone. For analysing such datasets, including spatiotemporal correlations using traditional models based on Gaussian processes does not scale with data volume and requires strong assumptions about stationarity, separability, and distance measures of covariance functions that are often unrealistic for global data. Only very few modeling approaches suitably model spatiotemporal correlations while addressing both computational scalability as well as flexible covariance models. In this paper, we provide an extension to the multi-resolution approximation (MRA) approach for spatiotemporal modeling of global datasets. MRA has been shown to be computationally scalable in distributed computing environments and allows for integrating arbitrary user-defined covariance functions. Our extension adds a spatiotemporal partitioning, and fitting of complex covariance models including nonstationarity with kernel convolutions and spherical distances. We evaluate the effect of the MRA parameters on estimation and spatiotemporal prediction using simulated data, where computation times reduced around two orders of magnitude with an increase of the root-mean-square prediction error of around five percent. This allows for trading off computation times against prediction errors, and we derive a practical strategy for selecting the MRA parameters. We demonstrate how the approach can be practically used for analyzing daily sea surface temperature and precipitation data on global scale and compare models with different complexities in the covariance function.

stat.CO↗