Search

Background: The mapping of tree species within Norwegian forests is a time-consuming process, involving forest associations relying on manual labeling by experts. The process can involve both aerial imagery, personal familiarity, or on-scene references, and remote sensing data. The state-of-the-art methods usually use high resolution aerial imagery with semantic segmentation methods. Methods: We present a deep learning based tree species classification model utilizing only lidar (Light Detection And Ranging) data. The lidar images are segmented into four classes (Norway Spruce, Scots Pine, Birch, background) with a U-Net based network. The model is trained with focal loss over partial weak labels. A major benefit of the approach is that both the lidar imagery and the base map for the labels have free and open access. Results: Our tree species classification model achieves a macro-averaged F1 score of 0.70 on an independent validation with National Forest Inventory (NFI) in-situ sample plots. That is close to, but below the performance of aerial, or aerial and lidar combined models.

Estimating building footprint maps from geospatial data is of paramount importance in urban planning, development, disaster management, and various other applications. Deep learning methodologies have gained prominence in building segmentation maps, offering the promise of precise footprint extraction without extensive post-processing. However, these methods face challenges in generalization and label efficiency, particularly in remote sensing, where obtaining accurate labels can be both expensive and time-consuming. To address these challenges, we propose terrain-aware self-supervised learning, tailored to remote sensing, using digital elevation models from LiDAR data. We propose to learn a model to differentiate between bare Earth and superimposed structures enabling the network to implicitly learn domain-relevant features without the need for extensive pixel-level annotations. We test the effectiveness of our approach by evaluating building segmentation performance on test datasets with varying label fractions. Remarkably, with only 1% of the labels (equivalent to 25 labeled examples), our method improves over ImageNet pre-training, showing the advantage of leveraging unlabeled data for feature extraction in the domain of remote sensing. The performance improvement is more pronounced in few-shot scenarios and gradually closes the gap with ImageNet pre-training as the label fraction increases. We test on a dataset characterized by substantial distribution shifts and labeling errors to demonstrate the generalizability of our approach. When compared to other baselines, including ImageNet pretraining and more complex architectures, our approach consistently performs better, demonstrating the efficiency and effectiveness of self-supervised terrain-aware feature learning.

During the past 20 years, numerous stellar streams have been discovered in both the Milky Way and the Local Group. These streams have been tidally torn from orbiting systems, which suggests that most of them should roughly trace the orbit of their progenitors around the Galaxy. As a consequence, they play a fundamental role in understanding the formation and evolution of our Galaxy. This project is based on the possibility of applying a technique developed by Binney in 2008 to various tidal streams and overdensities in the Galaxy. The aim is to develop an efficient method to constrain the Galactic gravitational potential, to determine its mass distribution, and to test distance measurements. Here we apply the technique to the Grillmair & Dionatos cold stellar stream. In the case of noise-free data, the results show that the technique provides excellent discrimination against incorrect potentials and that it is possible to predict the heliocentric distance very accurately. This changes dramatically when errors are taken into account, which wash out most of the results. Nevertheless, it is still possible to rule out spherical potentials and set constraints on the distance of a given stream., Comment: To appear in the proceedings of the IAU Symposium 289: Advancing the physics of cosmic distances, held in Beijing, China, August 2012, eds. Richard de Grijs and Giuseppe Bono, Cambridge Univ. Press

The direct detection of dark matter on Earth depends crucially on its density and its velocity distribution on a milliparsec scale. Conventional N-body simulations are unable to access this scale, making the development of other approaches necessary. In this paper, we apply the method developed in Fantin et al. 2008 to a cosmologically-based merger tree, transforming it into a useful instrument to reproduce and analyse the merger history of a Milky Way-like system. The aim of the model is to investigate the implications of any ultra-fine structure for the current and next generation of directional dark matter detectors. We find that the velocity distribution of a Milky Way-like Galaxy is almost smooth, due to the overlap of many streams of particles generated by multiple mergers. Only the merger of a 10^10 Msun analyse can generate significant features in the ultra-local velocity distribution, detectable at the resolution attainable by current experiments., Comment: 9 pages, 6 figures, accepted for publication in MNRAS

Various laboratory-based experiments are underway attempting to detect dark matter directly. The event rates and detailed signals expected in these experiments depend on the dark matter phase space distribution on sub-milliparsec scales. These scales are many orders of magnitude smaller than those that can be resolved by conventional N-body simulations, so one cannot hope to use such tools to investigate the effect of mergers in the history of the Milky Way on the detailed phase-space structure probed by the current experiments. In this paper we present an alternative approach to investigating the results of such mergers, by studying a simplified model for a merger of a sub-halo with a larger parent halo. With an appropriate choice of parent halo potential, the evolution of material from the sub-halo can be expressed analytically in action-angle variables, so it is possible to obtain its entire orbit history very rapidly without numerical integration. Furthermore by evolving backwards in time, we can obtain arbitrarily-high spatial resolution for the current velocity distribution at a fixed point. Although this model cannot provide a detailed quantitative comparison with the Milky Way, its properties are sufficiently generic that it offers qualitative insight into the expected structure arising from a merger at a resolution that cannot be approached with full numerical simulations. Preliminary results indicate that the velocity-space distribution of dark matter particles remains characterized by discrete and well-defined peaks over an extended period of time, both for single and multi-merging systems, in contrast to the simple smooth velocity distributions sometimes assumed in predicting laboratory experiment detection rates., Comment: 6 pages, 6 figures, accepted for publication in MNRAS