TorchGeo

Avalanches pose a significant threat to the population and infrastructure of mountainous regions. The mapping and documentation of avalanches in Austria is mostly done by experts during field observations and covers usually only specific localized areas. A comprehensive mapping of avalanches is, however, crucial for the work of local avalanche commissions as well as avalanche warning services to assess, e.g., the avalanche danger. Over the past decade, mapping avalanches from satellite imagery has proven to be a promising and rapid approach to monitor avalanche activity in specific regions. Several recent avalanche detection approaches use deep learning-based algorithms to improve detection rates compared to traditional segmentation algorithms. Building on the success of these deep learning-based approaches, we present the first steps to build a modular data pipeline to map historical avalanche cycles in Copernicus Sentinel-1 imagery of the Austrian Alps. The Sentinel-1 mission has provided free all-weather synthetic aperture radar data since 2014, which has proven suitable for avalanche mapping in a Norwegian test area. In addition, we present a roadmap for setting up a segmentation algorithm, in which a general U-Net approach will serve as a baseline and will be compared with the mapping results of additional algorithms initially applied to autonomous driving. We propose to train the U-Net using labeled training dataset of avalanche outlines from Switzerland, Norway and Greenland. Due to the lack of training and validation data from Austria, we plan to compile the first avalanche archive for Austria. Meteorological variables, e.g., precipitation or wind, are highly important for the release of avalanches. In a completely new approach, we will therefore consider weather station data or outputs of numerical weather models in the learning-based algorithm to improve the detection performance. The mapping results in Austria will be complemented with pointwise field measurements of the MOLISENS platform and the RIEGL VZ-6000 terrestrial laser scanner.

show abstract

“…"DEM" stands for digital elevation model. Where the ML algorithms are implemented in PyTorch with TorchGeo (Stewart et al, 2022).…”

Section: Figurementioning

confidence: 99%

Automated snow avalanche monitoring for Austria: State of the art and roadmap for future work

Kapper

Goelles

Muckenhuber

et al. 2023

Front. Remote Sens.

View full text Add to dashboard Cite

show abstract

“…Crop recognition from remote sensing data has been a significant area of research for several decades now. Several methods based on machine learning algorithms have been developed for crop recognition using geospatial and remote sensing data [32,33]. The products developed for crop recognition are generally based on state statistics geodata, which enables the creation of a training sample distributed throughout the study area [34].…”

Section: Introductionmentioning

confidence: 99%

“…Factors such as growth stage, weather conditions, and agricultural practices contribute to the unique spectral signatures of crops. This variability makes it difficult to accurately establish universal algorithms and models to identify crops [32,33]. Certain crops possess similar spectral signatures, leading to spectral signature overlap [35].…”

Section: Introductionmentioning

confidence: 99%

“…Discriminating between crops with similar spectral characteristics, such as wheat and barley, requires advanced algorithms that consider additional contextual information [36]. Remote sensing data from satellites often need to improve their spatial resolution [32,33]. This limitation poses challenges when differentiating between crops occupying small areas, or when adjacent crops share the same pixel [35].…”

Section: Introductionmentioning

confidence: 99%

“…Accounting for the temporal dynamics and capturing the phenological variations of crops is essential accurate and reliable identification [33,36]. Various factors, such as cloud cover, atmospheric haze, and shadows, can interfere with the quality and availability of satellite imagery used for crop identification [32]. These interferences introduce noise and artifacts in the data, which affect the accuracy of crop classification algorithms [32].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Advancing Agricultural Crop Recognition: The Application of LSTM Networks and Spatial Generalization in Satellite Data Analysis

Gafurov,

Mukharamova,

Saveliev

et al. 2023

Agriculture

View full text Add to dashboard Cite

This study addresses the challenge of accurate crop detection using satellite data, focusing on the application of Long Short-Term Memory (LSTM) networks. The research employs a “spatial generalization” approach, where a model trained on one geographic area is applied to another area with similar vegetation conditions during the growing season. LSTM networks, which are capable of learning long-term temporal dependencies, are used to overcome the limitations of traditional machine learning techniques. The results indicate that LSTM networks, although more computationally expensive, provide a more accurate solution for crop recognition compared with other methods such as Multilayer Perceptron (MLP) and Random Forest algorithms. The accuracy of LSTM networks was found to be 93.7%, which is significantly higher than the other methods. Furthermore, the study showed a high correlation between the real and model areas of arable land occupied by different crops in the municipalities of the study area. The main conclusion of this research is that LSTM networks, combined with a spatial generalization approach, hold great promise for future agricultural applications, providing a more efficient and accurate tool for crop recognition, even in the face of limited training data and complex environmental variables.

show abstract

MobilityDL: a review of deep learning from trajectory data

Graser,

Jalali,

Lampert

et al. 2024

Geoinformatica

View full text Add to dashboard Cite

Trajectory data combines the complexities of time series, spatial data, and (sometimes irrational) movement behavior. As data availability and computing power have increased, so has the popularity of deep learning from trajectory data. This review paper provides the first comprehensive overview of deep learning approaches for trajectory data. We have identified eight specific mobility use cases which we analyze with regards to the deep learning models and the training data used. Besides a comprehensive quantitative review of the literature since 2018, the main contribution of our work is the data-centric analysis of recent work in this field, placing it along the mobility data continuum which ranges from detailed dense trajectories of individual movers (quasi-continuous tracking data), to sparse trajectories (such as check-in data), and aggregated trajectories (crowd information).

show abstract

TorchGeo

Cited by 21 publications

References 41 publications

Automated snow avalanche monitoring for Austria: State of the art and roadmap for future work

Automated snow avalanche monitoring for Austria: State of the art and roadmap for future work

Advancing Agricultural Crop Recognition: The Application of LSTM Networks and Spatial Generalization in Satellite Data Analysis

MobilityDL: a review of deep learning from trajectory data

Contact Info

Product

Resources

About