Vegetation encroachment near power transmission lines poses risks such as outages, fres, and increased maintenance costs. This study presents a method for estimating vegetation height using public satellite imagery and convolutional neural networks (CNNs). The approach involves segmenting dense vegetation areas, calculating height, and a new method for generating georeferenced alerts for risk analysis. Height estimates using Sentinel-2 and GEDI data showed a root mean square error (RMSE) of 7.7 meters and a mean absolute error (MAE) of 5 meters compared to high-resolution LiDAR data. The results demonstrate the originality of this study by identifying risk regions associated with the presence of vegetation near to power transmission lines, contributing to the improvement of vegetation management using public data in the electricity sector.

Vegetation encroachment is a major issue to the reliability of overhead power distribution networks, particularly in Brazilian urban areas where networks maintenance is still largely manual and the management uses the reactive approach instead of proactive solutions. This paper presents a decision-rule-based pipeline designed to automatically detect overhead power lines and vegetation contact areas using high-density mobile LiDAR data in Brazilian urban environments. The proposed method classifies point clouds into four primary classes: low-voltage cables, medium-voltage cables, poles, and trees in proximity to the power distribution network systems. The pipeline leverages geometric features derived from eigenvalue-based tensor analysis, height and density filters, Hough Transform, and region-growing techniques, to effectively segment and classify electric components and surrounding vegetation. The method was tested in two distinct urban scenarios, including suburban and downtown areas in Presidente Prudente, São Paulo, Brazil, with point densities exceeding 2 million points per square meter. Evaluation against reference datasets from a utility company demonstrated high precision and F-scores above 0.85 for power lines detection. Despite limitations related to parameter tuning, leak of reference data for tree detection evaluation, the pipeline offers a promising approach for semi-automatic annotation of LiDAR datasets. This process can support future applications in deep learning model training for urban asset monitoring and vegetation management. It is suggested that future works focus on reducing parameter dependency and enhancing vegetation classification reliability.

Understanding the stability of land intended for construction is crucial for the successful execution and long-term preservation of any project. This study compiles comprehensive data on geology, geomorphology, geotechnics, climate, precipitation, soil stability, soil taxonomy, and slope classification to establish a preliminary assessment of the Reina del Cisne sector in Paccha. Field investigations included the collection of soil samples for laboratory testing, as well as three-dimensional scans of a dwelling conducted on two separate occasions to assess structural deformation. By utilizing Scene and ReCap software to merge point clouds and analyzing the resulting data in CloudCompare, we identified significant deformation within the house's structure. The laboratory results, combined with the observed deformations, indicate that the terrain is susceptible to potential landslides, posing risks of human and economic loss. This study underscores the importance of integrating remote sensing techniques into urban disaster management and response strategies, emphasizing the need for sustainable development practices in vulnerable areas.

The Popocatepetl volcano (19.02° N, 98.62° W, 5425 masl) began its current eruptive phase at the end of 1994 and since then it has presented an eruptive history characterized by low and medium intensity events (VEI 1 to 3) [1]. Due to the elevation of the volcano's crater, these types of eruptive events place volcanic products such as ash at altitudes between 6 km and 8 km corresponding to the flight level ranges between FL180 and FL260. Volcanic ash placed at these flight levels is rapidly dispersed by wind, causing a large area of airspace used by air navigation to be frequently affected. During the years 1999 to 2023, the Washington VAAC reported a total of 2381 days with the presence of volcanic ash in the Popocatepetl region, as a result of eruptive events of which about 90% reached the airspace region previously mentioned, representing a serious risk to commercial aviation over this area.

To generate a preventive tool that can be used for the mitigation of risks in aviation due to the presence of volcanic ash in the airspace region, combined tools of Remote Sensing and Mathematical Modeling were used to identify the regions around the Popocatepetl volcano. We focus most likely on areas to be affected in the event of an eruption, taking into account the time of year in which it occurs.

First, an upper-level wind characterization study was carried out in the region of the Popocatepetl volcano to identify the behavioral patterns over the months of the year. Wind profiles in the atmosphere's vertical structure above the volcano crater were obtained from the Real-time Environmental Applications and Display sYstem (READY) web-based [2], using NOAA (National Oceanic and Atmospheric Administration) NCEP/NCAR (National Center for Environmental Prediction/National Center for Atmospheric Research) Reanalysis 1 data 4 times per day over the period 2000 to 2021.

Secondly, MODIS images concurrent to the development of the eruption were collected from the Terra and Aqua platforms. It was possible to identify 60% of the eruptive events reported by the Washington VAAC. Each image was analyzed for ash emission signature using the brightness temperature difference (BTD) between bands 31 (11 μm) and 32 (12 μm). The brightness temperature was obtained by the rearranged version of the Planck radiative transfer function formula [3].

Finally, to normalize the data of the regions identified in each eruptive event, we identified the MODIS images containing information about the start of the eruption and those with clouds without connection between the emission and the volcano crater. Then, the volcanic ash cloud dispersion model HYSPLIT developed by NOOA was used. With the help of HYSPLIT, the displacement pattern of the ash cloud is identified by comparing it with the satellite image. All eruptions will be standardized to identify the development of the ash cloud at a time of 8 hours after the eruption, rebuilding in cases where information was missing due to the lack of connection between the cloud and the crater of the volcano. Afterwards, the complete area affected by the volcanic event is identified.

The combination of these tools made it possible to identify patterns of ash dispersion emitted in the eruptive events of the Popocatepetl volcano. One pattern with displacement between NNW and ESE was identified for the months of November to May, while another pattern of dispersion between SSW and WNW occurred from July to September. This information can be used to create volcanic ash risk mitigation maps used in aviation safety for the Popocatepetl volcano region.

References

1. Jiménez-Escalona, J. C., Poom-Medina, J. L., Roberge, J., Aparicio-García, R. S., Avila-Razo, J. E., Huerta-Chavez, O. M., & Da Silva, R. F. (2022). Recognition of the Airspace Affected by the Presence of Volcanic Ash from Popocatepetl Volcano Using Historical Satellite Images. Aerospace, 9(6), 308.

2. Rolph, G.; Stein, A.; Stunder, B. Real-time Environmental Applications and Display system: READY. Environ. Model. Softw. 2017, 95, 210–228.

3. Wen, S.; Rose,W.I. Retrieval of sizes and total mass of particles in volcanic clouds using AVHRR bands 4 and 5. J. Geophys. Res. 1994, 99, 5421–5431.

In machine learning-based classifications, attribute selection plays a key role in building more efficient models for two main reasons: first, the selected attributes must adequately discriminate between classes; second, they must balance informational value and operational/computational cost. This raises a central question: What attributes and how many are truly essential to ensure an effective and resource-efficient classification? In the context of emergency mapping, particularly during disasters, attribute selection becomes even more critical. This work proposes an assessment of the dimensional trade-offs involved in feature selection, aiming to determine whether models using reduced sets of attributes can achieve performance comparable to more complex models that rely on extensive variable sets. The results indicate that it is possible to reduce the number of attributes, as well as the computational/operational costs associated with flood mapping, without significantly compromising the quality of the final product. This reduction allows to keep the efficiency of the mapping, making it easier its application to emergency contexts where agility and resources economy are essential