Search

[Display omitted] • A novel UAV-based snow depth retrieval method combining MSI imagery and direct georeferencing was tested in Northern Sweden. • The preprocessing routine improved the automatic feature detection in homogeneous snow areas generating 10× more key points. • The importance of red and NIR bands for 3D reconstruction of snow areas has been highlighted and quantified through PCA. • Snow depth maps were estimated with high accuracy (RSME 11.52 cm) for optimal illumination conditions. • The data set composed by UAV multispectral imagery, GNSS and IMU data, and ground validation is openly and freely available. More accurate snow quality predictions are needed to economically and socially support communities in a changing Arctic environment. This contrasts with the current availability of affordable and efficient snow monitoring methods. In this study, a novel approach is presented to determine spatial snow depth distribution in challenging alpine terrain that was tested during a field campaign performed in the Tarfala valley, Kebnekaise mountains, northern Sweden, in April 2019. The combination of a multispectral camera and an Unmanned Aerial Vehicle (UAV) was used to derive three-dimensional (3D) snow surface models via Structure from Motion (SfM) with direct georeferencing. The main advantage over conventional photogrammetric surveys is the utilization of accurate Real-Time Kinematic (RTK) positioning which enables direct georeferencing of the images, and therefore eliminates the need for ground control points. The proposed method is capable of producing high-resolution 3D snow-covered surface models (< 7 cm/pixel) of alpine areas up to eight hectares in a fast, reliable and affordable way. The test sites' average snow depth was 160 cm with an average standard deviation of 78 cm. The overall Root-Mean-Square Errors (RMSE) of the snow depth range from 11.52 cm for data acquired in ideal surveying conditions to 41.03 cm in aggravated light and wind conditions. Results of this study suggest that the red components in the electromagnetic spectrum, i.e., the red, red edge, and near-infrared (NIR) band, contain the majority of information used in photogrammetric processing. The experiments highlighted a significant influence of the multi-spectral imagery on the quality of the final snow depth estimation as well as a strong potential to reduce processing times and computational resources by limiting the dimensionality of the imagery through the application of a Principal Component Analysis (PCA) before the photogrammetric 3D reconstruction. The proposed method is part of closing the scale gap between discrete point measurements and regional-scale remote sensing and complements large-scale remote sensing data and snow model output with an adequate validation source. [ABSTRACT FROM AUTHOR]

Accurate burned area information is needed to assess the impacts of wildfires on people, communities, and natural ecosystems. Various burned area detection methods have been developed using satellite remote sensing measurements with wide coverage and frequent revisits. Our study aims to expound on the capability of deep learning (DL) models for automatically mapping burned areas from uni-temporal multispectral imagery. Specifically, several semantic segmentation network architectures, i.e., U-Net, HRNet, Fast-SCNN, and DeepLabv3+, and machine learning (ML) algorithms were applied to Sentinel-2 imagery and Landsat-8 imagery in three wildfire sites in two different local climate zones. The validation results show that the DL algorithms outperform the ML methods in two of the three cases with the compact burned scars, while ML methods seem to be more suitable for mapping dispersed burn in boreal forests. Using Sentinel-2 images, U-Net and HRNet exhibit comparatively identical performance with higher kappa (around 0.9) in one heterogeneous Mediterranean fire site in Greece; Fast-SCNN performs better than others with kappa over 0.79 in one compact boreal forest fire with various burn severity in Sweden. Furthermore, directly transferring the trained models to corresponding Landsat-8 data, HRNet dominates in the three test sites among DL models and can preserve the high accuracy. The results demonstrated that DL models can make full use of contextual information and capture spatial details in multiple scales from fire-sensitive spectral bands to map burned areas. Using only a post-fire image, the DL methods not only provide automatic, accurate, and bias-free large-scale mapping option with cross-sensor applicability, but also have potential to be used for onboard processing in the next Earth observation satellites. [ABSTRACT FROM AUTHOR]