Search

Numerous semi- and fully-automatic algorithms have been developed for individual tree detection from airborne laser-scanning data, but different rates of falsely detected treetops also accompany their results. In this paper, we proposed an approach that includes a machine learning-based refinement step to reduce the number of falsely detected treetops. The approach involves the local maxima filtering and segmentation of the canopy height model to extract different segment-level features used for the classification of treetop candidates. The study was conducted in a mixed temperate forest, predominantly deciduous, with a complex topography and an area size of 0.6 km × 4 km. The classification model's training was performed by five machine learning approaches: Random Forest (RF), Extreme Gradient Boosting, Artificial Neural Network, the Support Vector Machine, and Logistic Regression. The final classification model with optimal hyperparameters was adopted based on the best-performing classifier (RF). The overall accuracy (OA) and kappa coefficient (κ) obtained from the ten-fold cross validation for the training data were 90.4% and 0.808, respectively. The prediction of the test data resulted in an OA = 89.0% and a κ = 0.757. This indicates that the proposed method could be an adequate solution for the reduction of falsely detected treetops before tree crown segmentation, especially in deciduous forests. [ABSTRACT FROM AUTHOR]

ESA CCI SM products have provided remotely-sensed surface soil moisture (SSM) content with the best spatial and temporal coverage thus far, although its output spatial resolution of 25 km is too coarse for many regional and local applications. The downscaling methodology presented in this paper improves ESA CCI SM spatial resolution to 1 km using two-step approach. The first step is used as a data engineering tool and its output is used as an input for the Random forest model in the second step. In addition to improvements in terms of spatial resolution, the approach also considers the problem of data gaps. The filling of these gaps is the initial step of the procedure, which in the end produces a continuous product in both temporal and spatial domains. The methodology uses combined active and passive ESA CCI SM products in addition to in situ soil moisture observations and the set of auxiliary downscaling predictors. The research tested several variants of Random forest models to determine the best combination of ESA CCI SM products. The conclusion is that synergic use of all ESA CCI SM products together with the auxiliary datasets in the downscaling procedure provides better results than using just one type of ESA CCI SM product alone. The methodology was applied for obtaining SSM maps for the area of California, USA during 2016. The accuracy of tested models was validated using five-fold cross-validation against in situ data and the best variation of model achieved RMSE, R2 and MAE of 0.0518 m3/m3, 0.7312 and 0.0374 m3/m3, respectively. The methodology proved to be useful for generating high-resolution SSM products, although additional improvements are necessary. [ABSTRACT FROM AUTHOR]

The inventory of woody vegetation is of great importance for good forest management. Advancements of remote sensing techniques have provided excellent tools for such purposes, reducing the required amount of time and labor, yet with high accuracy and the information richness. Sentinel-2 is one of the relatively new satellite missions, whose 13 spectral bands and short revisit time proved to be very useful when it comes to forest monitoring. In this study, the novel spatio-temporal classification framework for mapping woody vegetation from Sentinel-2 multitemporal data has been proposed. The used framework is based on the probability random forest classification, where temporal information is explicitly defined in the model. Because of this, several predictions are made for each pixel of the study area, which allow for specific spatio-temporal aggregation to be performed. The proposed methodology has been successfully applied for mapping eight potential forest and shrubby vegetation types over the study area of Serbia. Several spatio-temporal aggregation approaches have been tested, divided into two main groups: pixel-based and neighborhood-based. The validation metrics show that determining the most common vegetation type classes in the neighborhood of 5 × 5 pixels provides the best results. The overall accuracy and kappa coefficient obtained from five-fold cross validation of the results are 82.97% and 0.75, respectively. The corresponding producer's accuracies range from 36.74% to 97.99% and user's accuracies range from 46.31% to 98.43%. The proposed methodology proved to be applicable for mapping woody vegetation in Serbia and shows a potential to be implemented in other areas as well. Further testing is necessary to confirm such assumptions. [ABSTRACT FROM AUTHOR]