PFG-Journal of Photogrammetry Remote Sensing and Geoinformation Science最新文献

Remote sensing approaches using Unmanned Aerial Vehicles (UAVs) have become an established method to monitor agricultural systems. They enable data acquisition with multi- or hyperspectral, RGB, or LiDAR sensors. For non-destructive estimation of crop or sward traits, photogrammetric analysis using Structure from Motion and Multiview Stereopsis (SfM/MVS) has opened a new research field. SfM/MVS analysis enables the monitoring of plant height and plant growth to determine, e.g., biomass. A drawback in the SfM/MVS analysis workflow is that it requires ground control points (GCPs), making it unsuitable for monitoring managed fields which are typically larger than 1 ha. Consequently, accurately georeferenced image data acquisition would be beneficial as it would enable data analysis without GCPs. In the last decade, substantial progress has been achieved in integrating real-time kinematic (RTK) positioning in UAVs, which can potentially provide the desired accuracy in cm range. Therefore, to evaluate the accuracy of crop and sward height analysis, we investigated two SfM/MVS workflows for RTK-tagged UAV data, (I) without and (II) with GCPs. The results clearly indicate that direct RTK-georeferenced UAV data perform well in workflow (I) without using any GCPs (RMSE for Z is 2.8 cm) compared to the effectiveness in workflow (II), which included the GCPs in the SfM/MVS analysis (RMSE for Z is 1.7 cm). Both data sets have the same Ground Sampling Distance (GSD) of 2.46 cm. We conclude that RTK-equipped UAVs enable the monitoring of crop and sward growth greater than 3 cm. At greater plant height differences, the monitoring is significantly more accurate.

Tropical montane forests (TMFs) are highly valuable for their above-ground biomass (AGB) and their potential to sequester carbon, but they remain understudied. Sentinel-1, -2, biophysical data and Machine Learning were used to estimate and map the AGB and above-ground carbon (AGC) stocks in Benguet, Philippines. Non-destructive field AGB measurements were collected from 184 plots, revealing that pine forests had 33.57% less AGB than mossy forests (380.33 Mgha⁻¹), whilst the grassland summit had 39.93 Mgha⁻¹. In contrast to the majority of literature, AGB did not decrease linearly with elevation. NDVI, LAI, fAPAR, fCover and elevation were the most effective predictors of field-derived AGB as determined by Random Forest (RF) feature selection in R. WEKA was used to evaluate and validate 26 Machine Learning algorithms. The results show that the Machine Learning K star (K*) (r = 0.213–0.832; RMSE = 106.682 Mgha⁻¹–224.713 Mgha⁻¹) and RF (r = 0.391–0.822; RMSE = 108.226 Mgha⁻¹–175.642 Mgha⁻¹) exhibited high modelling capabilities to estimate AGB across all predictor categories. Consequently, spatially explicit models were carried out in Whitebox Runner software to map the study site’s AGB, demonstrating RF with the highest predictive performance (r = 0.982; RMSE = 53.980 Mgha⁻¹). The study area’s carbon stock map ranged from 0 to 434.94 Mgha⁻¹, highlighting the significance of forests at higher elevations for forest conservation and carbon sequestration. Carbon-rich mountainous regions of the county can be encouraged for carbon sequestration through REDD + interventions. Longer wavelength radar imagery, species-specific allometric equations and soil fertility should be tested in future carbon studies. The produced carbon maps can help policy makers in decision-planning, and thus contribute to conserve the natural resources of the Benguet Mountains.