Search for: All records

Amazon forests are characterized by rich structural diversity. However, the influence of factors such as topography, soil attributes, and external disturbances on structural variability is not always well characterized, and traditional structural metrics may be inadequate to capture this type of complexity. While LiDAR offers expanded structural insights, traditional parameters used in LiDAR analysis, such as mean or maximum canopy height, are not always well directly linked to environmental variables like topography. Emerging approaches merge LiDAR with machine learning to uncover deeper structural complexities. However, work to date may fail to fully utilize the potential of fine-scale LiDAR information. Here we introduce a novel approach, leveraging 2D point cloud images derived from a profiling canopy LiDAR (PCL). The technique targets intricate details within LiDAR point clouds by using deep learning algorithms. With a dataset from the Central Amazon comprising 18 multitemporal transects of 450 m in length, our objective was to detect structural "fingerprints" of varied topographical types along a hillslope, comprising: Riparian, White-sand, and Plateau, and to detect any gradient of structural shifts based on terrain variations here represented by the height above the nearest drainage (HAND). The dataset was trained and tested using a leave-one-group-out approach (LOGO) in which, for each iteration, a complete 450 m multitemporal transect was excluded from training and tested after each iteration. The fast.ai platform and a ResNet-34 architecture, coupled with transfer learning, were used to perform a classification to distinguish between three topographical types. Furthermore, a hybrid model combining a Convolutional Autoencoder, and Partial Least Square (PLS) regression was designed to detect forest structural gradient correlations with HAND variation. Cross-validation achieved a promising high weighted F1 score of 0.83 to classify forests based on the topographical types. Additionally, a combined Convolutional Autoencoder and PLS regression revealed a strong correlation (R2 = 0.76) between actual and predicted HAND. Innovatively combining deep learning with ground-based PCL LiDAR, our study revealed unique Amazon Forest structures connected to topographic variation. Our findings underscore the transformative potential of such integrative approaches for investigating forest dynamics and promise a powerful new tool for understanding climate-related forest structure change. 

Abstract. In humid tropical regions, irregular illumination and cloud shadows can complicate near-surface optical remote sensing. This could lead to costly and repetitive surveys to maintain geographical and spectral consistency. This could have a significant impact on the regular monitoring of forest ecosystems. A novel correction method using deep learning is presented here to address the issue in high-resolution canopy images. Our method involves training a deep learning model on one or a few well-illuminated/homogeneous reference images augmented with artificially generated cloud shadows. This training enables the model to predict illumination and cloud shadow patterns in any image and ultimately mitigate these effects. Using images captured by multispectral and RGB cameras, we evaluated the method across multiple sensors and conditions. These included nadir-view images from two sensors mounted on a drone and tower-mounted RGB Phenocams. The technique effectively corrects uneven illumination in near-infrared and true-color RGB images, including non-forested areas. This improvement was also evident in more consistent normalized difference vegetation index (NDVI) patterns in areas affected by uneven illumination. By comparing corrected RGB images to the original in a binary classification task, we evaluated the method's accuracy and Kappa values. Our goal was to detect non-photosynthetic vegetation (NPV) in a mosaic. The overall accuracy and Kappa were both significantly improved in corrected images, with a 2.5% and 1.1% increase, respectively. Moreover, the method can be generalized across sensors and conditions. Further work should focus on refining the technique and exploring its applicability to satellite imagery and beyond.