Abstract
The availability of remote-sensing multisource data from optical-based satellite sensors has created new opportunities and challenges for forest monitoring in the Amazon Biome. In particular, change-detection analysis has emerged in recent decades to monitor forest-change dynamics, supporting some Brazilian governmental initiatives such as PRODES and DETER projects for biodiversity preservation in threatened areas. In recent years fully convolutional network architectures have witnessed numerous proposals adapted for the change-detection task. This paper comprehensively explores state-of-the-art fully convolutional networks such as U-Net, ResU-Net, SegNet, FC-DenseNet, and two DeepLabv3+ variants on monitoring deforestation in the Brazilian Amazon. The networks’ performance is evaluated experimentally in terms of Precision, Recall, F1-score, and computational load using satellite images with different spatial and spectral resolution: Landsat-8 and Sentinel-2. We also include the results of an unprecedented auditing process performed by senior specialists to visually evaluate each deforestation polygon derived from the network with the highest accuracy results for both satellites. This assessment allowed estimation of the accuracy of these networks simulating a process “in nature” and faithful to the PRODES methodology. We conclude that the high resolution of Sentinel-2 images improves the segmentation of deforestation polygons both quantitatively (in terms of F1-score) and qualitatively. Moreover, the study also points to the potential of the operational use of Deep Learning (DL) mapping as products to be consumed in PRODES.
Highlights
Deforestation is one of the most serious environmental problems today
The PRODES methodology is not based on a pixel-by-pixel classification, but uses (i) visual interpretation and manual vector editing for historical reasons and to provide high accuracy products, (ii) a minimum mapping area to maintain the historical series consistent, and (iii) a mapping scale of 1/75,000, because of the large extension of the Brazilian Amazon, the time-consuming visual interpretation and because a finer scale would not significantly improve the detection of deforestation polygons higher than the minimal mapping area
The assessment of the accuracy of Deep Learning (DL) models by the original PRODES map is relevant, the methodological differences between the two mapping techniques may lead to underestimations in accuracy metrics, as it assumes that the reference map represents the absolute truth at the pixel level
Summary
As the largest tropical forest in the world, the Amazon Rainforest has particular importance. It plays an essential role in carbon balance and climate regulation, provides numerous ecosystem services, and is among the most biodiverse biomes on earth [1]. With about 5 million km , the Brazilian Amazon occupies the largest area of the Amazon. Forest, covering about 65% of the total area. Until 1970, deforestation in the Brazilian Amazon comprised about 98,000 km , while in the last 40 years, the deforested area covered 730,000 km , which corresponds to twice the German territory and comprises nearly 18% of the area formerly covered by vegetation [2,3]
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have