An approach for accurate identification and monitoring of species in mangrove forests based on multi-source spectral data and deep learning

The mangroves conservation is essential due to the wide range of goods and ecosystem services. Despite their importance, mangrove distribution has drastically decreased in recent decades, highlighting the need to enhanced conservation efforts through innovative monitoring methods. Accordingly, this study presents an approach that integrates remote sensing data in a study area with a diverse range of ecological scenarios, comprising monospecific mangrove forests, which are dominated by a single species and mixed mangroves with flooded freshwater forested wetlands. First, the integration of photogrammetric analysis from unmanned aerial vehicles (UAVs) enables the precise determination of the spatial distribution of mangrove forests. Furthermore, the multispectral data obtained from the Sentinel-2 satellite was employed for the training of multilayer perceptron models (MLP) that are capable of accurately mapping mixed and monospecific mangrove forests. The computational experiments for MLP models involve different number of neurons and hidden layers. The best models for detecting mangrove forest achieved an accuracy of 99.95 % for training and 99.8 % for test, while for monospecific mangrove forest, at the species level, it attained an accuracy of 98.73 % for training and 96.84 % for test, in both cases the models demonstrated a great performance test for segmentation of different kind of mangrove forests. By leveraging multi-source data and the capabilities of remote sensing and artificial intelligence technologies, this approach offers a groundbreaking solution for monitoring mangrove ecosystems. Also, this minimizes the economic and human efforts and reduces risk of error, thus leading to more efficient and effective ecosystem conservation. • Effective Mangrove monitoring through deep learning and remote sensing. • Spectral and spatial identification of Mangrove communities. • Classification by type of mangrove: Rhizophora mangle , Avicennia germinans , and Laguncularia racemosa. • Low-cost methodology and high scalability. • Mangrove characterization and diagnosis.