Mangrove ecosystems are important in tropical and subtropical coastal zones, contributing to marine biodiversity and maintaining marine ecological balance. It is crucial to develop more efficient, intelligent, and accurate monitoring methods for mangroves to understand better and protect mangrove ecosystems. This study promotes a novel model, MangroveNet, for integrating multi-scale spectral and spatial information and detecting mangrove area. In addition, we also present an improved model, AttCloudNet+, to identify the distribution of mangrove species based on high-resolution multispectral drone images. These models incorporate spectral and spatial attention mechanisms and have been shown to effectively address the limitations of traditional methods, which have been prone to inaccuracy and low efficiency in mangrove species identification. In this study, we compare the results from MangroveNet with SegNet, UNet, and DeepUNet, etc. The findings demonstrate that the MangroveNet exhibits superior generalization learning capabilities and more accurate extraction outcomes than other deep learning models. The accuracy, F1_Score, mIoU, and precision of MangroveNet were 99.13%, 98.84%, 98.11%, and 99.14%, respectively. In terms of identifying mangrove species, the prediction results from AttCloudNet+ were compared with those obtained from traditional supervised and unsupervised classifications and various machine learning and deep learning methods. These include K-means clustering, ISODATA cluster analysis, Random Forest (RF), Support Vector Machines (SVM), and others. The comparison demonstrates that the mangrove species identification results obtained using AttCloudNet+ exhibit the most optimal performance in terms of the Kappa coefficient and the overall accuracy (OA) index, reaching 0.81 and 0.87, respectively. The two comparison results confirm the effectiveness of the two models developed in this study for identifying mangroves and their species. Overall, we provide an efficient solution based on deep learning with a dual attention mechanism in the acceptable real-time monitoring of mangroves and their species using high-resolution multispectral drone imagery.

Vegetation biomass and soil organic carbon (SOC) pools for the three representative forest types, i.e. conifer forest (CF), mixed conifer and broad-leaf forest (CBF), evergreen broad-leaf forest (EBF) in South China were investigated. We found that SOC stock of the three chief forest ranged from 55.54 to 151.16 MgCha^-1, and it increased with increasing vegetation biomass under the same type forest within 100cm depth. The organic carbon contents at an equivalent level of forest maturity tended to be in the following decreasing order: EBF > CBF > CF, various active organic carbon (AOC) fractions in the 0-20cm topsoil layer tended to be in the following decreasing order: light fraction carbon (LFC) ≈ particulate organic carbon (POC) > easily oxidisable carbon (EOC) > microbial biomass carbon (MBC) > water-soluble carbon (WSC). At an equivalent level of forest maturity, there was a trend that each of these five AOC fractions increased from CF to CBF to the EBF.

Mangrove ecosystems are important in tropical and subtropical coastal zones, contributing to marine biodiversity and maintaining marine ecological balance. It is cru-cial to develop more efficient, intelligent, and accurate monitoring methods for mangroves to understand better and protect mangrove ecosystems. This study promoted a novel model, MangroveNet, for integrating multi-scale spectral and spatial information and de-tecting mangrove area. In addition, we also presented an improved model—AttCloudNet+ to identify distribution of mangroves species based on high-resolution multispectral drone images. These models incorporate spectral and spatial attention mechanisms and have been shown to effectively address the limitations of traditional methods, which have been prone to inaccuracy and low efficiency in mangrove species identification. In this study, we compare the results from MangroveNet with the SegNet, UNet, and DeepUNet models. The findings demonstrate that the MangroveNet exhibits superior generalization learning capabilities and more accurate extraction outcomes than other deep learning models. The accuracy, F1_Score, mIoU, and precision of MangroveNet were 99.13%, 98.84%, 98.11%, and 99.14%, respectively. In terms of identifying mangrove species, the prediction results from AttCloudNet+ were compared with those obtained from traditional supervised and unsupervised classifications and various machine learning and deep learning methods. These included K-means clustering, ISODATA cluster analysis, random forest (RF), sup-port vector machines (SVM), and others. The comparison demonstrated that the mangrove species identification results obtained using AttCloudNet+ exhibited the most optimal performance in terms of the Kappa coefficient and the overall accuracy (OA) index, reach-ing 0.81 and 0.87, respectively. The two comparison results confirm the effectiveness of the two models developed in this study for identifying mangroves and their species. Overall, we provided an efficient solution based on deep learning with a dual attention mecha-nism in the acceptable real-time monitoring of mangroves and their species using high-resolution multispectral drone imagery.

Monitoring lake area changes with a higher spatial and temporal resolution can facilitate a more detailed analysis of climate-related changes in the Tibetan Plateau. In the Landsat era, optical remote sensing observation with water body index-based methods mainly contributed to alpine lake investigation. However, monitoring the seasonal or monthly change of a lake area is challenging since optical data are easily contaminated by the high cloud cover in the Tibetan Plateau. To cope with this, we generated new time series datasets including Sentinel-1 Synthetic Aperture Radar (SAR) and the Landsat-8 Operational Land Imager (OLI) observations. Meanwhile, we presented an improved deep learning model with spatial and channel attention mechanisms. Based on these datasets, we compared several deep learning models and found that the CloudNet+ had better performance. Taking this architecture as a baseline, we added spatial and channel attention mechanisms to generate our AttCloudNet+ for extracting the lake area. The results revealed that AttCloudNet+ had a better performance compared with the CloudNet+ and other CNNs (e.g., DeepLabv3+, UNet). For the accuracy of the lakeshore prediction, results from AttCloudNet+ demonstrated closer distance to the truth-value than other models. The obtained mean RMSE and MAE were 21.6 and 16.6 m, respectively. In contrast, the mean RMSE and MAE of the DeepLabv3+ were 99.5 and 76.0 m, while the corresponding RMSE and MAE for UNet were 91.1 and 64.9 m. In addition, we found our AttCloudNet+ was more robust than UNet and DeepLabv3+ because AttCloudNet+ is less influenced by the input optical images compared with DeepLabv3+ and UNet. By combining the results from different seasons and satellite sensors, we are capable of generating the complete lake area seasonal dynamics of the 15 largest lakes. The mean correlation coefficient (R²) between our seasonal lake area time series and the water level of LEGOS is 0.81, which is much better than the previous study (0.25). This indicates that our method can be used to monitor lake area seasonal variation, which is important for understanding regional climate change in the Tibetan Plateau and other similar areas.