Synergistic hyperspectral and SAR imagery retrieval of mangrove leaf area index using adaptive ensemble learning and deep learning algorithms

Abstract

Accurate mapping of leaf area index (LAI) is essential for mangrove conservation and restoration. This study proposes a new approach to the retrieval of the mangrove LAI by combining a one-dimensional convolutional neural network (1D-CNN) with adaptive ensemble learning regression (AELR) and deep learning regression (DNNR) algorithms. We further evaluated the performance of OHS (Zhuhai-1) hyperspectral and GF-3 SAR images in mapping the spatial distribution of the LAI in mangroves. Finally, the outputs of the AELR and DNNR models were interpreted, and the interactions between different image features were clarified to select the sensitive spectral ranges and vegetation indexes for estimating the mangrove LAI using SHAP (Shapley additive explanation). We confirmed that 1D-CNN + DNNR provided an effective approach to estimating the mangrove LAI, as it produced a higher-accuracy inversion (R2 = 0.8685) than that of the AELR model. It was found in this study that the 1D-CNN improved the retrieval accuracy (R2) of the mangrove LAI from 0.097 to 0.1297 when compared with the traditional data dimension reduction (DDR) method, which demonstrated that the 1D-CNN was able to improve the inversion accuracy of the mangrove LAI. This study revealed that the synergistic use of OHS hyperspectral and GF-3 SAR images (R2 = 0.8685, RMSE = 0.134) outperformed any of the lone datasets in the inversion of the mangrove LAI. Finally, this study provided explanations and interpretations of the outputs of the DNNR and AELR models, and it was found that the optimal spectral ranges for estimating the mangrove LAI are 637–671 nm and 802–822 nm; H19 ((NIR/Red)/Red), the NDVI, H14 (Red edge/Red), and the EMVI ((Green-SWIR2)/(SWIR1-Green)) provided important contributions for mapping the mangrove LAI. These results provide a scientific foundation for the preservation and restoration of coastal mangroves.