Improved mangrove α-diversity estimation by coupling multispectral satellite images, canopy traits and soil properties

Abstract

Precise monitoring of spatial patterns and dynamic changes in mangrove biodiversity promote the sustainable development of mangrove ecosystems. However, there is a lack of a comprehensive knowledge about the role of canopy traits and soil properties on the remote estimation of mangrove species biodiversity. This study investigated 27 modeling strategies, encompassing three regression models (eXtreme gradient boosting, XGBoost; random forest, RF; partial least squares regression, PLSR), three data splitting methods (random splitting, RS; Kennard-Stone algorithm, KS; sample set partitioning based on joint x-y distance, SPXY), and three types of remote sensing datasets (high spatial resolution imagery from WorldView-2 (WV2), medium spatial resolution imagery from Sentinel-2 (S2), and their combination) in estimating the mangrove α-diversity indices (Simpson diversity index, SDI; Shannon-wiener diversity index, SHDI; Pielou evenness index, PEI) in Qinglan Provincial Nature Reserve, China. Moreover, this study aimed to examine whether the additional incorporation of plant-soil parameters (six canopy traits and six soil properties) could enhance estimation accuracy compared to the optimal model using image features alone. Among the 27 modeling strategies, the results demonstrated that the combination of WV2 and S2 images led to the XGBoost model using the KS splitting, the RF model using the SPXY splitting, and the XGBoost model using the SPXY splitting achieving the best performance in estimating SDI (R²_val = 0.731, RRMSE_val = 0.182, RPD = 1.735, RPD stands for residual prediction deviation in the validation), SHDI (R²_val = 0.631, RRMSE_val = 0.384, RPD = 1.646), and PEI (R²_val = 0.856, RRMSE_val = 0.170, RPD = 2.592), respectively. Based on these optimal models, the inclusion of canopy height and leaf SPAD (relative leaf chlorophyll content) further improved the accuracy of SDI estimation, while the addition of canopy height, soil C/N ratio in the 0–20 cm layer, leaf SPAD, and leaf TN enhanced SHDI estimation accuracy. Additionally, the incorporation of soil C/N ratio in the 20–40 cm layer notably increased the accuracy of PEI estimation. The inclusion of these variables led to an increase in R²_val by 6.07–17.65 %, a decrease in RRMSE_val by 10.00–17.58 %, and an improvement in RPD by 10.88–32.86 % compared to the original model in estimating the three α-diversity indices. We concluded that coupling multispectral satellite images, canopy traits and soil properties holds great potential in improving α-diversity estimation. The findings could provide methodological insights into precise mapping of biodiversity indices in mangrove forests and enhance understanding of the interaction between plant biodiversity, canopy traits, and soil properties.