UAV hyperspectral estimation of Suaeda salsa carotenoid content and response to soil water-salinity stress in degraded coastal wetlands following Spartina alterniflora removal

Carotenoids play a crucial role as accessory pigments in plant photosynthesis and are closely associated with physiological responses to environmental stress. Accurate, rapid, and non-destructive retrieval of carotenoid content (Car) and characterization of its spatial distribution are essential for understanding vegetation physiological status and its responses to environmental stress. In this study, Suaeda salsa (S. salsa), a dominant native species used in ecological restoration of Spartina alterniflora–cleared coastal wetlands, was investigated. Based on simulated and field-measured canopy spectral data, a combination of a three-band conceptual model and fractional-order derivative (FOD) methods was employed to develop and select carotenoid-sensitive spectral indices. On this basis, a data-fusion strategy integrating simulated data with different proportions of field observations was adopted. Four machine learning algorithms, including partial least squares regression (PLSR), support vector machine (SVM), particle swarm optimization–random forest (PSO-RF), and multilayer perceptron (MLP), were used to develop carotenoid estimation models. The optimal model was further applied to UAV hyperspectral imagery to generate pixel-level maps of carotenoid distribution. Soil physicochemical properties were incorporated to analyze the response of carotenoid spatial patterns to soil water–salinity stress. The results show that: (1) Three-band indices and the proposed GG types of fractional-order normalized difference indices such as FOD_NDVI_GG_0.25, FOD_NDVI_GG_0.50, FOD_NDVI_GG_0.75, FOD_NDVI_GG_1.00, and FOD_NDVI_GG_1.25 exhibited higher sensitivity to canopy carotenoid variation of S.salsa; (2) under a modeling strategy with 30% field data integration, the MLP model achieved the best performance (R2 = 0.853, RMSE = 1.727 μg cm−2, RPD = 2.532); (3) carotenoid content exhibits pronounced spatial heterogeneity and is strongly associated with soil water–salinity gradients, with lower values observed under higher stress conditions. Further analysis indicates that soil water–salinity conditions are a key factor constraining the recovery of S. salsa following Spartina alterniflora removal. UAV-based carotenoid retrieval effectively captures spatial variations in vegetation physiological stress, providing valuable support for physiological diagnosis, restoration assessment, and spatially explicit management of coastal wetland restoration.