Combining spectral and texture indexes predicts plant diversity in temperate riverine wetlands of the lower Yellow River, China

• Optimized texture step size/direction boosted plant diversity correlation. • The combination of spectral and texture indexes improved the prediction accuracy. • Scale effect is an important factor to be considered in predicting plant diversity. • The results of UAV RGB image in predicting plant diversity were encouraging. • GF-1D multispectral image also had a good performance in estimating plant diversity. Remote sensing technology offers a cost-effective and efficient solution for the large-scale monitoring of wetland plant diversity. This study introduces a methodological framework that integrates spectral and texture indexes for predicting plant diversity by systematically quantifying the impacts of direction and step size on texture-diversity relationships. Firstly, we evaluated the correlation between diversity and spectral indexes derived from UAV RGB and GF-1D multispectral data. Then, GLCM texture indexes for 4 directions (0°, 45°, 90° and 135°) and different step sizes (1, 3, 5, 7, 9, 11 and 15 pixels for UAV RGB image; 1, 3 and 5 pixels for GF-1D image) were computed, and the influence of direction and step size on the correlation between texture and diversity indexes was examined. After index screening, we investigated the performance of the combination of spectral and texture indexes in predicting plant diversity at 2-m and 10-m quadrat scales. Results showed that: 1) Most texture indexes exhibited sensitivity to both direction and step size. Their correlation with the diversity indexes varied due to differing directional influences, or it may show a gradual increase or decrease in correlation as step size increases. In some cases, a negative correlation can even transition into a positive one. 2) Compared to using spectral indexes alone, the integration of spectral and texture indexes improved prediction R 2 by an average of 0.06 (2-m quadrat scale with UAV RGB imagery), 0.21 (10-m quadrat scale with UAV RGB imagery), and 0.25 (10-m quadrat scale with GF-1D multispectral imagery). 3) The Erigeron canadensis community and Tamarix chinensis + Phragmites australis community exhibited divergent responsiveness to spectral and texture indexes. The combined spectral-texture model demonstrated higher prediction accuracy for plant diversity indexes in single-type communities (e.g., mean R 2 = 0.64 for the Erigeron canadensis community and 0.58 for the Tamarix chinensis + Phragmites australis using GF-1D data) compared to the merged community (e.g., mean R 2 = 0.33 using GF-1D data). 4) The linear fitting of UAV RGB spectral-texture integration demonstrated progressive improvement as the quadrat scale expanded from 2 m to 10 m.