Grassland ecosystems play a vital role in biodiversity and carbon sequestration, but assessing these ecosystem services accurately is challenging due to their inherent spatial and temporal variability. Conventional field-based methods are often labor-intensive and may not capture this heterogeneity effectively. Recent progress in assessing grassland ecosystems has utilized a combination of structural and spectral data from Unmanned Aerial Vehicles (UAVs), showing promise for a thorough understanding of vegetation behavior. However, this method frequently overlooks an important factor — the horizontal variability within the vegetation, which significantly influences the precision of estimating plant characteristics, particularly in diverse ecosystems. Our study aims to fill this gap by incorporating texture analysis, a critical but often overlooked element in UAV-based assessments. Our research explored the potential of integrating various UAV-derived features to improve the estimation of above-ground biomass (AGB) and species richness in heterogeneous grasslands, key indicators of ecosystem health and productivity. This research investigated the efficacy of combining UAV-derived canopy height, multispectral data, and texture features for AGB and species richness estimation. The study was conducted in a heterogeneous wet grassland ecosystem, using a UAV equipped with multispectral sensors to capture high-resolution imagery. The imagery was processed to extract a range of features, including spectral indices, canopy height models, and textural information using Grey Level Co-occurrence Matrix methods. These features were then used to develop predictive models for AGB and species richness using advanced machine learning techniques, including Random Forest. Model performance was evaluated based on their predictive accuracy and ability to handle the spatial heterogeneity of grassland ecosystems. The study found that models integrating texture analysis with traditional spectral and structural data significantly improved predictive accuracy. For AGB estimation, the best models achieved an R² value of up to 0.84, with a relative root mean square error (rRMSE) of 26.58%. In predicting species richness, the most effective models reached an R² of 0.54 and a relative rRMSE of 31.95%. These results indicate an enhancement in estimation precision compared to models using traditional structural and spectral data types alone. This research demonstrated that UAV-based remote sensing, combined with a fusion of spectral, structural, and textural data, can improve the assessment of grassland characteristics such as AGB and species richness. The findings underscore the potential of integrated UAV-derived datasets in ecological monitoring and highlight the importance of advanced data processing and machine learning techniques in environmental research. This approach offers a promising avenue for more effective grassland management and conservation strategies, contributing to a deeper understanding of ecosystem dynamics.