This study addresses the issue of accurate and rapid aboveground biomass estimation in rangeland ecosystems, as traditional grazing methods are labor-intensive, while modern remote sensing techniques often require expensive equipment and controlled conditions. The goal of this work is to develop an efficient and accessible approach for biomass estimation of natural pastures based on ground-level RGB images captured with smartphones. For this purpose, a dataset consisting of 1196 field images and corresponding biomass values collected from 40 districts in southern Kazakhstan was used, and a wide + tiles architecture based on the DINOv3 model of Vision Transformer was proposed. The model utilized attention pooling and feature fusion mechanisms to integrate both global and local features, and various preprocessing and augmentation strategies were comparatively examined. Experimental results demonstrated that the proposed method exhibits high accuracy (with the best result being R² = 0.733, MAE ≈ 0.779 c/ha), where the DINOv3 model showed clear advantages over ConvNeXtV2. Furthermore, the impact of preprocessing strategies was minimal, and the importance of high-resolution images was clearly established. The obtained results show that the proposed method performs consistently under heterogeneous field conditions and allows for reliable biomass estimation without the need for specialized equipment. This makes it a practical tool for monitoring pastures, planning forage supply, and supporting agronomic decision-making.

Pasture degradation assessment is important for sustainable rangeland management, while conventional field surveys can be labor-intensive and remote-sensing approaches may require specialized sensing platforms or additional environmental data. This study proposes EcoPastureNet, a field-guided deep learning framework for four-level pasture degradation classification from ground-level RGB images. The framework uses DINOv3-L/16 as the visual backbone and incorporates overall vegetation cover, pasture productivity, grassland type, soil type, and salinity as privileged ecological information available only during training. Ecological knowledge is transferred through auxiliary prediction tasks and representation-level alignment, while inference requires only an RGB image. The model was evaluated on 419 pasture sampling points from 14 administrative districts and territories across four regions of southern and southeastern Kazakhstan using a strictly district-level geographic split. Among the image-only baselines, DINOv3-L/16 achieved a Macro-F1 of 0.6575. The full EcoPastureNet improved Macro-F1 to 0.7378, balanced accuracy to 0.7214, and achieved a quadratic weighted kappa of 0.8633. The largest class-specific improvement was observed for slight degradation, where the F1-score increased from 0.3333 to 0.6250. These results indicate that field-derived ecological supervision can improve image-based pasture degradation classification while preserving RGB-only inference.