Abstract: Agricultural machinery is a significant source of non-road mobile emissions, yet accurate estimation of its carbon and pollutant emissions is hindered by the lack of real-world fuel consumption data and localized emission factors. To address this issue and support the "Dual Carbon" goals, this study developed a comprehensive predictive framework for tractor fuel consumption,CO_2, and NO_x emissions from tractors by integrating massive Beidou trajectory big data with Portable Emission Measurement System (PEMS) measured data, and estimated the national operational emission inventory. Firstly, using samples with complete fields from the Beidou trajectory data and incorporating multidimensional features such as tractor speed, direction, engine speed, and torque, a feature set for fuel consumption prediction was constructed. Using a dataset of trajectory samples with complete fields from the 2023 summer harvest, a feature engineering process was conducted. The trajectories were segmented into kinematic fragments with a time step of 100s. A Residual Deep Neural Network (ResDNN) model with a 21-layer structure was employed to predict fuel consumption, achieving an R² of 0.75 and a MAPE of 24.46%. For data with missing feature fields, fuel consumption prediction features were further constructed based on operational duration and distance, and various machine learning algorithms were used to establish predictive models. Comparative results indicated that the Light Gradient Boosting Machine (LightGBM) model performed best (R²=0.77, MAPE=39.28%). Secondly, to establish localized emission models, PEMS tests were conducted on four typical China IV standard tractors with rated powers of 51.5, 58.8, 88.2, and 177 kW under varying engine loads. Based on the experimental data, machine learning algorithms (Ridge Regression, Random Forest, Decision Tree, SVM, and LightGBM) were compared to construct emission prediction models using multi-dimensional input features. The model evaluation results indicated that that the ResDNN model achieved the best performance for fuel consumption prediction with detailed features (R²=0.75, MAPE=24.46%), effectively resolving the gradient degradation problem in deep networks. For the simplified feature set, the LightGBM model performed optimally (R²=0.77, MAPE=39.28%). regarding emission prediction, the Support Vector Machine was identified as the optimal model for CO₂ emissions (R²=0.95, MAPE=5.98%), while LightGBM performed best for NO_x emissions (R²=0.86, MAPE=13.78%). Feature importance analysis revealed that instantaneous fuel consumption and engine speed were the most critical determinants for CO₂, whereas environmental parameters showed a higher contribution to NO_x prediction. Finally, applying these models to the trajectory data of 870,192 tractors equipped with Beidou terminals during the 2023 summer harvest (May 25 to June 25), a high-resolution national emission inventory was estimated. The results showed that the total fuel consumption was 47.01×10⁶ L, generating 128.21×10³t of CO₂ and 281.9t of NO_x. The average daily emissions per tractor were 147.33 kg for CO₂ and 0.32 kg for NO_x. Spatially, emissions were concentrated in major agricultural provinces such as Jiangsu, Henan, and Heilongjiang. Temporally, the emission peaks coincided perfectly with the busy farming rhythm. This research can provide methodological support and data foundation for managing carbon and pollutant emissions from large-scale agricultural machinery operations, promoting China's agricultural green transformation and the realization of the "Dual Carbon" goals.