Net ecosystem exchange (NEE) and evapotranspiration (ET) are important indicators for characterizing carbon and water cycling capacity in semi-arid areas ecosystems. This study presents an accurate modeling of the dynamics of carbon and water fluxes and an in-depth analysis of their driving mechanisms. It helps to clarify the functions of grassland ecosystems and their responses to climate change in the semi-arid regions of the Loess Plateau. Based on daily-scale flux observations of Artemisia sacrorum grassland ecosystem in the Loess Plateau from 2018 to 2022, we used a multiple linear regression model, machine learning models (Random Forest, Support Vector Machine and Artificial Neural Network model) and ecological knowledge-machine learning (EML) models to fit NEE and ET, respectively. Among EML models, six models based on different ecological assumptions were used to fit NEE, and seven models based on different ecological assumptions were used to fit ET. We then constructed the best-fitting and best-explained EML model and investigated the effects of environmental and vegetation factors on NEE and ET. The results showed that: (1) The EML model incorporating meteorological, soil moisture and vegetation factors had the best fit to NEE and ET. The R2 and RMSE of the EML model were 0.81 and 0.70 g C m-2 d-1, 0.83 and 0.48 mm/d, and the MRE and MAE of the EML model were 1.72 and 0.48 g C m-2 d-1, 0.29 and 0.30 mm/d, respectively. The fitting effect of this model on NEE and ET increased by 24.62% and 12.16% compared with the multiple linear regression model, and increased by 13.02% and 6.87% on average compared with machine learning models. (2) Air temperature was the primary influencing factor for NEE and ET, with importance values proportions being 63.12% and 60.38% respectively. 6℃ and 22℃ were the thresholds of the daily average air temperature of grassland NEE. NEE was in a downward trend between 6 and 22℃, and became a stable trend after 22℃. 0℃ and 22℃ were the thresholds of daily average air temperature of grassland ET. When the air temperature was greater than 22℃, ET transitioned from an upward trend to a stable trend. (3) Soil moisture factors accounted for 17.13% and 5.66% of the importance values on NEE and ET respectively. NEE was more sensitive to soil moisture than ET. The results contribute to improving the simulation method of carbon and water fluxes, and clarifying their responses to environmental and vegetation factors in grassland ecosystems in semi-arid areas.