In order to realize the dynamic monitoring of farmland ecosystem carbon flux, a method for estimating farmland ecosystem carbon flux based on Landsat series multi-source remote sensing data was proposed. Three experimental fields of agricultural Research and Development Center of University of Nebraska, northeastern United States were selected as the study area, and the corresponding flux site data published by AmeriFlux was used for subsequent modeling analysis. Based on the comprehensive analysis of climate variables, soil properties and plant traits, remote sensing factors closely related to carbon flux of farmland ecosystem were selected, and a full remote sensing factor data set covering key links of farmland ecological process was constructed. Then, the farmland carbon flux regression prediction model based on random forest was constructed. Compared with the ridge regression model and the lasso model, the model was more effective in estimating farmland ecosystem carbon flux, with a coefficient of determination of 0.94 and a root mean square error of 4.281g/（m2·d）. According to the importance analysis of factors based on random forest model, the contributions of DVI, NDWI, MSAVI, NRI and NDVI to carbon flux estimation were 35.6%, 25.8%, 12.2%, 7.8% and 5.2%, respectively. On the basis of above research, through the farmland ecosystem carbon balance space-time evolution characteristics analysis, the farmland carbon sink capacity was the strongest in 2013 when the crop growth was in the period of July and August in Nebraska, at the beginning of the planting soybeans and corn were rendered weak carbon source, and the carbon source ability was stronger for corn, in growth peak of corn and soybeans were in carbon sink, and the carbon sequestration ability was stronger for corn. The research result can provide theoretical support for accurately estimating the carbon budget of farmland ecosystems and guiding agricultural production.