Abstract: 【Objective】Forest biomass is closely related to forest production, and it is also the basic data reflecting forest ecological system function. 【Method】In this study, Xixia County in Henan Province, China, was selected as the study area, while Landsat TM/ETM+/OLI imagery in 1993, 1998, 2003, 2008, 2013 and fixed plot data from continuous forest inventory in the same five years were collected as the main information to build four remote sensing based models to predict the spatio-temporal dynamics of forest aboveground biomass during 1993-2013. 【Result】① The prediction accuracy of random forest is the highest, k-NN and bagging remain medium, while the accuracy of MLR is the lowest; ② Elevation, slope, brightness, wetness, vertical vegetation index and effective leaf area index are the key factors affecting forest aboveground biomass in the study area; ③ During 1993-2013, the unit forest biomass in the study area first decreased from 34.68 Mg/hm² in 1993 to 32.59 Mg/hm² in 2003, then increased to 44.65 Mg/hm² in 2013; ④ The global Moran' I index, which characterizes the degree of spatial autocorrelation, decreased continuously from 1993 to 2013, indicating that the spatial distribution pattern of forest aboveground biomass had changed from high aggregation to fragmentation. 【Conclusion】 Forest clearance for farmland, deforestation, returning farmland to forest, natural forest protection policy, as well as construction of ecological corridor and village greening project, are the main driving factors of spatio-temporal dynamics of forest aboveground biomass in the study area.