Several efforts have been made to model forest structure and estimate global aboveground biomass (AGB) magnitude and distribution. However, AGB models still show large uncertainties, especially over tropical mountains that are difficult to survey and understudied. Although forest disturbance and land use greatly influence forest structure and AGB accumulation, they are rarely considered in forest structure and AGB models. Including the effect of forest disturbance and land use could improve model performance. In this study, we used Landsat time series analysis and forest inventory data to assess and model the effect of small‐scale disturbance and land use on tropical montane forest structure and AGB. We explored two approaches: abrupt forest change detection, using the algorithm Breaks For Additive Seasonal Trend (BFAST), and gradual vegetation change by calculating the variation of remote sensing vegetation indices over time. We found that tropical montane landscapes are very dynamic and heterogenous. Out of the 284 forest plots analyzed, 27% had abrupt disturbances at least once between 1993 and 2014, and some of them showed abrupt disturbances several times. Forest plots with abrupt disturbances exhibit statistically significant lower basal area, tree height, and AGB than plots without abrupt disturbances. Using linear mixed‐effects models, we found that the number of breakpoints, annual Normalized Difference Vegetation Index SD, and Normalized Difference Water Index minimum values over time were the most informative variables for predicting forest structure, particularly basal area (adjusted R2 = 0.61). These variables are easy to calculate and could add significant power to forest structure and AGB models. In conclusion, incorporating small‐scale disturbance to model forest structure and AGB at regional scales is feasible through the identification of recent abrupt and gradual forest change using remote sensing time series. Including recent forest disturbance variables as predictors in forest structure models can greatly improve biomass predictability and mapping in dynamic landscapes.