Automatically detecting forest disturbances as they occur can be extremely challenging for certain types of environments, particularly those presenting strong natural variations. Here, we use a generic structural break detection framework (BFAST) to improve the monitoring of forest cover loss by combining multiple data streams. Forest change monitoring is performed using Landsat data in combination with MODIS or rainfall data to further improve the modelling and monitoring. We tested the use of the Normalized Difference Vegetation Index (NDVI) from the Moderate Resolution Imaging Spectroradiometer (MODIS) with varying spatial aggregation window sizes as well as a rainfall derived index as external regressors. The method was evaluated on a dry tropical forest area in lowland Bolivia where forest cover loss is known to occur, and we validated the results against a set of ground truth samples manually interpreted using the TimeSync environment. We found that the addition of an external regressor allows to take advantage of the difference in spatial extent between human induced and naturally induced variations and only detect the processes of interest. Of all configurations, we found the 13 by 13 km MODIS NDVI window to be the most successful, with an overall accuracy of 87%. Compared with a single pixel approach, the proposed method produced better time-series model fits resulting in increases of overall accuracy (from 82% to 87%), and decrease in omission and commission errors (from 33% to 24% and from 3% to 0% respectively). The presented approach seems particularly relevant for areas with high inter-annual natural variability, such as forests regularly experiencing exceptional drought events.