While remote sensing is a logical means to monitor the world remaining primary rainforests, the cloudiness of those areas is a huge obstacle to timely detection of deforestation events. Working in northeastern Madagascar with the Ministry of Environment, Ecology and Forests we observed 75% of all Landsat and Sentinel-2 pixels over a three-year period to be contaminated by clouds, leading to large latency in potential deforestation alerts. Madagascar’s forests harbor stunning levels of endemism created by millions of years of isolation. However, these forests are also a hotspot for deforestation, which affects over 100,000 hectares per year. Early detection of illegal cutting is needed for enforcement of forest protection laws.Cloud-penetrating radar is an option for augmenting Landsat and Sentinel-2, but radar signal is highly sensitive to topography, and our experience in the mountainous forests of Madagascar suggests that several backscatter images may be needed to accurately identify deforestation. Daily high-resolution Planet imagery offers a unique opportunity to supplement the data record if radiometry can be sufficiently stabilized across Planet’s many sensors to allow automated change detection. The proposed research investigates two strategies for assimilating Planet, Landsat, and Sentinel-2 into time series analysis designed to quickly detect forest loss. First, we investigate a previously published dynamic calibration approach called CESTEM, in which machine learning matches Planet acquisitions to radiometry from Landsat and MODIS.CESTEM processing has never been applied in a deforestation detection context. Second, we will explore a new approach based upon sensor fusion using the Kalman filter (KF). The KF approach minimizes the need for pre-processing and is designed to quickly fuse information from different sensors to support real-time decisions. We identify several research questions that need to be answered to maximize the value of the Planet record in monitoring deforestation hotspots. We will use what we learn to build, in conjunction with the Ministry, a workflow that flags potential areas of deforestation across Madagascar’s primary forest . In addition, we will develop a daily Landsat-like product called Landsat-K (for “Kalman”) that represents the posterior estimate for each Landsat band after assimilation of new acquisitions from the three sensors (with Planet only contributing to visible and near-infrared bands). This daily product may in the future be useful in agricultural, disaster, or other land cover-related applications that require high-frequency change analysis. In this way, we hope to extend what we learn about data assimilation and deforestation hotspot mapping to a broader monitoring setting.