Abstract
Constellations of CubeSats are emerging as a novel observational resource with the potential to overcome the spatiotemporal constraints of conventional single-sensor satellite missions. With a constellation of more than 170 active CubeSats, Planet has realized daily global imaging in the RGB and near-infrared (NIR) at similar to 3 m resolution. While superior in terms of spatiotemporal resolution, the radiometric quality is not equivalent to that of larger conventional satellites. Variations in orbital configuration and sensor-specific spectral response functions represent an additional limitation. Here, we exploit a Cubesat Enabled Spatio-Temporal Enhancement Method (CESTEM) to optimize the utility and quality of very high-resolution CubeSat imaging. CESTEM represents a multipurpose data-driven scheme for radiometric normalization, phenology reconstruction, and spatiotemporal enhancement of biophysical properties via synergistic use of CubeSat, Landsat 8, and MODIS observations. Phenological reconstruction, based on original CubeSat Normalized Difference Vegetation Index (NDVI) data derived from top of atmosphere or surface reflectances, is shown to be susceptible to large uncertainties. In comparison, a CESTEM-corrected NDVI time series is able to clearly resolve several consecutive multicut alfalfa growing seasons over a six-month period, in addition to providing precise timing of key phenological transitions. CESTEM adopts a random forest machine-learning approach for producing Landsat-consistent leaf area index (LAI) at the CubeSat scale with a relative mean absolute difference on the order of 4-6%. The CubeSat-based LAI estimates highlight the spatial resolution advantage and capability to provide temporally consistent and time-critical insights into within-field vegetation dynamics, the rate of vegetation green-up, and the timing of harvesting events that are otherwise missed by 8- to 16-day Landsat imagery.