The potential for mapping of turbidity in inland and coastal waters using imagery from the PlanetScope (PS) and RapidEye (RE) constellations is evaluated. With > 120 PS and 5 RE satellites in orbit these constellations are able to provide metre scale imagery on a daily basis and could significantly enhance high spatial resolution monitoring of turbidity worldwide. The Dark Spectrum Fitting (DSF) atmospheric correction is adapted to the PS and RE imaging systems to retrieve surface reflectances. Due to the large amount of imagery and the limited band sets on these sensors, automated pixel classification is required. This is here performed using a neural network approach, which is able to classify water pixels for clear to moderately turbid waters. Due to the limited band set and sensor performance, some issues remain with classifying extremely turbid waters and cloud shadows based on a spectral approach. Surface reflectance data compares well with in situ measurements from the AERONET-OC network. Turbidity is estimated from the Red, RedEdge (RE only) and NIR bands and is compared with measurements from autonomous stations in the San Francisco Bay area and the coastal waters around the United Kingdom. Good performance is found for Red band derived turbidity from PS data, while the NIR band performance is mediocre, likely due to calibration issues. For RE, all three turbidity products give reasonable results. A high revisit density allows for the mapping of temporal variability in water turbidity using these satellite constellations. Thanks to the RedEdge band on RE, chlorophyll a absorption can be avoided, and perhaps even estimated. |
