Qualification of a new short-term cloud forecasting method for the optimization of Earth observation satellite programs

Abstract

Cloud coverage is an important issue for Earth observation satellite programs as remote sensing images are useless if their cloud fraction is too high or if specific targets are not visible. These flawed acquisitions must be reintroduced in the waiting list, leading to significant delays. Agile satellites are able to choose between several targets according to various priority parameters, including cloud coverage forecasting. This paper assesses a short-term cloud forecasting method which uses time series acquired by a geosynchronous satellite and a cloud motion vector processing technique to predict the evolution of the cloud coverage up to six hours in the future. Using several SPOT/Pleiades acquisition dates from 2015 to 2017, and the corresponding Meteosat-8 datasets, cloud fraction forecasts have been produced for time horizons ranging from fifteen minutes to six hours, and compared to other forecast results obtained using two classical numerical weather prediction (NWP) models: GFS and ICON. Satellite-based forecast showed several advantages compared to these methods. First, the algorithm is much faster – a few minutes, whereas NWP models need several hours to build a forecast – allowing to get a result quickly from fresh dataset. Moreover, forecasts obtained with satellite images have a better time resolution – 15 minutes instead of one hour – and outperform GFS and ICON results in terms of prediction accuracy for horizons less than 3.5 hours.