Cloud detection and removal is a well-known critical issue for land remote sensing in the optical domain. Despite the vast literature on the subject, uncertainties still remain on the used cloud detection approaches that can have significant impact for a wide range of applications.
In the case of Proba-V cloud detection is particularly challenging, considering the limited number of spectral bands (Blue, Red, NIR and SWIR) and the lack of TIR channels or dedicated cirrus band (as the 1.38 micron band). The legacy approach for Proba-V cloud detection was inherited from SPOT-VGT [RD-1] and based on static thresholds on the Blue and SWIR channels. The main drawback of such threshold method is that its performances are strongly dependent on the amount of contrast in radiometry between the clouds and the underlying surface.
Recent improvements were proposed for both SPOT-VGT [RD-2] and Proba-V [RD-3], using climatology of reflectances to define a dynamic threshold on the Blue band, which depends on the surface cover and on vegetation conditions. Additional improvements could be obtained using statistically-sound approaches, such as Nerual Network, allowing to optimally use all the information from the four Proba-V channels.
In order to inter-compare the performances of different cloud detection algorithms for Proba-V, the European Space Agency (ESA) and the Belgian Science Policy Office (BELSPO) decided to organize a dedicated Round Robin exercise.
This Proba-V Cloud Detection Round Robin (PV-CDRR) will be open to any interested algorithm's provider. The main goal is to learn advantages and drawbacks of the different approaches for various cloud and surface conditions. As an outcome of the study recommendations will be provided to ESA on the potential new cloud detection algorithm for the operational processing chain.
[RD-1] Lisens, G., P. Kempeneers, F. Fierens, and J. Van Rensbergen. Development of Cloud, Snow, and Shadow Masking Algorithms for VEGETATION Imagery. Proceedings of Geoscience and Remote Sensing Symposium, IGARSS 2000, Honolulu, HI 2: 834–836.
[RD-2] Wolters, E., Swinnen, E., I. Benhadj, Dierckx, W., PROBA-V cloud detection evaluation and proposed modification, QWG Technical Note, 17/7/2015
[RD-3] Hagolle, O., et al. Quality assessment and improvement of temporally composited products of remotely sensed imagery by combination of VEGETATION 1 and 2 images. Remote Sensing of Environment 94.2 (2005): 172-186.
Round Robin Registration
The benefits in participating in the Proba-V Cloud Detection Round Robin include:
Note that a dedicated fixed price reward is allocated to each participant, allowing to cover the effort spent for algorithm development, testing and running. Access to the reward is granted to each participant upon contribution to the Round Robin with an original algorithm and provision of the required deliverables.
As a participant in the PV-CDRR you are committing to the following:
Note that the working environment will be provided by ESA, through the RSS Cloud Toolbox service. Thanks to this tool, the participant will have access to a remote virtual machine with all needed input data, some relevant tools and the required computing resources, the virtual machine can be tailored to the user needs in terms of available SW, as well as CPU and RAM.
Algorithm Providers that want to participate to the Round Robin can register here.
Please provide in the Registration Form your users name, e-mail and affiliation as well as information on the type of algorithm you are going to use for cloud detection, e.g., Neural Network, Bayesian.