Representativity of cloud‐profiling radar observations for data assimilation in numerical weather prediction

Abstract An algorithm is presented that aims to enhance the use of satellite‐based cloud‐profiling radar (CPR) data for the purpose of NWP data assimilation. It resembles the EarthCARE mission's scene construction algorithm: off‐nadir passive radiances get spectrally matched with nadir passive radiances, and the latter's collocated CPR profile gets replicated at the former's location. This process gets repeated until all passive pixels that cover an NWP domain D have a proxy column of CPR data. Only domain‐averaged profiles of CPR reflectivity ⟨ Z dB ⟩ and cloud fraction A c are sought after for NWP domains measuring, nominally, 25 × 25 km. If CPR measurements intersect D , then in addition to using just the intersecting values to represent ⟨ Z dB ⟩ and A c , the full array of proxy values get used. This is referred to as local estimation. If, however, CPR measurements do not intersect D , but are not too distant, this is referred to as non‐local estimation; estimates of ⟨ Z dB ⟩ and A c rest entirely on proxies. Current and planned satellite‐based CPRs have nadir‐pointing narrow fields‐of‐view, so it is difficult to see how to verify the algorithm with anything other than synthetic observations. Thus, it was assessed here with simulated cloudy atmospheres and 2D and 3D distributions of associated passive radiances and CPR reflectivities. In general terms, for local domains the algorithm performs slightly better than simply averaging intersecting CPR profiles. This at least demonstrates that the proxies are not detrimental. For non‐local domains, where there are no intersecting measurements to average, the algorithm appears to perform well enough to include two or three 25 × 25 km NWP domains either side of sequences of local domains.