Supercooled liquid water cloud classification using lidar backscatter peak properties

Abstract

The use of depolarization lidar to measure atmospheric volume depolarization ratio (VDR) is a common technique to classify cloud phase (liquid or ice). Previous work using a machine learning framework, applied to peak properties derived from co-polarized attenuated backscatter data, has been demonstrated to effectively detect supercooled-liquid-water-containing clouds (SLCCs). However, the training data from Davis Station, Antarctica, include no warm liquid water clouds (WLWCs), potentially limiting the model's accuracy in regions where WLWCs are present. In this work, we apply the same framework used on the Davis data to a 9-month micro-pulse lidar dataset collected in Ōtautahi / Christchurch, Aotearoa / New Zealand, a location which includes WLWC. We then evaluate the results relative to a reference VDR cloud-phase mask. We found that the Davis model performed relatively poorly at detecting SLCC with a recall score of 0.18, often misclassifying WLWC as SLCC. The performance of our new model, trained using data from Ōtautahi / Christchurch, displays recall scores as high as 0.88 for identification of SLCC, although it generally underestimates SLCC occurrence. The overall performance of the new model highlights the effectiveness of the machine learning technique when appropriate training data relevant to the location are used.