Abstract
Cloud base height (CBH) is an important parameter to describe cloud state and is highly related to the vertical motions in the atmosphere. CBH information is critical for both aviation safety and synoptic analysis. In this study, daytime CBH is estimated directly from Geostationary Operational Environmental Satellite-R Series (GOES-16) Advanced Baseline Imager (ABI) level 1b data and the European Centre for Medium-Range Weather Forecasts' (ECMWF) fifth generation reanalysis (ERA5) data using the Gradient Boosted Regression Trees (GBRT) machine learning technique. The CBH estimate algorithm, which is named as GETCBH, covers the same areal extent as the full disk of the ABI/GOES-16 and only for single-layer clouds. The 2-years of CBH measurements from the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) aboard Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) satellite is used as the label (which is the true value/class of the model output for regression/classification problem in machine learning terminology). A quality flag algorithm using another machine learning technique, the Gradient Boosted Decision Trees machine learning technique is developed to provide a confidence level for the CBH estimate. The evaluations show an overall root mean square error (RMSE) of 1.87 km and Pearson's correlation coefficient (Pearson's r) of 0.92 before any quality control. After excluding CBH estimates with low confidence (19.2% of all samples), the RMSE is reduced to 1.14 km, Pearson's r increases to 0.97, and 96% of the estimates are within 2 km of the CALIOP results. By analyzing model bias and feature importance, cloud phase information has the biggest impact on the CBH estimate, although all input features have positive impact on the estimate accuracy. Limited by the penetrability of CALIOP, GETCBH is valid for clouds with COD < 8.5. The CBH estimates have reduced accuracy (Pearson's r of 0.88) for optically thin clouds (clouds with cloud optical depth [COD] < 0.1) where little cloud information is contained in the ABI measurements, as well as for optically thick clouds (clouds with COD ≥ 3) where a larger proportion of opaque clouds is excluded. Furthermore, for the GBTCBH model using 9 months of CloudSat measurements as label, the CBH estimates are improved with an RMSE of 1.41 km and Pearson's r of 0.92. In a case study of Hurricane Dorian, CBHs for most of the single-layer clouds are successfully estimated with small errors and flagged with high confidence, for both high and low clouds. Deep convective clouds and multi-layer clouds, both of which are not included in the training, are reasonably flagged as low confidence with large CBH estimate errors. In this particular case, 65% of cloudy pixels have CBH estimate with high confidence in the scene. Daytime CBH with high spatial (2 km) and temporal (10 min) resolution can be derived from ABI measurements using this methodology.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have