Ensemble Calibration and Uncertainty Quantification of Precipitation
Forecasts for a Risk-based UAS Navigation
Mounir Chrit
Department of Atmospheric Sciences, University of North Dakota, Grand Forks, ND 4149, USA, Department of Atmospheric Sciences, University of North Dakota, Grand Forks, ND 4149, USA
Corresponding Author:[email protected]
Author ProfileAbstract
Uncertainty on precipitation forecasts results in major high
cancellation rate in Unmanned Aircraft Systems operations and reduces
the benefits of BVLOS operations in terms of riskbased contingency
planning. Hence, quantifying and reducing the uncertainty on
precipitation forecasts will reduce mission uncertainties, avoid
accidents and make the integration of UAS into the National Airspace
System more efficient and reliable. To achieve this goal, the
Member-ByMember post-processing technique is used to calibrate a
probabilistic forecast composed of 20 members of precipitation rate over
South Florida during summer period. The Continuous Ranked Probability
Score (CRPS) of the ensemble is minimized to achieve the optimal
regression between ensemble members without any assumption on the
forecasted parameter. The radar data from the Multi-Radar/Multi-Sensor
(MRMS) is used to correct ensemble spread every 10 min and reduce
forecasting uncertainty. A multi-physics ensemble was used to generate
high-resolution, convection resolving/allowing 48-hours forecasts. The
calibration was obtained over a learning process over the 48-h simulated
period over 3 years. The comparison between the raw and calibrated
ensemble from unseen data is presented in terms of bias correction and
ensemble reliability. The calibration was able to correct the bias found
in raw probabilistic forecasts relative to MRMS data. The comparison
with precipitation data from tipping buckets over four airports over
South Florida revealed that the calibrated ensemble tends to
overestimate the precipitation rates mainly because of the particles
evaporation that is taking place under radar beam.