Thursday, 31 August 2023
Boundary Waters (Hyatt Regency Minneapolis)
Handout (5.5 MB)
Quantitative precipitation estimation (QPE) is crucial for hydrological, climatological and meteorological studies. Weather radars provide remote observations of hydrometeors with a wide-range coverage. However, an accurate retrieval of precipitation intensity from weather radar observations remains a challenge.
An approach using a random forest algorithm to derive QPE from operational polarimetric radar observations was recently presented by Wolfensberger et al. (2021). Random forest regression is a machine learning approach based on an ensemble of decision trees. The model “Rainforest” (RF) is trained with a database containing six years (January 2016 to December 2021) of collocated observations from 288 rain gauges and polarimetric radar observations from five dual polarization Doppler C-band radars (Swiss weather radar network). This model was thoroughly evaluated by Wolfensberger et al. (2021).
In this study, we further develop RainForest by applying new methods to derive probabilistic estimates. A quantile random forest is used to derive various quantiles associated with the rain intensity estimate at each Cartesian grid cell and each time step. The resulting interquantile ranges are evaluated by calculating the percentage of observations that fall within these interquantile ranges. By means of two case studies, a stratiform and a convective event, we further investigate the accuracy of the quantile estimates.
Random forest is a machine learning algorithm that is based on an ensemble of decision trees. Hence, we propose to use the individual decision trees to create an ensemble of rain intensity fields correlated in space and time. By means of the continuous ranked probability score and a rank histogram, the spread of these members is evaluated. First results show a high potential for such an approach.
In summary, we present a novel model for probabilistic QPE at a 10 min temporal and a 1x1 km2 spatial resolution for Switzerland. Our approach shows great potential for meteorological and hydrological studies that require accurate and reliable estimates.
An approach using a random forest algorithm to derive QPE from operational polarimetric radar observations was recently presented by Wolfensberger et al. (2021). Random forest regression is a machine learning approach based on an ensemble of decision trees. The model “Rainforest” (RF) is trained with a database containing six years (January 2016 to December 2021) of collocated observations from 288 rain gauges and polarimetric radar observations from five dual polarization Doppler C-band radars (Swiss weather radar network). This model was thoroughly evaluated by Wolfensberger et al. (2021).
In this study, we further develop RainForest by applying new methods to derive probabilistic estimates. A quantile random forest is used to derive various quantiles associated with the rain intensity estimate at each Cartesian grid cell and each time step. The resulting interquantile ranges are evaluated by calculating the percentage of observations that fall within these interquantile ranges. By means of two case studies, a stratiform and a convective event, we further investigate the accuracy of the quantile estimates.
Random forest is a machine learning algorithm that is based on an ensemble of decision trees. Hence, we propose to use the individual decision trees to create an ensemble of rain intensity fields correlated in space and time. By means of the continuous ranked probability score and a rank histogram, the spread of these members is evaluated. First results show a high potential for such an approach.
In summary, we present a novel model for probabilistic QPE at a 10 min temporal and a 1x1 km2 spatial resolution for Switzerland. Our approach shows great potential for meteorological and hydrological studies that require accurate and reliable estimates.

