Random forest retrievals of DSD properties for NPOL radar.
Added to GitHub for reproduction of results of Conrick et al. (2019).
If you use these objects, please cite:
Conrick, R., J.P. Zagrodnik, and C.F. Mass, 2020: Dual-polarization radar retrievals of coastal Pacific Northwest rain drop size distribution parameters using random forest regression. J. Atmos. Oceanic Technol., 37, 229–242.
Update October 15, 2024 If the pickle files are needed, please request them by emailing me.
Update January 26, 2022 Please ensure scikit-learn version 0.21.3 is installed prior to use. Newer versions of scikit-learn have changed the way that random forests are called, so loading the pickle files will throw an error.
Update: August 29, 2019 GitHub does not allow large files, so the TAR.GZ file is now hosted on UW Atmospheric Sciences Servers: https://www.atmos.washington.edu/~rconrick/RF/ Download it there and follow the instructions below.
File Format The tar.gz file contains three Python pickle files (RF_D0.pickle, RF_RR.pickle, RF_LWC.pickle) each for retrieving D0, Rain rate (RR), and LWC, respectively, from NPOL observations of Z,ZDR.
Use
- Unpack into a working directory.
- Load *.pickle files using Python.
- Each pickle file contains an instance of Scikit-learn's Random Forest Regressor object.
- Consult the Scikit-learn site for appropriate methods to use: https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.RandomForestRegressor.html To produce a retrieval from radar data, use: RF.predict(ZDR,Z) , where RF is the previously loaded instance for D0 or LWC. where ZDR is the NPOL Differential reflectivity. Z is the NPOL reflectivity.
For Support or Questions, contact Robert Conrick ([email protected])