Skip to content

Latest commit

 

History

History
77 lines (68 loc) · 5.33 KB

README.md

File metadata and controls

77 lines (68 loc) · 5.33 KB
Raw

Finnish Meteorological Institute Probabilistic Precipitation Nowcasting system (FMI-PPN)

FMI-PPN is a modular weather-radar-based nowcasting system built for research and operational usage.

Subsystems

Precipitation motion

Current implementation of FMI-PPN uses Lucas-Kanade optical flow method to estimate precipitation movement from radar measurements.

Nowcasting

Currently FMI-PPN uses pysteps to generate ensemble nowcasts.

Known issues and limitations

  • Parameter SEED must be None or an integer between 0 and 2**32 - 1. This is a limitation in numpy.random.
  • OpenMPI conflicts with dask when both are installed, leading to significant decrease FMI-PPN performance. Workaround is to set OpenMPI use only one thread. In Linux you can set environment variable OMP_NUM_THREADS=1.

Usage

Installation

  1. Install conda (Miniconda is recommended)
  2. Clone this repository
  3. Change directory to FMI-PPN folder
  4. Setup conda environment: $ conda env create -f environment.yml
  5. Modify pystepsrc file: Add your data source configuration under data_sources (see pysteps documentation for details)
  6. Replace the value for DOMAIN parameter under defaults with your data source configuration's name
  7. Activate conda environment: $ conda activate fmippn
  8. Run FMI-PPN with default settings: $ python run_ppn.py

Running FMI-PPN

Before running FMI-PPN, you should configure pysteps (via pystepsrc file) and PPN by adding your parametrisations to ppn_config.py.

How to run:

  1. Activate your conda environment
  2. Run FMI-PPN with your settings: $ python run_ppn.py -c your_config_parametrisation

Configuration

Adding new parametrisations

Create a json file with parameters you want to change from defaults and put it in config folder. The file's name (without file type extension) will be used to select the settings.

The ppn_config.py module has a utility function dump_defaults() for creating a configuration file based on default settings.

Parametrisations

Parameter Explanation Default value
DOMAIN Data source used from pystepsrc fmi
ENSEMBLE_SIZE Number of ensemble members 24
FFT_METHOD FFT method used in pysteps calculations pyfftw
FIELD_VALUES Select the units to store the nowcast (before scaling). Valid units are dbz for dBZ and rrate for mm/h. dbz
GENERATE_DETERMINISTIC Calculate extrapolation-only nowcast True
GENERATE_ENSEMBLE If True, then calculate ensemble members True
GENERATE_UNPERTURBED Calculate nowcast using pysteps, but without noise False
KMPERPIXEL Pixel size in kilometers 1.0
LOG_FOLDER Path where log files should be stored ../logs
LOG_LEVEL Logging level used by Python Logging Module 20
MAX_LEADTIME How long your nowcast will be (in minutes) 120
NORAIN_VALUE Value assigned to dry pixels during thresholding. Units depend on VALUE_DOMAIN parameter (dBZ or mm/h). Must be less than RAIN_THRESHOLD value! 1.5
NOWCAST_TIMESTEP Timestep between consecutive nowcast images 5
NUM_CASCADES How many cascade levels are used in cascade decomposition by pysteps 8
NUM_PREV_OBSERVATIONS Number of previous observations used in optical flow calculation 3
NUM_TIMESTEPS How many timesteps will be calculated in nowcasts (If this setting is None, this value is automatically calculated based on MAX_LEADTIME and NOWCAST_TIMESTEP parameters) None
NUM_WORKERS Number of worker threads used in parallel computing 6
OPTFLOW_METHOD Optical flow method (see pysteps docs) lucaskanade
OUTPUT_PATH If not None, then use this path to store output instead of setting in pysteprc None
OUTPUT_TIME_FORMAT Python datetime format for showing timestamps %Y-%m-%d %H:%M:%S
RAIN_THRESHOLD Thresholding value for rain. Pixels with values under this parameter are regarded as dry pixels. Units depend on VALUE_DOMAIN parameter (dBZ or mm/h). 6.5
REGENERATE_PERTURBED_MOTION Re-calculate motion for each ensemble member (requires SEED != None) False
SCALER Scaling coefficient for output 100
SCALE_ZERO Value for "0" after scaling the output. Setting to "auto" or None uses the minimum value found in data before scaling. auto
SEED Seed parameter for random number generation. Use None for unseeded nowcasts. None
STORE_DETERMINISTIC Write "deterministic" nowcast in output file True
STORE_ENSEMBLE Write all ensemble members in output file True
STORE_MOTION Write optical flow motion field in output file True
STORE_PERTURBED_MOTION Write optical flow motion for each ensemble member in output file True
STORE_UNPERTURBED Write "unperturbed" nowcast in output file True
VALUE_DOMAIN Choose if nowcasting is performed for data in dBZ or mm/h units. Valid parameters are dbz for dBZ and rrate for mm/h. dbz
VEL_PERT_KWARGS Parameters for velocity perturbation (see pysteps docs). Set this parameter to None to use default pysteps values. {'p_par': [2.20837526, 0.33887032, -2.48995355], 'p_perp': [2.21722634, 0.32359621, -2.57402761]}
VEL_PERT_METHOD Velocity perturbation method used in pysteps bps
WRITE_LOG If True, then generate a log file False
ZR_A Value for coefficient a in R(Z) relation 223.0
ZR_B Value for coefficient b in R(Z) relation 1.53