event.csv specify the precipitation events used in the training data.
| Column Name | Description |
|---|---|
| station | the 3-digit station code for each METAR station |
| start_time | the start time of the precipitation episode |
| end_time | the end time of the precipitation episode |
| lon | station longitude |
| lat | station latitude |
| median_time | the median time of the precipitation episode |
| year | the year in which the precipitation episode occurs |
| month | the month in which the precipitation episode occurs |
| label | '0' - rain, '1' - snow, '2' - freezing rain, '3' - ice pellets |
US_X_30_30_2.5_1.3_0.25_constrained.npy and US_y_30_30_2.5_1.3_0.25_constrained.npy can be used to train the LightGBMClassifier. In particular,
| Column Number | Description |
|---|---|
| 0 | Surface Temperature (K) |
| 1 - 16 | 1000 - 500 hPa Temperature (K) |
| 17 | Surface Relative Humidity (%) |
| 18 - 33 | 1000 - 500 hPa Relative Humidity (%) |
We note that 30_30_2.5_1.3 means that 30000 rain events, 30000 snow events, 2500 freezing rain events, and 1300 ice pellets events are used to train the algorithm. 0.25 denotes a horizontal resolution of 0.25 degree. constrained means that relative humidity is constrained to between 0 and 100%.
US_X_30_30_2.5_1.3_1deg_constrained.npy and US_y_30_30_2.5_1.3_1deg_constrained.npy are analogous to their 0.25 deg counterparts but should be used on model data with a horizontal resolution of about 1 degree.
import numpy as np
import lightgbm as lgb
X = np.load('US_X_30_30_2.5_1.3_0.25_constrained.npy')
y = np.load('US_y_30_30_2.5_1.3_0.25_constrained.npy')
clf = lgb.LGBMClassifier(num_leaves=27,
learning_rate=0.05,
max_depth=6,
min_child_samples=18,
max_bins=300,
verbosity=-1,
data_sample_strategy='goss',
importance_type='gain')
clf.predict(X_test) # X_test should have a shape of (N, 34)