Extended Variogram Mode¶

This notebook shows how to enable TerraFlow's extended kriging variogram search on a small synthetic station network. Extended mode evaluates the default spherical, exponential, and Gaussian candidates plus nested variogram candidates, then records candidate LOOCV scores in interpolation_cv.

In [1]:

import numpy as np
import pandas as pd

from terraflow.climate import ClimateInterpolator

climate_df = pd.DataFrame(
    {
        "lat": [39.97, 39.97, 39.97, 39.97, 40.00, 40.00, 40.00, 40.00],
        "lon": [-100.03, -100.01, -99.99, -99.97, -100.03, -100.01, -99.99, -99.97],
        "mean_temp": [17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0],
        "total_rain": [90.0, 100.0, 110.0, 120.0, 130.0, 140.0, 150.0, 160.0],
    }
)

interpolator = ClimateInterpolator(
    climate_df=climate_df,
    strategy="spatial",
    interpolation_method="kriging",
    variogram_mode="extended",
)

interpolator.cv_metrics["candidate_scores"]

import numpy as np import pandas as pd from terraflow.climate import ClimateInterpolator climate_df = pd.DataFrame( { "lat": [39.97, 39.97, 39.97, 39.97, 40.00, 40.00, 40.00, 40.00], "lon": [-100.03, -100.01, -99.99, -99.97, -100.03, -100.01, -99.99, -99.97], "mean_temp": [17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0], "total_rain": [90.0, 100.0, 110.0, 120.0, 130.0, 140.0, 150.0, 160.0], } ) interpolator = ClimateInterpolator( climate_df=climate_df, strategy="spatial", interpolation_method="kriging", variogram_mode="extended", ) interpolator.cv_metrics["candidate_scores"]

/Users/marupilla/workspace/projects/oss/AgroTerraFlow/terraflow/climate.py:347: OptimizeWarning: Covariance of the parameters could not be estimated
  params, _ = curve_fit(
INFO:terraflow.climate:Kriging variogram selected: 'nested_exponential_gaussian' (LOOCV RMSE=0.8174 for 'mean_temp')

INFO:terraflow.climate:ClimateInterpolator initialised: strategy='spatial', interpolation_method='kriging', variogram_mode='extended', records=8, variables=['mean_temp', 'total_rain']

Out[1]:

{'spherical': {'rmse': 2.618615, 'mae': 2.285714},
 'exponential': {'rmse': 1.774592, 'mae': 1.504604},
 'gaussian': {'rmse': 1.255237, 'mae': 0.805604},
 'nested_spherical_gaussian': {'rmse': 0.817436, 'mae': 0.591058},
 'nested_exponential_gaussian': {'rmse': 0.817434, 'mae': 0.591057}}

Use the default variogram_mode: standard for large station networks unless nested structures are specifically needed. Extended mode fits custom nested variograms before LOOCV, so it is intentionally more expensive than standard kriging.

In [2]:

cell_lats = np.array([39.985, 39.995])
cell_lons = np.array([-100.005, -99.985])

interpolator.interpolate(cell_lats, cell_lons)

cell_lats = np.array([39.985, 39.995]) cell_lons = np.array([-100.005, -99.985]) interpolator.interpolate(cell_lats, cell_lons)

Out[2]:

	mean_temp	mean_temp_krig_std	total_rain	total_rain_krig_std
0	20.223769	0.947425	122.237685	0.947425
1	22.839164	0.535845	148.391636	0.535845