For e.g. ocean temperature interpolation, split reconstruction into wet and dry points. Interpolate with creepfill.

[JanStreffing]

@fernandadialzira

[JanStreffing]

So far, I have not found a python lib with creepfill as it is available in YAC: https://dkrz-sw.gitlab-pages.dkrz.de/yac/d4/d7b/interp_method_creep.html

I think we might be able to use the YAC python interface to call the routine of yac:

import numpy as np
import yac

# Create a structured grid (regular grid) with sample data
lon_values = np.linspace(-180, 180, 360)  # Example longitude values
lat_values = np.linspace(-90, 90, 180)    # Example latitude values
structured_data = np.random.rand(len(lat_values), len(lon_values))  # Example data

# Define unstructured mesh points (where you want to interpolate)
unstructured_lon = [10.0, 20.0, 30.0]  # Example longitude values
unstructured_lat = [50.0, 60.0, 70.0]  # Example latitude values

# Create a YAC grid object
yac_grid = yac.Grid(structured_data, lon_values, lat_values)

# Interpolate using the "creepfill" method
interpolated_values = yac.interp_method_creep(yac_grid, unstructured_lon, unstructured_lat)

# Print the interpolated values
for lon, lat, value in zip(unstructured_lon, unstructured_lat, interpolated_values):
    print(f"Interpolated value at ({lon}, {lat}): {value:.4f}")

However installation is tricky. One would have to basically install the c libarary yac as normal with --enable-python-bindings configure flag. That might be a bit much for a small improvement. We shall try with NN first.