NAN values interpolation (or extrapolation)

[nmathewa]

Issue 1

• NAN values exist in the dataset (since sst is only over ocean)

Method 1

The missing values can be filled using linear extrapolation methods

def extrapolate_fields(dset):
    try :
        dset_n = dset.reindex(latitude=list(reversed(dset.latitude)
                                        )).interpolate_na(dim='latitude',
                                                          fill_value="extrapolate"
                                                          ).interpolate_na(
                                                              dim='longitude',
                                                              fill_value="extrapolate")

    except ValueError:
        dset_n = dset.interpolate_na(dim='latitude',fill_value="extrapolate"
                                                              ).interpolate_na(
                                                                  dim='longitude',
                                                                  fill_value="extrapolate")
    return dset_n

[nmathewa]

Issues (method1)

• The linear interpolation can cause outliers which will be learned by the model
• SST extrapolated over land does not make any sense physically but can be treated as a augmented method for this specific case (since the patterns are linear with the original data). [any research supports the air temperature?]

[nmathewa]

SST

1. Fill land values with land temperature
2. Remove the land near points

[nmathewa]

Curious landfalls

Q1. If land points are filled with land temperature the gradient will be high, but it may make the model learn the landfall conditions ? (There must be a tradeoff)

Q2. SST is not only the factor , humidity is also takes part here

[nmathewa]

Tests

• [ ] fill with surface temp