Curve fitting#

class eocrops.tasks.curve_fitting.CurveFitting(range_doy=(1, 365), q=0.5)[source]#

Bases: object

Methods

fit_whitakker(ts_mean, degree_smoothing[, ...])

Fit whitakker smoothing given a ndarray (1,T,D)

get_doy_period(eopatch)

Get day of the year acquisition dates from eopatch.timestamp

get_time_series_profile(eopatch, feature[, ...])

Get aggregated time series at the object level, i.e. aggregate all pixels not masked.

replace_values(eopatch, ts_mean, doy_o, ...)

Given an interpolated time series using a curve fitting method, replace interpolated values with the nearest observed one iif the image had less than a given cloud coverage

resample_ts(doy, ts_mean_[, resampling, ...])

Apply resampling over fixed period before applying whitakker smoothing

whittaker_zonal(eopatch, feature[, ...])

Apply whitakker smoothing over time series
get_time_series_profile(eopatch, feature, feature_mask='MASK', threshold=None, resampling=None)[source]#

Get aggregated time series at the object level, i.e. aggregate all pixels not masked

Parameters:
eopatchEOPatch

EOPatch saved

featurestr

Name of the feature

feature_maskstr

Name of the boolean mask for pixels outside field boundaries

Returns:
Aggregated 1-d time series (np.array)
get_doy_period(eopatch)[source]#

Get day of the year acquisition dates from eopatch.timestamp

Parameters:
eopatchEOPatch

eopatch to read

Returns:
np.arrayday of the years, np.arrayindex of image within the season (range_doy)
replace_values(eopatch, ts_mean, doy_o, ts_fitted, doy_r, resampling, cloud_coverage=0.05)[source]#

Given an interpolated time series using a curve fitting method, replace interpolated values with the nearest observed one iif the image had less than a given cloud coverage

Parameters:
eopatchEOPatch
ts_mean :np.array)the original observed time series at the field-level
doy_onp.array)

days of the year from the observed scene

ts_fittednp.array

fitted time series using a cure fitting method

doy_rnp.array

resampled day of the year

resamplingint

period length to resample time series (e.g. 8 for 8-day period starting on the 1st January)

cloud_coveragefloat

maximum percentage of cloud given an image (e.g. we keep only values if <5% cloudy)

Returns:
Fitted time series mixing interpolated and observed values
resample_ts(doy, ts_mean_, resampling=8, min_threshold=None, max_threshold=None)[source]#

Apply resampling over fixed period before applying whitakker smoothing

fit_whitakker(ts_mean, degree_smoothing, weighted=False, min_threshold=0)[source]#

Fit whitakker smoothing given a ndarray (1,T,D)

whittaker_zonal(eopatch, feature, feature_mask='MASK', degree_smoothing=1, min_threshold=0, max_threshold=None, weighted=False, resampling=8)[source]#

Apply whitakker smoothing over time series

Parameters:
eopatchEOPatch

eopatch to apply the smoothing method

featurestr

Feature to apply the smoothing (e.g. NDVI)

feature_maskstr

Feature that refers to the pixel masking outside the field polygon

degree_smoothingfloat

Degree of smoothing (0.5 works great)

min_thresholdfloat

Minimum threshold of the time series. If it is below, we mask as np.nan to avoid outlier values

max_thresholdfloat

Minimum threshold of the time sereis. If it is above, we mask as np.nan to avoid outlier values

weightedbool

Weight the smoother w.r.t. the maximimum value to better fit the enveloppe over high value data

resamplingint

Resample over fixed periods from the 1st of January

Returns:
np.arrayday of the year, np.arrayaggregated time series
class eocrops.tasks.curve_fitting.AsymmetricGaussian(**kwargs)[source]#

Bases: CurveFitting

Methods

execute(eopatch, feature[, feature_mask, ...])

Execute A-G curve fitting function

fit_whitakker(ts_mean, degree_smoothing[, ...])

Fit whitakker smoothing given a ndarray (1,T,D)

get_doy_period(eopatch)

Get day of the year acquisition dates from eopatch.timestamp

get_fitted_values(eopatch, feature[, ...])

Apply Asymmetric function to do curve fitting from aggregated time series.

get_time_series_profile(eopatch, feature[, ...])

Get aggregated time series at the object level, i.e. aggregate all pixels not masked.

replace_values(eopatch, ts_mean, doy_o, ...)

Given an interpolated time series using a curve fitting method, replace interpolated values with the nearest observed one iif the image had less than a given cloud coverage

resample_ts(doy, ts_mean_[, resampling, ...])

Apply resampling over fixed period before applying whitakker smoothing

whittaker_zonal(eopatch, feature[, ...])

Apply whitakker smoothing over time series
get_fitted_values(eopatch, feature, feature_mask='MASK', resampling=0, threshold=0)[source]#

Apply Asymmetric function to do curve fitting from aggregated time series. It aims to reconstruct, smooth and extract phenological parameters from time series

Parameters:
eopatchEOPatch
featurestr

name of the feature to process

feature_maskstr

name of the polygon mask

functionnp.function

function to aggregate pixels at object level

resamplingnp.array

dates from reconstructed time series, np.array : aggregated time series)

Returns:
np.arraydoy, np.arrayfitted values
execute(eopatch, feature, feature_mask='MASK', resampling=0)[source]#

Execute A-G curve fitting function

Parameters:
eopatchEOPatch
featurestr
feature_maskstr
resamplingint
Returns:
np.arraydoy, np.arrayfitted values
class eocrops.tasks.curve_fitting.DoublyLogistic(**kwargs)[source]#

Bases: CurveFitting

Methods

execute(eopatch, feature[, feature_mask, ...])

Execute the smoothing at the field level

fit_whitakker(ts_mean, degree_smoothing[, ...])

Fit whitakker smoothing given a ndarray (1,T,D)

get_doy_period(eopatch)

Get day of the year acquisition dates from eopatch.timestamp

get_fitted_values(eopatch, feature[, ...])

Apply Doubly Logistic function to do curve fitting from aggregated time series.

get_time_series_profile(eopatch, feature[, ...])

Get aggregated time series at the object level, i.e. aggregate all pixels not masked.

replace_values(eopatch, ts_mean, doy_o, ...)

Given an interpolated time series using a curve fitting method, replace interpolated values with the nearest observed one iif the image had less than a given cloud coverage

resample_ts(doy, ts_mean_[, resampling, ...])

Apply resampling over fixed period before applying whitakker smoothing

whittaker_zonal(eopatch, feature[, ...])

Apply whitakker smoothing over time series
get_fitted_values(eopatch, feature, feature_mask='MASK', resampling=0, threshold=0.2)[source]#

Apply Doubly Logistic function to do curve fitting from aggregated time series. It aims to reconstruct, smooth and extract phenological parameters from time series

Parameters:
eopatchEOPatch
featurestr

name of the feature to process

feature_maskstr

name of the polygon mask

functionnp.function

function to aggregate pixels at object level

resamplingnp.array

dates from reconstructed time series, np.array : aggregated time series)

Returns:
np.arraydoy, np.arrayfitted values
execute(eopatch, feature, feature_mask='MASK', resampling=0, threshold=0.2)[source]#

Execute the smoothing at the field level

Parameters:
eopatchEOPatch
featurestr
feature_maskstr
resamplingint
thresholdfloat
Returns:
np.arraydoy, np.arrayfitted values
class eocrops.tasks.curve_fitting.FourierDiscrete(omega=1.5, **kwargs)[source]#

Bases: CurveFitting

Methods

execute(eopatch, feature[, feature_mask, ...])

Execute FFT

fit_whitakker(ts_mean, degree_smoothing[, ...])

Fit whitakker smoothing given a ndarray (1,T,D)

get_doy_period(eopatch)

Get day of the year acquisition dates from eopatch.timestamp

get_fitted_values(eopatch, feature[, ...])

Apply Fourier function to do curve fitting from aggregated time series.

get_time_series_profile(eopatch, feature[, ...])

Get aggregated time series at the object level, i.e. aggregate all pixels not masked.

replace_values(eopatch, ts_mean, doy_o, ...)

Given an interpolated time series using a curve fitting method, replace interpolated values with the nearest observed one iif the image had less than a given cloud coverage

resample_ts(doy, ts_mean_[, resampling, ...])

Apply resampling over fixed period before applying whitakker smoothing

whittaker_zonal(eopatch, feature[, ...])

Apply whitakker smoothing over time series
get_fitted_values(eopatch, feature, feature_mask='MASK', resampling=0, fft=True, threshold=0.2)[source]#

Apply Fourier function to do curve fitting from aggregated time series. It aims to reconstruct, smooth and extract phenological parameters from time series

Parameters:
eopatchEOPatch
featurestr

name of the feature to process

feature_maskstr

name of the polygon mask

functionnp.function

function to aggregate pixels at object level

resamplingnp.array

dates from reconstructed time series, np.array : aggregated time series)

Returns:
np.arraydoy, np.arrayfitted values
execute(eopatch, feature, feature_mask='MASK', resampling=0, fft=True, threshold=0)[source]#

Execute FFT

Parameters:
eopatchEOPatch
featurestr
feature_maskstr
resamplingint
fftbool
thresholdfloat
Returns:
np.arraydoy, np.arrayfitted values