- "description": "# Whittaker\n\n## Description\n\nWhittaker represents a computationally efficient reconstruction method for smoothing and gap-filling of time series.\nThe primary function takes as input two vectors of the same length: the y time series data (e.g. NDVI) and the\ncorresponding temporal vector (date format) x, comprised between the start and end dates of a satellite image\ncollection. Missing or null values, as well as the cloud-masked values (i.e. NaN), are handled by introducing a\nvector of 0-1 weights w, with wi = 0 for missing observations and wi=1 otherwise. Following, the Whittaker smoother\nis applied to the time series profiles, computing therefore a daily smoothing interpolation.\n\nWhittaker's fast processing speed was assessed through an initial performance test by comparing different\ntime series fitting methods. The average runtime is 0.0107 seconds to process a single NDVI temporal profile.\n\nThe smoother performance can be adjusted by tuning the lambda parameter, which penalises the time series roughness:\nThe larger the lambda, the smoother the time series, but at the cost of the fit to the data getting worse. We found a lambda of\n10000 is adequate for obtaining more convenient results. A more detailed description of the algorithm can be\nfound in the original work of Eilers 2003.\n\n\n\n",
0 commit comments