Temporal descriptive metrics as new features
Descriptive statistics are quantitative parameters which provide simple summaries about the time series or parts of the time series. The following metrics could be implemented:
- mean: for identifying the central value of a time series;
- standard deviation: to quantify the amount of variation within the pixel values of a time series profile;
- minimum: to identify extrema in the time series;
- maximum: to identify extrema in the time series;
- minimum-maximum range: to identify the interval in which the pixel values of a time series are located;
- sum: to quantify the overall throughput of the time series in a given time length;
- median: to identify the “middle” value which separates the higher half of the pixel values in the time series from the lower one;
- 10 th percentile: to identify the pixel value below which 10% of the pixel values of the time series can be found; and
- 90 th percentile: identify the pixel value below which 90% of the pixel values of the time series can be found
- Sums over partial periods
This could be implemented as a new application or as an option for iota2FeatureExtraction.
For many of these features, the value but also the date of the value (ex. max ndvi and its date) are needed. Other features can also be computed:
- number of valid dates
- the length of temporal gaps (the longest, its length and the starting date)
- slope information (may need spline gapfilling instead of linear)
Edited by Jordi Inglada