Temporary smoothing is used in order to eliminate outliers in the time series.
smootH(x, na.rm = FALSE, method.interp = "na.interp")
| x | Numeric, matrix. |
|---|---|
| na.rm | If |
| method.interp | Four interpolation methods are presented, "na.interp", "na.approx" and "na.spline". By default is the method "na.interp". |
Vegetation indices or fractions of photosynthetic activity generally present noise in the time series that in some way or another hinder subsequent analyzes. This noise can be due to two reasons: i) no algorithm was used to mask the atmospheric noise before analyzing the time series and ii) negative outliers that were not detected by the masking algorithm. To minimize this problem, it is possible to eliminate negative outliers using the method proposed by Hamunyela et al. (2013).
This method is not able to remove consecutive outliers. The mathematical approach of this method of removing outliers implies the non-modification of the first and last values of the historical series, so that the near real-time detections of ecosystem disturbances will not be affected in any way.
Tarazona, Y., Mantas, V.M., Pereira, A.J.S.C. (2018). Improving tropical deforestation detection through using photosynthetic vegetation time series (PVts-\(\beta\)). Ecological Indicators, 94, 367 379.
Hamunyela, E., Verbesselt, J., Roerink, G., & Herold, M. (2013). Trends in spring phenology of western European deciduous forests. Remote Sensing,5(12), 6159-6179.
library(ForesToolboxRS) # Here a vector with an outlier x <- c(80, 78, 75, 76, 79, -100, 82, 76, 81, 77, 76) plot(x, type = "o", col = "red") # Apply a smoothing smth <- smootH(x) plot(x, type = "o", ylab = "Reflectance %", xlab = "Time") lines(smth, col = "blue", type = "o")