Intro

When breaking down Bfast outputs by season some strange breaks were noticed. An important number of breaks had very wide confidence intervals (CIs) associated with the time of break. Further, some positive breaks (i.e. abrupt increase in greenness) occurred in the winter which is quite strange.

Below I explore further these breaks.

Breaks with very wide CIs

Positive breaks

Positive breaks with CI ranges bigger than 60 months are shown below.

Negative Breaks

Negative breaks with CI ranges bigger than 80 months are show below.

Breaks with very narrow CIs

Positive breaks

The time step in the EVI time series is 16-days. How many negative breaks have CIs narrower than three data points (i.e. 48 days or 1.5 months)?

## [1] "N breaks:45"

## [[1]]
## NULL

Negative breaks

The time step in the EVI time series is 16-days. How many negative breaks have CIs narrower than three data points (i.e. 48 days or 1.5 months)?

## [1] "N breaks:116"

## [[1]]
## NULL

Positive breaks in Winter

There are more than 1000 positive breaks that occurred in the winter. But how reliable are these given the presence of wide CI intervals? To explore this issue I focused on relatively big positive breaks with narrow CI as these are likely to be more reliable.

Winter breaks with relatively narrow CIs (< 3 months) have relative small magnitudes compared to breaks with similar CIs in other seasons.

##   winter_break      mean count
## 1           no 1527.8461   109
## 2          yes  536.7728     6
  • It seems that the current Bfast model set-up does not allow for a robust seasonal analysis unless we focus on breaks with narrow CIs (< 3 months).

The relationship between break magnitude and CIs

What is the relationship between break magnitude and range of CIs (time of break)?