indices based on annual climatology of percentiles
This is an overall issue aiming at splitting the "issue of percentile based indices" into more tangible and concrete sub-issues.
-
Indices counting the number of occurrences (days) above/below a temporally and spatially variable threshold. I.e. for each point (gridcell) there is a threshold value for each day of the year (complication: leap days!). Example indices are `tn10p, tx10p, tg10p, tx90p, tg90p, txgt50p, txgt{PRC}p, tngt{PRC}p, tmgt{PRC}p, txlt{PRC}p, tnlt{PRC}p, tmlt{PRC}p. Cf. #196 (closed) (M 0.13) -
Same as for point 1, but the climatology of threshold values is based on days meeting a threshold condition. E.g. for r75p, r95p, r99p, r{PRC}p
only "rainy days" (e.g. days with precipitation of at least X mm day-1) are included in the percentile calculation. Cf. #197 (closed). -
Indices giving the total precipitation, in mm day-1, above a percentile threshold as a percentage of total wet-day precipitation, e.g. r95ptot, r99ptot, r{PRC}ptot
. -
Indices counting the total number of days belonging to temperature spells of length at least ND above/below a percentile-based threshold (as per details under 1). Example indices include wsdi, wsdi{ND}, csdi, csdi{ND}
. In these indices the percentile threshold is fixed at 90 for wsdi and 10 for csdi, but this should be possible to specify as a parameter. -
A few other indices involving two input variables, each one with its own threshold array. Examples are txDtnD, ...
, but let's leave them for now as they need further specification (as to whether it is all days or only the longest spell).
The threshold array is calculated as a daily climatology over a reference period, typically 30 years. For each day the percentile is calculated based all days in all years of the reference period and falling under a centred 5-day sliding window. This a bit involved as
- the window should extend into the previous/next year (including extending outside the reference period)
- handling of leap years
- use the right percentile method, see wiki page
- use bootstrapping as per Zhang et al, 2005
For thresholded data, e.g. indices based on "rainy days" include only RR >= 1 mm day-1 in percentile calculation.