Use of SeriesAggreg when TimeLag >= 3600

The SeriesAggreg() function does not work when TimeLag >= 3600.

Error in `*tmp*`[1:(length(Ind2) - 1)] : 
   only 0's may be mixed with negative subscripts

added BUG R labels

This is a very bad function. I think it might be a good idea to rewrite it entirely...

created branch SeriesAggreg to address this issue

mentioned in commit 7a804e87

mentioned in commit 81fccf1d

mentioned in commit 50d0ad9f

mentioned in commit 25d44cab

First tests:

library(airGR)

## loading catchment data
data(L0123002)

## preparation of the initial time series data frame at the daily time step
TabSeries <- BasinObs[, c("DatesR", "P", "E", "T", "Qmm")]

## yearly time series
NewTabSeries <- SeriesAggreg2(TabSeries = TabSeries,
                              Format = "%Y",
                              ConvertFun = c("sum", "sum", "mean", "sum"))
plot(Qmm ~ DatesR, data = NewTabSeries, type = "l")

## monthly time series
NewTabSeries <- SeriesAggreg2(TabSeries = TabSeries,
                              Format = "%Y%m",
                              ConvertFun = c("sum", "sum", "mean", "sum"))
plot(Qmm ~ DatesR, data = NewTabSeries, type = "l")

## daily times series
NewTabSeries <- SeriesAggreg2(TabSeries = TabSeries,
                              Format = "%Y%m%d",
                              ConvertFun = c("sum", "sum", "mean", "sum"))
plot(Qmm ~ DatesR, data = NewTabSeries, type = "l")

mentioned in commit 344a99d0

The first tests on the new function seem to indicate that it works.

For SeriesAggreg and SeriesAggreg2 documentation, I think we should add the following in the Details:

TabSeries first column must be a vector of class 'POSIXct' or 'POSIXlt'

and

Only 'sum' and 'mean' functions are allowed for the ConvertFun argument

In SeriesAggreg and SeriesAggreg2,

09 for hydrological year starting in September

should be replaced with

10 for hydrological year starting in October

to be more consistent with what we usually define as hydrological year.

SeriesAggreg2 seems to work well when one column is full of NA.

SeriesAggreg2 does not work when all functions are the same (only mean or only sum) or when there is only one column to aggregate.

We should specify that YearFirstMonth must be comprised between 1 and 12.

If YearFirstMonth is different from 1, with the example data, the aggregated E value differs between years, which is not the case is YearFirstMonth == 1. Not sure why as E is apparently recycled between years...

SeriesAggreg2 seems to work well when one column is full of NA.

Yes, it is. In the old function, this was due to a bad check. Here, there's none at the moment, so it's okay.

SeriesAggreg2 does not work when all functions are the same (only mean or only sum) or when there is only one column to aggregate.

I am aware of this. It is because I calculate on the one hand the variables for which a sum is calculated, and on the other hand the variables for which an average is calculated. I haven't written a condition yet, so it blocks as soon as there is only one type of calculation (which is the case when there is only one column).

We should specify that YearFirstMonth must be comprised between 1 and 12.

That's the plan. It will be done with all checks.

If YearFirstMonth is different from 1, with the example data, the aggregated E value differs between years, which is not the case is YearFirstMonth == 1. Not sure why as E is apparently recycled between years...

I don't think there's anything wrong with E, it's just an artifact due to the time lag.

t(sapply(1:12, function(iMonth) {
  colMeans(SeriesAggreg2(TabSeries = TabSeries,
                Format = "%Y",
                ConvertFun = c("sum", "sum", "mean", "sum"), YearFirstMonth = iMonth)[, -1], na.rm = TRUE)
}))

             P       E        T      Qmm
 [1,] 1306.659 773.637 4.397387 834.0093
 [2,] 1307.660 773.637 4.394651 837.2398
 [3,] 1301.804 773.637 4.399010 837.1611
 [4,] 1297.990 773.637 4.397400 834.4623
 [5,] 1295.416 773.637 4.407993 836.4688
 [6,] 1296.348 773.637 4.405374 830.5675
 [7,] 1296.186 773.637 4.402236 820.5473
 [8,] 1295.703 773.637 4.405294 818.0329
 [9,] 1294.788 773.637 4.401807 817.5955
[10,] 1297.892 773.637 4.403590 817.5614
[11,] 1300.759 773.637 4.404807 818.1412
[12,] 1300.401 773.637 4.397562 818.2841

plot(E ~  DatesR, data = TabSeries, type = "l", col = "green3")

mentioned in commit 34e7de42

I think that all checks are now written.

I tested the last version.

I corrected the help page (attached).

I also tested the checks. I made modifications in the code directly. I also corrected minor typos.

If we put a wrong value in NewTimeFormat, which is deprecated, then the error might not be 100% clear.

+                               NewTimeFormat = "%d",
+                               ConvertFun = c("sum", "sum", "mean", "sum"))
Error in match.arg(Format, choices = listFormat) : 
  'arg' should be one of “%Y%m%d”, “%Y%m”, “%Y”, “%m”, “%d”

Maybe check just before line 10 that NewTimeFormat respects the correct possibilities?

  TimeStep <- c("hourly", "daily", "monthly", "yearly")
  NewTimeFormat <- match.arg(NewTimeFormat, choices = TimeStep)

The function now runs when all ConvertFun are the same, but not yet when there is only one column to convert.

> NewTabSeries <- SeriesAggreg2(TabSeries = TabSeries[, 1:2],
+                               Format = "%Y%m%d",
+                               ConvertFun = c("sum"))
 Error in apply(TabSeries[, -1L], MARGIN = 2, FUN = function(iCol) { : 
  dim(X) must have a positive length 

SeriesAggreg2.R

SeriesAggreg2.Rd

mentioned in commit 605281fc

mentioned in commit 3fba4b08

mentioned in commit c2f98a6f

mentioned in commit 7f5b7e56

mentioned in commit 9b1db179