frametoLme(), modelFrame(),
summary.modelFrame(), print.summary.modelFrame(), plot.modelFrame(),
anova.modelFrame(), muleMan() and plot.muleMan() fit the same
well-posed random parts (a tree intercept, with a tree-level log(csx)
slope for the time models; a plot intercept for the aridity index), so
their printed numbers change. The frametoLme() example no longer
compares with the default random part. The muleMan() examples correlate
the processed tree-biomass series with the aridity-index fluctuations.frametoLme(): residual series in $fluc could be attached to the
wrong rows, because row labels were used as row positions when the residual
frame was built. Residuals are now matched to model-data rows by label, and
fitting stops with an error if any row cannot be matched. Model fits were not
affected. The defect was introduced in development versions 0.9.2-0.9.4
(April 2026); the CRAN release 0.9.1 matched rows with
merge(..., by = 'row.names') and is not affected.$fluc.wlai(): with label_year = "end", hydrological years were still
labeled by the year in which they start, so October 2000 to September 2001
came out as 2000 instead of 2001 and lag 0 pointed to the year before ring
formation. The output now has a year column: with "end" it is the year in
which each hydrological year ends, with "start" the year in which it
starts. hydro_year_id keeps its meaning and is always the start
year. label_year now also applies when the input already has a
hydro_year_id column, as returned by moveYr(); if the input year column
follows the other convention, wlai() warns and its own label_year
decides. With calendar years label_year has no effect. label_year was
added in development version 0.9.2; the CRAN release 0.9.1 is not affected.frametoLme(): when the input had more than one time column (for
example year and the time column added by rtimes()) and the model used
one other than the first detected, building $fluc stopped with
"object 'year' not found". $fluc is now built on the time variable of the
fitted fixed formula. Fits that use year are unchanged. The code that
picked the first detected time column was introduced in development version
0.9.2; versions 0.9 and 0.9.1 built $fluc differently.shiftFrame(): dot-separated wide column names were read in the
wrong order, so P16106.17.a gave plot = a and sample = P16106, and
climate columns such as P16106.Apr gave plot = Apr and
month = P16106. Column names are read from the highest level to the
lowest again, as in 0.9.1: P16106.17.a is core a of tree 17 in plot
P16106. This also corrects modelFrame() on wide chronologies, which
fitted the grouping structure on the swapped labels (for Pchron, 8 trees
instead of 4). The defect was introduced in development version 0.9.2; the
CRAN release 0.9.1 is not affected.moveYr() then wlai() route through ringApply(): the start
month recorded by moveYr() was lost when ringApply() combined the
blocks, so wlai() ordered each hydrological year January to December
instead of October to September and returned different aridity indexes
(for PTclim05, P44005, October 1998 to September 1999: 1.3774 instead of
1.5276). ringApply() now keeps the start month in the data's metadata and
wlai() uses it, so this route agrees with wlai(ini.mnt = "Oct") on every
block. The defect was introduced in development version 0.9.2; the CRAN
release 0.9.1 is not affected.label_year in wlai() and moveYr() from
"start" to "end", so that by default each hydrological year is labeled
by the year in which it ends (October 1998 to September 1999 is labeled
1999), as in the CRAN release 0.9.1. "start" remains available.
hydro_year_id still holds the start year.scacum() with a vector of
reference radii in sc.c scaled some cores to another tree's radius (with
Prings05 and Pradii03, 6 of the 8 cores). Every core is now scaled to its
own tree's radius at rf.t. Fixed since development version 0.9.2; added a
regression test.tdForm() to the behavior of the
CRAN release 0.9.1: with on.time = TRUE (the default) the relative time
column time from rtimes() is used when present, otherwise a year-like
column. In development version 0.9.4 the default had switched to year
whenever the data carried metadata, which ringApply() adds automatically
(checked with the 0.9.2, 0.9.3 and 0.9.4 builds). This affects the
ringApply() help route and modelFrame(Pchron), which again fit
log(x) ~ log(csx) + time. A time variable declared with
defineDataRoles(time = ...), on.time = FALSE, or a formula string still
decides as before.muleMan() help example (modelFrame() on
Prec and Temp with moveYr() and wlai(), then lmeForm) and
muleMan() itself. The route failed in development versions 0.9.2 to 0.9.4
(checked with their builds), and in later development code the
aridity-index model used hydro_year_id, the start year, instead of the
hydrological-year label year. lmeForm() now uses a time variable from
the data's metadata only when it was declared with
defineDataRoles(time = ...), so the model is AI ~ year as in the CRAN
release 0.9.1, and muleMan() pairs ring year t with the hydrological year
ending in t. muleMan() also no longer fails when the tree-ring data have a
single time column or when both inputs have a residual column of the same
name.muleMan(): the plot and core labels of its output could be swapped
(for Pchron, core a of tree 17 in plot P16106 came out as
plot = a, sample = P16106), because the labels were parsed from block
names in an order that depended on the column order of the tree-ring data.
Each correlogram now takes its labels from its own rows, whatever the
column order. The CRAN release 0.9.1 labeled them correctly; in the
0.9.2 to 0.9.4 builds the help-example route failed before reaching
muleMan().plot() now works on frametoLme() fits made with res.data = FALSE: the
residual fluctuations are rebuilt from the model and show the same points
as the plot of the same fit with $fluc. The plot method added in
development version 0.9.4 stopped with "subscript out of bounds" on such
fits, and earlier versions, including the CRAN release 0.9.1, had no plot
method for them.amod(): the first increment of each series was the whole
transformed size, which is wrong when the cumulative size is scaled (with
scacum() and radii in sc.c). On the ringApply() help route, the first
basal-area increment of core a of tree 17 in plot P16106 (1897) was
5545 instead of 183. The first increment is now the change from the size
before the first ring (csx minus the first ring width) to the first
transformed size; later increments and unscaled series are unchanged. The
CRAN release 0.9.1 returned NA for first increments.wlai(): on long calendar data without ini.mnt, every 12-month
block started at the month of the first row of cd, so the aridity indexes
depended on the row order (shuffling the rows of PTclim05 changed them by
about 20% on average). Blocks now start at the first month of the earliest
year. Without a year column the first row's month is still used, so rows
should then be in time order. Results on time-ordered data are unchanged.
The CRAN release 0.9.1 computed one index over all rows of long data.levexp(): its help call levexp(Pradii03, Prings05) returned an
empty vector. The integer year column was taken as a grouping column, and
a name with two levels such as P16106.17 could not match a core label such
as a.17.P16106. Time columns are no longer taken as grouping columns, and
a name now matches a split label when its dot-separated parts are a
contiguous run of the label's parts, read in either direction; each core now
gets its own tree's radius. Single-part names match as before. The CRAN
release 0.9.1 returned only the two cores of tree 7 in plot P44005.ringApply(): named arguments that vary by block, such as
sc.c = Pradii03, were handed to the blocks by position, so their names were
ignored and the result depended on the order of the vector's elements (with
rev(Pradii03) every core of the help route was scaled to another tree's
radius). A named vector whose names match the split labels is now given to
each block by name through levexp(); other arguments are recycled over
blocks as before. If the names match some blocks but not all, a warning
names the argument and the unmatched blocks, and the values are given by
position. The CRAN release 0.9.1 was also affected: it handed sc.c out by
position, and with rev(Pradii03) 6 of the 8 cores got another tree's
radius.cClass()), and character columns are now
also treated as groups (0.9.1 ignored them); whole-number numeric columns
are time candidates, and other numeric columns are responses. The month
column (named month, or a factor whose levels are month names) is never a
group. The development versions 0.9.2-0.9.4 guessed roles from column names
(lists such as plot, tree, zone, site, year, hydro_year_id); those lists are
removed. Roles declared with defineDataRoles() override the type rule;
metadata that ringApply() creates automatically no longer does. On the
bundled data the documented results are unchanged, except that month in
PTclim05 is no longer a grouping column (lmeForm(PTclim05) is
pr ~ year | plot, and slitFrame(PTclim05) splits by plot only). Year can
still be a random grouping level: declare it with
defineDataRoles(groups = ...) or add a factor copy of it.rtimes(), scacum(), moveYr() and wlai() find their time columns by
the package's column names: year (or hydro_year_id from moveYr()),
month and time; a month column may also be a factor whose levels are
month names. The development versions 0.9.2-0.9.4 also read aliases (yr,
calendar_year, hydro_year, water_year, mon, mnt, mo) and other
time units (week, day, hour, minute, rel_time, ring_time, cambial_age, age);
these are no longer read. moveYr() stops if there is no column named
year; scacum() stops if rf.t is given and there is no year,
hydro_year_id or time column (it used to fall back silently to each
block's last value); rtimes() and wlai() warn when none of year,
hydro_year_id, month or time is present. The CRAN release 0.9.1 never
had the aliases.scacum() reads its values from the column x, as in the CRAN release
0.9.1; a response declared with defineDataRoles() takes precedence, and
without either scacum() stops. wlai() reads precipitation from pr and
temperature from tm, the package's column names; declared inputs take
precedence (declared pr and tm by name, otherwise the declared order,
precipitation first). Without pr or tm, wlai() uses the first two
numeric columns and warns which ones; this fallback follows 0.9.1, which
read columns 1 and 2 by position. The name lists added in development
versions 0.9.2-0.9.4 (trw, rw, rwi, abi; pm, prec, ppt,
precip, rain, temp, tmean, tavg and others) are removed.frametoLme() and modelFrame() now say that rows with
missing values in the fitted variables or grouping columns are dropped
without a message (e.g. the rows where rtimes() gives no time), and
describe the default random part and how to pass random =; rtimes()
says which rows get NA time; moveYr() has a Value section explaining that
the calendar year is relabeled and how to recover it.frametoLme(): when a column was both a grouping level and a time
column (for example year declared in groups with defineDataRoles()),
$fluc carried it twice, the second copy as year.1. It now carries it
once.defineDataRoles() example uses the bundled data's own columns (it
named zone and health, which no bundled dataset has, and stopped with an
error).frametoLme(), summary(), print()
and the printed summary of frametoLme objects.frametoLme() and modelFrame() fits: the
"Groups" line also listed nlme's counts for the model matrix and response
(X 1, y 1); it now lists only the grouping levels.frametoLme() and modelFrame() fits: the
fixed-effects table was printed in scientific notation, including the DF
column (9.8200e+02); it is now formatted column by column as in nlme's
summary, with whole DF and p-values rounded to 4 decimals.modelFrame() (fitted model,
Friedman smoother and processed data only) and for summary(), print()
and the printed summary of modelFrame objects.frametoLme(): refitting a fit's model data (fit$model$data) with
tdForm or lmeForm reversed the nesting (sample/tree/plot), because the
fitting copy listed the grouping columns in reverse. The copy now keeps the
input's column order.frametoLme(): the model data stored in the fit (fit$model$data)
kept only the time variable of the fitted model, so refitting a year model
with tdForm(log.t = TRUE), as in the BIOdry 0.8 example, silently used
log(year) instead of log(time). The model data now keep all time
columns; rows are still dropped only for missing fitted variables and
grouping columns, and $fluc is unchanged.frametoLme() examples compare the two Zeide time forms (linear and
logarithmic time, log.t) by maximum likelihood under a random part the
bundled data support (a tree intercept and a tree-level log(csx) slope),
refit the preferred form with REML, and show the default random part's
near-zero variance components for comparison; the frametoLme(),
tdForm() and modelFrame() help pages explain the time forms, singular
fits and how to pass a simpler random part.moveYr() example on the bundled PTclim05: hydrological-year
labels, recovering the calendar year, label_year and ini.mnt.frametoLme() and modelFrame() fits: nlme's
default aspect = "xy" collapsed the panels to thin strips, and the axes
carried internal column names (.mf_resid__, .mf_fitted__,
.model_resid__). The panels now fill the plotting area and the axes read
"Pearson residuals" and "Fitted values" (or the time variable). The plot of
processed data (modelFrame(form = NULL)) also fills its panels. aspect,
xlab and ylab given by the user are kept.plot() of frametoLme and
modelFrame objects (residuals over time; residuals against fitted values,
by core and by plot; processed series with form = NULL).plot() of frametoFriedman objects and of
modelFrame() Friedman fits): the panels now fill the plotting area, and
the axes read "Ratio index" (or "Difference index" with difference = TRUE) and "Fitted smooth" or the time variable instead of the column names
x.idx and x.fit; aspect, xlab and ylab given by the user are kept.frametoFriedman() and modelFrame() Friedman fits: the printed status
table no longer carries the note column, which wrapped into a mostly empty
block; notes are listed below the table, one short line per series with its
number of NA values (the full note and the count n_na are kept in the
series table). frametoFriedman() lists the series in the order of their
first rows in the data, as the modelFrame() path does, instead of
factor-level order.plot() of frametoFriedman objects with levels coarser than the
series (e.g. levels = 'plot'): one line was drawn through all points of a
panel, joining different cores; each series now has its own line within the
panel, with a key. The default plot is unchanged.FriedmanForm(), frametoFriedman()
and the summary(), print() and plot() methods of frametoFriedman
objects (ratio and difference indices, series status, panels by plot).muleMan() and its plot() method
(biomass and aridity-index fluctuations, Mantel correlograms per core).anova() example for modelFrame objects: sequential F tests of
one biomass model, and AIC/BIC of the time forms t and ln t fitted by ML.plot() of muleMan objects orders the panels by plot, tree (as numbers)
and core, and labels the axes "Lag" and "Mantel r" unless xlab or
ylab are given (they were ecodist's lag and mantelr).anova() of frametoLme objects: description, arguments and
a runnable example (F tests of one fit; ML comparison of the time forms t
and ln t).update() of frametoLme objects: the refitted object's call shows the
original data expression instead of the internal placeholder ..rd...
Fits are unchanged.normalizeGroupLevels() on fitted objects (frametoLme, modelFrame)
without groups orders the grouping levels of $fluc: the grouping
variables come from the metadata of $fluc, else from the model's grouping
formula. Before, the call left $fluc unchanged (tree levels 18, 7, 56, 17
in the frametoLme example fit; now 7, 17, 18, 56).normalizeGroupLevels() orders a grouping column as numbers only when all
its non-missing labels read as numbers, and keeps the labels themselves as
levels (e.g. "07" stays "07" and comes before 17). Before, a label that
did not read as a number ("T7" next to 17, 18, 56), or that changed when
read as one ("07", "1.0"), became NA.update() of frametoLme objects: description, arguments and
a runnable example (ML refit; the time forms t and ln t).normalizeGroupLevels(): description, arguments and a
runnable example (fitted object and data frame).plot() of wlai objects: description, details, arguments
and an example on the bundled PTclim05 (it simulated random climate data).wlai() example uses the bundled PTclim05 (one calendar year; one
index per plot and hydrological year; a diagram) instead of random data.moveYr() help page: descriptions of the arguments cd, ini.mnt,
keep_hydro_id, hydro_id_name and validate.plot() of wlai objects: the block label printed after the aridity index
follows the year column of the block, so it follows label_year: with
the default 'end', October 2000 to September 2001 of plot P44005 is
labeled P44005.2001 (it showed the start year, P44005.2000). Block names,
used by which, still hold hydro_year_id.frametoLme, modelFrame and
frametoFriedman objects, is declared in Imports.\code markup (plot() and summary() of
frametoLme objects), double-escaped markup in scacum(), the links to
the 'nlme' plot methods for grouped data (plot.nfnGroupedData,
plot.nmGroupedData), and the cClass() title.|>.AIC, BIC, as.formula, ave, fitted, logLik,
resid and setNames from 'stats'.wlai() help page: the Value section describes the returned data frame
(it described a numeric index and a plot).frametoLme() drops only rows with missing values in the fitted variables
or the grouping columns. The CRAN release 0.9.1 dropped rows with a missing
value in any column, so fits can now use more rows (20 more in the
modelFrame() biomass example).mUnits() help page: "milli" spelled correctly.Strengthened the internal helper chain used by modelFrame() and related workflows for multilevel tree-ring and climate data.
Improved class and hierarchy detection in cClass().
Date, and POSIXt columns.Improved formula construction in lmeForm() and tdForm().
lmeForm() now resolves response, time covariates, and grouping levels more defensively on multilevel long data.tdForm() now preserves the intended formulation log(x) ~ log(csx) + f(time).tdForm() now checks tree-growth inputs more explicitly and warns or stops more clearly when the input looks climatic or does not contain the expected growth variables.Improved argument and level expansion helpers.
arguSelect() now validates target functions more safely, handles MoreArgs more defensively, and warns when direct arguments or MoreArgs entries are dropped.levexp() now avoids the previous slitFrame() + grep() bottleneck, uses token-based matching, and behaves more predictably on multilevel labels.slitFrame() now detects grouping columns more robustly and keeps split metadata for downstream helpers.Improved hydrological-year handling in moveYr().
label_year = c("start", "end").hydro_year_id so downstream functions can group annual climatic blocks independently of the displayed year label.Improved Walter--Lieth aridity workflows.
wlai() now works more safely on long multilevel climate data by splitting into annual blocks instead of treating long concatenated tables as one continuous series.wlai() now preserves hydrological month order inside each block.plot.wlai() now supports multi-block wlai objects more safely, allows block selection without leaving the standard S3 plot() interface, and defaults to plotting the first available block when no selector is provided.plot.wlai() now uses a black 100 mm line, a blue compressed precipitation band above 100 mm, and automatic top headroom for legends and labels.plot.wlai() to document multilevel block selection, plotting defaults, and Walter--Lieth styling.Improved split/apply helpers for multilevel tree-ring and climate data.
ringApply() now respects the resolved lv more consistently, orders rows within each split block on the remaining time variables before applying fn, and returns columns in the order: records, time variables, then factor hierarchy.rtimes() now works both as a block-level helper and on multilevel data frames, including grouped time reconstruction through lv.scacum() now works both as a block-level helper and on multilevel data frames, including block-wise ordering before cumulative growth reconstruction.amod() now works on block-level series and multilevel data frames, applies the allometric model to cumulative growth internally, returns both allometric x and csx, and avoids introducing an initial NA in the reconstructed increments.Improved reshaping in shiftFrame().
split_pattern while keeping dot splitting as the default.tm and pm instead of dropping one variable during reshaping.Improved model fitting wrappers.
frametoLme() now orders multilevel long data before fitting, rebuilds residual series from real time and grouping columns more safely, and wraps residual outputs more consistently.modelFrame() now distinguishes long data frames from non-data-frame lists more safely, handles staged fn/lv recursion more defensively, applies mUnits() to record-like numeric columns more conservatively, and passes validation settings through the helper chain.modelFrame() handling of long multilevel inputs by resolving stage levels against the current hierarchy without silently dropping valid grouping columns.Added or expanded inlinedocs details in patched helpers so the internal contracts are documented more explicitly.
ringApply(), shiftFrame(), amod(), and plot.wlai().anova.modelFrame.Rd: linked anova.lme to the nlme package.frametoLme.Rd: linked lme, nlme, and groupedData to nlme.lmeForm.Rd: linked groupedData to nlme.modelFrame.Rd: linked lme, nlme, and groupedData to nlme.muleMan.Rd: linked groupedData to nlme, mgram to ecodist, and ecodist correctly.shiftFrame.Rd: linked groupedData to nlme.summary.modelFrame.Rd: linked summary.lme to nlme.Pchron.Rd, Prec.Rd, and Temp.Rd.shiftFrame: The internal implementation of data.frame() now uses stringsAsFactors = TRUE.Developed five new S3 methods:
anova.modelFrameplot.modelFrameplot.muleManplot.wlaisummary.modelFrameChanges in modelFrame:
Changes in muleMan:
New binary datasets added to data/ folder:
Pchron.RData, Prec.RData, and Temp.RData (contain dendroclimatic chronologies).New function mUnits: Converts metric units in dendroclimatic chronologies.
Changes in amod:
New function cClclass:
colclass function.Changes in tdForm:
Changes in ringLme:
frametoLme() to improve lme implementation from the nlme package.Changes in shiftFrame:
lev.nm = c("plot", "tree", "sample")
Changes in muleMan:
groupedData object.plot.man = TRUEpsig = 0.05groupedData object using other trellis-graph functions. (See function examples.)