EOF()'s rotate argument now will take a function to apply to the EOF loadings. rotate = TRUE is still supported but deprecated and will default to using function(x) stats::varimax(x, normalize = FALSE).
Contour functions gain a proj.latlon to decide if the projection needs to go to latlon coordinates or to projected coordinates.
GetSMNData() function is defunct. The functionality was a hack and hard to maintain and it's also very specific and should be out of scope for a general package.trans argument of scales has been renamed to transform following ggplot2's change.scale_mag() internals are now much simpler and inherits all the functionality from common continuous scales. (thanks @teunbrand for the suggestion, #186)step computed variable in geom_streamline() now goes from 0 to the total number of steps instead of from negative half to positive half.clip argument to only show contours in an area defined by a polygon.kriging argument of the contour functions now can be a numeric to control de number of pixels used.FitWave() and friends improved (#184, @pascaloettli).proj argument in geom_contour_fill() and friends now operate over the isolines returned by the isoband package. This might break code that used a custom function to proj.DivideTimeseries() function is now defunct.FitLm() gains a new intercept argument that allows you to remove the intercept term that is automatically added.Laplacian() when computing the laplacian of a single variable. (#170, Thanks @pascaloettli)geom_relief() and geom_shadow() now don't have a size aesthetic, since they shouldn't have had it anyway.scale_mag() is now chosen to be a "pretty" number close to the maximum. This will prevent guides with many unnecesary decimal places (thanks, @PanfengZhang #161).expand value in sale_x_longitude() and friends (thanks, @tamas-ferenci, #167).Smooth2D() smooths a 2D field (hence the name). There are two smoothing methods. smooth_svd() computes the SVD of the field and reconstructs it keeping only the leading values that ensures a maximum variance lost. smooth_dct() computes the Discrete Cosine Transform of the field and sets a proportion of the components to zero.Fixed a bug in geom_streamline() when plotting multiple fields on the same panel.
Fixes a bug with GetTopography() on Windows.
Updates documentation to use valid HTML5 per CRAN's new format.
Fixes the "The following aesthetics were dropped during statistical transformation" warning for contours.
Removes ReadNetCDF() file and URL checking because it didn't work in some cases.
(#164, thanks @pascaloettli)
ReadNetCDF() not longer uses it to parse dates and times. The homebrewed parser might be more limited, and the wild variety of netCDF files in the wild makes it hard to fully test. Please report any issues you have!Adds example of scale_y_level() (@paocorrales, #153).
ReadNetCDF() now should parse times correctly even if the use non-standard calendars.
This now makes udunits2 and PCICt required to parse time.
Arrays returned by ReadNetCDF(…, out = "array") gain a "dimvalues" attribute which is analogous to dimnames but has the correct types (dates are dates, numerics are numerics, etc...).
EOF() gains a new engine argument to chose the function to compute the singular value decomposition.
sdev component in EOF() when using base::svd() in some cases.seasonally() now returns a Date object even if the input is datetime. This avoids issues when the time component of the input was not all the same.ImputeEOF() in which the algorithm tried to compute 0 EOFs.geom_contour2() gains the ability to draw --and leave space for-- labels!
This is finally proper labelling support without having to use a different geom (geom_text_contour()).
Thanks to the isoband package for this.
Following isoband's naming convention, the family of functions that decide where to place labels has been renamed as label_placer_ and the corresponding argument is now label.placer .
The label_placement_ family of functions will be deprecated in future releases.
discretised scales now work better when passing user-supplied limits.
New functions to use the International Standard Atmosphere to get height from pressure and vice versa, as well as to use as secondary axis.
See ?standard_atmosphere.
scale_y_level() and scale_x_level() now print more breaks by default.
These functions will probably use this transformation instead of the reverse log transformation in a future release.
Translation! Thanks to @MichaelChirico priceless guidance, metR messages are now translatable and already translated to Spanish. If you are using R in a Spanish locale you should be getting messages and error in Spanish. Partial translation to Portuguese is also included.
During the process of translating messages, many messages were improved and made more consistent.
New Function ResidLm() that returns the residuals of a linear fit.
New function Detrend() that, you guessed it, returns a (linearly) detrended version of the input vector.
In ReadNetCDF(), the "vars" argument now can take a function.
(#142)
Discretised scales now support user-defined breaks.
seasonally() result will be on the 15th of the centre month of each season instead of on the 1st. This makes the date more representative of the time span and also solves a bug in which dates on the 31st would return NA. This is a potentially breaking change.ReadNetCDF() doesn't fail when reading variables with no dimensions (thanks to @paocorrales, #141).geom_contour_tanaka() in R >= 4.1.0.as.discretised_scale() is a quick way of created a discretised version of any continuous scale.
stroke.colour is now an accepted aesthetic for geom_text_contour().
The computations in MixingRatio() were wrong.
Now they are fixed.
I really wanted geom_arrow() to automatically add the arrow legend, but the workarounds I managed to write were brittle and couldn't handle even slight changes in people's code (see https://github.com/eliocamp/metR/issues/130).
The definitive answer is that this is simply not possible due to limitations on how ggplot2 works (see https://github.com/tidyverse/ggplot2/issues/4291).
This release, then, backtracks those workarounds and tries to accept the things I cannot change.
I'm very happy with discretised_scale(), which is a type of scale which takes a discrete values that are the result of some discretisation and treats them as continuous.
It's, in a sense, the inverse of the new ggplot2::binned_scale().
Whereas.
ggplot2::binned_scale() takes continuous values and then discretises them, discretised_scale() takes discrete values which where the result of some discretisation procedure (such as the levels of geom_contour_fill()/ggplot2::geom_contour_filled()) and allows you to treat them as continuous.
Related to that, geom_contour_fill() now gains a new computed aesthetic called level_d, which is the same as level but forces ggplot2 to automatically use the new discretised scale.
AnchorBreaks() gains a bins argument to mimic the default functionality of MakeBreaks().
New label_placement_minmax() to label contours at the maximum and minimum points of a contour (mimicking isoband's behaviour)
geom_contour_tanaka() now has a (rather experimental) argument smooth which allows to smooth the transition between segments.
Fixes error introduced in previous version when geom_arrow() had mappings other than dx and dy.
(Thanks Santiago!)
The level derived aesthetic from geom_contour_fill() now returns and ordered factor with the correct labels that can be interpreted by ggplot2::guide_colorsteps().
This might a breaking change!
geom_label_contour() lives!
The previous release rewrote much of the way geom_text_contour() worked, but I messed up and didn't realised that the new code had broken geom_label_contour() (to be honest, I'd almost totally forgotten about it :P).
(fixes #126, thanks @kongdd)
geom_text_contour() placement of labels is completely redesigned.
It gains an argument label.placement which takes a function that is in charge of positioning the labels.
See label_placement_flattest() for more details on the possible placement methods and how to build your own.
The default method is label_placement_flattest, which places the label where the product of the curvature and the angle of the contour is minimised.
This aims to put labels in a straight segment that is also as horizontal as possible.
If more than one point satisfy this condition, then it chooses the closest to the midpoint.
This is a breaking change, as it will change the label position of previous plots.
The contour functions also gain a proj argument.
It can be a proj4 string to project the contours of or an arbitrary function to alter the contours after they are computed.
This now makes it possible to compute contours for data that is on a regular grid on projected coordinates but want to plot in lon-lat coordinates (or vice versa).
Bear in mind that contours that cross the dateline will probably end up mangled.
Contour functions also gain a kriging argument.
If TRUE, will perform ordinary kriging to interpolate irregularly placed data into a regular grid.
FitLm() handles NAs better.
GetSMNData returns the date parameter with the correct time zone.
WaveFlux() now only returns the value of the horizontal fluxes.
That is, it will not return lon and lat.
This is a potentially breaking change.
The contour family of functions now use isoband to compute contours (thanks to @clauswilke for the awesome package) instead of my ugly hack/workaround.
As a result, contours are faster and much more reliable.
EOF() will now work even if {irlba} is not installed.
GetTopography() is now updated to the new ETOPO server.
It now requires {raster} to work.
FitWave() and related functions return NAs when the inputted signal has NAs.
FitLm() accepts a weights argument to perform weighted regression.
ReadNetCDF() now can read files directly from DAP servers and other urls, and objects returned by $$ncdf4::nc_open()$$.
ReadNetCDF() won't try to parse "time" dimensions that are not dates and will try to parse as time all dimensions.seasonaly() is now correctly called seasonally(). This proves that you don't put an ESL person in charge of naming stuff.EPflux() computes Eliassen-Palm fluxes (experimental).
geom_arrow() and geom_vector() should plot faster.
New functions is.full_season() and seasonaly().
FitLm() returns model $r^2$ and adjusted $r^2$.
FitLm() adds names to unnamed terms.
New function WaveEnvelope() that computes...
the wave envelope.
geom_contour2(), geom_contour_fill() and geom_text_contour() now accept a global.breaks argument that controls whether breaks should be computed once using the range of the whole dataset or once for every grouping (e.g. faceting).
TRUE (the default) ensures that intervals between contours are comparable between panels.
Setting it to FALSE computes contours compatible with ggplot2::geom_conotur() (#109, thanks @freeseek)
A reworked non-equispaced derivative gives better results in Derivate().
ReadNetCDF() will not fail if the first variable was called "v" (yeah, I know.. weird error related to data.table's non standard evaluation).
Subsets in ReadNetCDF() won't fail if no element is named (#107, thanks @m-saenger)
Fixed bug in WaveFlux() (#110, thanks @salvatirehbein)
ReadNetCDF() supports using NA in subset to refer to max or min value.
ReadNetCDF()'s subset argument supports more complex queries.
(see the help section at?ReadNetCDF()).
ReadNetCDF() now has a simple date-time parser that is tried if the udunits2 package is not installed.
GetSMNData() gains the ability to cache results in a file.
Derivate() now can derive in a non-equispaced grid.
geom_contour_fill() imputation method is fixed for some special cases (#96; thanks @bergmul).
predict.eof() handles complex value svd correctly.
Accommodates new grid implementation of units (#105 thanks @pmur002).
New GlanceNetCDF() function that is an alias for ReadNetCDF(out = "vars") but now prints a human readable summary of the contents of the NetCDF file.
geom_streamline() now uses 4th order Runge-Kutta instead of plain old Euler.
It also draws arrows in the middle of the streamline.
ReadNetCDF() is slightly faster and should use less memory.
ReadNetCDF() supports more complex subsetting operations now.
The df element returned by FitLm() now has the same length as the rest.
ReadNetCDF() result will print correctly.Anomaly() has a new baseline argument.
New function Trajectory() that computes trajectories in time-varying velocity fields.
ReadNetCDF() now accepts dates as elements for subset.
ReadNetCDF() will read all dates correctly (#91; thanks to @m-saenger).
The es argument from MixingRatio() changes to e, to be consistent with the rest of the variables.
Sorry, thermodynamics is not my forte!
(thanks @PaoCorrales)
Arrow heads in geom_arrow() are now scaled correctly in faceted plots (fixes, #92; thanks to @m-saenger)
There has been some changes in the interface of some functions for the sake of consistency.
In Derivate() (and it's derived functions --see what I did there?), the data argument has been moved back.
This is because this function is intended to be called inside a data.table of mutate() call, where you don't need to explicitly specify the data.
In EOF() the dcast-style formula interface has been removed.
The data argument was also moved back so you can use the n argument more easily without naming it.
ImputeEOF() follows the same conventions.
The dcast-style interface has been removed and the data argument has been moved after the max.eof argument.
BuildQsWave() and FitQsWave() have been removed and should had never even existed.
The default skip argument for geom_text_contour() is now 0.
Removed hemisphere argument from season() since the function returns the trimester so it made no sense.
Contour functions now compute breaks globally (using all the data) instead of per panel.
This means default intercomparabilty between faceted plots but also a considerable deviation from ggplot2::geom_contour().
The Between operators are removed since they were already implemented in data.table.
The default geom for stat_na() is changed to point for consistency with stat_subset()
geom_arrow() and geom_vector() scale with vector magnitude.geom_streamline() for visualizing vector fields.dlon(), dlat(), dx(), dy() for converting physical units into spherical units.geom_contour_tanaka() that plots illuminated contours.Interpolate() for bilinear interpolation.FitWave() with wavenumber 0. Now it returns the mean.FitWave() runs slightly faster and BuildWave() runs much faster.GeomContourFill object since it was just a polygon.EOF() now use factors instead of numbers to identify each PCscale_mag() and guide guide_vector() for controlling and communicating the scale of vectors. These are highly experimental and will change in the future, but provide some very needed functionality so I decided to export them as they are.geom_arrow() gains new pivot argument to control point of rotation and preserve.dir to tell if angle should be preserved.stat_contour_fill() and stat_contour2() print a warning when no contours can be made.EOF() now supports estimation of confidence intervals via bootstrap.EOF() supports varimax rotation. Rotated components are labelled accordingly.geom_relief() is much faster now (but see package rayshader).geom_shadow() for casting shadows in topographic maps.StatContour2 are memoised so they are only computed once even adding several layers with the same contours (geom_contour() + geom_text_contour()) or running the same plot while tweaking it's appearance.FitLm() for getting fast estimates of linear fits.FilterWave() for filtering waves.RepeatCircular() was deprecated in favour of WrapCircular() which is more robust, and then WrapCircular() was deprecated in favour of the ggperiodic package.stat_contour2(), stat_contour_fill() and geom_streamlines() handle circular dimensions has changed. Now you need to use xwrap/ywrap to explicitly set the domain. This makes the implementation more robust and also allow to easily wrap to an arbitrary domain.ggperiodic, which can handle this stuff better. The above mentioned arguments still work but are not documented and will be deprecated.EOF() gains a fill argument for sparse-ish data.geom_text_contour() now has the ability to draw a stroke around text.geom_contour_fill() now can impute missing values with bivariate interpolation.BuildField() actually renamed BuildWave() (which should've happen in the previous release according to this News).as.path() and added functionality to Interpolate() that allows to interpolate values along a path of locations.Impute2D() which is an exported version of the method used by geom_contour_fill() to (soft) impute missing values.subset argument in ReadNetCDF() now accepts character vectors for time components.GetTopography()ReadNetCDF() now can output arrays and vectors.pkgdown.coriolis() and coriolis.dy().Anomaly() (can't believe I did that).EOF() principal components are numeric instead of "PC1", etc...stat_contour() created geom_contour_fill() which works just like stat_contour_fill().stat_na() for easy masking of NA values.geom_arrow() now is geom_vector() with better documentation and more parameters and geom_arrow() is a new geom for arrows that preserve direction.ReadNetCDF() now handles more date formats.Laplacian(), Divercence() and Vorticity()DerivatePhysical() is removed since it was made redundant by the extended functionality in Derivate()geom_text_contour() and geom_label_contour() for labelling contours.GeostrophicWind().ReadNetCDF() and as.POSIXct.ReadNetCDF() now supports time-zones via de udunits2 package.stat_contour_fill().MakeBreaks() and AnchorBreaks().guide_colorstrip() displays discretized values of a continuous colour or fill scale.stat_contour_fill()geom_relief() for generating relief shading.ReadNetCDF() gains a subset argument for subsetting data.\AnchorBreaks() is a new way of generating breaks.geom_contour2() that takes a function as argument in breaks and geom_contour_fill() also does the same.stat_subset() that makes subsetting data on the fly easier.ReadNetCDF() can return a keyed data.table.FitQsWave() and BuildQsWave() renamed to FitWave() and BuildWave(), respectively.GetSMNData() is updated to include the new SMN website and data types.geom_contour2() and geom_contour_fill() gain a circular argument to specify a circular dimension.RepeatCircular() renamed to RepeatCircular().LonLabel() and LatLabel() aid labelling latitude and longitude.EOF() and ImputeEOF().