User’s guide

Miquel De Cáceres, Victor Granda

2018-10-22

Introduction

Purpose

Reliable meteorological data are a basic requirement for hydrological and ecological studies at the landscape scale. Given the large spatial variation of meteorology over complex terrains, meteorological records from a single weather station are often not representative of entire landscapes. Studies made on multiple sites over a landscape require different meteorological series for each site; and other studies may require meteorological data series for all grid cells of a landscape, in a continuous way. In these cases, spatial correlation between the meteorology series of different sites or cells must be taken into account. For example, the sequence of days with rain of contiguous cells will normally be the same or very similar, even if precipitation amounts may differ. Finally, studies addressing the impacts of climate change on forests and landscapes require downscaling coarse-scale predictions of global or regional climate models to the landscape scale. When downscaling predictions for several locations in a landscape, spatial correlation of predictions is also important.

With the aim to assist research of climatic impacts on forests, the R package meteoland provides utilities to estimate daily weather variables at any position over complex terrains:

  1. Spatial interpolation of daily weather records from meteorological stations. }
  2. Statistical correction of meteorological data series (e.g. from climate models).}

Using meteoland package is easy, but some ideas and concepts must be addresed to make the experience easy for new users. This vignette is intended as a working example to explain all the steps needed to get comfortable with meteoland workflow. We assume that the reader is more or less familiar with S4 classes of package sp. You will find detailed information regarding the package structure, its functions, and how calculations are done in vignette “Package meteoland”.

Installing and loading the package

First of all, before starting to work with the package, we must install and load the library. meteoland stable version can be found at CRAN (https://CRAN.R-project.org/package=meteoland), and it can be installed and loaded as any other R package:

install.packages("meteoland")


library(meteoland)

Alternatively, you can install the development version located at GitHub using the devtools package:

library(devtools)


install_github("miquelcaceres/meteoland", build_vignettes = TRUE)


library(meteoland)

Interpolation of daily weather

Spatial interpolation is required when meteorology for the area and period of interest cannot be obtained from local sensors. The nearest weather station may not have data for the period of interest or it may be located too far away to be representative of the target area.

Preparing weather data for interpolation

Before starting using the package, you need to have access to the elevation (in m) and daily weather data series corresponding a set of locations (normally weather stations). Elevation is needed because interpolation routines perform corrections for differences in elevation between the reference locations and the target point. The initial format of your data will be different depending on the format used by your data provider (the package has also tools to access weather data). For our example, we will assume you have data from a set of 38 stations in your study area. On one hand, you should have a data.frame with the coordinates and elevation of each location:

str(st_data)


#> 'data.frame':    38 obs. of  3 variables:


#>  $ X_UTM    : num  347104 358240 382829 350600 377232 ...


#>  $ Y_UTM    : num  4614634 4615380 4645816 4616925 4628439 ...


#>  $ elevation: int  428 555 689 429 632 490 445 687 330 329 ...


head(st_data)


#>     X_UTM   Y_UTM elevation


#> C7 347104 4614634       428


#> C8 358240 4615380       555


#> CA 382829 4645816       689


#> VD 350600 4616925       429


#> VP 377232 4628439       632


#> W5 360730 4659875       490

On the other, you should have at least three matrices of meteorological data (one for minimum temperature, one for maximum temperature and the last one for precipitation) with stations in rows and dates in columns. In our example we also add relative humidity (in percent), so that other derived variables can be calculated:

dim(tmax)


#> [1]   38 1461


dim(tmin)


#> [1]   38 1461


dim(prec)


#> [1]   38 1461


dim(rhum)


#> [1]   38 1461


tmax[1:6,1:6]


#>    2000-01-01 2000-01-02 2000-01-03 2000-01-04 2000-01-05 2000-01-06


#> C7        8.8        8.8       10.7        2.9        1.3       -0.2


#> C8        7.9        8.0        9.9        3.2        3.3       -0.4


#> CA       11.5       13.4       13.5       10.6       14.9       12.7


#> VD        7.4        8.0        9.3        2.8        1.4       -0.2


#> VP        4.8        8.1        8.9        2.3        9.0       -1.4


#> W5         NA         NA         NA         NA         NA         NA

Units should be in degrees Celsius for temperature and mm for precipitation.

Building an interpolation data object

Meteoland stores weather series for reference locations and interpolation parameters in a single object of class MeteorologyInterpolationData. There are several ways of building such objects, but we will first illustrate how to do it from the data we just presented.

First we need to create an object of class SpatialPoints (see package sp) with the spatial coordinates of our stations and the coordinate system (here UTM 31N):

sp = SpatialPoints(st_data[,c("X_UTM", "Y_UTM")],


                   proj4string = CRS("+proj=utm +zone=31 +ellps=WGS84 +datum=WGS84 +units=m +towgs84=0,0,0"))


head(sp)


#> SpatialPoints:


#>     X_UTM   Y_UTM


#> C7 347104 4614634


#> C8 358240 4615380


#> CA 382829 4645816


#> VD 350600 4616925


#> VP 377232 4628439


#> W5 360730 4659875


#> Coordinate Reference System (CRS) arguments: +proj=utm +zone=31


#> +ellps=WGS84 +datum=WGS84 +units=m +towgs84=0,0,0

We can now build an object MeteorologyInterpolationData using:

interpolator <- MeteorologyInterpolationData(sp, elevation = st_data$elevation,


                                             MinTemperature = tmin,


                                             MaxTemperature = tmax,


                                             Precipitation = prec,


                                             RelativeHumidity = rhum)


class(interpolator)


#> [1] "MeteorologyInterpolationData"


#> attr(,"package")


#> [1] "meteoland"

The resulting object is now ready to be used to perform interpolation on a set of target locations (see next section). We can inspect the amount of data in our interpolation object using:

spatial_coverage <- interpolation.coverage(interpolator, type = 'spatial')


head(spatial_coverage)


#>          coordinates MinTemperature MaxTemperature Precipitation


#> C7 (347104, 4614634)           1456           1454          1456


#> C8 (358240, 4615380)           1450           1449          1450


#> CA (382829, 4645816)           1444           1444          1444


#> VD (350600, 4616925)           1461           1461          1461


#> VP (377232, 4628439)           1461           1461          1461


#> W5 (360730, 4659875)           1288           1288          1288


#>    RelativeHumidity Radiation WindSpeed WindDirection


#> C7             1444         0         0             0


#> C8             1405         0         0             0


#> CA             1378         0         0             0


#> VD             1461         0         0             0


#> VP             1455         0         0             0


#> W5             1287         0         0             0


temporal_coverage <- interpolation.coverage(interpolator, type = 'temporal')


head(temporal_coverage)


#>            MinTemperature MaxTemperature Precipitation RelativeHumidity


#> 2000-01-01             13             13            14                6


#> 2000-01-02             13             13            14                6


#> 2000-01-03             13             13            14                6


#> 2000-01-04             13             13            14                6


#> 2000-01-05             13             13            14                6


#> 2000-01-06             13             13            14                6


#>            Radiation WindSpeed WindDirection


#> 2000-01-01         0         0             0


#> 2000-01-02         0         0             0


#> 2000-01-03         0         0             0


#> 2000-01-04         0         0             0


#> 2000-01-05         0         0             0


#> 2000-01-06         0         0             0

In the first call, function interpolation.coverate returns the number of non-missing observations for each station and variable. The second call produces a similar output, but by dates instead of stations.

Interpolation parameters are also stored in the same object (do not worry if you do not know their meaning now):

names(interpolator@params)


#>  [1] "initial_Rp"                "iterations"               


#>  [3] "alpha_MinTemperature"      "alpha_MaxTemperature"     


#>  [5] "alpha_DewTemperature"      "alpha_PrecipitationEvent" 


#>  [7] "alpha_PrecipitationAmount" "alpha_Wind"               


#>  [9] "N_MinTemperature"          "N_MaxTemperature"         


#> [11] "N_DewTemperature"          "N_PrecipitationEvent"     


#> [13] "N_PrecipitationAmount"     "N_Wind"                   


#> [15] "St_Precipitation"          "St_TemperatureRange"      


#> [17] "pop_crit"                  "f_max"                    


#> [19] "wind_height"

A parameter that is important to understand is ìnitial_Rp`, which specifies the initial radius for the truncated spatial Gaussian kernel. By default its value is:

interpolator@params$initial_Rp


#> [1] 140000

The value of initial_Rp must be set in relation to the units of the spatial coordinates of weather data. Our data was in meters, so the default radius is 140 km. In general, the initial radius should be large enough to include a reasonable number of stations (~20-40), but the kernel radius is adjusted for each interpolation target point.

Calibration and cross-validation of the interpolation data

Once we already have weather stations data in shape, we can start calibrating the model in order to obtain the optimal parameters for the meteorological variables we want to interpolate. Parametere calibration has to be done for each variable separately. For example for minimum temperature:

tmin_cal <- interpolation.calibration(interpolator, variable = "Tmin",


                                      N_seq = 20,


                                      alpha_seq = seq(5, 10, by = 1),


                                      verbose = TRUE)


#> Total number of stations: 38


#> Number of stations with available data: 15


#> Number of stations used for MAE: 15


#> Number of parameter combinations to test: 6


#> 


#> Evaluation of parameter combinations...


#>    N: 20 alpha: 5 MAE = 1.21918531810085


#>    N: 20 alpha: 6 MAE = 1.18871267576827


#>    N: 20 alpha: 7 MAE = 1.17521992792415


#>    N: 20 alpha: 8 MAE = 1.16871617692956


#>    N: 20 alpha: 9 MAE = 1.16559427384805


#>    N: 20 alpha: 10 MAE = 1.16431324430726


#> 


#> Minimum MAE value: 1.16431324430726 N: 20 alpha: 10

This function returns an interpolation.calibration class object which contains several items:

The result of the calibration needs to be manually stored in the interpolation params:

interpolator@params$N_MinTemperature = tmin_cal$N


interpolator@params$alpha_MinTemperature = tmin_cal$alpha

In general, we recommend conducting calibration exercises at least once for each variable and each data set used as reference for interpolation, and more than once periods differ in the number of stations available.

Before using the object for interpolations, we also need to assess its performance. This is done by leave-one-out cross-validation in function interpolation.cv:

#> Station #1 C7
#> Station #2 C8
#> Station #3 CA
#> Station #4 VD
#> Station #5 VP
#> Station #6 W5
#> Station #7 WA
#> Station #8 XT: No observations.
#> Station #9 9713
#> Station #10 9713A: No observations.
#> Station #11 9650X: No observations.
#> Station #12 0166: No observations.
#> Station #13 9717
#> Station #14 9717A
#> Station #15 9718C
#> Station #16 9712N: No observations.
#> Station #17 9712O
#> Station #18 9720N: No observations.
#> Station #19 9638E: No observations.
#> Station #20 9638: No observations.
#> Station #21 0130: No observations.
#> Station #22 9640: No observations.
#> Station #23 9649
#> Station #24 9649A: No observations.
#> Station #25 0133: No observations.
#> Station #26 0132: No observations.
#> Station #27 0131U
#> Station #28 9720X
#> Station #29 9720: No observations.
#> Station #30 9720A
#> Station #31 9647X: No observations.
#> Station #32 9718X: No observations.
#> Station #33 0166I: No observations.
#> Station #34 9648: No observations.
#> Station #35 9644: No observations.
#> Station #36 9645: No observations.
#> Station #37 9712: No observations.
#> Station #38 9712E: No observations.

And the results are inspected using:

summary(cv)


#>                                n          r        MAE sd.station.MAE


#> MinTemperature             20618  0.9719910   1.164313      0.4365641


#> MaxTemperature             20616  0.9841445   1.130107      0.4711833


#> TemperatureRange           20612  0.8479033   1.621617      0.5729868


#> RelativeHumidity           10148  0.8775180   6.228231      2.8613220


#> Radiation                      0         NA        NaN             NA


#> Station.rainfall              16  0.8045007 179.693633    150.1331305


#> Station.rainfall.relative     16         NA  11.825926     14.4411686


#> Station.precdays              16 -0.3374503  99.937500     98.1247293


#> Station.precdays.relative     16         NA  29.180756     29.5429033


#> Date.rainfall                689  0.9935306   4.640109             NA


#> Date.rainfall.relative       689         NA  27.099078             NA


#> Date.precstations            689  0.9717731   1.085631             NA


#> Date.precstations.relative   689         NA  23.439052             NA


#>                            sd.dates.MAE         Bias sd.station.Bias


#> MinTemperature                0.4188641   0.18036684       0.8095216


#> MaxTemperature                0.4915072   0.03094508       0.7161885


#> TemperatureRange              0.5510869  -0.14945685       1.1035713


#> RelativeHumidity              2.0755297  -1.05109149       5.5615346


#> Radiation                            NA          NaN              NA


#> Station.rainfall                     NA -21.26181706     237.6977276


#> Station.rainfall.relative            NA   1.64067057      18.8375189


#> Station.precdays                     NA -33.43750000     138.1636560


#> Station.precdays.relative            NA  -1.19840973      42.1845407


#> Date.rainfall                 7.1813374  -0.11319169              NA


#> Date.rainfall.relative       22.7324069 -15.50877004              NA


#> Date.precstations             1.0158252  -0.36574746              NA


#> Date.precstations.relative   23.7813635 -15.01030698              NA


#>                            sd.dates.Bias


#> MinTemperature                 0.1765283


#> MaxTemperature                 0.2530024


#> TemperatureRange               0.3116582


#> RelativeHumidity               2.2965499


#> Radiation                             NA


#> Station.rainfall                      NA


#> Station.rainfall.relative             NA


#> Station.precdays                      NA


#> Station.precdays.relative             NA


#> Date.rainfall                  8.5510632


#> Date.rainfall.relative        31.8012292


#> Date.precstations              1.4416115


#> Date.precstations.relative    29.8346132

Interpolation on a grid

The target for weather interpolation in meteoland can be a set of points, pixels or a whole grid. Again, the initial format of data can be very different. Here we assume you have a small grid of 400 (20x20) cells of 1ha in size, with elevation data in form of class SpatialGridDataFrame (see method read.asciigrid in package sp):

summary(elev)


#> Object of class SpatialGridDataFrame


#> Coordinates:


#>        min     max


#> s1  359950  361950


#> s2 4638950 4640950


#> Is projected: TRUE 


#> proj4string :


#> [+proj=utm +zone=31 +ellps=WGS84 +datum=WGS84 +units=m


#> +towgs84=0,0,0]


#> Grid attributes:


#>    cellcentre.offset cellsize cells.dim


#> s1            360000      100        20


#> s2           4639000      100        20


#> Data attributes:


#>    elevation    


#>  Min.   :418.0  


#>  1st Qu.:484.0  


#>  Median :518.0  


#>  Mean   :511.3  


#>  3rd Qu.:539.0  


#>  Max.   :601.0

Note that the coordinate reference system needs to be the same as that of interpolator, which in this case it is. Before performing the interpolation over this grid, we need to reshape this data in a class called SpatialGridTopography:

sgt = SpatialGridTopography(as(elev, "SpatialGrid"), elevation = elev$elevation,


                            proj4string = elev@proj4string)


sgt


#> Object of class SpatialGridTopography


#> Grid topology:


#>    cellcentre.offset cellsize cells.dim


#> s1            360000      100        20


#> s2           4639000      100        20


#> Coordinate Reference System (CRS) arguments: +proj=utm +zone=31


#> +ellps=WGS84 +datum=WGS84 +units=m +towgs84=0,0,0 


#> 


#> Topography summary:


#>    elevation         slope             aspect        


#>  Min.   :418.0   Min.   : 0.2265   Min.   :  0.9392  


#>  1st Qu.:484.0   1st Qu.: 5.7382   1st Qu.:100.8848  


#>  Median :518.0   Median : 8.8257   Median :156.2479  


#>  Mean   :511.3   Mean   : 9.4178   Mean   :172.7948  


#>  3rd Qu.:539.0   3rd Qu.:13.1438   3rd Qu.:242.3809  


#>  Max.   :601.0   Max.   :23.2798   Max.   :347.0054

As you can see in the result, meteoland has calculated for us slope and aspect (both in degrees) from elevation data 1. Slope and aspect are important for radiation calculations, which also requires relative humidity data. We can display elevation over the grid using:

spplot(sgt, "elevation")

Before we call the interpolation routine, we need to define the dates (i.e. days) for which we want weather to be interpolated, for example:

dates = as.Date(c("2001-02-03", "2001-06-03"))

Of course, we need to be sure that the interpolator object has data corresponding to this dates. We can check if there is any missing date using:

sum(!(dates %in% interpolator@dates))


#> [1] 0

The name of interpolation functions depend on the target spatial structure. For grids we need to use function interpolationgrid:

ml <- interpolationgrid(interpolator, sgt, dates)


#> Date 2001-02-03 (1/2) - done.


#> Date 2001-06-03 (2/2) - done.

This function works processing each date at a time. Since calculations can take some time, the console output shows the progress. The output of the function is an object of class SpatialGridMeteorology:

ml


#> Object of class SpatialGridMeteorology


#> Dates:  2  (initial: 2001-02-03 final: 2001-06-03)


#> Grid topology:


#>    cellcentre.offset cellsize cells.dim


#> s1            360000      100        20


#> s2           4639000      100        20


#> Coordinate Reference System (CRS) arguments: +proj=utm +zone=31


#> +ellps=WGS84 +datum=WGS84 +units=m +towgs84=0,0,0

We can display interpolated grids in a map using function spplot:

spplot(ml, 2, "MinTemperature")


spplot(ml, 2, "MaxTemperature")

Objects of class SpatialGridMeteorology include a list of data frames, one per date. We can access the interpolated data for a given date using:

df_1 = ml@data[[1]]


head(df_1)


#>   MeanTemperature MinTemperature MaxTemperature Precipitation


#> 1        6.772026      -1.311487       12.02764             0


#> 2        6.803369      -1.255197       12.04277             0


#> 3        6.866746      -1.141037       12.07313             0


#> 4        6.850004      -1.171677       12.06542             0


#> 5        6.929330      -1.028796       12.10343             0


#> 6        6.806007      -1.251788       12.04490             0


#>   MeanRelativeHumidity MinRelativeHumidity MaxRelativeHumidity Radiation


#> 1             92.96434            65.25483                 100  10.30587


#> 2             92.76410            65.18977                 100  10.92906


#> 3             92.36071            65.05943                 100  12.03064


#> 4             92.46702            65.09245                 100  11.18735


#> 5             91.96424            64.92961                 100  12.30161


#> 6             92.74708            65.18046                 100  10.91869


#>   WindSpeed WindDirection      PET


#> 1        NA            NA 1.340116


#> 2        NA            NA 1.456203


#> 3        NA            NA 1.662892


#> 4        NA            NA 1.507910


#> 5        NA            NA 1.719131


#> 6        NA            NA 1.454781

Some columns are missing (e.g. wind speed) because we did not include weather station data regarding these variables. If we wants to add grid coordinates to this data frame, we can use the spatial information stored in the ml object:

sgdf_1 = SpatialGridDataFrame(grid = ml@grid, data = ml@data[[1]], proj4string = ml@proj4string)


summary(sgdf_1)


#> Object of class SpatialGridDataFrame


#> Coordinates:


#>        min     max


#> s1  359950  361950


#> s2 4638950 4640950


#> Is projected: TRUE 


#> proj4string :


#> [+proj=utm +zone=31 +ellps=WGS84 +datum=WGS84 +units=m


#> +towgs84=0,0,0]


#> Grid attributes:


#>    cellcentre.offset cellsize cells.dim


#> s1            360000      100        20


#> s2           4639000      100        20


#> Data attributes:


#>  MeanTemperature MinTemperature    MaxTemperature  Precipitation


#>  Min.   :6.175   Min.   :-2.3274   Min.   :11.70   Min.   :0    


#>  1st Qu.:6.504   1st Qu.:-1.7491   1st Qu.:11.86   1st Qu.:0    


#>  Median :6.649   Median :-1.5001   Median :11.95   Median :0    


#>  Mean   :6.628   Mean   :-1.5320   Mean   :11.93   Mean   :0    


#>  3rd Qu.:6.770   3rd Qu.:-1.2829   3rd Qu.:12.00   3rd Qu.:0    


#>  Max.   :7.077   Max.   :-0.7434   Max.   :12.16   Max.   :0    


#>                                                                 


#>  MeanRelativeHumidity MinRelativeHumidity MaxRelativeHumidity


#>  Min.   :91.04        Min.   :64.68       Min.   :100        


#>  1st Qu.:92.98        1st Qu.:65.36       1st Qu.:100        


#>  Median :93.76        Median :65.59       Median :100        


#>  Mean   :93.90        Mean   :65.66       Mean   :100        


#>  3rd Qu.:94.70        3rd Qu.:65.96       3rd Qu.:100        


#>  Max.   :96.87        Max.   :66.70       Max.   :100        


#>                                                              


#>    Radiation        WindSpeed   WindDirection      PET        


#>  Min.   : 4.482   Min.   : NA   Min.   : NA   Min.   :0.2716  


#>  1st Qu.: 9.582   1st Qu.: NA   1st Qu.: NA   1st Qu.:1.1952  


#>  Median :10.966   Median : NA   Median : NA   Median :1.4527  


#>  Mean   :10.962   Mean   :NaN   Mean   :NaN   Mean   :1.4454  


#>  3rd Qu.:12.293   3rd Qu.: NA   3rd Qu.: NA   3rd Qu.:1.6941  


#>  Max.   :16.254   Max.   : NA   Max.   : NA   Max.   :2.3792  


#>                   NA's   :400   NA's   :400

Results can be retrieved in this way and saved using write.asciigrid for its use outside R. The package also provides function for reading/writing NetCDFs (see function writemeteorologygrid) from SpatialGridMeteorology objects.

Interpolation on a set of points

If you want to interpolate on a set of target locations, the starting point will normally be a data.frame. In our example this points come from the grid, but we have reshaped them so the starting format is familiar:

points_df


#>    X_UTM   Y_UTM elevation


#> 1 363500 4640900       524


#> 2 362700 4640700       509


#> 3 362900 4640400       544


#> 4 360300 4638600       425

Analogously with the grid, we need to transform this data into an object SpatialPointsTopography:

spt = SpatialPointsTopography(as.matrix(points_df[,c("X_UTM", "Y_UTM")]),


                              elevation = points_df$elevation,


                              proj4string = CRS("+proj=utm +zone=31 +ellps=WGS84 +datum=WGS84 +units=m +towgs84=0,0,0"))


spt


#> Object of class SpatialPointsTopography


#>         coordinates elevation slope aspect


#> 1 (363500, 4640900)       524    NA     NA


#> 2 (362700, 4640700)       509    NA     NA


#> 3 (362900, 4640400)       544    NA     NA


#> 4 (360300, 4638600)       425    NA     NA

In this case we only have elevation (in m), but slope and aspect should also be included if possible. Let us assume you want to interpolate on this points for the whole time series available in object interpolator. Since we are dealing with points, the function to interpolate is called interpolationpoints:

mp = interpolationpoints(interpolator, spt)


#> Processing point '1' (1/4) - done.


#> Processing point '2' (2/4) - done.


#> Processing point '3' (3/4) - done.


#> Processing point '4' (4/4) - done.

This function works processing one point at a time. The output of the function is an object of class SpatialPointsMeteorology:

mp


#> Object of class SpatialPointsMeteorology


#> Dates:  1461  (initial: 2000-01-01 final: 2003-12-31)


#> SpatialPoints:


#>    X_UTM   Y_UTM


#> 1 363500 4640900


#> 2 362700 4640700


#> 3 362900 4640400


#> 4 360300 4638600


#> Coordinate Reference System (CRS) arguments: +proj=utm +zone=31


#> +ellps=WGS84 +datum=WGS84 +units=m +towgs84=0,0,0

And the time series for a given can be plotted using function meteoplot. For example, we show here the precipitation series of point #1:

meteoplot(mp, 1, "Precipitation", ylab="Precipitation (mm)", xlab="")

Objects of class SpatialPointsDataFrame include a list of data frames, one per point. We can access one of them using:

df_1 = mp@data[[1]]


head(df_1)


#>            DOY MeanTemperature MinTemperature MaxTemperature Precipitation


#> 2000-01-01   1        4.707261     -0.8748933       8.336582             0


#> 2000-01-02   2        5.329918     -1.0196346       9.458175             0


#> 2000-01-03   3        5.571347     -1.6584762      10.271926             0


#> 2000-01-04   4        2.741383     -1.4220586       5.448307             0


#> 2000-01-05   5        2.992349     -1.1386497       5.678180             0


#> 2000-01-06   6        1.742828     -1.9828008       4.165102             0


#>            MeanRelativeHumidity MinRelativeHumidity MaxRelativeHumidity


#> 2000-01-01             92.14027            71.74532                 100


#> 2000-01-02             93.40416            70.41263                 100


#> 2000-01-03             93.82730            68.10807                 100


#> 2000-01-04             93.67873            77.43982                 100


#> 2000-01-05             88.90175            73.62767                 100


#> 2000-01-06             81.30430            68.45843                 100


#>            Radiation WindSpeed WindDirection PET


#> 2000-01-01  1.702528        NA            NA   0


#> 2000-01-02  1.626303        NA            NA   0


#> 2000-01-03  1.567222        NA            NA   0


#> 2000-01-04  1.957039        NA            NA   0


#> 2000-01-05  1.971969        NA            NA   0


#> 2000-01-06  2.073756        NA            NA   0

This data frame can now be written into a file for its analysis outside R. The package also provides its own functions to write/read point meteorology data in different formats. If we are interested in inspecting the interpolation result by date, instead of by point, we can use function extractpointdates, which returns objects of class SpatialGridDataFrame:

dt_4 = extractpointdates(mp, as.Date("2001-01-04"), verbose = FALSE)


dt_4


#>         coordinates DOY MeanTemperature MinTemperature MaxTemperature


#> 1 (363500, 4640900)   4        8.775443       3.971453       11.89883


#> 2 (362700, 4640700)   4        8.823484       3.970469       11.97874


#> 3 (362900, 4640400)   4        8.732016       4.025805       11.79183


#> 4 (360300, 4638600)   4        9.131380       4.062162       12.42721


#>   Precipitation MeanRelativeHumidity MinRelativeHumidity


#> 1      2.110921             84.88668            68.89553


#> 2      2.118662             84.61249            68.53432


#> 3      2.091609             85.13838            69.38540


#> 4      2.179343             82.87795            66.54663


#>   MaxRelativeHumidity Radiation WindSpeed WindDirection PET


#> 1                 100  2.331560        NA            NA   0


#> 2                 100  2.329258        NA            NA   0


#> 3                 100  2.337099        NA            NA   0


#> 4                 100  2.323822        NA            NA   0

Statistical correction of daily weather

Correcting the biases of a meteorological data series containing biases using a more accurate meteorological series is necessary when the more accurate series does not cover the period of interest and the less accurate series does. The less accurate series may be at coarser scale, as with climate model predictions or climate reanalysis data. In this case one can speak of statistical correction and downscaling. However, one may also correct the predictions of climate models using reanalysis data estimated at the same spatial resolution.

In the following example we will correct the predictions of Regional Climate Model (CCLM4-8-17; driving global model CNRM-CERFACS-CNRM-CM5) on the same area of the interpolation example. RCM data includes 3 model cells. Meteorological data covers an historical (reference) period (2000-2003) and a future (projection) period (year 2023), the latter simulated under rcp4.5 scenario.

Preparing weather data to be downscaled/corrected

One needs several data items to perform downscaling and statistical correction. First, we need a matrix or data frame with the central coordinates of the RCM cells (here in longitude/latitude format):

pt_coords


#>       long      lat


#> 1 1.226384 41.79279


#> 2 1.368943 41.81727


#> 3 1.335890 41.92448

Second, we need the uncorrected meteorological data (here RCM outputs) for both a reference period (that will be matched with our more accurate series) and a projection period (that our more accurate series does not cover). Both should be arranged in lists of data frames, one per RCM cell:

length(ref_data)


#> [1] 3


length(proj_data)


#> [1] 3


head(ref_data[[1]])


#>            DOY MeanTemperature MinTemperature MaxTemperature Precipitation


#> 2000-01-01   1        7.008813       5.613855       9.692926  2.273821e-01


#> 2000-01-02   2        7.628625       6.728754       8.123132  1.979953e+01


#> 2000-01-03   3        7.204065       3.538507       9.701288  4.990360e+00


#> 2000-01-04   4        4.317285       1.941400       8.395471  0.000000e+00


#> 2000-01-05   5        4.235284       1.565973       8.663568  1.820105e-09


#> 2000-01-06   6        4.464075       2.094171       7.982202  2.747191e-03


#>            SpecificHumidity MeanRelativeHumidity Radiation WindSpeed


#> 2000-01-01      0.005365001             87.24593  6.479796  5.300286


#> 2000-01-02      0.006007533             93.63294  2.333908  4.713901


#> 2000-01-03      0.005704192             91.52707  4.233040  2.108916


#> 2000-01-04      0.004497202             88.16652  6.957025  1.168019


#> 2000-01-05      0.004520838             89.14193  6.588341  1.993287


#> 2000-01-06      0.004539691             88.08730  6.849565  1.955944


head(proj_data[[1]])


#>            DOY MeanTemperature MinTemperature MaxTemperature Precipitation


#> 2023-01-01   1      0.52730713     -0.9837708       1.922754  5.914489e-10


#> 2023-01-02   2      2.08501587     -1.0495361       6.185968  2.419252e-01


#> 2023-01-03   3     -0.76285400     -3.2190002       2.428308  1.936977e-12


#> 2023-01-04   4      0.02242432     -1.7950195       1.537469  1.174330e-01


#> 2023-01-05   5      1.30117188     -1.0007385       3.657709  3.518532e+00


#> 2023-01-06   6     -1.27564087     -3.2251038       1.087396  1.464739e-09


#>            SpecificHumidity MeanRelativeHumidity Radiation WindSpeed


#> 2023-01-01      0.003752435             96.40739  6.263537  1.353957


#> 2023-01-02      0.003748596             86.08917  5.417012  2.057226


#> 2023-01-03      0.003032080             85.58024  7.274775  1.688474


#> 2023-01-04      0.003270915             87.17539  3.522371  1.115798


#> 2023-01-05      0.003813237             92.64220  5.621084  1.008287


#> 2023-01-06      0.003074167             90.09795  7.251414  1.606562

Note that these data frames should follow the conventions of meteoland for variable names and units. Finally, we need to specify the projection period, in this case one year:

proj_dates = seq(as.Date("2023-01-01"), as.Date("2023-12-31"), by="day")

Building the uncorrected data object

Statistical correction needs an object of class MeteorologyUncorrectedData, analogous to the MeteorologyInterpolationData of the previous section. To build this object we need first to express coordinates in an object SpatialPoints:

sp = SpatialPoints(pt_coords, CRS("+proj=longlat +datum=WGS84 +ellps=WGS84 +towgs84=0,0,0"))


sp


#> SpatialPoints:


#>          long      lat


#> [1,] 1.226384 41.79279


#> [2,] 1.368943 41.81727


#> [3,] 1.335890 41.92448


#> Coordinate Reference System (CRS) arguments: +proj=longlat


#> +datum=WGS84 +ellps=WGS84 +towgs84=0,0,0

Assuming the weather data to be corrected is in the proper format, the object is constructed using:

uncorrected = MeteorologyUncorrectedData(sp, ref_data, proj_data, proj_dates)

Target point weather meteorology taken as reference

Downscaling/statistical correction is done on a set of target spatial points. meteoland will first find the RCM cell to which each point is nearest (using the central coordinates of RCM cells). Once this matching is done, statistical relations are build between the meteorology of RCM cells and that of target points for the reference period (here 2000-2003) which are used to correct RCM data for the projection period (year 2023). For this process to be done, we need the coordinates and weather series for the reference period. In our case, we will employ a copy the object of class SpatialPointsMeteorology that resulted from interpolation in the previous section:

historical  = mp


historical


#> Object of class SpatialPointsMeteorology


#> Dates:  1461  (initial: 2000-01-01 final: 2003-12-31)


#> SpatialPoints:


#>    X_UTM   Y_UTM


#> 1 363500 4640900


#> 2 362700 4640700


#> 3 362900 4640400


#> 4 360300 4638600


#> Coordinate Reference System (CRS) arguments: +proj=utm +zone=31


#> +ellps=WGS84 +datum=WGS84 +units=m +towgs84=0,0,0

We named it historical to remind that we are using the historical period as reference and want to correct the RCM series for the projection period There are several ways to shape data into objects of class SpatialPointsMeteorology. For example, one can use function readmeteorologypoints.

Tuning correction parameters

Objects of class MeteorologyUncorrectedData have a slot with parameters:

uncorrected@params


#> $varmethods


#> $varmethods$MeanTemperature


#> [1] "unbias"


#> 


#> $varmethods$MinTemperature


#> [1] "quantmap"


#> 


#> $varmethods$MaxTemperature


#> [1] "quantmap"


#> 


#> $varmethods$Precipitation


#> [1] "quantmap"


#> 


#> $varmethods$MeanRelativeHumidity


#> [1] "unbias"


#> 


#> $varmethods$Radiation


#> [1] "unbias"


#> 


#> $varmethods$WindSpeed


#> [1] "quantmap"


#> 


#> 


#> $qstep


#> [1] 0.01


#> 


#> $fill_wind


#> [1] TRUE


#> 


#> $allow_saturated


#> [1] FALSE


#> 


#> $wind_height


#> [1] 10

Importantly, varmethods is a named list that specifies which correction method has to be used for each variable. Since we do not have reference data for wind speed, we must turn the method for wind speed to "none":

uncorrected@params$varmethods$WindSpeed="none"

Conducting statistical correction

Once we have all the data objects that we need (the hard part), conducting statistical correction is straightforward:

projected = correctionpoints(uncorrected, historical)


#> Points to correct: 4


#> All points inside boundary box.


#> Correcting point '1' (1/4) - ipred = 3 done.


#> Correcting point '2' (2/4) - ipred = 3 done.


#> Correcting point '3' (3/4) - ipred = 3 done.


#> Correcting point '4' (4/4) - ipred = 3 done.

As mentioned above, correction proceeds point by point. The result is again an object of class SpatialPointsMeteorology:

projected


#> Object of class SpatialPointsMeteorology


#> Dates:  365  (initial: 2023-01-01 final: 2023-12-31)


#> SpatialPoints:


#>    X_UTM   Y_UTM


#> 1 363500 4640900


#> 2 362700 4640700


#> 3 362900 4640400


#> 4 360300 4638600


#> Coordinate Reference System (CRS) arguments: +proj=utm +zone=31


#> +ellps=WGS84 +datum=WGS84 +units=m +towgs84=0,0,0

The following code displays the minimum/maximum temperatures before and after correction:

#Plot predicted mean temperature for point 1


meteoplot(projected, 1, "MinTemperature", ylab="Temperature (Celsius)", ylim=c(-5,40), col="blue")


meteoplot(projected, 1, "MaxTemperature", add=TRUE, col="red")


#Add uncorrected mean temperature data (cell #3)


lines(uncorrected@dates,


      uncorrected@projection_data[[3]]$MinTemperature,


      col="blue", lty=3)


lines(uncorrected@dates,


      uncorrected@projection_data[[3]]$MaxTemperature,


      col="red", lty=3)


legend("topright", legend=c("corrected","uncorrected", "Maximum", "Minimum"), 


       col=c("black","black", "red","blue"), lty=c(1,3,1,1), bty="n")

  1. Objects of class SpatialGridTopography can be initialized with user input values for slope and aspect too, but meteoland has its own routines when this are missing↩