Weather Regimes

Weather regimes can be calculated using the k-means clustering algorithm, typically applied to the PCs of anomalies from selected variables, after smoothing and detrending.

The work flow will be something like:

• Load data: Load().
• Calculate the anomalies: Ano().
• Smooth the anomalies: Smooth().
• Detrend the smoothed anomalies: Trend().
• Calculate the PCs, the kmeans clusters and reconstruct the field as Cluster center x EOFs: WeatherRegime().
• Visualize the results: PlotEquiMap().

The loading, detrending and calculation of anomalies needs to be done before applying the WeatherRegime() function. The calculation of the EOFs/PC (including the latitudinal area weighting) and the reconstruction of the field (Cluster centers * EOFs) is done within the WeatherRegime() function.

North Atlantic Weather Regimes

Below is a simple example of calculating the weather regimes for the North Atlantic from JRA55 SLP for January (with detrending disabled to reduce the computation time).

```r
library(s2dverification)
library(reshape2)
JRA55 <- '/esnas/recon/jma/jra55/$STORE_FREQ$_mean/$VAR_NAME$_f6h/$VAR_NAME$_$YEAR$$MONTH$.nc'
psl <- "psl"
year.start <- 1981 
year.end <- 2016
#Euro-Atlantic region
lat.max <- 81 
lat.min <- 27 
lon.max <- 45 
lon.min <- 274.5
firstmonth <- 1                              
leadmonth <- 0      
psl <-  Load(var = psl, exp = NULL, obs = list(list(path=JRA55)), paste0(year.start:year.end,'0101'), 
             storefreq = 'daily', leadtimemax = 366, output = 'lonlat', latmin = lat.min,       
             latmax = lat.max, lonmin = lon.min, lonmax = lon.max, nprocs=8)
lon <- psl$lon
lat <- psl$lat
psl <- psl$obs
#psl <- Trend(psl, posTR = 3)$detrended 
psl <- psl[1, 1, , , , ]  # Drop the dataset and members dimensions
psl <- Subset(psl, 2, 1 : 31) # Select the month of January
psl <- psl / 100 #(Convert to hPa)
psl_ano <- psl - InsertDim(Mean1Dim(psl, 1), 1, 36)
psl_ano_smoothed <- Smoothing(psl_ano, 7, 2)
psl_ano_smoothed <- psl_ano_smoothed[, 4 : 28, , ]
names(dim(psl_ano_smoothed)) <- c("sdate", "ftime", "lat", "lon")
WR <- WeatherRegime(psl_ano_smoothed, lon = lon, lat = lat, ncenters = 4)
PlotLayout(fun = c('PlotEquiMap', 'PlotEquiMap', 'PlotEquiMap', 'PlotEquiMap'),
                 plot_dims = c('lat', 'lon'), 
                 var = list(WR$field[1, , ], WR$field[2, , ],
                 WR$field[3, , ], WR$field[4, , ]), 
                 lon = lon, lat = lat, 
                 titles = paste0("Frequency of occurence ", round(unlist(WR$frequency)[1 : 4], digits = 1), "%"), 
                 drawleg = 'E', units = 'hPa', units_scale = 2, 
                 width = 12, height = 10, res = 200,      fileout = 'weather_regimes.png')
            
```

Hierarchical clustering

Hierarchical clustering methods can also be used through the method option. For example the hierarchical clustering method with mean linkage can be applied via method = average on the same dataset provided above:

r WR <- WeatherRegime(psl_ano, lon = lon, lat = lat, ncenters = 4, method = "average") PlotEquiMap(WR$field$composite[,,1], lon = lon, lat = lat, dots = WR$field$pvalue[,,4 ] < 0.05) #Plot the 1st regime