Insurance Risk Indices

Insurance Risk Indices are an ensemble of indices relevant for insurance industry. These indices are based on Expert Team on Climate Change Detection Indices (ETCCDI). There are currently 5 available indices to compute for extreme: heat (tx90p), cold (tn10p), wind (wx), drought (ccd) and flooding (rx5day). The individual indices can be combined into a single index with or without weighting for each component. This combined index is roughly analogous to the Actuaris Climate Risk Index.

Note: This vignette has been writen to process and plot the output of an ESMValTool namelist configured with a specific set of parameters. If you wish to run and plot the outputs for another set of parameters, you need to re-run the corresponding namelist with the desired parameters, interpolate the output NetCDF files to a common grid, and modify the vignette code accordingly.

The following exemple will compute the 5 different indices in the North Atlantic - European Sector [60ºW-40ºE; 30ºN-70ºN] using the CMIP5 projection for the 8.5 scenario.

1- Load dependencies

This example requires the following system libraries:

  • ŀibssl-dev
  • libnecdf-dev
  • cdo

You will need to install specific versions of certain R packages as follows:

library(devtools)
install_git('https://earth.bsc.es/gitlab/es/startR', branch = 'develop-hotfixes-0.0.2')

Too functions should be load by running the following lines in R, until integrated into current version of s2dverification

source('https://earth.bsc.es/gitlab/es/s2dverification/raw/develop-Climdex/R/Climdex.R')
source('https://earth.bsc.es/gitlab/es/s2dverification/raw/develop-Climdex/R/Threshold.R')
source('/home/Earth/nperez/git/s2dverification/vignettes/risk_index/DailyAnomaly.R')
source('https://earth.bsc.es/gitlab/es/s2dverification/raw/develop-Magic_WP6/R/WeightedMean.R')

All the other R packages involved can be installed directly from CRAN and loaded as follows:

library(s2dverification)
library(startR)
library(multiApply)
library(ggplot2)
library(climdex.pcic)
library(parallel)

2- Computing the Extreme Heat Index

The Extreme Temperature Index (tx90p) is defined as the percentage of days when maximum temperature exceed the 90th percentile.

2.1- Extreme Heat Index Climatology

In order to evaluate the future projections, it is necessary to compute the index climatology during a reference historical period. In this example, the period 1971⨪2000 has been selected. The next steps should be followed:

  • First is loaded the historical period with the function Start(). The full description and examples of its use can be found here: https://earth.bsc.es/gitlab/es/startR. The object returned by Start will have at least one time and two spatial dimensions.
var0 <- 'tasmax'
start_climatology <- '1971-01-01'
end_climatology <- '2000-12-31'
lat.min <- 30;  lat.max <- 70
lon.min <- -60; lon.max <- 40
climatology_filenames <- paste0("/esarchive/scratch/pbretonn/cmip5-cp4cds/historical/day/tasmax/",
                                "tasmax_day_IPSL-CM5A-LR_historical_r1i1p1_19500101-20051231.nc")

reference_data <- Start(model=climatology_filenames, 
                        var = var0, var_var = 'var_names',
                        time = values(list(as.POSIXct(start_climatology), 
                                           as.POSIXct(end_climatology))),
                        lat = values(list(lat.min, lat.max)),
                        lon = values(list(lon.min, lon.max)),
                        lon_var = 'lon',
                        lon_reorder = CircularSort(0, 360),
                        return_vars = list(time = 'model', lon = 'model', lat = 'model'),
                        retrieve = TRUE)
metadata <- attributes(reference_data)
> dim(reference_data)
model   var  time   lat   lon 
    1     1 10957    21    28
  • To remove seasonality effects, the anomaly is computed with the next code lines. First, the day of the year, from 1 to 365, is identify for the 30 years period. Then, the anomaly for each day and gridpoint is computed by applying the dailyAno function. Finally, the names of the dimensions are restablished.
dates <- attributes(ret = values(list(lat.min, lat.max)),
                        lon = values(list(lon.min, lon.max)),
ference_data)$Variables$dat1$time
time_dim <- which(names(dim(data)) == "time")

dates <- as.PCICt(dates, cal = attributes(dates)$variables$time$calendar, format = "%Y%m%d")
jdays <- as.numeric(strftime(dates, format = "%j"))

anomaly_data <- Apply(list(reference_data), target_dims = which(names(dim(reference_data)) == "time"), AtomicFun = "dailyAno", jdays = jdays, ncores = 4)$output1
names(dim(anomaly_data))[1] <- "time"
  • This data can be detrended by applying Trend function. In order to remove the trend from the reference_data, the correction is calculated by substracting the detrended_data to the anomaly_data. Again, the original dimensional information is restored.
detrended_data <- Trend(anomaly_data, posTR=which(names(dim(anomaly_data)) == "time"))
diff <- anomaly_data - detrended_data$detrended
diff <- aperm(diff, c(2,3,1,4,5))
detrended_data <- reference_data - diff
attributes(detrended_data) <- metadata
  • For each gridpoint and day of the year (from the 1st of January to the 31st of December), the maximum temperature on the position of the 90 percent of the series will be calculated as the threshold.
quantile <- 0.9
base_range <- as.numeric(c(substr(start_climatology, 1, 4), substr(end_climatology, 1, 4)))
thresholds <- Threshold(detrended_data, base_range=as.numeric(base_range),
                            qtiles = quantile, ncores = detectCores() -1)
> dim(thresholds)
jdays model   var   lat   lon 
  365     1     1    21    28
  • By indicating the metric and introducing the threshold, Climdex() function will return the extreme heat index during the reference periode.
base_index <- Climdex(data = detrended_data, metric = 'tx90p',
                          quantiles = thresholds, ncores = detectCores() - 1)

The output of ´Climdex´ function will be a ´list()´ object. In the ´base_index$result´ label will be save the index values.

> str(base_index)
List of 2
 $ result: num [1:30, 1, 1, 1:21, 1:28] 8.22 12.33 4.38 8.49 9.59 ...
 $ years : chr [1:30] "1971" "1972" "1973" "1974" ...
> dim(base_index$result)
 year model   var   lat   lon 
   30     1     1    21    28
  • Finally, the index will standarized, by substracting the mean and dividing by the standard deviation, in order to facilitate the comparison with the future projection. Notice that, by definition, the mean of the percentage of the number of days exceeding the 90th percentile is 10. Only standard deviation is computed.
base_sd <- Apply(list(base_index$result), target_dims = list(c(1)), AtomicFun = "sd")$output1

2.2 Extreme Heat Index Climate Projection

The Extreme Heat Index is computing for the period 2020⨪2040 by following the next steps:

  • Loading data by using Start() function.
start_model <- '2020-01-01'
end_model <- '2040-12-31'
model_filenames <- paste0("/esarchive/scratch/pbretonn/cmip5-cp4cds/rcp85/day/tasmax/",
                          "tasmax_day_IPSL-CM5A-LR_rcp85_r1i1p1_20060101-22051231.nc")

rcp_data <- Start(model = model_filenames,
                  var = var0, var_var = 'var_names',
                  time = values(list(as.POSIXct(start_model), as.POSIXct(end_model))),
                  lat = values(list(lat.min, lat.max)),
                  lon = values(list(lon.min, lon.max)),
                  lon_var = 'lon', 
                  lon_reorder = CircularSort(0, 360),
                  return_vars = list(time = 'model', lon = 'model', lat = 'model'),
                  retrieve = TRUE)

The object returned by Start() contains, among others, four arrays with the requested daily maximum temperature data for each gridpoint. summary() function allows to check their content.

> dim(rcp_data)
model   var  time   lat   lon 
    1     1  7671    21    28 
> summary(rcp_data)
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  221.6   276.7   283.6   283.4   291.2   321.1
  • Now the index can be computed by considering the threshold obtain for the reference period.
projection_index <- Climdex(data = rcp_data, metric = 'tx90p',
                                  quantiles = thresholds, ncores = detectCores() - 1)

  • Its normalized with mean 10 and the standard deviation of the reference period.

    projection_mean <- 10
base_sd_proj <- InsertDim(base_sd, 1, dim(projection_index$result)[1])
projection_index_standardized <- (projection_index$result - projection_mean) / base_sd_proj

2.3 Visualitzation of Extreme Heat Index

A Map of the mean index values during the examined period 2020-2040 is obtained and save in PNG format in the working directory with the name: “tx90p_IPSL-CMIP5_85rpc_2020-01-01_2040-12-31.png”

lat <- attr(rcp_data, "Variables")$dat1$lat
lon <- attr(rcp_data, "Variables")$dat1$lon
title <- paste0("Index for tx90p ", substr(start_model, 1, 4), "⨪",
                  substr(end_model, 1, 4), " 85rcp CMIP5-IPSL")
breaks <- -8 : 8
PlotEquiMap(Mean1Dim(projection_index_standardized, which(names(dim(projection_index_standardized)) ==
            "year")), lon = lon, lat = lat, filled.continents = FALSE, toptitle = title, brks=breaks, 
            fileout = paste0('tx90p_IPSL-CMIP5_85rpc_', start_model, "_", end_model, ".png"))

The inland average of the Extreme Heat Index can be computed to plot its time evolution.

To visualize the time evolution of models agreement, the spatial average is performed by a grid pixel size weithed mean. Smoothing() returns the smoothed time series for a 5 years moving window.

masc <- Start(model = '/esarchive/exp/ipsl-cm5a-lr/cmip5-rcp85_i1p1/constant/r0i0p0/fx/sftlf/sftlf.nc', 
    var = 'sftlf', var_var = 'var_names', lat = values(list(lat.min, lat.max)),
        lon = values(list(lon.min, lon.max)), lon_reorder = CircularSort(0,360), 
    return_vars = list(lon = 'model', lat = 'model'),retrieve = TRUE)

temporal <- WeightedMean(projection_index_standardized, lon = lon, lat = lat, mask = drop(masc))
temporal_3ysmooth <- Smoothing(temporal, runmeanlen = 3, numdimt=1)

The next code should be runned to plot and save the original average and the 3 year smoothed data.

png("Temporal_Inland_ExtremeHeatIndex_IPSL-CMIP5_85rcp_2020-2040.png", width = 8, height = 5, units= 'in', res = 100, type="cairo")
plot(substr(start_model,1,4):substr(end_model,1,4), temporal, type="l", lty=5, lwd=2, bty='n', ylab="Time (years)", xlab="Extreme Heat Inde", main= "Inland average Extreme Heat Index")
lines(substr(start_model,1,4):substr(end_model,1,4), temporal_3ysmooth, col="darkgreen",lwd=2)
legend('bottomright', c('Anual', '3 years smooth'), col= c(1, 'darkgreen'), lty=c(5,1), lwd=2, bty='n')
dev.off()