In leonawicz/snapclim: API to SNAP Climate Data Collections

knitr::opts_chunk$set(
  collapse = TRUE, comment = "#>", message = FALSE, warning = FALSE, error = FALSE, tidy = TRUE
)

Available climate data

To begin using snapclim effectively, take a look at the climate data collections available in your current version of the package.

library(snapclim)
climate_collections()

This prints a table of available data sets, one per row. The columns provide useful metadata regarding each data set.

Available locations

For most data sets, a request returns data for a specific location. Given that there are 86 defined climate regions and 3,867 point locations across Alaska and western Canada in the ar5stats collection, it is helpful to consult a table of available locations.

climate_locations()

The table is truncated here, but contains nearly 4,000 options under the Location column. There is a corresponding column providing the set that each location is grouped under. Every location belongs to a group. The table above contains all regions and point locations. climate_locations can be used to list a subset of only one or the other.

climate_locations(type = "region")
climate_locations(type = "point")

Some locations share the same name. For example, there is Galena, Alaska and Galena, British Columbia.

library(dplyr)
climate_locations() %>% filter(Location == "Galena")

It is good practice to avoid ambiguity when requesting data, though it is permitted since familiarity with the available locations you are interested in can make your data requests simpler.

Data requests

snapclim provides access to a large amount of SNAP climate data, far more than would be stored locally within an R package. The data collections are stored on Amazon Web Services (AWS). snapclim interfaces with AWS to bring the specific data you need into your R session, as if it were a native package data set.

SNAP climate data sets are accessed with climdata. If at any time you get stuck with using climdata, see the function documentation. It provides detailed descriptions and usage for the available function arguments. The first argument, id, specifies a unique data collection. See the id column in climate_collections above. Next, a location is specified. A simple call to climdata for SNAP 2-km downscaled AR5/CMIP5 climate data summary statistics for Anchorage, Alaska looks like the following.

climdata("ar5stats", "Anchorage")

In subsequent examples, arguments are named for additional clarity.

The data includes SNAP's downscaled historical, observation-based Climatological Research Unit (CRU) 4.0 data and both downscaled historical and projected climate model outputs for all five of the General Circulation Models (GCMs) utilized by SNAP. All three CMIP5 emissions scenarios, or Representative Concentration Pathways (RCPs) are included. The data cover the entire available time period at a monthly time step.

By default, all available climate variables are returned: precipitation and mean, minimum and maximum temperature. These refer to monthly precipitation totals and monthly means of mean, minimum and maximum daily temperatures. This can be reduced to a specific variable in the initial call to climdata with variable = "pr" for example, or the table can be filtered subsequently.

What if the location is not unique, like Galena? climdata will throw a warning and let you know it is assuming the first group found in the list of available locations.

climdata(id = "ar5stats", area = "Galena")

The following example avoids the ambiguity, hence no warning.

climdata(id = "ar5stats", area = "Galena", set = "British Columbia")

Regional statistics

The climate variable values given in the tables obtained so far have all pertained to specific points in space. For regional climate data, a broad set of statistics is available that summarizes the distribution of climate values over a spatial domain defined by a polygon. For example, the table for the Arctic Tundra contains the mean, standard deviation, minimum, maximum and a set of distribution quantiles. Since the columns are truncated when printed in the display below, the first seven columns of ID variables are dropped in this example using select in order to show more of the additional statistics.

x <- climdata(id = "ar5stats", area = "Arctic Tundra")
select(x, -c(1:7))

Note that these statistics summarize values across space. They do not also summarize values over months or years, or across climate models and scenarios. Distributional information is available for each point in time and under each combination of other available factors.

Seasonal and annual data

More highly aggregated data sets are available as well. The previous data sets were returned using the default argument time_scale = "monthly". If you simply change this to seasonal or annual, climdata will return the respective data set. The first few columns have been dropped:

x <- climdata(id = "ar5stats", area = "Arctic Tundra", time_scale = "seasonal")
select(x, -c(1:3))

It is important to note that seasonal and annual aggregate statistics are not simple means of monthly statistics. Each of these collections is independently derived from climate variable spatial probability distributions at their respective temporal resolutions.

This means that, for example, monthly temperature quantiles for the Arctic Tundra are calculated from monthly spatial temperature distributions and winter temperature quantiles are calculated from the applicable 3-month period spatial temperature distributions. While the mean is invariant to this difference, other statistics are not. The 95th percentile winter temperature across space during the three month period does not result from taking the average of three monthly 95th percentile values.

A final note on season and annual statistics is that, like monthly statistics, these remain period totals for precipitation and period averages for temperature variables.

Decadal data

In contrast to monthly, seasonal and annual resolution statistics, all three of which are computed across space at their respective temporal resolutions, decadal statistics are in fact simple decadal averages of monthly, seasonal and annual data. For example, the decadal mean of the 95th percentile monthly temperature across a region is just that; the mean of the ten annual 95th percentile monthly values in a decade.

For this reason, decadal data is not requested with climdata by specifying it with time_scale, which always pertains to annual and intra-annual (monthly or seasonal) time steps. Instead, use decavg = TRUE. This is FALSE by default so it did not previously need to be specified. When requesting decadal averages, there is still the choice of whether those averages should be of monthly, seasonal or annual resolution statistics.

x <- climdata(id = "ar5stats", area = "Arctic Tundra", time_scale = "seasonal", decavg = TRUE)
select(x, -c(1:3))

Mulitple locations

It is possible to obtain climate data sets that include multiple locations using climdata, but this is only available for smaller data sets where it would not lead to a cumbersome data download. Currently, the only data set for which multiple locations can be returned at once is the decadal averages data set in the ar5stats collection. By specifying area = "points" rather than a specific point location, a table is returned containing data for all 3,867 point locations.

There are two requirements that help to ensure this does not lead to an excessive download size or waiting time. As mentioned, this is only available for the highly aggregated decadal data. Without setting decavg = TRUE, attempting to specify area = "points" will throw an error. The second requirement is that only a single climate variable will be returned. You can always call climdata multiple times for additional variables if desired. Therefore, you should also specify the variable argument. If you do not, mean temperature (tas) is assumed.

x <- climdata(id = "ar5stats", area = "points", time_scale = "annual", decavg = TRUE, variable = "tas")

To show the result more effectively, filter the table to a specific combination of other factors.

filter(x, RCP == "6.0" & Model == "GFDL-CM3" & Decade == 2050)

Analyzing SNAP climate data

The snapclim package is essentially a data package. It provides a simplified and convenient interface in R enabling easy access to a large amount of SNAP climate data spread over multiple collections, stemming from different sources and existing for different purposes. It does not provide functionality for performing statistical analysis and graphing, which is provided by R in general. Useful stock functions pertaining specifically to analyzing SNAP data, including climate data accessed with snapclim, are available in the snapstat package (under development).

Climate distributions

For more information on the climate probability distributions from which regional climate statistics are calculated, see the snapdist package (under development) or SNAP's Climate Analytics Shiny app for working examples.

leonawicz/snapclim documentation built on May 15, 2019, 5:02 a.m.