Variable names, units and timestamps
Streamflow is runoff routed along a drainage network, in m3/s, also known as discharge, which is the variable name used in the files. Water temperature is given in units of Kelvin. Filenames include the variable name, GCM, scenario (hist for historical, or one of the RCPs) and the time period (years). The timestamps in the files reflect the last date of the period over which the output was averaged, so the first timestamp of the weekly averages is January 7th 1976.
Ecologically-relevant variables
The ecologically-relevant streamflow and water temperature variables derived from the weekly values are established based on a combination of classification frameworks, i.e., indicators of hydrologic alteration19, terrestrial bioclimatic variables in the worldclim dataset20 as well as the CMCC-BioClimInd dataset21, aggregated accordingly: 1976–2005 (1979–2005 for E2O); 2021–2040; 2041–2060; 2061–2080; 2081–2099. The scripts used to compute these derived variables can be found under Code Availability.
For files containing information on timing (see Tables 2–3), note that the counting is 0-indexed. So week numbers run from 0 through 51, months from 0 to 11. For timing of quarters, 0 is DJF, 1 is MAM, 2 is JJA, 3 is SON. The week number (for WT-wmin, WT-wmax, Q-wmin, Q-wmax) is determined as the mode, i.e. the most frequent week number within a period. For each period (20, 25 or 30 years) we looked for the week number in which the minimum or maximum water temperature or discharge occurs. If that happens most often in week X, that week number is stored. It can however occur that a certain minimum/maximum temperature or discharge occurs equally often in multiple weeks – then we assign a missing value.
The variables Q-bfi and Q-vi are calculated according to Pastor et al.30. The baseflow index is an indicator of the importance of stored sources; a high index indicates that flow is mostly sustained by stored sources such as groundwater.
Scripts used to create the derived variables are available through the FutureStreams GitHub repository (see Code Availability below).
Multi-model set-up
We provide future scenarios for four RCPs (representative concentration pathways; 2.6, 4.5, 6.0 and 8.5 W/m2 in 2100) for the five ISI-MIP GCMs. Projections differ across RCPs due to differences in greenhouse gas forcing, and across GCMs due to differences in e.g model parameterization and resolution. Generally the spread across GCMs is larger than that across RCPs7,31. When interested in the general effect of climate change, users are advised to use the mean or median across the GCMs, rather than selecting a specific GCM. When interested in the spread across GCMs, users can explore or represent that in various ways, such as color intensity indicating agreement amongst models5, bar or violin plots7 etc.
Warming levels
To facilitate assessments and comparisons of streamflow and water temperature at a certain air temperature rise rather than specific years5,7, we provide a table with the years in which each GCM/RCP reaches the global mean temperature rises 1.5°, 2.0°, 3.2°, 4.5° compared to pre-industrial temperatures (as used by Barbarossa et al.7) with our scripts (see Code Availability). These years represent the central value of a 30-year running mean, so users should evaluate the 30-year mean (or other statistic) of discharge or water temperature centered around the year that a certain warming level is reached, which is specific to each RCP and GCM combination. For instance, if 1.5° warming is reached in 2040, the 30-year period 2025–2054 should be considered.
GCMs, bias-correction and reanalysis data
The majority of our simulations are forced with meteorological time series from GCMs. Those are bias-corrected27 before being applied to impact models such as PCR-GLOBWB, which corrects for systematic deviations of the simulated historical data from observations. For instance, for temperature the offset in average temperature in the historical GCM simulation with respect to observations is subtracted from temperatures in all scenarios of that GCM. The bias-corrected GCM forcing should thus well represent climatology, but not necessarily timing of actual events such as floods and droughts. Reanalysis data is created by assimilating observations into weather models, to obtain consistent and globally complete time series. The output of the simulation forced with meteorological time series from the (E2O) reanalysis data should therefore reflect not only the average streamflow and water temperatures, but also timing of actual events such as droughts.
If users want to check for themselves how the GCM-forced historical simulations discussed here deviate from reanalysis-forced simulations, they can use the output from the E2O-forced simulation provided here, the monthly output linked to Wanders et al.13 (see also Code Availability) or the daily output of those simulations which are available from Niko Wanders upon request. The latter are forced with ERA-40/ERA-Interim reanalysis data.
Notes of caution
Beware of temperature in grid cells where streamflow is low, which can cause temperatures to become unrealistically high due to strong fluctuations in the water level. The computational timesteps currently implemented in DynWat are not sufficiently small to provide stable solutions for these conditions. For some lakes and reservoirs we observe a similar problem when lakes expand or shrink as a result of water levels changes. These locations can be masked and we can assume that water temperature follows the air temperature for these very shallow water layers. A file with locations of lakes and reservoirs is provided in the data repository (under indicators/mask) so users can mask these if desired.
Furthermore, we provide masks for each GCM-RCP-period which users can apply to the derived variables if desired. These masks are based on Q-mean and WT-mean and thresholds of 10 m3/s and 350 K, respectively. They can be found in the data repository (i.e. indicators/waterTemperature/WT-mask). The scripts used to create these masks are provided through the FutureStreams GitHub repository (see Code Availability below), which can be used to create masks with different thresholds. These scripts are called mask_unrealistic_values.py and maskFunctions.py.
We also provide scripts to mask out unrealistic values directly in the weekly Q and WT files, these scripts are mask_unrealistic_values_weekly.py and maskFunctions_weekly.py. In all these scripts the threshold for discharge is set to 10 m3/s and for water temperature to 350 K, but users can change those to their preferred values. The threshold value will be included in the resulting output file name.
Furthermore, we encountered spin-up issues in some pixels for the future RCP simulations. Instead of following the temperatures from the end of the historical simulation, temperatures drop at the beginning of the future simulation, so the first few weeks of 2006 temperatures can be unrealistically low. In Fig. 2, output of the year 2007 is used for the year 2006 .
Water temperature [°C] anomaly. The maps show the difference between the mean water temperature over the period 2070–2099 (RCP8p5) and the historical period 1975–2005. The map shows values only for rivers with streamflow greater than 50 m3/s and the width of the rivers is scaled based on the streamflow values for clarity of representation. Insets below the map show the original gridded resolution at 5 arcminute for cells with streamflow values greater than 10 m3/s. The bottom insets show water temperature time series sampled at specific grid-cell locations (white crosses in the insets) for the Amazon (−57.2083° longitude, −2.625° latitude), Danube (20.125° lon, 45.2083° lat) and Ganges (88.375° lon, 24.375° lat). Time series are represented for each GCM and RCP available within FutureStreams; thin lines represent weekly values, thick lines represent 10 year rolling means.
Streamflow [m3/s] anomaly. The maps show the difference between the log10 transformed mean streamflow over the period 2070–2099 (RCP8p5) and the log10 transformed mean streamflow over historical period 1975–2005. The map shows values only for rivers with streamflow values greater than 50 m3/s and the width of the rivers is scaled based on the streamflow values for clarity of representation. Insets below the map show the original gridded resolution at 5 arcminute for cells with streamflow values greater than 10 m3/s. The bottom insets show water temperature time series sampled at specific grid-cell locations (white crosses in the insets) for the Amazon (−57.2083° longitude, −2.625° latitude), Danube (20.125° lon, 45.2083° lat) and Ganges (88.375° lon, 24.375° lat). Time series are represented for each GCM and RCP available within FutureStreams; thin lines represent weekly values and thick lines represent 10 year rolling means.
Anomalies for selected ecologically relevant derived variables (bioclimatic indicators) for the same areas in the Amazone (left), Danube (middle) and Ganges (right) basins as used in Figs. 2 and 3. Differences are shown between RCP8.5 2080–2099 and 1976–2005. WT-cq is the water temperature of the coldest quarter, WT-range is temperature range, Q-max is maximum streamflow, Q-dm is streamflow of the driest month (see also Tables 2 and 3 below). For streamflow we show the difference between log10-transformed flow.
Source: Ecology - nature.com
