GloFAS

Attention

This module has the following additional requirements:

cdsapi
cfgrib

These can be installed as follows:

python -m pip install ocha-anticipy[glofas]

Background

The Global Flood Awareness System (GloFAS) is part of the Copernicus Emergency Management Service (CEMS).. It provides a global model and forecast of river flow for large catchments and riverine flooding, intended to be complementary to national hydrological and meteorological services.

The processed and curated data, plus several other products, are available to view through the GloFAS map viewer. While the map viewer makes image graphs of modelled and forecasted river discharge available, the actual data are needed to perform more in-depth analyses, such as computing specific return periods, knowing when a certain river discharge threshold was exceeded in the past, or detailed forecast performance at a specific location,

The raw data in raster format can be downloaded from the Climate Data Store. However, the API can be tricky to navigate, with query size limits and long download times which depend on the query structure. Furthermore, the data are in the less commonly used GRIB file format, and the river discharge data at specific reporting points (those found on the map viewer) still need to be extracted. This GloFAS Python module takes care of all of these steps.

Both the latest version (version 4, released summer 2023) and the previous version (version 3, released May 2021 ) are supported. We do not support version 2 at this time.

Before describing the module usage, we will outline the different GloFAS datasets available: reanalysis, forecast, and reforecast. All datasets have a spatial resolution of 0.05° in version 4, and 0.1° for version 3.

Reanalysis

GloFAS-ERA5 reanalysis (Harrigan et al.) is a global gridded dataset of river discharge with a daily timestep, available from 1 January 1979 to the present day (updated daily). It is based on ECMWF’s latest global atmospheric reanalysis (ERA5, Herbasch et al.) combined with the Lisflood hydrological and channel routing model, and calibrated against observed river discharge (Alfieri et al.). The data is available to download here on CDS.

Forecast

The GloFAS forecast (Harrigan et al.) uses the above model, applied to the ECMWF integrated forecast system (IFS) ensemble forecast. The IFS has 51 ensemble members out to a lead time of 15 days. The ensemble is composed of a single control member, generated from the most accurate estimate of current conditions, and 50 other members that have their initial conditions perturbed.

Twice per week (Mondays and Thursdays), the IFS is extended to run up to 46 days ahead at a coarser resolution (~36 km). GloFAS makes available days 16 to 30 for this extended forecast.

Both versions 3 and 4 of the forecast data are available to download from 26 May 2021 until present day (updated daily) here. While CDS does have version 3 pre-release data from 2020-2021, we understand that there were some small issues that were fixed in the operational version, so at this point in time this module only retrieves data from the operational model.

Reforecast

The GloFAS reforecast (Harrigan et al.) is similar to the forecast, but uses the ECMWF IFS reforecast. The reforecast is initialized twice per week (Monday and Thursdays) and has 11 ensemble members. The data runs from March 2003 to August 2022, and is only available for the months March to August. (For version 3 it runs from January 1999 to December 2018, for all months). The data can be downloaded from CDS here.

Usage

This module performs three basic steps:

  1. Download the raw GloFAS data, in GRIB file format. Each file is a raster based on a user-specified region of interest, and contains data for a full day, month, or year, depending on the GloFAS product.

  2. Process the data. For each raw GRIB file, river discharge for a set of user-provided pixel locations is extracted and stored in NetCDF format.

  3. Load the data. The processed files, which are split by time in the same way as the raw data, are read in as a single xarray.DataSet.

CDS API

The download step makes use of the CDS API. You will need to register an account on CDS, then once logged in go to your user page and note down your UID and API key.

The how-to page has instructions for each operating system, but we summarize the instructions below:

Windows: Create a file called %USERPROFILE%.cdsapirc where, %USERPROFILE% is usually located in your C:UsersUsername folder.

Mac and Linux: Create the file ~/.cdsapirc.

Then, for all operating systems, add the following to the created file:

url: https://cds.climate.copernicus.eu/api/v2
key: ${UID}:${API-key}

where $UID is your UID and $API-key is your API key.

ecCodes

To read in .GRIB files using xarray, you’ll need to install ecCodes on your machine. For all operating systems, this can be done through the source distribution, or in a conda environment with the command:

conda install -c conda-forge eccodes

Some OS-specific binaries are also available. For Linux, python3-eccodes can be found as a .deb and .rpm (check your specific Linux distribution for the latest version). For Mac, according to the ecCodes-Python documentation <https://github.com/ecmwf/eccodes-python#system-dependencies>_, ecCodes can be installed using brew:

brew install eccodes

Reporting points

Next, if it doesn’t already exist, you need to create a country configuration for the country you would like to analyze.

An example country config for Bangladesh is:

iso3: bgd
glofas:
  reporting_points:
  - name: Bahadurabad
    lon: 89.65
    lat: 25.15
  - name: Hardinge Bridge
    lon: 89.05
    lat: 24.05

The reporting points indicate the raster file coordinates used to extract the river discharge for a particular location. Those in the above example have been taken from the GloFAS map viewer. If you select “Reporting Points” from the “Hydrological” menu at the top, they will appear as dots on the map. If you then click on one of the points, you are able to see information such as the station name, and LISFLOOD X and Y, which are the respective longitude and latitude used in the configuration file.

Reporting point coordinates are manually selected by the GloFAS team to be representative of physical gauge locations, and to be located on a river in the model raster file. In principle, one could specify any set of coordinates that exists on the raster, but caution is advised when doing so.

Running the code

You can initialize a built-in country config as follows:

from ochanticipy import create_country_config

country_config = create_country_config(iso3="bgd")

Another required input is the geographic area of interest, which will define the bounds of raw raster data to be downloaded. A simple way to identify the area around the chosen country is to use the COD administrative boundaries. You will need to download the data, and extract a geo bounding box:

from ochanticipy import CodAB, GeoBoundingBox

codab = CodAB(country_config=country_config)
codab.download()
admin0 = codab.load()
geo_bounding_box = GeoBoundingBox.from_shape(admin0)

Note that the reporting points in the configuration file need to lie within the geographic area of interest.

Next you need to instantiate the GloFAS class with the country config. For this example, we will use the GloFAS forecast, however the steps are nearly identical for the the reanalysis and reforecast – the only differences are the acceptable date ranges, and that leadtime_max is not an input parameter to the reanlaysis.

In this case, we would like the data for the past month. In general, we suggest specifying dates explicitly, as using e.g. date.today() will not produce the same results when run on a different day.

from datetime import date

from ochanticipy import GlofasForecast

glofas_forecast = GlofasForecast(
    country_config=country_config,
    geo_bounding_box=geo_bounding_box,
    leadtime_max=15,
    start_date=date(year=2022, month=9, day=22)
    end_date=date(year=2022, month=10, day=22),
 )

We then need to download the GloFAS data. The module will download all the data between start_date and end_date (inclusive). The raw files are in .GRIB format, and are separated by day (forecast), month (reforecast), or year (reanalysis) depending on the data type, due to CDS query size limits.

glofas_forecast.download()

The downloading process works by generating all requests required for the specified timeframe, sending them to CDS, and saving the request numbers in memory. You can see all your requests on the CDS website, and cancel any pending ones in case the run is interrupted (we hope to implement request number caching in a later version).

The module then pings the CDS API every minute to check which requests have completed, and downloads those that have. This continues until all requests have been downloaded.

This process can unfortunately take a long time, however, with this module we’ve tried to optimize the queries to be as fast as possible. In our experience, downloading the full reanalysis (from 1979 to today) is the fastest and takes a couple of hours, while the full reforecast (1999 to 2018) is the slowest and takes around a day to complete. It also depends how busy the queue is, which you can check here.

The next step is to process the files. First, we want to convert from the less used GRIB format to the more common and flexible NetCDF. Furthermore, rather than having the full raster, we extract the river discharge data at the reporting poitns specified in the configuration file. This can be done in a single step:

glofas_forecast.process()

Note that each individual raw GRIB file is converted to a corresponding processed NetCDF file. This is to simplify the downloading and processing of addition data, i.e. for adding new dates.

To load all of the processed files into a single dataframe (which can then be saved to a single NetCDF file or other compatible format), execute:

bgd_glofas_forecast_reporting_points = glofas_forecast.load()

The full codde snippet is below:

from datetime import date

from ochanticipy import create_country_config, CodAB, \
    GeoBoundingBox, GlofasForecast

country_config = create_country_config(iso3="bgd")
codab = CodAB(country_config=country_config)
codab.download()
admin0 = codab.load()
geo_bounding_box = GeoBoundingBox.from_shape(admin0)

glofas_forecast = GlofasForecast(
    country_config=country_config,
    geo_bounding_box=geo_bounding_box,
    leadtime_max=15,
    start_date=date(year=2022, month=9, day=22)
    end_date=date(year=2022, month=10, day=22),
 )
glofas_forecast.download()
glofas_forecast.process()

bgd_glofas_forecast_reporting_points = glofas_forecast.load()