394
A. Amani et al.
and class maps with a spatial resolution of 250 m (Hengl et al. 2017). The soil maps
were produced using machine learning algorithms based on about 150,000 global
soil profiles and 158 remote sensing soil covariates. The maps consist of about
280 raster layers of global soil information, including organic carbon, bulk density,
cation exchange capacity (CEC), pH, soil texture fractions and coarse fragments. Soil
properties are provided at seven standard depths (0, 5, 15, 30, 60, 100 and 200 cm).
Soil texture data was converted into soil hydraulic properties needed by the land
surface model using standard pedotransfer function relationships and coarsened to
5 km using bilinear interpolation.
Land cover data is obtained from GlobeLand30, which consists of open-access,
high-resolution maps of land cover across the globe (Chen et al. 2015). It is generated
using a machine learning algorithm based on the integration of pixel- and objectbased methods, by the use of more than 20,000 Landsat and Chinese HJ-1 satellite
images. The accuracy assessment, using nine types and over 150,000 test samples,
shows that the overall accuracy of the product is 80.33%. The spatial resolution is
as high as 30 × 30 metres, but the land cover is aggregated into tiles of different
land cover types at the 5 km modelling grid scale. Other model parameters that are
specific to the VIC model and its parameterizations are taken from the AFDM and
interpolated to 5 km.
c. Short-term (GEFS) and seasonal (NMME) forecasts
Short-term forecasts are based on the U.S. Global Ensemble Forecast System (GEFS)
which produces global weather forecasts out to 16 days operationally. We use GEFS
precipitation, temperature and wind speed forecasts out to 7 days to drive the VIC
model to produce weather and hydrological forecasts for the LCB; evaluations for
some example rainfall events are shown in Sect. 3.3. Weather forecast data are downscaled and bias-corrected to align with the statistics of the historic CHADFDM data,
using a quantile mapping approach (Wood et al. 2002). Seasonal forecasts are taken
from the North American Multi-Model Ensemble (NMME; Kirtman et al. 2014),
which consists of hindcasts (historic forecasts) and operational forecasts from a set
of six global climate models. The hindcasts are for approximately the past 30 years,
depending on the climate model, with 10–20 ensemble members per model per
monthly forecast, with lead times of up to 12 months. The forecasts of precipitation
and temperature are downscaled and bias-corrected based on the ensemble mean
forecasts for each model.
d. Web interface
The website (Fig. 2) provides an intuitive interactive interface for visualizing,
analysing and accessing data. The interface is built on Google Maps, which provides
the functionality to pan, zoom and interrogate maps, including animation of maps
of historic, current and forecast data and indices. The user can select among daily,
monthly and annual data, and derived statistics. Users can also display time series
of data at any time step for a chosen location selected via a mouse click or latitudelongitude selection. The data can also be downloaded in standard formats such as
Netcdf, Arc-ASCII and csv to be used with other software.
A. Amani et al.
and class maps with a spatial resolution of 250 m (Hengl et al. 2017). The soil maps
were produced using machine learning algorithms based on about 150,000 global
soil profiles and 158 remote sensing soil covariates. The maps consist of about
280 raster layers of global soil information, including organic carbon, bulk density,
cation exchange capacity (CEC), pH, soil texture fractions and coarse fragments. Soil
properties are provided at seven standard depths (0, 5, 15, 30, 60, 100 and 200 cm).
Soil texture data was converted into soil hydraulic properties needed by the land
surface model using standard pedotransfer function relationships and coarsened to
5 km using bilinear interpolation.
Land cover data is obtained from GlobeLand30, which consists of open-access,
high-resolution maps of land cover across the globe (Chen et al. 2015). It is generated
using a machine learning algorithm based on the integration of pixel- and objectbased methods, by the use of more than 20,000 Landsat and Chinese HJ-1 satellite
images. The accuracy assessment, using nine types and over 150,000 test samples,
shows that the overall accuracy of the product is 80.33%. The spatial resolution is
as high as 30 × 30 metres, but the land cover is aggregated into tiles of different
land cover types at the 5 km modelling grid scale. Other model parameters that are
specific to the VIC model and its parameterizations are taken from the AFDM and
interpolated to 5 km.
c. Short-term (GEFS) and seasonal (NMME) forecasts
Short-term forecasts are based on the U.S. Global Ensemble Forecast System (GEFS)
which produces global weather forecasts out to 16 days operationally. We use GEFS
precipitation, temperature and wind speed forecasts out to 7 days to drive the VIC
model to produce weather and hydrological forecasts for the LCB; evaluations for
some example rainfall events are shown in Sect. 3.3. Weather forecast data are downscaled and bias-corrected to align with the statistics of the historic CHADFDM data,
using a quantile mapping approach (Wood et al. 2002). Seasonal forecasts are taken
from the North American Multi-Model Ensemble (NMME; Kirtman et al. 2014),
which consists of hindcasts (historic forecasts) and operational forecasts from a set
of six global climate models. The hindcasts are for approximately the past 30 years,
depending on the climate model, with 10–20 ensemble members per model per
monthly forecast, with lead times of up to 12 months. The forecasts of precipitation
and temperature are downscaled and bias-corrected based on the ensemble mean
forecasts for each model.
d. Web interface
The website (Fig. 2) provides an intuitive interactive interface for visualizing,
analysing and accessing data. The interface is built on Google Maps, which provides
the functionality to pan, zoom and interrogate maps, including animation of maps
of historic, current and forecast data and indices. The user can select among daily,
monthly and annual data, and derived statistics. Users can also display time series
of data at any time step for a chosen location selected via a mouse click or latitudelongitude selection. The data can also be downloaded in standard formats such as
Netcdf, Arc-ASCII and csv to be used with other software.