Figure 6 . Boxplots of dispersion of three model parameters
before (Initial) and after the single-variable calibration (Q –
discharge; h – water level; A – flood extent; TWS – total water
storage anomalies; ET - vegetation ET; W – soil moisture), and
multi-variable calibration (All – variables except discharge; h+W –
water level and soil moisture). The spread of the values in the boxplots
stems from 300 model runs (100 for each calibration experiment).
Description of parameters is presented in Supporting Information Table
S2. A complete figure with boxplots for all parameters is presented in
Supporting Information Figure S2.
Spatial Evaluation
For model calibration, we used one streamflow gauge for discharge, one
virtual station for water level, and averaged RS data for the whole
basin for TWS, ET and soil moisture. However, many recent studies
investigated the potential for using RS spatially distributed
information in model calibration, for instance with bias-insensitive
metrics (Demirel et al., 2018; Zink et al., 2018; Dembele et al., 2020).
Here we further analyze how the lumped calibration affected the
simulated spatial patterns (Figure 7; Figure S3 in Supporting
Information).
For discharge, water level, flood extent and TWS, spatial patterns are
well reproduced even when running the model with the initial parameter
set, because the spatial patterns of these variables are determined by
intrinsic characteristics of the basin. Nonetheless, for ET, the spatial
patterns are completely different between the initial parameter set and
the calibrated setup. In this case, the calibration with spatially
aggregated ET was able to recover the spatial representation of MOD16. A
similar result was found for soil moisture spatial representation by
Demirel et al. (2019), that calibrated a model with spatially aggregated
soil moisture and TWS data.
In summary, these results highlight the overall model capability to
retrieve the ET spatial pattern even by using a lumped calibration
approach. However, for other variables, the spatial pattern was not
considerably affected by the differing model calibration strategies.