RESULTS
SWAT model calibration and
validation
The SWAT model successfully simulated both streamflow and sediment
regime of the watershed (Fig. 3 and 4). For the streamflow, the
calibration results showed a very good agreement between simulated and
observed data, with high R2, NSE, and good PBIAS in
all available hydrometric stations (Fig. 4). Similarly, the statistical
indices for the validation procedure were within the range of a “very
good” model performance according to the model’s performance criteria
established by Moriasi et al. (2007) (Table S3). Considering the
sediment load calibration, despite using randomly distributed sediment
data, simulated and observed data showed a “very good” performance
match for calibration and validation in both Monte Da Ponte (Fig. 4a)
and Oeiras (Fig. 4b) stations. Regarding the streamflow simulation, the
calibration/validation phases showed more than 70% of data bracketed by
the 95PPU (P-factor ≥ 0.70) with R-factors ranging from 0.15 to 0.3. For
sediment, both Monte da Ponte and Oeiras reached a P-factor ≥ 0.6 with
an R-factor of 0.32 and 0.41, respectively. These results remarked a
high model performance and efficiency for both variables.
A global sensitivity analysis was implemented to identify those
parameters that strongly influenced the simulated flow and soil losses
within those listed in Table 1. The significance of the sensitivity test
was evaluated using the statistical index “p-value”, automatically
generated through the application of SUFI-2 algorithm. The results of
the sensitivity analysis (Table 1, where the bold parameters represent
the sensitive ones), indicated that the streamflow simulation was
strongly dependent from DEEP_IMP, GW_DELAY, RCHGR_DP, and ALPHA_BF,
while SLSUBBSN, USLE_K, and USCLE_C were the main parameters
influencing the sediment loss simulation.