Predicting spatiotemporal variation in runoff in a data-sparse region:
analyses of a whole-country model for Panama
Abstract
Panama faces seasonal floods and droughts as well as rising freshwater
demands ranging from domestic consumption to hydro-power and the
operation of the Panama Canal. A process-based hydrological model of the
country is a desirable scenario planning tool to complement the existing
national water security plan. In Panama as in much of the Global South,
sufficient observed data do not exist for all watersheds to calibrate
complex hydrological models. Understanding and improving the performance
of uncalibrated hydrological models could greatly expand their utility
in such regions. In this study, we build and validate an uncalibrated
Soil and Water Assessment Tool (SWAT) model for Panama. We extend the
default precipitation submodel and demonstrate the importance of
sufficiently accounting for for spatial autocorrelation patterns in
precipitation inputs: we found large improvements over the default
model, not only for monthly means (NSE = 0.88, from NSE= 0.69 for
default SWAT), but especially for standard deviations (NSE = 0.59, from
0.27) and maxima (NSE = 0.51, from 0.21) of discharge across locations
and months. We found a strong seasonal trend and regional differences in
the spatial autocorrelation of rainfall, suggesting that this phenomenon
should not be modeled statically. The resulting precipitation and
hydrology models provide important baseline information for Panama,
especially on variability and extremes, and could serve as a template
for other regions with limited data.