The role of cross-correlation between precipitation and temperature in basin-scale simulations of hydrologic variables

Uncertainty in climate forcings causes significant uncertainty in estimating streamflow and other land-surface fluxes in hydrologic model simulations. Earlier studies primarily analyzed the importance of reproducing cross-correlation between precipitation and temperature (P-T cross-correlation) using various downscaling and weather generator schemes, leaving out how such biased estimates of P-T cross-correlation impact streamflow simulation and other hydrologic variables. The current study investigates the impacts of biased P-T cross-correlation on hydrologic variables using a fully coupled hydrologic model (Penn-state Integrated Hydrologic Model, PIHM). For this purpose, a synthetic weather generator was developed to generate multiple realizations of daily climate forcings for a specified P-T cross-correlation. Then, we analyzed how reproducing/neglecting P-T cross-correlation in climate forcings affect the accuracy of a hydrologic simulation. A total of 50 synthetic data sets of daily climate forcings with different P-T cross-correlation were forced into to estimate streamflow, soil moisture, and groundwater level under humid (Haw River basin in NC, USA) and arid (Lower Verde River basin in AZ, USA) hydroclimate settings. Results show that climate forcings reproducing the P-T cross-correlation yield lesser root mean square errors in simulated hydrologic variables (primarily on the sub-surface variables) as compared to climate forcings that neglect the P-T cross-correlation. Impacts of P-T cross-correlation on hydrologic simulations were remarkable to low flow and sub-surface variables whereas less significant to flow variables that exhibit higher variability. We found that hydrologic variables with lower internal variability (for example: groundwater and soil-moisture depth) are susceptible to the bias in P-T cross-correlation. These findings have potential implications in using univariate linear downscaling techniques to bias-correct GCM forcings, since univariate linear bias-correction techniques reproduce the GCM estimated P-T cross-correlation without correcting the bias in P-T cross-correlation.