Multisite Downscaling of Seasonal Predictions to Daily Rainfall Characteristics over Pacific-Andean River Basins in Ecuador and Peru Using a Nonhomogeneous Hidden Markov Model

The seasonal predictability of daily rainfall characteristics is examined over 21 hydrologic units in the Pacific-Andean region of Ecuador and Peru (PAEP) using a nonhomogeneous hidden Markov model (NHMM) and retrospective seasonal information from general circulation models (GCMs). First, a hidden Markov model is used to diagnose four states that play distinct roles in the December-May rainy season. The estimated daily states fall into two wet states, one dry state, and one transitional dry-wet state, and show a systematic seasonal evolution together with intraseasonal and interannual variability. The first wet state represents regionwide wet conditions, while the second one represents north-south gradients. The former could be associated with the annual moisture offshore of the PAEP, thermally driven by the climatological maximum of sea surface temperatures in the Nino-1.2 region. The latter corresponds with the dynamically noisy component of the PAEP rainfall signal, associated with the annual displacement of the intertropical convergence zone. Then, a four-state NHMM is coupled with GCM information to simulate daily sequences at each station. Simulations of the GCM-NHMM approach represent daily rainfall characteristics at station level well. The best skills were found in reproducing the interannual variation of seasonal rainfall amount and mean intensity at the regional-averaged level with correlations equal to 0.60 and 0.64, respectively. At catchment level, the best skills appear over catchments south of 4 degrees S, where hydrologically relevant characteristics are well simulated. It is thus shown that the GCM-NHMM approach provides the potential to produce precipitation information relevant for hydrological prediction in this climate-sensitive region.