Downscaling humidity with Localized Constructed Analogs (LOCA) over the conterminous United States

Humidity is important to climate impacts in hydrology, agriculture, ecology, energy demand, and human health and comfort. Nonetheless humidity is not available in some widely-used archives of statistically downscaled climate projections for the western U.S. In this work the Localized Constructed Analogs (LOCA) statistical downscaling method is used to downscale specific humidity to a 1 degrees/16 degrees grid over the conterminous U.S. and the results compared to observations. LOCA reproduces observed monthly climatological values with a mean error of similar to 0.5 % and RMS error of similar to 2 %. Extreme (1-day in 1- and 20-years) maximum values (relevant to human health and energy demand) are within similar to 5 % of observed, while extreme minimum values (relevant to agriculture and wildfire) are within similar to 15 %. The asymmetry between extreme maximum and minimum errors is largely due to residual errors in the bias correction of extreme minimum values. The temporal standard deviations of downscaled daily specific humidity values have a mean error of similar to 1 % and RMS error of similar to 3 %. LOCA increases spatial coherence in the final downscaled field by similar to 13 %, but the downscaled coherence depends on the spatial coherence in the data being downscaled, which is not addressed by bias correction. Temporal correlations between daily, monthly, and annual time series of the original and downscaled data typically yield values >0.98. LOCA captures the observed correlations between temperature and specific humidity even when the two are downscaled independently.