Improving the Representation of Historical Climate Precipitation Indices Using Optimal Interpolation Methods

Abstract

ABSTRACT Defining a reference climate for precipitation is an important requirement in the development of climate change scenarios to support climate adaptation strategies. It is also important for many hydrological and water resource applications. This, however, remains a challenge in regions that are poorly covered by meteorological stations, such as northern Canada or mountainous regions. Reanalyses may represent an interesting option to define a reference climate in such regions. However, these need to be validated and corrected for bias before they can be used. In this paper, two data assimilation methods, Optimal Interpolation (OI) and Ensemble Optimal interpolation (EnOI), were used to combine four reanalysis datasets with observations in order to improve the representation of various precipitation indices across Canada. A total of 986 meteorological stations with minimally 20-year precipitation records over the 30-year reference period (1980–2009) were used. Annual values of ten Climate Precipitations Indices (CPIs) were estimated for each available dataset and were then combined (reanalysis plus observations) using OI and EnOI. A cross-validation strategy was finally applied to assess the relative performance of these datasets. Results suggest that combining reanalysis and observations through OI or EnOI improves CPI estimates at sites where no recorded precipitation is available. The EnOI dataset outperformed OI applied to each reanalysis independently. An evaluation of the gridded interpolated observational dataset from Natural Resources Canada showed it should be used with considerable caution for extreme CPIs because it can underestimate annual maximum 1-day precipitation, as well as overestimate the annual number of wet days.