Comparison of Gridding Methods for Precipitation Over Canada and Assessment of Station and Data Density Effects on Gridding Results

This study compares the results from applying different gridding methods to different precipitation variables, including total and normal monthly precipitation amounts as well as anomalies and relative anomalies of monthly precipitation for representation of precipitation climate and regional mean precipitation trends. We applied three gridding models, including Optimal Interpolation (OI), currently used to produce the Canadian Gridded (CanGRD) data, Thin-Plate Smoothing Splines, and ordinary kriging. Two observation-based precipitation source datasets were used to derive gridded benchmark and pseudo-observational datasets, with the latter being sampled at full and subsets of a typical long-term precipitation station network in Canada to be then gridded and evaluated against the corresponding benchmark. The results show that the best regional mean precipitation trend estimates are obtained through gridded total precipitation data generated from combined grids of separately kriged relative precipitation anomalies and normal precipitation amounts. This scheme was then used to assess the impact of different station and data densities on the gridded data. The results also indicate that CanGRD-OI is the least accurate model in representing the precipitation climate and trends, and the CanGRD-based trend estimations notably overestimate the trend of regional mean precipitation amounts in the North while underestimating it in the South.