Reconstruction of a Blended Monthly Precipitation Dataset for the Pre-Satellite Era

There are less than 400 stations across Canada that have long-term precipitation records, which makes it very challenging to produce a robust long-term gridded precipitation dataset for Canada. This study developed an algorithm to produce a robust long-term gridded precipitation dataset for Canada. The algorithm is based on the multivariate regression relationship between two sets of leading principal components (PCs), the leading PCs of the predictor fields and of the predictand fields. The predictor fields consist of a set of core stations of long-term homogenized precipitation data records, and the predictand fields, the Canadian blended monthly precipitation data set version 1 (CanBPmlyV1) derived by blending in situ precipitation data and satellite precipitation estimates. To evaluate the model skill, the model was trained using three training sets of stations representing sparse, low, and medium network densities of the predictand fields. It was evaluated using an independent evaluation data set that is of much higher station density (it consists of all the other available stations in the region, excluding those in the training set). The results show that the algorithm can produce a much better gridded dataset than a kriging analysis of the long-term station data alone. The model was then calibrated using a set of core stations of long-term homogenized monthly precipitation data records and used to extend the Canadian blended precipitation data set, CanBPmlyV1, to the pre-satellite era of 1948-1978.