Analyses of Global Monthly Precipitation Using Gauge Observations, Satellite Estimates, and Numerical Model Predictions

An algorithm is developed to construct global gridded fields of monthly precipitation by merging estimates from five sources of information with different characteristics, including gauge-based monthly analyses from the Global Precipitation Climatology Centre, three types of satellite estimates the infrared-based GOES Precipitation Index, the microwave (MW) scattering-based Grody, and the MW emission-based Chang estimates, and predictions produced by the operational forecast model of the European Centre for Medium-Range Weather Forecasts. A two-step strategy is used to: 1) reduce the random error found in the individual sources and 2) reduce the bias of the combined analysis. First, the three satellite-based estimates and the model predictions are combined linearly based on a maximum likelihood estimate, in which the weighting coefficients are inversely proportional to the squares of the individual random errors determined by comparison with gauge observations and subjective assumptions. This combined analysis is then blended with an analysis based on gauge observations using a method that presumes that the bias of the gauge-based field is small where sufficient gauges are available and that the gradient of the precipitation field is best represented by the combination of satellite estimates and model predictions elsewhere. The algorithm is applied to produce monthly precipitation analyses for an 18-month period from July 1987 to December 1988. Results showed substantial improvements of the merged analysis relative to the individual sources in describing the global precipitation field. The large-scale spatial patterns, both in the Tropics and the extratropics, are well represented with reasonable amplitudes. Both the random error and the bias have been reduced compared to the individual data sources, and the merged analysis appears to be of reasonable quality everywhere. However, the actual quality of the merged analysis depends strongly on our uncertain and incomplete knowledge of the error structures of the individual data sources.