An assessment of blended satellite-gauge precipitation products for Australia

[摘要] This work examines the potential role that satellite-derived rainfall products may play in improving daily rainfall estimation for Australia.The motivation was the hypothesis that blending the less accurate yet spatially-comprehensive coverage of satellite-derived rainfall products, with the more accurate yet point observation of rain gauges, would compensate for the deficiencies in the respective data sets and result in enhanced rainfall products for the continent.A number of rainfall interpolation and estimation methods were examined to generate gauge-only and blended satellite-gauge rainfall analyses for the study period 31 May 2009 – 31 May 2010.Data used were daily rainfall observations from the near real-time network of rain gauges across Australia, and rainfall accumulations derived from the Tropical Rainfall Measuring Mission (TRMM) multi-satellite precipitation analysis (TMPA) system rain rates. Assessment of the rainfall estimates was based on the gauges reporting daily rainfall post real-time (i.e. an independent sets of fitting and validation gauges).For the national average performance metrics, we observed the blended satellite-gauge rainfall estimates to be better than the satellite daily rainfall product alone. However, the satellite data did not appear to improve rainfall estimates compared to methods that used gauge data alone.Exploring why this is the case, our investigations indicated that with more than ~1,000 gauges across Australia there is marginal difference in performance statistics between the gauge-only and blended satellite-gauge rainfall estimates. The value of satellite rainfall only became in local assessments of performance (based on cells, or stratification zones, over Australia). We observed an improvement in performance of the blended product over the gauge-only analysis by as much as 20% in parts of the country where the gauge density is less than approximately 4 gauges per 10,000km2. While rainfall estimation algorithm intercomparison was of secondary importance to our main aim, we found that, of the methods we controlled, the Li and Shao (2010) kernel-based double-smoothing algorithm performed best (comparable to, but not quite as good as, the AWAP rainfall estimates). However limitations in traditional approach to performance evaluation (based on national average error statistics) prompted an exploratory look at the potential for stochastic simulations to provide the spatially-explicit rainfall uncertainty that the modelling community desires, as well as to serve as performance evaluation data. We conclude that satellite rainfall products are likely to be most useful in improving rainfall estimation in gauge-sparse regions of Australia. We recommend that future research efforts be directed towards demonstrating the improvement, which may include:•Assessing different satellite-based rainfall products (e.g. CMORPH, GSMaP) to help examine the effect of spatial resolution on rainfall estimation at local scales. •Assessing quantitative precipitation forecasts for use as spatial covariates to blend with gauge observations and satellite data.•Examining the effect of gauge configuration with each cell on local performance statistics. •Exploring the use radar rainfall as (i) an indirect verification for continental products and (ii) to provide tighter constraint on rainfall estimation around cities and towns. •Continuing to examine the use of stochastic simulation to provide rainfall uncertainty for the continent.

[发布日期] 2011-02-01 [发布机构] CSIRO

[学科分类] 地球科学(综合)