01984nam a2200157 a 450000100080000000500110000800800410001910000220006024500860008226000920016852014990026065300190175965300180177870000150179670000150181120742462017-08-31       2004    bl uuuu     u00u1 u    #d1 aMAIA, A. de H. N.  aAssessment of probabilistic forecast skill using p-values.h[electronic resource]  aIn: INTERNATIONAL CROP SCIENCE CONGRESS, 4., 2004, Brisbane, Austrália. Anais...c2004  aThe establishment and communication of climatological forecast ?skill? are complex issues requiring simple approaches. The major issues are: (a) inappropriate use of significance testing to quantify signal intensity, (b) skewed probability distributions of time series of bio-physical data (such as rainfall, crop or pasture production) rendering parametric skill measures based on Normal distribution inadequate, (c) for a spatial assessment of forecast skill, the use of skill measures derived from parametric tests require location-by-location checking of assumptions about underlying distributions, making the process cumbersome and expensive, (d) the level of significance required for forecast skill to be useful depends on the user and the application rather than on an arbitrary, pre-determined significance level and (e) signal intensity varies temporally and spatially. Hence, we propose the use of p-values derived from non-parametric tests such as the Log-Rank test as direct indicators of signal intensity. This method does not require any knowledge of the underlying data structure, nor does it require any arbitrarily chosen level of significance. Further, given adequate spatial coverage, p-values can be mapped using interpolation methods, providing a powerful and intuitive means of communicating the spatial variability of signal intensity. We illustrate this method by assessing the ability of a three-way ENSO classification in forecasting winter rainfall across Australia.  aForecast skill  aProbabilistic1 aHOLGER, M.1 aLENNOX, S.