On an extension of the L-moment approach to modelling distributions which include trend

In this paper, extensions of existing L-moment procedures are considered, aiming for ways of assessing changes over time in the distributions of quantities such as annual maxima while avoiding subdivision of data records into possibly overlapping sub-periods. Distributional changes of many types are included: location, scale, skewness, etc. Although direct application of some simple ideas is unsuccessful at providing good estimates of trends, less direct application of the same ideas allows for the implementation of significance tests for changes in shape, for the construction of confidence regions for the quantiles of a distribution that might change with time and for an indirect use of L-moments in estimating trends.