| description abstract | AbstractInference about time series in weather data can be made in several ways. Current practice focuses on computing summary measures, such as mean and variance, or constructing a reference year from small subsets of data derived from multiple years. Some applications require the selection of an instance of observed data over a fixed time frame, typically a year or more, for modeling. In addition, many current methods do not include rainfall as a parameter of interest. This paper reviews and refines existing methods for determining a reference year by creating a metric that measures the (abstract) distance between observed patterns of rainfall. The reference year is then chosen from a group of potential reference years. This method is computationally efficient, easily explained, and robust against differences in the index of reporting, to include leap years. Application of the distance metric to data from Philadelphia, Pennsylvania, and Norfolk, Virginia, shows that it appropriately identifies not only the years that are most typical for a location, but extreme years as well. Both are particularly useful in applications related to urban hydrology, which formed the basis for development of this method. Results also demonstrate that the proposed method is functionally different than existing methods. The distance metric represents an evolutionary step forward, overcomes some difficulties present from other approaches, and would be applicable to a number of cross-disciplinary applications. | |
