Challenges in Mapping Daily Temperature and Precipitation across the Conterminous United States

In 2010, The PRISM Climate Group at Oregon State University (OSU) partnered with the Risk Management Agency (RMA) of the U.S. Department of Agriculture to deliver detailed weather and climate estimates for every farm, every day, across the conterminous US. The RMA oversees the nation's crop insurance program, which insures over $120 billion worth of agricultural production each year. Given that most crop damage claims are weather related, the value of knowing what weather events happened, and where, is important to being able to evaluate and pay claims quickly and accurately. Previous to this partnership, the PRISM climate mapping system was used to develop normal and time series of gridded climate data at the monthly time step. But the crop insurance industry required data at a daily time step. PRISM has now developed more than thirty years of daily minimum and maximum temperature and precipitation grids, but the challenges in doing so were many, and much work still remains to improve the datasets further. The first challenge was to define a “day” across the country. Given the 1200–1200 UTC “hydrologic day” used by the National Weather Service, and the tendency of one-per-day observers to favor mornings in recent years, this definition was adopted. Questions arising from this decision were: How are stations that observe in the afternoon handled? What do we do when the time of observation is unknown? How do we handle multi-day accumulations? When interpolating these station data to a regular grid, what form of PRISM is most appropriate for interpolating the station data to a grid? PRISM monthly time series datasets are created using Climatologically-Aided Interpolation (CAI), where the 1981-2010 PRISM normal serve as the predictor grids in PRISM's local regression function. Is this suitable at the daily time step? Can RADAR enhancement be used to improve results for precipitation? Does the best method vary with region and season? The signal-to-noise ratio when interpolating at the daily time step, especially for precipitation, is quite large, and can lead to chronic underestimation in areas where precipitation is spotty. Should the daily grids be altered to conform to the grids interpolated at the monthly time step, where the signal-to-noise ratio is larger? When estimating the uncertainty of the interpolated product, standard statistics such as mean absolute error and bias are of little use, because the daily values are typically very small and highly variable. Are there more appropriate metrics, such as the rate of correctly identifying wet and dry days, more useful for estimating uncertainty at the daily time step? Our approaches to meeting these challenges, and those remaining, will be discussed.