Spatial Analysis of Mean Annual Maxima for Precipitation Frequency Estimation in Atlas 14/15

Grids (rasters) of mean annual maximum (MAM) precipitation based on observing station data have long been the basis for gridded precipitation frequency estimates at different frequencies and durations for NOAA Atlas 14. For Atlas 14 Volumes 1-12, MAM grids were produced in collaboration with the PRISM Climate Group at Oregon State University. To support future Atlas volumes the Hydrometeorological Design Studies Center (HDSC) at the NWS Office of Water Prediction (OWP) is developing an in-house MAM capability.

The current capability consists of two components. First, a stepwise multiple linear regression procedure generates a background, or "first guess," gridded MAM analysis using a variety of geography-based (e.g., elevation, latitude, distance to coast, height above local terrain) and climatology-based (e.g., mean annual precipitation) quantities as covariates (independent variables). Second, an ordinary kriging step spatially interpolates and corrects errors in the background analysis. Cross-validation analyses of results produced using the current capability indicate accuracy similar to or improved relative to that of PRISM-based estimates for Atlas 14 Volume 12.

This presentation will describe the current state of the MAM methodology, discuss the spatial covariates that are typically prominent in regression, present results for several different spatial domains and storm durations, and share the latest developments in the methodology, including a third step in the spatial analysis to correct residual errors. Finally, we will address the challenge of producing MAM grids for the full coterminous United States (CONUS) as is needed for Atlas 15, versus the smaller and thereby more homogeneous regions associated with Atlas 14 Volumes.