Improvements in Daymet Continental-Scale Gridded Daily Precipitation and Temperature Estimates

Michele M Thornton,  ORNL DAAC,  thorntonmm@ornl.gov (Presenter)
Rupesh Shrestha,  Oak Ridge National Laboratory,  shresthar@ornl.gov
Peter E. Thornton,  Oak Ridge National Laboratory,  thorntonpe@ornl.gov
Shih-Chieh Kao,  Oak Ridge National Laboratory,  kaos@ornl.gov
Yaxing Wei,  Oak Ridge National Laboratory,  weiy@ornl.gov
Bruce E Wilson,  Oak Ridge National Laboratory,  wilsonbe@ornl.gov

The NASA TE Program and NACP Project have long supported the development of the Daymet dataset which now ranks as one of the top 10 in NASA Earthdata downloads. This poster describes the current Version 4 distribution. The Daymet dataset provides high-resolution daily gridded weather parameters for the entire North American spatial extent for the time period 1980 – 2022 (https://daymet.ornl.gov/). Daymet uses gauge data from multiple surface weather observation networks compiled from the National Centers for Environmental Information (NCEI) Global Historical Climate Network (GHCN) Daily. Combined with digital elevation data and a three-dimensional gradient estimation method, Daymet estimates continuous surfaces of minimum and maximum temperature and precipitation occurrence and amount. Some stations suffer from a “time-of-day reporting bias”, wherein stations with daily reporting times earlier than mid-day are actually reporting the maximum temperature (Tmax) observed from the previous day. Likewise, the time of observation for precipitation events results in daily occurrence bias.
We used cross-validation analysis to demonstrate that shifting the recorded Tmax for these stations back by one day significantly reduced the mean prediction errors for Tmax. We applied the same logic to the daily total precipitation records and explored several alternatives for how to redistribute precipitation amounts. It was found that shifting the entire daily amount back by one day for stations with reporting times earlier than noon reduced prediction errors while maintaining observed daily event size frequency distributions. The precipitation shifting logic was evaluated with a regression analysis using radar-based hourly NCEP Stage IV QPE. It was shown that this method is effective in identifying biases in the timing of daily events. These improvements became available in late fall 2020 in a new dataset release of Daymet V4 daily meteorological data for North America for year 1980 – present. The results of the methodologies toward data fusion in precipitation data products show progress toward fulfilling the need for improved high-resolution estimates of daily gridded weather data across North America.

Poster Location ID: 2-24

Presentation Type: Poster

Session: Poster Session 2

Session Date: Wed (May 10) 5:15-7:15 PM

CCE Program: TE