Documentation Revision Date: 20160930
Data Set Version: V3
Summary
There are a total of 111 files in this data set. There are 108 shape files and 3 csv files. The data are distributed as shape files that represent the 2degree by 2degree tile structure in which the Daymet model estimates are derived. The annual crossvalidation statistics are provided as a separate shape file for the North American domain for each of the three variables for each year of available Daymet input data (i.e., 3 files/year for 36 years).
Also provided are the complete time series of annual summary crossvalidation statistics for the three Daymet input parameters in comma separated files (*.csv). There is one file for each of the three parameters for each tile.
Citation
Thornton, P.E., M.M. Thornton, and R.S. Vose. 2016. Daymet V3: Annual Tile Summary CrossValidation Statistics for North America, Hawaii. ORNL DAAC, Oak Ridge, Tennessee, USA. http://dx.doi.org/10.3334/ORNLDAAC/1348
Table of Contents
 Data Set Overview
 Data Characteristics
 Application and Derivation
 Quality Assessment
 Data Acquisition, Materials, and Methods
 Data Access
 References
 Data Set Revisions
Data Set Overview
Project: Daymet
This data set provides annual summary crossvalidation statistics for minimum temperature (tmin), maximum temperature (tmax), and daily total precipitation (prcp) of "Daymet: Daily Surface Weather Data on a 1km Grid for North America, Version 3" (Thornton et al., 2016). The crossvalidation statistics were generated by the Daymet model algorithm from the stationbased daily observations and predictions and are summarized for each of the 2degree by 2degree tiles in the regimen in which Daymet is derived. Data are available for the temporal period 1980 through 2015, the most recently processed calendar year of Daymet Version 3.
Summarized by tile are average and periodofrecord mean absolute error (MAE) and bias statistics for the input weather observations of tmin, tmax, and prcp. Also available are tilewide values of number of ground weather stations evaluated, total stationdays evaluated, and mean observed input parameter values. Summary statistics are also available for the Gaussian distribution functions, used in the Daymet interpolation method, as mean and standard deviations of the radius of the kernel weights and x, y, and z components of the 3dimensional regression formula.
Related Data Sets:
Thornton, P.E., M.M. Thornton, B.W. Mayer, Y. Wei, R. Devarakonda, R.S. Vose, and R.B. Cook. 2016. Daymet: Daily Surface Weather Data on a 1km Grid for North America, Version 3. ORNL DAAC, Oak Ridge, Tennessee, USA. http://dx.doi.org/10.3334/ORNLDAAC/1328
Data Characteristics
The annual crossvalidation statistics are provided for North America as a separate shape file for each of the three variables for each year of available Daymet input data (i.e., 3 files/year for 36 years).
Also the complete time series of annual crossvalidation statistics for a variable is provided in a single comma separated file (*.csv). There is one file for each of the three variables.
Spatial Coverage: North America and Hawaii: including Canada, Mexico, the United States of America, Puerto Rico, and Bermuda.
Spatial Resolution: 2degree x 2degree
Temporal Coverage: 19800102 to 20151231
Temporal Resolution: Annual
Site boundaries: (All latitudes and longitudes given in decimal degrees)
Site  Westernmost Longitude  Easternmost Longitude  Northernmost Latitude  Southernmost Latitude  Geodetic Datum 

North America, Puerto Rico, and Hawaii 
180  52  84  14  WGS_1984 
Data File Information
Shape Files
There are 108 shape files associated with this data set. Daymet crossvalidation data are available as shape files covering North America (Canada, United States, Mexico, Puerto Rico) and Hawaii – with a spatial resolution of 2 degrees.
The shape files geometric polygon structure represents the 2 degree x 2 degree tile “grid” in which the Daymet model is processed and output. Note that this vector file is an approximation of the Daymet 2degree raster tile grid.
Three shape files with crossvalidation statistical information for each of the three Daymet daily weather input variables minimum temperature (tmin), maximum temperature (tmax), and total precipitation (prcp) are available each year of available Daymet data.
The North American shape files are zipped for convenience and contain five files (*.dbf,*.prj,*.shp,*.shp.xml, and *.shx).
File names follow this syntax: DaymetV3_xval_pppp_yyyy.shp (*.zip)
Where:
xval distinguishes these as Daymet tile summary crossvalidation data files;
pppp is the respective Daymet input meteorological variable (tmin, tmax, and prcp); and
yyyy is year.
Data Dictionary:
Fields within each shape file contain the tilewide summary crossvalidation statistics.
Shape files for temperature (tmin and tmax) CrossValidation Statistics have these attributes.
Field 
Units/format 
Description 
Xmin 
decimal degrees 
Approximate minimum longitude of tile 
Xmax 
decimal degrees 
Approximate maximum longitude of tile 
Ymin 
decimal degrees 
Approximate minimum latitude of tile 
Ymax 
decimal degrees 
Approximate maximum latitude of tile 
year 
YYYY 
Daymet processing year 
tileid 
Daymet Tile ID 

nstns 
stations 
number of stations evaluated (tileid) 
nstndays 
days 
number of stationdays evaluated (tileid) 
rad90mean 
meter 
mean: radius capturing 90% of filter kernel weight 
rad90std 
meter 
standard deviation: radius capturing 90% of filter kernel weight 
daymae 
degrees Celsius 
mean absolute error for singleday predictions 
pormae 
degrees Celsius 
mean absolute error for periodofrecord predictions 
bias 
degrees Celsius 
mean prediction bias 
tamean 
degrees Celsius 
mean observed temperature (tmin and tmax) 
xlrmean 
degrees C/meter 
3d regression: mean xcomponent 
xlrstdv 
degrees C/meter 
3d regression: amongstation std dev of xcomponent 
ylrmean 
degrees C/meter 
3d regression: mean ycomponent 
ylrstdv 
degrees C/meter 
3d regression: amongstation std dev of ycomponent 
zlrmean 
degrees C/meter 
3d regression: mean zcomponent 
zlrstdv 
degrees C/meter 
3d regression: amongstation std dev of zcomponent 
Shape files for precipitation (prcp) CrossValidation Statistics have these attributes.
Field 
Units/format 
Description 
Xmin 
decimal degrees 
Approximate minimum longitude of tile 
Xmax 
decimal degrees 
Approximate maximum longitude of tile 
Ymin 
decimal degrees 
Approximate minimum latitude of tile 
Ymax 
decimal degrees 
Approximate maximum latitude of tile 
year 
YYYY 
Daymet processing year 
tileid 
Daymet Tile ID 

nstns 
stations 
number of stations evaluated (tileid) 
nstndays 
days 
number of stationdays evaluated (tileid) 
rad90mean 
meter 
mean: radius capturing 90% of filter kernel weight 
rad90std 
meter 
standard deviation: radius capturing 90% of filter kernel weight 
daymae 
cm/day 
mean absolute error for singleday predictions 
pormae 
cm/day 
mean absolute error for periodofrecord predictions 
pormpae 
% 
mean absolute error as a percentage, for period of record predictions 
bias 
cm/day 
mean prediction bias 
ppmean 
cm/day 
mean observed daily total precipitation 
xlrmean 
1/meter 
3d regression: mean xcomponent 
xlrstdv 
1/meter 
3d regression: amongstation std dev of xcomponent 
ylrmean 
1/meter 
3d regression: mean ycomponent 
ylrstdv 
1/meter 
3d regression: amongstation std dev of ycomponent 
zlrmean 
1/meter 
3d regression: mean zcomponent 
zlrstdv 
1/meter 
3d regression: amongstation std dev of zcomponent 
User’s Notes
 When “nstns” is zero (0), no input weather station data are available within that tile. All attributes are recorded as nodata (9999) or "nan".
 When “nstns” have very low values (e.g. 1, 2, or 3) denoting limited input data available for that tile, values for the 3dimensional regression components may be set to “nan” where the regressions algorithm failed.
 Floating point precision has been carried forward from the Daymet model for all attributes.
Spatial Data Properties
Spatial Representation: vector
Vector Format: shape file
Nodata Value: 9999
Spatial Reference Properties
Type: Geographic
"GEOGCS['GCS_WGS_1984',
DATUM['WGS_1984',
SPHEROID['WGS_84',6378137.0,298.257223563]],
PRIMEM['Greenwich',0.0],
UNIT['Degree',0.0174532925199433]]"
Comma Separated Files
There are 3 comma separated files with this data set. The complete time series (19802015) of annual crossvalidation statistics for a variable is provided in a single comma separated file (*.csv)  one file for each of the three variables.
File names follow this syntax: DaymetV3_xval_pppp_yyyyyyyy.csv
Where:
xval distinguishes these as Daymet crossvalidation data files;
pppp is the respective Daymet input meteorological variable (tmin, tmax, and prcp); and
yyyyyyyy is the range of annual summary statistics included in the file.
User’s Notes
 When “nstns” is zero (0), no input weather station data are available within that tile. All attributes are recorded as nodata (9999) or "nan".
 When “nstns” have very low values (e.g. 1, 2, or 3) denoting limited input data available for that tile, values for the 3dimensional regression components may be set to “nan” where the regressions algorithm failed.
 Floating point precision has been carried forward from the Daymet model for all attributes.
Application and Derivation
The Daymet crossvalidation analysis are used to characterize the sensitivity of Daymet model methods to the variation of parameters and to estimate the prediction errors associated with the final selected parameters. The general crossvalidation protocol is to withhold one observation at a time from a sample, generating a prediction error for the withheld case by comparing with the observed value, and repeating over all observations in the sample to generate an average prediction error. The mean absolute error and bias are the basic error prediction error statistics. MAE does not exaggerate the influence of outliers as would a root mean square error and provides a more robust parameterization framework. Both the absolute value and sign of the prediction are considered in the generation of MAE and bias, respectively.
Quality Assessment
Occurrence of No Data and Not A Number (nan) field values
For tiles that had no input weather stations located within the 2 degree by 2 degree tile processing extent (e.g. nstns = 0), there are no crossvalidation data available. For these tiles, the nodata values are represented with 9999 or "nan" values. For tiles with very low weather station inputs (e.g. nstns <= 3), it is often the case that the 3dimensional regression components calculations failed. In those cases, the regression values are represented with “nan” values in the attribute fields.
Data Acquisition, Materials, and Methods
Crossvalidation Protocol
The Daymet crossvalidation summary statistics are used to test the sensitivity of Daymet model methods to the variation of parameters and to estimate the prediction errors associated with the final selected parameters (Thornton, 1999).
The general crossvalidation protocol is to withhold one observation at a time from the sample, generating a prediction error for the withheld case by comparing with the observed value, and repeating over all observations in the sample to generate an average prediction error. The mean absolute error and bias are the basic error prediction error statistics. MAE does not exaggerate the influence of outliers as would a root mean square error and provides a more robust parameterization framework. Both the absolute value and sign of the prediction are considered in the generation of MAE and bias, respectively.
The mean absolute error for single prediction days, or "daymae" is determined as below:
The bias for the single prediction days is determined as below:
The mean absolute error for the period of record predictions, or pormae, is determined as below:
Version 3.0 Daymet model inputs of spatially referenced ground observations of daily maximum and minimum temperature and precipitation were obtained from the NOAA National Centers for Environmental Information's Global Historical Climatology Network (GHCN)Daily data set from Version 3.22 of the data distribution (Menne et al., 2012). The sparse network of Mexican stations available through the GHCNDaily network was augmented with additional stations provided by the Servicio Meteorológico Nacional of Mexico. These data were obtained through an agreement with NOAA/GHCNDaily in which GHCND acquired the station data directly from the Servicio Meteorológico Nacional and processed this data through the same QA/QC measures as all GHCND data are subject ensuring credibility to the data provenance of these additional data. The additional Mexican data were then provided to the NASA Daymet processing group.
Data Access
These data are available through the Oak Ridge National Laboratory (ORNL) Distributed Active Archive Center (DAAC).
Daymet V3: Annual Tile Summary CrossValidation Statistics for North America, Hawaii
Contact for Data Center Access Information:
 Email: uso@daac.ornl.gov
 Telephone: +1 (865) 2413952
References
Menne, M.J., I. Durre, R.S. Vose, B.E. Gleason, and T.G. Houston, 2012: An overview of the Global Historical Climatology NetworkDaily Database. Journal of Atmospheric and Oceanic Technology, 29, 897910, doi:10.1175/JTECHD1100103.1. http://dx.doi.org/10.1175/JTECHD1100103.1
Menne, M.J., I. Durre, B. Korzeniewski, S. McNeal, K. Thomas, X. Yin, S. Anthony, R. Ray, R.S. Vose, B.E.Gleason, and T.G. Houston, 2012: Global Historical Climatology Network Daily (GHCNDaily), Version 3.22. NOAA National Climatic Data Center. http://doi.org/10.7289/V5D21VHZ February 29, 2016.
Thornton, P.E. and S.W. Running. 1999. An improved algorithm for estimating incident daily solar radiation from measurements of temperature, humidity, and precipitation. Agriculture and Forest Meteorology. 93:211  228. http://dx.doi.org/10.1016/S01681923(98)001269
Thornton, P.E., M.M. Thornton, B.W. Mayer, Y. Wei, R. Devarakonda, R.S. Vose, and R.B. Cook. 2016. Daymet: Daily Surface Weather Data on a 1km Grid for North America, Version 3. ORNL DAAC, Oak Ridge, Tennessee, USA. http://dx.doi.org/10.3334/ORNLDAAC/1328
Data Set Revisions
The ORNL DAAC is publishing Version 3.0 of the Tile Summary Cross Validation Statistics and will update the Daymet V3.0 products annually. Version and change history documentation are provided.
ORNL DAAC Version Record for Version 3.0:
Daymet Product Version 
ORNL DAAC Release Date  Description 

Version 3, Tile Summary Cross Validation  September 30, 2016  ORNL DAAC archived and released Version 3 of Daymet Tile Summary Cross Validation Statistics 
ORNL DAAC Version Record for Version 2.0:
Daymet Product Version  ORNL DAAC Release Date  Description 

Version 2, Tile Summary Cross Validation Revision  July 6, 2016  ORNL DAAC released the final revision of the Version 2 Daymet Tile Summary Cross Validation Statistics 
Version 2, Tile Summary Cross Validation  April 29, 2016  ORNL DAAC released the Version 2 Daymet Tile Summary Cross Validation Statistics 