Comparison of three interpolation schemes for reducing three-hourly North American Regional Reanalysis data to hourly data |
The three-hourly NARR data must be reduced to hourly data for use in a spatial interpolation scheme that uses other hourly observations. The following information shows results of a comparison between interpolation techniques, including a local harmonic analysis, a cubic spline interpolation, and simple linear interpolation. The Asheville Regional Airport provides a continuous hourly time series of all observations between 0000 UTC 1 January 1985 and 0000 UTC 1 January 1986. A subset of these observations every three hours (beginning at 0000 UTC) forms the basis for the interpolation. Scalar accuracy measures compare the interpolated values with the observed values at each hour. Note that each interpolation scheme provides an exact match to the observations at each three-hour time step. Some quality assurance checks include the following: 1) Cloud cover is bound by 0% to 100%, 2) The dewpoint cannot exceed the temperature. Also, since precipitation is a three-hourly total in the NARR data, the interpolation procedure does not apply. The total will instead be spread evenly over the three hours for those data. You may also wish to view graphical summaries. Accuracy measures using only interpolated data points (NOT including the 3-hourly correct points for all schemes)n= 5842 Mean error (bias) Variable Harmonics Spline Linear tair 0.0098 0.0115 0.0116 (K) dewp -0.0278 -0.0143 -0.0062 (K) wspd -0.0469 -0.0470 -0.0469 (m/s) ugrd 0.0039 0.0041 0.0040 (m/s) vgrd 0.0162 0.0155 0.0157 (m/s) prec 0.0399 0.0219 0.0121 (mm) cloud 0.0168 0.1503 0.2846 (%)Mean absolute error Variable Harmonics Spline Linear tair 0.6540 0.5454 0.5953 (K) dewp 0.6224 0.5267 0.5106 (K) wspd 1.1030 1.0284 0.9682 (m/s) ugrd 1.2307 1.1461 1.0725 (m/s) vgrd 1.3501 1.2410 1.1743 (m/s) prec 0.1415 0.1234 0.1147 (mm) cloud 8.5497 7.7891 7.7021 (%)Mean squared error Variable Harmonics Spline Linear tair 0.7906 0.6042 0.7132 (K2) dewp 0.7753 0.5957 0.5802 (K2) wspd 2.0508 1.8019 1.6141 (m2/s2) ugrd 2.6901 2.3884 2.1070 (m2/s2) vgrd 3.2389 2.8060 2.5171 (m2/s2) prec 0.5269 0.4714 0.3983 (mm2) cloud 212.7661 193.0462 187.2902 (%2)Root mean squared error Variable Harmonics Spline Linear tair 0.8892 0.7773 0.8445 (K) dewp 0.8805 0.7718 0.7617 (K) wspd 1.4321 1.3424 1.2705 (m/s) ugrd 1.6401 1.5454 1.4516 (m/s) vgrd 1.7997 1.6751 1.5866 (m/s) prec 0.7259 0.6866 0.6311 (mm) cloud 14.5865 13.8941 13.6854 (%)********************************* Accuracy measures using ALL data points (including the 3-hourly correct points for all schemes) n= 8763 Mean error (bias) Variable Harmonics Spline Linear tair 0.0065 0.0076 0.0077 (K) dewp -0.0185 -0.0095 -0.0041 (K) wspd -0.0313 -0.0314 -0.0313 (m/s) ugrd 0.0026 0.0027 0.0027 (m/s) vgrd 0.0108 0.0103 0.0104 (m/s) prec 0.0266 0.0146 0.0081 (mm) cloud 0.0112 0.1002 0.1897 (%)Mean absolute error Variable Harmonics Spline Linear tair 0.4360 0.3636 0.3968 (K) dewp 0.4149 0.3511 0.3404 (K) wspd 0.7354 0.6856 0.6455 (m/s) ugrd 0.8204 0.7640 0.7150 (m/s) vgrd 0.9001 0.8273 0.7829 (m/s) prec 0.0944 0.0823 0.0765 (mm) cloud 5.6998 5.1927 5.1348 (%)Mean squared error Variable Harmonics Spline Linear tair 0.5271 0.4028 0.4755 (K2) dewp 0.5168 0.3971 0.3868 (K2) wspd 1.3672 1.2013 1.0761 (m2/s2) ugrd 1.7934 1.5923 1.4047 (m2/s2) vgrd 2.1593 1.8706 1.6781 (m2/s2) prec 0.3513 0.3143 0.2655 (mm2) cloud 141.8441 128.6975 124.8601 (%2)Root mean squared error Variable Harmonics Spline Linear tair 0.7260 0.6347 0.6895 (K) dewp 0.7189 0.6302 0.6219 (K) wspd 1.1693 1.0960 1.0373 (m/s) ugrd 1.3392 1.2618 1.1852 (m/s) vgrd 1.4695 1.3677 1.2954 (m/s) prec 0.5927 0.5606 0.5153 (mm) cloud 11.9098 11.3445 11.1741 (%)********************************* |