Spatial interpolation of Daily temperatures using an advection scheme

1. Spatial interpolation of Daily temperatures using an advection scheme Kwang Soo Kim

2. Outline • Introduction • Weather data • Spatial interpolation • Natural neighbor • Pseudo Advection Scheme • Results • Conclusions

3. Introduction • Disease warning systems depend on weather data • Site-specific weather estimates can be used as inputs to disease warning systems • Site-specific estimates have been obtained using spatial interpolation

4. Sites of Interest • Soil Climate Analysis Network (SCAN) of National Resources Conservation Service (NRCS) • Distributed over the USA. • Represent various climate conditions. • 53 % of SCAN stations were established after 2002. • 64 sites were included as validation sites

5. Neighbor stations • The US National Climate Data Center (NCDC) Global Surface Summary of the Day (GSOD) database • Daily temperatures and precipitation • Long term weather records are available • Free access to the public

6. Site distribution • On average, about 1600 neighbor stations were used for spatial interpolation

7. Spatial interpolation products • DAYMET • Daily estimates of weather variables • Truncated Gaussian filter • From 1980 to 2003 • PRISM • Monthly estimates of weather variables • Various covariate variables are used • Both products are based on National Weather Service (NWS) Cooperative Observers Network stations

8. Voronoi tessellation

9. Delaunay triangulation

10. Natural Neighbors Watson, 1999

11. Conventional VS Natural neighbor interpolation

12. Local coordinates Sukuma, 2003

13. Pseudo Advection Scheme • Natural neighbor interpolation can be used to solve the partial differential equation (Sukuma, 2003). • Advection scheme can be used for spatial interpolation • ∂y/∂t + u·y = 0

14. Daily minimum temperature Environmental Lapse Rate Empirical Lapse Rate R2 = 0.895 Y = 1.07 + 0.914 X RMSE = 3.55 R2 = 0.889 Y = 0.61 + 0.920 X RMSE = 3.60

15. DAYMET R2 = 0.917 Y = 1.25 + 0.93 X RMSE = 3.20

16. PAS Environmental Lapse Rate Empirical Lapse Rate R2 = 923 Y = 0.35 + 0.953 X RMSE = 2.96 R2 = 0.926 Y = 0.86 + 0.942 X RMSE = 2.96 17% 18%

17. Conventional VS PAS

18. Monthly temperature Environmental Lapse Rate Empirical Lapse Rate R2 = 0.963 Y = 0.96 + 0.927 X RMSE = 2.00 R2 = 0.963 Y = 0.43 + 0.943 X RMSE = 1.90

19. DAYMET and PRISM R2 = 0.971 Y = 0.99 + 0.946 X RMSE = 1.80 R2 = 0.970 Y = 1.21 + 0.933 X RMSE = 1.92

20. PAS Environmental Lapse Rate Empirical Lapse Rate R2 = 0.964 Y = 0.96 + 0.927 X RMSE = 1.99 R2 = 0.963 Y = 0.43 + 0.943 X RMSE = 1.89

21. Conventional VS PAS

22. Conclusions • PAS can improve accuracy of site-specific estimates of daily minimum temperature • Natural neighbor interpolation can provide accurate estimates of monthly weather variables • Accuracy of PAS/Natural Neighbor method can be improved when COOP data are used

23. What’s next • Sub daily interpolation of weather data using PAS • Advection scheme is better suited for sub daily data

24. Thank you! • Questions? • luxkwang@yahoo.com • kwang.kim@plantandfood.co.nz