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

Spatial interpolation of Daily temperatures using an advection scheme

Kwang Soo Kim

Outline

Introduction Weather data Spatial interpolation

Natural neighbor Pseudo Advection Scheme

Results Conclusions

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

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

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

Site distribution

• On average, about 1600 neighbor stations were used for spatial interpolation

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

Voronoi tessellation

Delaunay triangulation

Natural Neighbors

Watson, 1999

Conventional VS Natural neighbor interpolation

Local coordinates

Sukuma, 2003

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 ∂/∂t + u· = 0

Daily minimum temperatureEnvironmental Lapse Rate Empirical Lapse Rate

R2 = 0.895Y = 1.07 + 0.914 XRMSE = 3.55

R2 = 0.889Y = 0.61 + 0.920 XRMSE = 3.60

DAYMET

R2 = 0.917Y = 1.25 + 0.93 XRMSE = 3.20

PASEnvironmental Lapse Rate Empirical Lapse Rate

R2 = 0.926Y = 0.86 + 0.942 XRMSE = 2.96

R2 = 923Y = 0.35 + 0.953 XRMSE = 2.9617% 18%

Conventional VS PAS

By Site

Gest VLR DM GestAdv VLRAdv

RM

SE

(oC

)

1

2

3

4

5

6

7

Monthly temperatureEnvironmental Lapse Rate Empirical Lapse Rate

R2 = 0.963Y = 0.96 + 0.927 XRMSE = 2.00

R2 = 0.963Y = 0.43 + 0.943 XRMSE = 1.90

DAYMET and PRISM

R2 = 0.970Y = 1.21 + 0.933 XRMSE = 1.92

R2 = 0.971Y = 0.99 + 0.946 XRMSE = 1.80

PAS

R2 = 0.963Y = 0.43 + 0.943 XRMSE = 1.89

R2 = 0.964Y = 0.96 + 0.927 XRMSE = 1.99

Environmental Lapse Rate Empirical Lapse Rate

Conventional VS PAS

Gest VLR DM PM GestAdv VLRAdv

RM

SE

(o C

)

0

1

2

3

4

5

6

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

What’s next

Sub daily interpolation of weather data using PAS Advection scheme is better suited for

sub daily data

Thank you! Questions?

[email protected] [email protected]