Presentation on theme: "Interpolation of real-time ozone measurements in Europe Results of a feasibility exercise for the Neighbourhood project Bill Oates"— Presentation transcript:
Interpolation of real-time ozone measurements in Europe Results of a feasibility exercise for the Neighbourhood project Bill Oates firstname.lastname@example.org
Objectives Review of available interpolation method and operational constraints Tests on archive data Tests on real-time data
Review of Methods and Constraints What methods can be used What is the most feasible for EEA Numbers and types of stations: What is the minimum density of stations needed? How many stations per country does this represent? Do we need to make a distinction between rural and urban stations?
Tests on Archive Data Datasets (from AirBase) 12th August 2003: greatest number of stations exceeding the alert threshold 1st April 2003: contrasting, spring concentrations Methods tested Simple – nearest neighbour Interpolation – IDW, Kriging Station selection All Random 3 densities: 1 station per 100x100km / 200x200km / 300x300km Exclusion of urban and roadside concentrations Validation Internal (Jack-knife) and External (test dataset) RMSE
Results (Simple Interpolation) Number of Stations RMSE (ug/m 3 ) Cell Size: 50km 10255.73 15749.75 26148.10 46545.58 Number of Stations RMSE (ug/m 3 ) Cell size (km) 102550 10241.7841.3938.96 15740.4039.7236.74 26137.6135.7534.01 46536.9335.0933.59 Nearest Neighbour IDW
Results (Kriging Interpolation) DAYTIMEDENSITYSELECTION Number of stations Cross- validation RMSE ug/m3 External Valdiation RMSE ug/m3 01 April2 PMall valid pointsall149116.7 01 April2 PM100kno urban no roadside26018.217.5 01 April2 PM100kall31120.117.5 01 April2 PM300kall3720.420.2 01 April2 PM200kno urban no roadside6720.820.2 01 April2 AM300kno urban no roadside3122.124.7 01 April2 AM100kall30322.121.9 12 August2 PM200kall13322.428.8 01 April2 PM300kno urban no roadside3423.018.8 01 April2 PM200kall7623.118.5 01 April2 AM100kno urban no roadside25023.223.0 01 April2 AM200kno urban no roadside12423.624.0 01 April2 AM300kall3423.726.3 01 April2 AM200kall13224.023.1 12 August2 PM100kno urban no roadside26424.625.8 12 August2 PM200kno urban no roadside12426.028.2 12 August2 PM100kall31426.425.1 12 August2 AM200kall12931.336.3 12 August2 AM300kall2631.640.3 12 August2 PM300kall3731.934.0 12 August2 AM100kall30933.633.2 12 August2 AM200kno urban no roadside12136.137.9 12 August2 PM300kno urban no roadside3337.231.4 12 August2 AM300kno urban no roadside3237.238.2 12 August2 AM100kno urban no roadside25738.532.7
Methods and Constraints – Findings #1 Accuracy levels can be improved from 50 ug/m3 to 30 ug/m3 (and better) by selection of improved interpolation methods Kriging techniques delivered highest accuracy results Station number and density has less impact on accuracy than the time of year / day: Optimum station spacing of 200km No consistent pattern from the different type of station included
Methods and Constraints – Findings #2 Geostatistical Analyst methods not available in automation routines No de-trending functions Spatial Analyst methods are available in automation routines Kriging proven as the most accurate from the tests Choice of automatic vs. manual determination of variogram parameters
Real-time data objectives How much real-time data exists? Is real-time interpolation for ozone feasible and practical? What are the accuracy levels? Automated vs. manual determination of variogram parameters?
Methods Manual data scraping from existing AQ sites Existing OzoneWeb stations Interpolation Spatial Analyst within ArcGIS Ordinary Kriging Semi-variogram: Spherical, self-optimising nugget and sill Lag-size: 50,000m Search Radius: variable, 12 nearest neighbours 10km grid
1st September Tests Earlier tests and results discussed with ETC – suggestions for further analysis: Stratify by type of station (NB all background sites) Urban Rural Suburban Run tests within the class of station, and between the classes of stations e.g. Rural stations for interpolation map, test the accuracy at the urban stations 10 Daylight Hours Test #1: Average of the RMSE Test #2: Sum of the Maximum Residuals
Test #1: Average RMSE (ug/m 3 ) The least accurate: rural interpolation – urban test site These can be mitigated for by including suburban sites in additional to the rural sites. This does however slightly decrease the accuracy for the rural sites themselves. Tested Sites Interpolated SitesALLRURALSUBURBANURBAN ALL16.316.417.116.5 RURAL -15.720.523.0 RURAL & SUBURBAN -18.317.616.0
Test #2: Sum of Max. Residuals (ug/m 3 ) Tested Sites Interpolated SitesALLRURALSUBURBANURBAN ALL477386419377 RURAL -442530559 RURAL & SUBURBAN -552517468 Same pattern as the RMSE results
Overall Findings Interpolation of real-time Ozone concentration data is feasible Even from a relatively small number of monitoring stations, Results of acceptable accuracy can readily be generated using the standard interpolation techniques found within the ESRI software selected for the Neighbourhood Project Next steps: Improved accuracy through meteorological parameters Increased resolution through differential interpolation for station groups