University of Twente Student Theses
Geostatistical analysis of air pollution using models, in situ and remote sensed data
Mweda, Leons Peter (2011) Geostatistical analysis of air pollution using models, in situ and remote sensed data.
|
PDF
3MB |
| Abstract: | Understanding the status of air pollution concentration is of great importance due to severe effects to human health and environments. Different countries in Europe have several monitoring stations that collect information about pollutants continuously in time. These stations use different measurement techniques and different calibration practices are usually applied to measurements. However due to cost implication these stations are limited in space and therefore knowing air pollution at every point in space requires interpolation techniques. Geostatistical methods have been reported in various research to produce accurate maps because they incorporate spatial variability in observations. Recent research have shown that incorporating information that are correlated with in situ observations using appropriate techniques increases accuracy of resulting maps. Country by country PM10 daily annual mean concentration for 2006 were explored to understand effect of combining data from different countries. Two groups of data were explored; combined data from different countries measured by all techniques and combined data from different countries measured by one technique (beta ray attenuation) followed by country to country data exploration. Results shows that there are slightly differences in variability between two groups but relatively have similar spatial structures. It was difficult to obtain reliable spatial structure for some country due to small number of measurements. Geostatistical methods known as regression kriging (RK) and cokriging (CK) were applied to integrate in situ measurements (PM10), models (PM2.5) and remotely sensed data (AOT) to predict PM10 daily annual mean concentration for the year 2003. Accuracy assessment done at validation points has shown that regression kriging (RK) gave better results having lower RMSE equals 0.096 as compared to RMSE 0.099 obtained by CK. RK increased R from 0.40 to 0.71. Comparing to performance of ordinary kriging (OK) and universal kriging (UK), results shows that both RK and UK gave similar results of RMSE (0.096) and correlation (0.72). But RK was less biased as compared to UK. These results were obtained using exponential model. Hole effect model was used in this study due to hole effect emerged on estimated variograms. Hole effect model fitted better the estimated variograms than exponential model at shorter distance but gave poor prediction results as compared to exponential model. RK of PM10 on AOT and PM2.5 using exponential model resulted to RMSE equals 0.096 as compared to RMSE value equals to 0.105 when hole effect model were used. However hole effect model was less biased as compared to exponential model. Change of support was evaluated using universal block kriging. The results showed that lower RMSE were obtained for block sizes less or equal to10 km by 10 km and high RMSE for block sizes greater than 10 km by 10 km. SSE was more sensitive to change of block size as compared to RMSE. Key words: Air pollution, Cokriging, Regression kriging, hole effect model, exponential model, block kriging, universal kriging, ordinary kriging. |
| Item Type: | Essay (Master) |
| Faculty: | ITC: Faculty of Geo-information Science and Earth Observation |
| Programme: | Geoinformation Science and Earth Observation MSc (75014) |
| Link to this item: | https://purl.utwente.nl/essays/92775 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page