AYBE- İklim ve Deniz Bilimleri Lisansüstü Programı - Doktora
Bu koleksiyon için kalıcı URI
Gözat
Konu "Atmospheric models , Atmospheric pollution , Regional transport model , Gas emission , Air pollution ,Sulfur dioxide emissions , Spatio-temporal modelling , Environmental pollution , Environmental uncertainty, Multivariate statistic" ile AYBE- İklim ve Deniz Bilimleri Lisansüstü Programı - Doktora'a göz atma
Sayfa başına sonuç
Sıralama Seçenekleri
-
ÖgeQuantification of the impact of uncertainty in emissions on air quality model estimates(Eurasia Institute of Earth Sciences, 2019-12-20) Özdemir, Ümmügülsüm Alyüz ; Ünal, Alper ; 601122007 ; Climate and Marine Sciences ; İklim ve Deniz BilimleriThe Air Quality Model, especially Chemical Transport Model, prediction represents mean concentration over the entire grid volume. Predictions of CTMs may differ from observations due to four reasons; 1) inherent or stochastic variability in the observations, 2) errors in model physics and chemistry assumptions, 3) errors due to uncertainties in model input variables, and 4) numerical errors. Here, variability is a description of the range of spread of the values, and it is often expressed by statistical metrics such as variance and standard deviation. Therefore, inherent uncertainty can be considered as variability. Uncertainty refers to lack of knowledge regarding the true value of a quantity. Uncertainty can be reduced or eliminated with more or better data, where variability cannot be reduced. Among the four reasons of uncertainty, provided above, inputs are regarded to have the largest levels of uncertainty. The aim of this study is to evaluate and quantify the contribution of uncertainties in input dataset to AQM estimates. For this purpose, it is necessary to define the problem that poor performance of the model is caused mostly by unfit data. In literature, models perform poor in the Eastern European countries. However, a more detailed study is needed to say that this poor performance is mostly due to model inputs. Because, as it is known, the poor performance of the models may also have other reasons. In the first part of this study, inter-model variability is defined quantitatively by participating in an international project. In the second part of the study, contribution of uncertainties to this problem is quantified by being part of a national project. In the second part, a sample of the solution is presented which includes development of country specific emission factors and compiling a probabilistic emission inventory. As a part of an international project (AQMEII-3), 12 modelling groups were cooperated from different countries of Europe and conducted 18 model runs on Europe domain (covers 34 Europe countries) for 2010 by using 7 different AQMs, 3 meteorology models and 2 emission inventories. This study, for the first time in Turkey, contributed to AQMEII-3 which is organized by the joint leading of U.S. EPA and European JRC. One of the most important benefits of this project is that the model results of all groups can be reached through a common platform. In this dissertation, performance metrics were calculated and mapped for each of 1431 stations of Europe, and for each model for evaluation of model performances. Taylor diagrams were also used for seasonal evaluation. Up to now, there are several air quality modelling studies for Turkey, however they are developed for a specific city or region of Turkey for a timescale starting from days to a few months, or by using just one type of AQM. Thanks to its wide coverage domain (Europe continent) and multi-model contributions from AQMEII-3 project, this study looks to the problem from a large perspective in order to define the problem and recommends a solution by representing a sample of the solution. Thus, an inventory study was conducted to overcome this problem by adopting a deep statistical approach which is not encountered in Turkish inventory studies yet. To this end, country specific EFs are calculated for the energy production industry of Turkey, an inventory has been created for the energy production industry of the Marmara Region. Monte Carlo and Bootstrap approaches are used for uncertainty calculations at these stages. According to results of modelling part of this dissertation, correlations between models and PM10 observations are 8% less in Eastern European countries when compared to Western European countries. BIAS of Eastern European countries is 2.5-fold of Western European countries, when all countries are considered. RMSE of Eastern countries is 90% more than Western countries average, where MAE is 99% and MNE is 25% more. From these results it is clear that, model predictions are significantly beyond the observations in Eastern European countries. Turkey, which is located in the Eastern Europe, has one of the worst results calculated by all models. All models predict PM10 concentrations with an average of -40 ug/m3 BIAS in stations of Turkey, where it is the worst value within 34 countries of Europe considered in this study. Moreover, models predict close to each other but quite far from the observations in 80% of the stations. MAE is over 20 ug/m3 in 80% of all stations in Turkey. Remaining 20% of the stations encounters 18 over 101, mostly in Istanbul and some other big cities. In fact, when the results of the models are examined, it is seen that models generally make better predictions in big cities compared to the small cities. This may be due to the fact that inventory compilers have more information on emission sources in large cities. In seasonal evaluation, it is seen that emissions in Winter cannot be well predicted, but in Summer it is relatively better predicted. This difference can be caused by inadequate representation of increased emissions (in the model inputs) in Winter months from residential heating and traffic emissions when compared to other months. In this case, it would not be unreasonable to suspect that the inputs to the models significantly affect predictions. Model inputs are considered as a reason for poor model predictions in this study. However, problems caused by the model itself or erroneous measurements, or combination of all, may also cause this. In this study, problems due to the model itself are out of consideration since 6 different AQMs were used by 13 modelling groups where same models were also considered by different groups. The fact that all models give close CDFs in Western Europe despite they have different modelling configurations, where they are not close to observations in Eastern Europe countries even in same models, shows that problems in the models are not dominant in prediction errors. Since the number of observation stations included in the scope of this study is very high, measurement errors are not considered to be predominant in poor model estimates. Also, systematic errors are not thought to occur at all stations at the same time. The quality of an emission inventory that will be used in air quality modelling is associated with its low-level uncertainty and adequate coverage of the sources. Emission inventories approach to the ultimate result as in-situ measurements and full activity data are available. In this study, in-situ measurements were conducted within the scope of the national KAMAG project in order to generate country-specific EFs, and an emission inventory was prepared in the light of the most consistent information possible. Besides, official emission measurement reports (EMRs), whose reliability is controversial as they were prepared by the companies under authorization of the emission emitting plants, were also used for comparison with in-situ EFs. Country-specific dust, CO, SO2, NO, NO2 and NOx EFs are calculated in this part of the study for each of coal combusting large wet/dry bottom boilers, coal combusting large size fluid bed boilers, coal combusting large wet and dry bottom boilers, natural gas combusting medium size boilers and gaseous fuels combusting gas turbines. EFs are typically assumed to be representative of an average emission rate from a population of pollutant sources in a specific category. However, there may be uncertainty in the average emissions from population because of three reasons: random sampling error, measurement errors, or when the sample population is not representative for EF development. First two factors typically lead to imprecision in the estimate of the population average. The third factor may lead to possible biases or systematic errors in the estimated average. In order to avoid errors, it is important to understand and account for the uncertainty in the inventory. In the relevant part of this study, a probabilistic emission inventory is developed by considering statistical analysis of variability and uncertainty. The development of a consistent procedure for the uncertainty evaluation is still a challenge for the scientific community. In this study a deep uncertainty analysis technique is applied in EF development, which is including Monte Carlo method and Bootstrap simulation. The uncertainty analysis described in this study can be used as a basis for developing probabilistic emission inventories. When the probability range of emissions to be given as input to air quality is known, it is possible to determine the probability of the model result. Thus, for example, the probability of achieving an air quality management goal can also be calculated. In statistics, sampling error is a type of error caused by investigating a small part of the population rather than examining the whole population. It is calculated by the difference of a sample statistic used to estimate a population parameter and the actual but unknown value of the parameter. Since uncertainty is expressed as lack of knowledge regarding to true value of a quantity, random sampling error can be represented by a sampling distribution. In order to calculate uncertainty of EFs, a distribution is fitted (F^) to the EF dataset (x) where actual underlying distribution (F) is unknown. The goodness-of-fit is evaluated by some techniques. Then Monte Carlo method is applied in order to generate random datasets from assigned distribution, F^. In Bootstrap simulation part of the study, each of the alternative probability models generated by Monte Carlo approach (Bootstrap replicates) are simulated to develop a reasonably stable characterization of the percentiles of the distribution. Then parameters, θ^*, are estimated. In this study, uncertainty in the estimate of θ is reflected by dispersion of θ^*, which also gives random sampling error. A confidence interval for a statistic is a measure of the lack of knowledge regarding the true value of the statistic. The θ^* data is sorted then, in order to calculate confidence interval for the fitted cumulative distribution function. Consequently, the results are compared to the original dataset by generating probability bands. Then results are compared to EMEP and EPA EFs. At the end, dust EFs obtained from in-situ measurements are significantly lower than the literature for coal combusting plants. The reason of these large differences between in-situ measurements and literature EFs may be due to wide usage of dust abatement technologies in Turkish energy production plants. CO and SO2 EFs are significantly larger than EMR, EMEP and EPA EFs in large coal combusting plants and in plants combusting gaseous fuels with gas turbines. But in all EFs, uncertainty is low when compared to EMEP EFs. Country specific NOx EFs are generally larger than all other studies and range of confidence interval is narrow when compared to them. This situation indicates low uncertainty in in-situ EFs. Since each stack measurement may differentiate from the real value due to variations in operating conditions, the overall uncertainty of the emission factors can also be referred as "uncertainty due to variability". After calculating country specific EFs, next step is preparing an emission inventory for power plants of Marmara region and comparing it with the existing emission inventories. The most common emission inventories currently used by CTMs are the TNO-MACC and EDGAR-HTAP emission inventories. These two inventories are mainly used in AQMEII-3 models. EDGAR-HTAP emission inventory contains much more plants (34 plants) than TNO-MACC (19 plants) but is still far from the actual number of power plants (57 plants) that considered in this study for Marmara region of Turkey. Furthermore EDGAR-HTAP emission inventory has more plants than TNO-MACC in all regions of Turkey. From this point of view, it is clear that EDGAR-HTAP emission inventory is more inclusive than TNO-MACC emission inventory in Turkey in terms of number of plants. Also, it is more inclusive in Eastern Anatolian regions of Turkey where TNO-MACC emission inventory has almost no plants for public electricity and heat production sector. There are missing plants in EDGAR-HTAP and TNO-MACC emission inventories where there some unidentified plants in those emission inventories. As a result of emission inventory calculations, NOx emissions calculated in this study is 93,000 ton/year with lower CI as 69,000 ton/year and upper CI as 114,000 ton/year. When same emission inventory is calculated with EMEP EFs 60,000 ton/year with lower CI as 33,000 and upper CI as 90,000 ton/year. The inventory compiled by this study beyond the upper CI of EMEP and it is considerably larger than TNO (24,000 ton/year) and EDGAR-HTAP (42,000 ton/year). SO2 emissions are calculated as 152,379 tonne/year in this study. Same activity data is used in calculation of EMEP emission inventory and resulted 170,596 tonne/year, because in-situ SO2 EF was smaller than EMEP EF for coal combustion plants. It is 69,000 ton/year in TNO and 125,00 ton/year in EDGAR-HTAP emission inventory. 4 large lignite combustion plants, which are not included in the TNO inventory, have resulted in 73,500 tons less SO2 emissions in TNO emission inventory when compared to this study. 1000 tonnes of SO2 emissions is also not included in the TNO inventory due to about 40 missing natural gas incineration plants. Uncertainty range of NOx emission inventory of this study is between 26 (lower bound of CI) to 23% (upper bound of CI). When same emission inventory is compiled with EMEP EFs, overall uncertainty range is 45 (lower) to 48% (upper). As it is clear, country specific EFs decrease uncertainty when compared to usage of EFs from literature. This situation is dominant in NOx emission inventory than SO2 and CO emission inventories, because number of natural gas combusting power plants are large (48 over 57 plants in Marmara region). TNO and EDGAR HTAP emission inventories are out of the uncertainty range of this study which proves their inadequacy for representing emissions of power plants in Marmara region. Generally, the data on energy facilities is among the most easily accessed by inventory compilers. Such large differences in emissions from power plants reinforce doubts about the reliability of the entire TNO-MACC and EDGAR-HTAP emission inventories. In this case, it is quantifically proved that poor emission inventories are primarily responsible for the poor air quality predictions in Turkey, and most probably in all Eastern European countries. No matter how many and high-quality measurements are conducted, no matter how good models are used, it is not possible for air quality models to predict accurate results without a good emission inventory. Therefore, consistent, low uncertainty and comprehensive emission inventories should be compiled for the Eastern European countries, including Turkey. Development country specific EFs is the preliminary step of emission inventory development. Access to activity data used in these studies should be facilitated in order to make room for calculation of the representative EFs easily.