تخمین تمرکز ذرات معلق (pm10) در جو با استفاده از داده‌های سنجش از دور ماهواره‌ای و زمین‌پایه و پراسنج‌های هواشناختی: کاربست شبکۀ عصبی مصنوعی

در مقالۀ حاضر، تمرکز روزانۀ ذرات معلق با قطر کمتر از 10 میکرون (pm10)با استفاده از نمایه‌های نورشناخت حاصل از داده‌هایسنجش از دور و پراسنج‌های هواشناختی تخمین زده شده‌ است. برای این پژوهش از داده‌های حاصل از سنجندۀ مادیس (ماهواره‌های آکوا و ترا) و داده‌های دستگاه نورسنج خورشیدی شامل عمق نوری هواویزها (aod)، نمای آنگستروم (α) و ضریب تیرگی آنگستروم (β) و همچنین داده‌های هواشناختی شامل فشار، دما، رطوبت، تندی و جهت باد و داده‌های مربوط به تمرکز pm10 برای دورۀ مطالعاتی دسامبر 2009 تا سپتامبر 2010 منطقۀ زنجان که دارای اقلیمی خشک به‌ویژه در تابستان است، استفاده شده ‌است. مقایسۀ نمایه‌های نورشناخت هواویز در دو فصل تابستان و زمستان نشان می‌دهد که اندازۀ متوسط ذرات و تیرگی جو در تابستان در مقایسه با زمستان بیشتر است. برای تخمین تمرکز pm10 با استفاده از نمایه‌های نورشناخت جو و پراسنج‌های هواشناختی، از دو روش همبستگی سادۀ چندمتغیره و شبکۀ عصبی مصنوعی با توابع پایۀ شعاعی استفاده شده است.نتایج نشان می‌دهد ضریب همبستگی بین مقادیر مشاهداتی با مقادیر پیش‌بینی‌شده برای روش همبستگی سادۀ چندمتغیره و شبکۀ عصبی به‌ترتیب برابر 0/62 و 0/82 است. ازاین‌رو استفاده از شبکۀ عصبی که قادر به پیش‌بینی روابط پیچیده بین پراسنج‌های ورودی و خروجی است، در مقایسه با روش همبستگی سادۀ چندمتغیره، برای برآورد تمرکز pm10مناسب‌تر است.

Estimation of atmospheric particulate matter (PM10) concentration based on remote sensing measurements and meteorological parameters: application of artificial neural network

Suspended aerosols in the atmosphere have strong impact on the global climate. They influence the earth’s radiation budget by scattering or absorbing both incoming and outgoing radiation. Aerosols in troposphere are caused by natural sources, such as dust, seaspray and volcanoes and also by anthropogenic sources, such as combustion of fossil fuels and biomass burning activities and from gastoparticle conversion processes. Those have been implicated in human health effects and visibility reduction in urban and regional areas. In this work, the aerosol optical indices were calculated by using the CIMEL sun photometer i.e. passive measurement. These indices have been monitored during December 2009 to September, 2010, in a semi urban area in the Zanjan region in Iran, which has a continental climate. Aerosol optical depth (AOD) is a dimensionless number that characterizes the total absorption and scattering effects of particles in the direct or scattered sunlight. The value of AOD was measured by means of a sunphotometer in a ground station, located at the University of Zanjan (36.7 N, 48.5 E). The information on the aerosol number distribution was defined by Angstrom in 1929. The wavelength exponent is calculated according to the Angstrom formula. Hence, the wavelength exponent may be calculated from the slope of a linear fit of lnAOD against lnλ. The value of 1.3 for α represents an average value for the mean atmospheric conditions. An empirical relationship between the wavelength exponent and the dominant geometric diameter of the aerosol particles was found by Angstrom. Besides such groundbased observations of AOD, which are pointbased, aerosol optical depth measurements taken by MODIS on board of the Terra and Aqua satellites are used for further analysis. Satellites are able to yield timely information on the atmospheric conditions at the regional and global scales inexpensively. The MODIS sensor onboard the Terra/Aqua Earth Observation System satellites captures the radiative energy from the target in 36 spectral bands over the visible light, near infrared and infrared spectra. The raw imagery has a spatial resolution ranging from 250 m to 1 km at a ground swath of 2,330 km. Standard meteorological variables, such as air pressure, relative humidity, wind speed and direction are also measured at Zanjan synoptic station. Moreover, the concentration of particle mass under 10 μm (PM10) which is measured hourly by the Zanjan environmental protection bureau, is also used. In this study, the relationship between the suspended particulate matter (PM10 ) concentration and aerosol optical indices such as AOD, Angstrom coefficients (α,β) and meteorological parameters such as wind speed and direction, and relative humidity were considered. Two forecasting techniques are presented in this paper for predicting the average hourly PM10 concentration. The first one is the Multivariate Linear Regression (MLR) and the second technique is an Artificial Neural Network (ANN) model, based on Radial Basis Function (RBF). Multiple linear regression models were developed with several sets of data (aerosol optical properties and meteorological data as predictor). The results show that correlation Coefficient between predicted values and observed values for MLR model and ANN model were 0.62 and 0.81, respectively. The impact of wind direction on PM10 concentration prediction is weak in MLR model. The results also show that MLR could not predict PM10 concentration as well as ANN model.