Updates on aerosol sampling methods

Since last last post I've made significant strides in improving the hourly aerosol prediction methods. That has been, in a nutshell, my job for the past two weeks: interpret aerosol scattering measures. Since my collected data has been thoroughly digested, I thought now might be an opportune time to regurgitate what I've done.

Airport location

First, the equation given below is currently how I'm assessing changes in PM2.5 levels:

M_PM2.5(t) = <PM2.5>*Rel bscat,p

where  Rel bscat,p = [b_sp/<b_sp>]/[b_sp,calc/<b_sp,calc>]


and b_sp,calc is determined using two separate methods, either empirically via a line of best fit reported by the IMPROVE team or from semi ab initio calculations done by a fortran program called ISORROPIA.


This light scattering-based  equation is also my guiding light. The first term <PM2.5> is the averaged BAM (beta attenuation monitor/mass) filter results. These filters come from BC's Abbotsford airport, as measured via Environment Canada's Met-One BAM 1020. EC's methodology can be found here.


The second part of the equation, Rel bscat,p, compares the amount of predicted nephelometer light scattering of PM2.5 aerosols with a reported value. If only the relative humidity changes though out the day (i.e. PM2.5 dry mass remains constant) then both numerator and denominator (b_sp and b_sp,calc, respectively) should fluctuate in tandem, assuming I got the equation right. If dry mass changes over the course of the day, then those changes should be reflected in the numerator changing more than the denominator causing an inflection with respect to the mean PM2.5.

I normalized both sets of data with their 24-hour means so that a systematic over prediction of bscat would cancel out.

The relative changes in bscat due to relative humidity changes are accounted for using empirical hygroscopic factors, i.e.

f_inorg(RH) = −0.18614 + 0.99211(1/(1 − RH))
f_org(RH) = 1 + 0.1/(1-RH)

and b_sp,calc is calculated as

b_{sp, IMP calc} = 3*f_inorg(RH)*[ASO4 + ANO3] + 4*f_org(RH)*[POM] + 10[LAC] + [Soil]

Or, if using the ISORROPIA approach (which we are also doing), we have 

b_{sp,ISO calc} = 3*[total calculated inorganic and water mass] + 4*f_org(RH)*[POM] + 10[LAC]

Clearly both methods rely on a semi-empirical organic equation; the only way of incorporating organics into ISORROPIA would be to effectively include this equation in the code, but unless we assume chemical interaction between organics and inorganics (thus opening a fresh can of worms), then it's better to just use f_org(RH) as given above.

Both scattering equations rely on similar -but not identical- inputs. To calculate b_{sp, IMP calc} we require aerosol sulfate and (Teflon-captured) nitrate concentrations SO4 and NO3 and total aerosol ammonium [NH4]. Ammonium nitrate (ANO3) and ammonium sulfate (ASO4) are then obtained implicitly via

[ANO3] = [NO3] + 18/62*[NO3]

[ASO4] = [SO4] + [NH4] - 18/62*[NO3]

Total organic carbon is measured and converted to total organic matter. LAC is identical to elemental carbon and soil is given in EC's data sheets (see list of assumptions below)

To calculate b_{sp, IMP calc} we require total nitrate (gaseous as well as aerosol) and total ammonia/ammonium. We need total sulfate as well, but assume this is identical with aerosol sulfate. The ISORROPIA model then requires knowledge of the ions Cl, K, Ca, and Mg (these may be high near coastal regions).

A few other assumptions made in both systems (ISORROPIA and IMPROVE)

1. No change in particle density. This isn't a perfect assumption since wet particles tend to weight more than dry ones (pure water has a density of 1.0 and sulphuric acid is 1.8; real aerosols lie between these values and approach unity as RH increases).

2. Particle count remains constant. In other words the number density does not change as particles grow in size (which leads to more surface area). This is a reasonable assumption over short time scales, however in time particles can conglomerate.

3. No change in index of refraction. I had mentioned before that b_ext  ~ [(m² - 1)/(m² + 2)]² so that changes in m (complex index of refraction) could have a strong effect on the total scattering. On short time scales we expect these changes to be slight enough to ignore.

4. Other forms of scattering are relatively minor and can be tentatively ignored. That is, b_sp,calc is only based on organics and inorganic material,  light-absorbing carbon LAC (aka EC) and soil.  We neglect changes in gaseous NO2 as it affects aerosol composition.


5. As recommended by Environment Canada, total particulate organic matter (POM) is obtained by multiplying the measured organic matter (OM) by 1.9.


6. Also recommended by EC, we assume aerosol soil remains constant at 0.45 ug/m3 (based on their speciation measurements). Soil could alternatively be estimated from four major soil components

[Soil] = 2.2[Al]+2.49[Si]+1.63[Ca]+2.42[Fe]+1.94[Ti]

Finally, some results:

A single day of replicated nephelometer scattering. The red line is an attempt to mimmic the blue line. The relative differences are taken to be changes in actual PM2.5 levels (overlaid in this plot).

This is how well the hourly PM2.5 is mimicked on this particular sampling day in Abbotsford (May 31, 2009)

Hourly PM2.5 sampling error averaged for matched hours from 35 sampling days spread over one year. No special hours of day in which errors increase or decrease

From the graph above you can tell there is little difference between either the purely empirical or semi-empirical approach (which is actually a good sign). The only issue is that Abbotsford isn't much of a stress test for our simulations; the weather is fairly mild and temperatures vary little year-round. As can be seen below, errors remain under control for any humidity below 96% (empirical bscat is not reported for 96-100% RH, hence variation above this threshold is unknown). 

The SPARTAN program is designed for coastal environments, so that means very hot/cold weather swings are not a significant concern. Yet I'm still hoping to see what happens in dry, dusty places like Saharan landscapes or with wild temperature swings in Siberian-type landlocked zones.