Lateral thinking. From Micro to Macro. (by Diego Fdez-Sevilla)
I am 40 years old. Since I was a child I have been always fascinated by understanding the world around us. Actually, more about how it works than how many names I can remember to name things. I began my career studying general Biology in my Bachelor degree to understand how life behaves and why. Then I carried on into Environmental Biology with a Mater’s degree to look at the dominant forces creating environments in which life develops. From there I moved into the field of Aerobiology, where in my thesis I looked into the methodologies applied to identify and understand the meaning of the presence of biological information carried in our atmosphere. Since then (2007), I have kept developing skills and broadening my capacity to identify interconnections between the players running our environment.
One lesson learnt has been the limitations on any assessment due to the implications of the perspective applied. This happens in environmental studies as well as in market research and policy design. The events happening at micro-scales are closely linked with those happening at macro-scale. Many times, even the same patterns emerge in both worlds. And the behaviour of one can not be assessed without the consideration of the other.
The definition of a micro or macro environment comes with the perception adopted. If we can not see the micro-world with our eyes then we need a microscope. And when we can not observe the macro-world with our eyes, we need a telescope. However, for both apply the same principle of perspective when we want to observe them. Distance. If it is too small from our distance we need to come closer through magnification. If it is too big from our distance, we need to move away to see it, e.g. sending satellites to observe the Earth.
Both systematic approaches (micro and macro) generate images which allow us to be the observer at the right distance to “see”. And then, the observer just need to know what it is that he/she is looking at to adapt the same skills and training to identify and analyse the image in front.
Now, with the era of satellite data, it seems just part of a natural growth, to take all lessons learnt and incorporate them into analysing the potential of all those new instruments carried by satellites. Their information could well help to integrate the world of the microcosmos in the macro-environments in which they exist.
Both, micro and macro observation, look at features as well as at the background around them. One field of study which relies strongly on microscopic observation is called Aerobiology. In this field it is studied the concentration of particles of biological origin that gets airborne in the atmosphere. In order to identify those particles of interest, we have to be able to differentiate them from the background in the sample, to observe the features that makes them different from other particles and even to identify the characteristics that makes them singular within the same family.
For some particles like pollen grains, we can see them as little planets through a telescope, on which, some superficial features are quite similar to features on planet Earth.
3D rendering of A. repens pollen imaged using wide-field microscopy. The first part is a series of optical sections, and the second part is a 3D projection of all slices after blind deconvolution. (more here)
Holes and pores are like volcanoes, and furrows can look like the borders of tectonic plates. Similarly as when looking at the surface of pollen grains, satellite imagery can show wrinkles in the topography of the Earth as well as patterns in the land surface configuration and distribution due to the roughness and geometry of land cover, use and topography.
In micro and macro observation, we look for ways to differentiate all of those features from their surroundings like when it is being done with topographic features, cyclonic events, land and sea surface covered by ice or snow, patches of vegetation and the different species within the same patch, etc.
There are several techniques which can be applied. Biological particles like Pollen grains, can be analysed by microscopy with white light using the set-up of mechanical elements such as a condenser, a diaphragm, etc and variations in your point of focus to obtain optical optimization and look through different depths obtaining a 3D composition. Also we can polarize the bean of light or use dark-fields enhancing contrast with the background regulating the angle of incidence from the light bean. Furthermore, biological tissues have the unique particularity of fluorescence under certain wave lengths that makes them stand up out from the rest of the material in the sample.
Similarly in microscopy as with satellite observation, images are compared based on the trained eye and skills of the observer, in order to identify and analyse similarities and differences.
Satellite observation applies techniques which can be based on optical imagery or wavelength spectrum analyses. And as well as with microscopy, the measurements are affected by the conditions of your sample; the thickness of the atmosphere, atmospheric conditions or the characteristics of the material under observation: water, land, ice and snow, gasses, plants and aerosols. Filtering the relevant part of information from the interference of other sources which interact with the depth of field and the accuracy of the measurements is a challenge.
Looking at the steps followed by the technological developments applied in satellite observation, it seems very probable from my experience that the next steps will move towards incorporating automated image analyses systems into the data being obtained.
Relying on the efficacy of digital measurements, algorithms and automated analysis systems to define and interpret natural variability carries some risks.
Therefore I want to try to make a point on what could be seen as a first approach to make an assessment over the limitations carried within applying automated analysis systems into satellite observation. And for this purpose I want to use some comments which I have previously made on assessing the limitations of applying automated processing on “Microscopy Particle Analysis”.
Can Advance Microscopy Particle Analysis replace other techniques as a primary technique of Particle Distribution and analysis?
I have some experience from trying to help setting up an digital automated particle counting system for pollen grains from microscopy images and also I used a different particle counting devices. Applying microscopy for an automated system required for the image analyses system to generate a dataset to be applied by the system in order to create a parametrization of those characteristics that define the identity of the particle to be counted. So the system takes one image from your sample and compares it with the database in order to identify your sample. But also, it required to create a database from images with nonpollen particles that could be used to discard untargeted particles. Furthermore there was the problem of interferences in the sample from the mounting media used to capture the particles. If you use fiber filters, they give you the best performance in reducing resistance against the flow rate and surface pressure but the fibers from the material can interfere with the image used by the system to identify the particles. If you use membrane filters or other type of impaction based media for your samples, you can have clusters of the same particle type and particles of different nature stick to each other making it difficult for the image system to recognise what it is.
In a different study I also used a particle size counter (don´t remember the name) using suction making particles go across a beam of light, in lab conditions. The measurements were aimed to characterise the range of sizes for pollen grains of a single pollen type from the same sample. There were some particles with a size too big to be single pollen grains giving some bias in the results. I identified those as clusters by analysing the sample microscopically. In a different study I measured the number of pollen grains contained per gram for two samples of the same family but different species of pollen. There was also a variability inside each sample due to a small range of sizes for the pollen grains from each specie but a significant difference between both species. You can only perform such observations applying microscopy, or combining microscopy and automatization.
About particle distribution. Through my thesis I worked with the German Weather service in Freiburg. There used glass fiber filters for Total Particle Mass measurements by weighting filters before and after sampling. I performed a wind tunnel study aimed to observe particle distribution across glass fibre filters for particle sampling using pollen as reference. The filter faced downwards (flow rate and the design of the outlet of the holding head it was supposed to collect particles homogeneously in the whole surface of the filter. Only by optically analysing under the microscope the distribution of the particles I discovered that particles were concentrated at one side of the filter following inertial movement while dragged into the filter. That means the inertial momentum of the particle created a bias in the measurements due to wind speed conditions and sampler design.
Why not to make microscopy method as primary technique? Because is highly time consuming and very dependant on technical training. So from my point of view, it all depends on what is that you are aiming for to obtain from your measurements and the margin of error you can assume.
I believe things are improving very fast so it is a matter of being updated about the new developments but knowing their limitations.
I am not sure my experience can be useful for your specific query but I thought it was worthy to share it. I hope it helps.
There is also the optical quality of your sample and the type of lightning technique you can use. I used Fluorescence with pollen grains since can be used with biological particles to highlight them in the visual field against dark background and inorganic particles. When I worked with inorganic particles, I used dark field microscopy improving resolution eliminating light scattering.
As David said, it all depends on the type of particle and the goal of the analysis, and I would add, the source of the sample. Samples from indoors and from outdoors might be treated differently. From my experience, in wind tunnel experiments and lab studies, the sample released for the sampling method to catch, conserve its purity against other particles. When outdoors, the sample gets contaminated by other factors which need to be considered. Even the characteristics of the particle might change outdoors due to meteorological conditions, physically and even chemically. Pollen grains shrink to different sizes and shapes in dry conditions from when fully hydrated.
Use of microscopy and the limitations of dimensionality
Just to add something about the use of microscopy and the limitations of dimensionality from my experience, In pollen identification (from 10-100 µm) there are many features which need to be identified to recognise the specie like shape and size but also, apertures e.g. pores and furrows, and the surface of the grain can also have a meshed, granular, grooved, spined or striated surface or can appear very smooth. The position of the pollen grain in the sample makes recognition challenging. Not all pollen grains are spherical, some are oval or disc-shaped, some are like rugby balloons. The position of a non spherical object projects different shapes in a 2 dimensional observation but furthermore, in pollen recognition, sometimes, the features you are studying are in different levels of depth in your visual field. So you have to be all the time scanning up and down all levels covering the whole grain. When trying to address same problem with an automated system, the answer was to replicate the same process of scanning. Meaning, the dataset created for each pollen type was build taking pictures at several levels creating a 3D image of each type of pollen grains and in different positions. And the process of automation was set with a mechanical system incorporating an stage which moved the up and down the focusing level of the lent creating multilevel 3D images.
I am describing the challenges for pollen identification because they are the ones I have more experience working with and there are similarities with other particles in those challenges . I have worked using transmission light microscopy with particles like pollen grains, dust mites and cellulose fibres, with embedding mediums of different refraction indexes like Immersion oil, gelatine or solar matrix, and with different mounting media support for the particles sampled like fibre filters, membrane filters as well as plastic impaction surfaces with active and passive samplers. Most of this experience comes from applying all those techniques to study the aerodynamic properties of pollen grains and their impact in sampling efficiency based on settling speed calculations performed by combining concentration (active methods) and deposition (sedimentation) measurements.
I had to use microscopy because in my research I had to be sure that the number of particles being counted for a determined size had to be the same “particle type” from all instruments. The accuracy of my results made it mandatory. In both cases, outdoors and in the wind tunnel. A particle sizer would count any object or cluster of the same size without differentiating the type of particle. For me, using microscopy, I had to learn to adapt all my settings to different visual challenges even when analysing across the same sample, changing depth of field with the diaphragm and condenser of the microscope, angle of light in the field and sometimes even using fluorescence filters. But as I said, microscopy was the most accurate method to be sure that the numbers applied to calculate concentrations and deposition rates applied for the same particle type despite being surrounded by others of similar size.
In the following link you will find the work published by the people I worked with in Freiburg.
* Invariants for Automated Pollen Recognition Olaf Ronneberger. (2007)
At risk of taking the subject sideways I would like to incorporate some points here. The fields addressing particle distribution are many and broad in applications. Taking my experience as Biologist and as Aerobiologist I have faced some of these situations in which everything depends on the interpretation you want to make of your measurements and data.
When charactering the data distribution of a sample from a unique type of particle which can be in different sizes and forms, you don´t need to see the particle because you know the origin and the properties associated so you can focus on particle distribution with a particle sizer. E.g. if I want to characterize the particle distribution of carbon particles from combustion being liberated by car exhausts. From this characterization of particle distribution you can identify health risk due to aerodynamic behaviour when entering the lungs. Same with studies aimed for particle exposure in Environmental Health Risk assessments indoors (e.g. quality of air inside factories). However, once you have characterized your sample under controlled conditions, if you want to integrate your results into “uncontrolled” environment you will have to identify also which other particle types are sharing same aerodynamic behaviour as the particle you are trying to aim which can add bias to your measurements. In environmental policies being implemented for Air Quality assessments outdoors and indoors, I have always defended that particle size distribution is not enough without identification. European Commission applies particle PM2,5 and PM10 only basically because they base their assessments in the characterization of particle size and aerodynamic behaviour identifying as health hazards only those particles being originated, in principle, within those ranges of size <2,5 and <10 microns. With no consideration for particles, biological and none biological, of bigger size but breaking or liberating smaller particles when exposed to the elements (e.g. pollen).
In another hand, aerosols play a role in clouds formation due to their drop nuclei activity. You can find aerosols of inorganic and organic nature, biological and non-biological, playing this role with similar aerodynamic behaviour, therefore, particle size distribution without image identification will not give you enough information to make an accurate assessment.
Basically I am trying to say that particle size distribution in an open environment gives quantity but has many limitations assessing the quality of your measurements. And last, comparing measurements from different sites based on particle size distribution alone, without optical characterization of the particle types under study, imply assuming limitations. In the coast there are particles being measured from salt crystals with a size distribution you will not have in locations inland, but rural areas give you particles from agricultural activities. You could compare measurements from both sites only with particle size distribution and, from the point of view of quality assessments, you would not know what are you comparing. You need microscopic identification also.
I know there are other applications in which these questions are not a matter of concern so I hope I haven´t taken my points too much off topic.
About using “dynamic imaging as a counting method”.
I remember somebody I met whom wanted to use a software for digital image analyses looking at the homogeneity of bubbles in a sample of foam. At conferences treating EHS I met somebody talking about controlling air quality and health environmental systems being developed to monitor concentration of nano particles being produced by the introduction of new materials in the market. And also, I have worked in studies looking at the relationship between asthma and atmospheric particle load with people looking into PM2.5 and PM10 as a whole. They only measured quantity using black carbon measurements with spectrophotometry.
So, for me, the limitations and applicability of any instrument or methodology, and the measurements obtained from them, have to be “considered” with the understanding of some questions:
What is that we want to characterize?. Particles defining “a product” or “an environment”.
Are we looking at Quantity or Quality?
Where is our sample coming from? A controlled environment or open conditions?
Is our sample homogeneous in particle source or a mixture?
Are we going to integrate our measurements with others obtained by other people/instruments? Are we sharing same margins of error and uncertainty?
Dynamic imaging is a very useful tool for counting. But, like any other automation, painting cars or screwing bolts, it has no brains attached.
I hope it helps.
Some more info about pollen identification: http://www-ist.massey.ac.nz/pollen/index.htm
I understand and value the usefulness of technology. I use it everyday and rely on it also for research. I use and have used sensors and instruments for my projects and even designed and implemented some methodologies in environmental studies. For instance, back in 2007 I wrote a whole thesis evaluating the impact of methodological designs in aerobiological studies. And since then I keep a methodologist attitude into any analytical study and subsequent assumptions.
With my input in the subject, I am only trying to bring forward the relevance of “bias” as a component which starts with our own perception of things. And by “things” in science I mean bias from the moment in which we identify what do we want to monitor and focus our attention into, the limitation of the instrument chosen, the meaning of the data and level of uncertainty within, the narrowing in representing a reality from the capabilities of our method to handle the data and the narrow perspective applied when we interpret the potential meaning of our results.
My point is that assumptions and assessments have to incorporate these uncertainties as part of the validation of the final result. And the most important role played by these considerations come when we have to decide how we can improve what we have or even incorporate something new which would fill up the gaps on existent protocols, instruments, etc.
One day, it might be interesting to have an International Symposium focused just on “bias and unification of criteria”.
Just a thought.