Design Tips
Our design framework focuses on three things: cost, ease of use, and portability. A traditional fluorometer is highly sensitive to fluorescence readings and can detect changes in the microseconds. This allows for specific fast biological mechanisms to be measured indirectly, however, highly increases the cost. To overcome this limitation, we use less precise measurements and rely on slower biological mechanisms.
Detecting flourescence is complicated. Detecting flourescence is even more complicated.
For a moment, let's revisit the FUSSy design. The FUSSy design uses a rather open facing LED array to illuminate and detect flourescence. Because it detects flourescence over an area, the canopy coverage (CC) strongly influences the reading it will report. If seedlings are growing and there is little or no canopy, flourescence will be very low. As the canopy increases, the flourescence value will artifically inflate until there is 100% canopy coverage. In certain circumstances, this may be preferable, but is worth taking into consideration. The PAM is able to remove this variable by always ensuring the canopy coverage is 100%, something that is difficult in our situation.
One way to potentially avoid this bias is by taking a flourescence ratio (the most common solution to all of our problems). If we assume that the relationship between canopy coverage and flourescence is linear, then we can take two measurements in a small time delta. Since time between measurements is small, we can assume that both values have the same leaf area coverage. Since
However, the plant health influences the equations of our variables inequally. In other words, plant health doesn't conviniently seperate out since the systems of the plant that respond to certain stimuli can be affected differently. We know that certain health condition affect different biological systems (ChlF) in different ways. We also know that biological systems react in different ways when hit with different intensities and wavelengths of light. So for a given species, we can say that
In addition, if we assume that
-
$f(C_T)$ and$f(CC)$ are the same function for all plants (does not matter what function) -
$λ_e$ , λ_{r}, and$I$ are known
Then we get the following ratio:
Which simplifies to:
You may have noticed that we are assuming no variation of response in plant species, which is likely not true. However, most plants function mostly the same, so we can take the risky leap that the difference is insignificant (this assumption may greatly affect the comparability of reported values between species, be warned). Then we take some more risky leaps and assume that
-
$dt$ $\simeq$ 0 -
$f(dt=0)$ $\simeq$ 1
Taking that into consideration, we get this:
This equation tells us that if we take two relatively similar flourescence measurements, the resulting ratio of the measurement is dependent on the health of the plant. We can rearrange the equation to make that more apparent:
However, there is a better way to notate this. We can say that the ChlF of a measurement at a given excitation wavelength, emission wavelength, and intensity correlates to the health of a plant.
Since the function
All of this to say something really simple: ratios of flourescence are one of the best ways to give meaning to a normally unitless value.
CC => canopy coverage in sensing area C_T => Canopy topography I => light intensity λ_e => excitation wavelength dt => time between measurements d_e => distance from the light to the canopy
There is one other thing I need to mention. If you read the other sections of this wiki, you might have noticed something called a flourescence factor. This is another way of giving the value meaning. In the research section I mentioned that the most intuitive measurement of flourescence is photons/m2 (or similar). Well, this is (mostly) what the flourescence factor is. It reflects the percentage of light flouresced by the plant. Determinaton of the equation for this measurement was determined via a computer simulation (also found in this Github). While the equation should work in theory, it has not (yet) been verified via testing. In addition, some of the issues listed below will still cause a great deal of headaches that will likely require some adjustments to this equation. Either way, here is the equation:
Where...
- k is a theoretical callibrated constant for the device (dependent on the devices maximum reading, output flux, and photodiode parameters)
- flux_{avg} is the total flux hitting the canopy in the detection area divided by the detection area (this is a rather large value that does not need to be very accurate)
- distance is the distance from the photodiode to the canopy surface (a more accurate equation might take into account the difference in height between the LED and photodiodes, but it should be relatively minimal)
Light is complicated. While photons come into the leaf at a relatively known angle, how they come out is just... not studied at all. We assume it is a lambartian source because we have nothing better to assume. However, this creates issues. For instance, a leaf turned at a 45 degree angle would produce 29.3% less luminance! This means our device introduces at least a 29.3% error on leaf canopies from 0 to 45 degrees, which is less than ideal. Tradiational PAM devices fix this issue by fixing the angle at some known theta (typically 45 or 90 to minimize the amount of far red excitation light the detector picks up), although this is a variable we cannot control with a remote detection device. This is a huge problem that still needs to be addressed. The only solution that has been proposed is switching to flourescence ratios as described above, potentially even taking ratios of the flourescence factor?
Just, don't forget about noise. Other devices like gigantic fans will be running in the greenhouse the device is used in, and if our rather sensitive device is not properly shielded, there will be error in our measurements. This is just here as a reminder to not get rid of the shielding we have and to make sure the future device is properly shielded.
You may have noticed that while calculating flourescence factor, we divide by the distance squared. This is just a basic application of the inverse squared distance law, which we have proven to be true in our testing. As the device moves further away, flourescence factor accounts for this and normalizes out the distance effect away. However, other measurements do not account for this, so moving the device away can artifically reduce or increase the reported values. This is another reason why ratios are good: ratios taken at the same distance will normalize out the effect of said distance automagically!
Our sensor has a limited range of values it can pick up in terms of raw irradiance. This makes things difficult. Different plants may flouresce more or less and the usage distance also strongly effects the reading. However, this limitation can be made into a good thing. We can specify a range of operating distances to optimize our detection range. It's just worth keeping in mind when attempting to reduce the light intensity. It might also be worth using the simulation to design any future theoretical light layouts and distance ranges (though it should be verified more as well).
Fluorescence is difficult to measure due to its low intensity. Because of this, wavelength optimization of both incoming and detection wavelengths is crucial. For most plants, fluorescence is at a maximum when hit with ~600nm and ~475nm light.
While sampling the device using Waveforms, we see a very nice sloped result as shown below. However, when sampling the device using the ESP32 and MicroPython code, this is not what we see at all. Samples are extremely noisy and variable, and we honestly are not sure why. This needs to be looked into further, as previous versions of the FUSSy also had a similar, though less profound, issue.
The PAM can detect in the nanoseconds, we cannot. This means that much of the PAM research is invalid for us to use, see the research section for more information.
I'll keep this rather complex issue simple: Plants undergo a variety of changes when exposed to intense pulsing light. his changes the flourescence response. The best we can do to minimize this effect on our readings during testing is to dark adapt the plant for at least ten minutes between measurements. In vito, this becomes even more complicated, as these high doses of light can permenantly damage the plant over long periods of time. For this reason, it is crutial that we operate at as low of an intensity as possible. For this reason, the first HFS design allows you to turn on groups of LEDs individually (this has not been implimented, but the ability to do so exists in the code through the four LED controllers).
Realistically, a plant's flourescence response does not immidiately stop when light stops shining on it. This means that between pulses, there is some small amount of time where the light will not be on, but some flourescence response may still be seen. However, this window is likely very small. This means that actual point samples might introkduce error if any measurement is taken between the pulses. This is overcome by using a lock-in-amplifier, which takes a running average of the signal over time, then sampling the lock-in-amplifier instead of directly taking samples from the photodiode.
The last complication is that different plant species react to stimuli in different ways. This makes a huge headache when trying to optimize the device or compare certain statistics between plant species. For this reason, out of all the parameters we highlighted, the flourescence factor is likely to be the most helpful.