Computational permaculture

Sorry, I have no idea where you want this to go, but it grew out of a greenhouse discussion.

In another thread, some AI topics came up with respect to scheduling things.  The name of the topic escapes me at the moment, but it is somewhat opensource and might be useful.

I'm a Canadian, and I live in the "north".  Most Canadians live close to the latitude 49, or int he case of southern Ontario, below latitude 49.  I live at 56N.  We get slightly more water in a year, than would allow us to say we live in a desert.  The "Peace Country" is the watershed of the Peace River in Alberta and British Columbia.  Dawson Creek (where I live), is where the Alaska Highway starts.  The Peace Country is about the area of Germany.  Germany is closing in on 90 million people, we are somewhere between 150,000 and 300,000.

Because "nobody lives here", there is no interest in having accurate weather forecasts (we have other problems in that regard).  In the last 2 months, the weather office has been wrong on a daily low predicted the day in question, 3 times at 10 more degrees (Celsius) error.  We are on the eastern slopes of the Rocky Mountains, which is a problem in predicting weather.  We get "chinooks" (aka foehn winds), and can warm by 30C or more in 12 hours, starting at any time in the winter.  I want to teach myself enough about downscaling, so that I can make use of global circulation model data to try and predict weather for 1, 2, 4, 8, ... years in the future.  Will we continue to have wind (I live 5 miles downwind of a 130 MW wind farm).  Will we get more or less precipitation?

There are not enough weather stations in the Peace Country, and I suspect most of those don't have long records of data.  What I thought was a starting point, was to try and model prediction variance as a function of location.  The idea being to find the places in (or upwind?) of the Peace, which have the largest variances in prediction.  But, if I am an idiot at predicting, that won't mean much.  But the hope is that if a person can find those locations, maybe a person can find someone near that spot who would like to buy a weather station (probably at cost).  In that manner, we clamp down on the variance in prediction.

The history of my farm is not long, but maybe 1/2 of it is as yet unknown.  We bought the farm in 1975 (I was going into grade 10 at the time), and while I worked on the farm a  bit the next 2 years, it was largely ignored ever since.  I have more problems than I may be able to fix.  But I think permaculture will work a heck of a lot better than monoculture.  Or doing nothing.

The most durable wood in the Peace is tamarack (larch).  I have posts on my land which have to be tamarack, that are 70 years old.  They can't last much longer.  The other trees we normally have here, if you made a fence post out of them, might last a couple of years (aspen, poplar, willow, birch).  So, I want to grow some fence posts.

There has been a dugout on the property since before we moved here.  We made it bigger, I will guess it is about 2 million litre.  It has no windbreak of trees or a berm.  It is on a north facing slope (average slope probably 7%).  All the data I seen on windbreak design, is for places on the prairies where things are flat.  WindNinja is a code from Montana which looks to model surface winds.  There is a well thought of package for doing PDFs in OpenSource (name escapes me), which might also be able to model this.  It's always best to get multiple codes to try and compare things.

As Hamming said, the object of computing is insight, not numbers.

But, if a person learns to model the winds for my farm, I can model the winds for somewhere else.  Which might be useful.

At the moment, I have 34 amd64 GPU cores, quite a bit of disk space (a lot of JBOD, 4TB of RAID-10, 6TB of RAID-10 waiting to be installed).  I have 2 Polaris GPUs now (RX-460 and RX-560).  I have room for 5 more, if the price would come down.  I have another Ryzen 5 processor waiting for sales on motherboard and RAM, which would push me to 46 amd64 cores (dual threaded cores).  I have asked a couple of recently retired professors who have worked on weather problems for a long time, whether I have enough hardware to attempt this, and they think I did.

Going back to AI.  Darn, I wish the name of this package would come to mind.  In any event, they talk about doing the speech analysis at your site, instead of sending it via the Internet to them.  Google designed special tensor processors for doing this.  You can "get by" with using nVidia GT-1080 processors with 8GB of RAM.  And as described, if you have AMD GPUs, you are out of luck.  Nobody has ported this code to AMD.

At the moment, it is hard to impossible to find Polaris GPUs, because they happen to be really good at some of the math that bitcoin type things find useful.  It's been this way for a while.  If you are on a budget (as I am), you wait.

Attacking things with GPUs is one approach.  If you have a lot of CPU cores, BOINC seems a reasonable approach.  You run a lot of single threaded jobs via BOINC who sends jobs  to computers via a server.  If at some point I get some traction with local politiicans, maybe municipal and county computers could donate cycles to a BOINC effort to predict weather (or even to preprocess for that).  But, my 34 cores will work as a starting point in that regard.

----

That is mostly a where I am now (which ignores a lot).  Arduino and Raspberry Pi have brought a lot of electronics stuff within reach of any body.

Electric fences can be useful.  By and large, you need enough voltage in a "pulse", so as to get the animals attention.  As the length of wire increases, the voltage of the pulse decreases.  So how big a pulse you need to be effective, depends on how long your fenceline is.  It's possible that to get a shock (accidentally) right next to the charger causes problems because it is too high.

If the fence system knew "where" the intruder was and there were multiple chargers on the fence, you could send a pulse from the two bracketing chargers.  At locations where the intruder is not thought to be, the pulse if too low in value to cause a problem.  The two pulses from different directions overlap at the known location, and hence add.  Providing sufficient charge to get the animals attention.  And making the fenceline safer everywhere.

This isn't done now, as far as I know.

If we are growing hay, we would like to make hay at the time it stores maximum nutrition (maybe).  According to a friend, the best time is somewhat before flowers blossom.  I will leave it at that for now.  If you have a small hay crop, it may be that all of the field hits this point at the same time.  Do you have a "meter" to tell you when to cut the hay?  So even if you have a sufficiently small field, you need to analyse the crop multiple times to know when to cut it.  If you have a large field, different locations will reach this point at different times.

If you have the big equipment, you look for the best average and harvest the field all at one time.  And usually the day after you cut everything, it rains and you lose a bunch of nutritional value to the rain.

Let's have robots which weigh almost nothing wander around the field, and take multi-spectral images of the crop.  They might (can?) tell when the crop at any particular location is ready to be cut.  And maybe we have robots that can cut (fairly easy) and condition (to me, this means crimping the stalks often enough) the hay.  So on any given day, robots may go out when moisture levels are best for cutting, and cut what has been assigned for cutting.

This is different from currant practice in a lot of ways.  One is that there are a lot of images to process.  I suspect your "prospector"  robots want to look at the field at times when it typically wouldn't want to be cut anyway (too much dew, ...).  So they examine what hasn't already been flagged or cut or gathered at a convenient time, accumulating a bunch of images (probably multiple per location).  They upload their images to the farm LAN, and computers crunch the numbers.

Something similar for analysing cereal crop, but you probably don't want to cut in this haphazard manner?

If you are growing fibre crops (such as flax), you want to harvest in a manner that doesn't introduce defects in the fibre (no kinks).

In a cereal or fibre crop, you might want to plant a cover crop before the cereal or fibre is ready to harvest.  So that when you harvest, now the cover crop can maximally take advantage of
the "open canopy" created by removing the plants just harvested.  Weeds don't in general get to know that.

The limiting factor in growing crops on the Great Plains is water.  How many farms have a single moisture sensor in the soil?  How many have multiple sensors?  Only on the fenceline?

Orchards and gardens could easily use this data with drip irrigation.  (Probably means emitters with on/of valves, don't water what doesn't need watering.)

We will be able to use lots of different kinds of sensors.  By and large these sensors are not only sensitive to what it is they are sold as.  What this means in practice, is that we may need to create "maps" of correction factors to the various sensors we employ.

---

This got created in greenhouses section, so maybe we need a greenhouse thing.

We are often interested in when seeds germinate.  I will take the example of a tiny seeded plant, because it requires little cover.

We carefully place tiny seeds on the surface (instead of just saying they are too small, and try to spread them out in some less labour intensive way).  Maybe we use a robot to place the seeds, and so the robot knows where it placed each seed (regardless of its size).  We can use multispectral imaging in reflection on tiny seeds, we might have to use it in transmission for larger seeds, but on some regular basis a camera takes one or more pictures of our seed beds.  And then the number crunching comes in the background, looking for differences.  This could reliably detect the first leaf emission by a seed more accurately than anything we do now (assuming I have read all literature on germinating seeds, which I haven't).

Who knows, maybe this thread is thought to be spam.

I don't think permaculture needs to go heavy into computational aspects.  It probably could continue to survive or thrive on holistic measures.  But I think it could benefit from computational aspects.

After all, the object of computing is insight, not numbers.

And I don't even see my copy of the Dover reprint I have where I get that quotation from.  Probably buried under something.  Google suggests Numerical Methods for Engineers and Scientists,
which sounds right to me.  Uggh, now I see it.

If you are looking for other old books, there is a book by a Statistical Engineer at the NBS (before NIST) which is also a wonderful book.  The Statistical Analysis of Experimental Data (Mandel).

Wow a lot of info there in the OP!

I happen to live in roughly similar latitude, and happen to work in tech. My most recent job is quite involved with AI so I totally get where you're line of  thinking is going.

There are a few things to always keep in mind when thinking of permaculture and technology. Things like resiliency, efficiency, and reliability. Remember too that the permaculturist is intimately interwoven with the land and everything in it. Is the hay ready? They know because they see it every day.

These machines that run AI are pretty energy hungry, and they typically only do one job. They're great for applications where a human could do it, but there's just too much workload to get through in the time available to put humans on it, or it's just not economical to do it with tons of humans. The permaculture farm is relatively small (not 100's of acres) and there are many many feedback loops within processes that drive, govern, and stabilize the many systems. So they're small because they don't need to be big. It's not an economy of scale thing here.

Many of the ideas you put forth make a lot of sense in you typical farm. They're even using AI for animal facial recognition systems so they can track movements, habits etc.

The electric fence thing is interesting for long runs of permanent fencing. whether or not the additional equipment would pay for itself I don't know but it's an interesting technical proposition.

The thing on crop detection... A few years back I was working for a company that was in the optical, RF, and nanophotonics field. They developed a spectrographic sensor so sensitive that it can detect all sorts of stuff. I was thinking about using this technology (because it's so small) incorporated into an AUV for monitoring crop growth, weed propagation, etc using the light spectra reflected from the plants.

At this point I think smart energy distribution is a good area of technology for permaculture homesteads. There is always excess energy is one form of another in one place, and a shortage in another at some other time.

Something I've been thinking about for a number of years is how annual cereal crops have changed since the industrial revolution. Wheat and barley have been bred to address many limitations in agro-technology facing people at the time. And today we have this technical debt in the form of biology. What if we could start again with modern technology? What if we could breed for soil health, people health, and multiple uses while not having to worry about the logistics of growing, harvesting, and processing?

Imagine a landrace of full height wheat, barley, rye, wild flowers, dill, clover, vetch etc all grown together in an artificial prairie. Then equipment capable of harvesting the grains, cleaning them, and sorting them into their appropriate bins without the need for tilling or spraying. I think AI could be very useful for separating wheat from barley from rye based on computer vision techniques.

But back to the OP. This weather thing is not easy in some places for sure. It's a big reason why the textbooks recommend living on your land for a year before doing anything. Just observing the energy flows through the land and the biome.
Again, the permaculturist spends a lot of time observing (which looks like doing nothing) and hopefully they learn the local weather patterns.

But yes, a network of weather stations is handy to have. And as far as forecasters is concerned, two specific people come to mind but I forget their names. One is a woman who is very good at long range weather trends for areas, and the other is an atmospheric geophysicist who used to forecast precise localized weather for things like nuclear bomb tests and rocket launches. He has a private company now that does the same thing for countries and corporations. I'll try and find their names. You never know they might give you some pointers.

My AI exposure is focused on extracting human insights. But I keep a pretty close eye on other developments too. I vaguely recall a group doing weather pattern recognition stuff with pretty good results. I'm also in touch with researchers etc who are building these things so if I come across one (maybe a phd student doing a thesis?) then I can make some inquiries for you if you like.

The coldest I can remember in Dawson Creek was -54C.  Two outlying communities about 20 miles away that morning were -64C,  Where I lived, some 200m higher in elevation and 1.5 miles south of Dawson Creek was only -49C.  I think that was the winter of 76/77.  In the last 5 years, I think Dawson Creek has only gotten below -30C 5 or 6 times, 3 of those times this year.  And Environment Canada screwed up the 12 hour forecast low by -10C or all on those occasions.

But I've been thinking for a couple of years now, that Dawson Creek is not Zone 2 any more, it has moved to the warmer Zone 3.  And my plans for planting longer lived plants, is tending to wormer things.

WeatherUnderground seems to be a good place for weather stuff.  There is a local-ish place which tries to place itself there, but they are run by a company which wants to do all the westher calculating themselves, and they want to won the data that people's equipment submits.  I think my ideas for a weather station come from OpenEngineering?  A boomberang shaped collector seems to be the basis.

In terms of observing, where I live I need to keep some things in mind if I walk through forest.  We can get a heavy snowfall, including 2 or more weeks of cool temperatures, at any time of the year.  There are a LOT of warped trees of various ages, which come from such events.  We get chinooks (foehn winds).  We can go from -30C to +10C in 12 hours.  We can stay above zero for a few hours, or more than a week.  Lots of trees get fooled into thinking it is springtime, and start "awakening".  And then it gets cold and trees get damaged or killed.

Part of my farm has trees (aspen) growing next to the road, and then the field is open pasture (largely weeds now) to the south.  If you walk through the trees, you can see partially damaged trees from the thawing and refreezing which chinooks can bring.  The damage is to the south, where the sunshine comes from.  My thinking is that if I plant a grove of sugar maple (and other trees as well), the thing I need to do is establish a stand of evergreens to the south.  To keep winter sunshine from getting to the sugar maples.

Just looked up Dawson Creek. You're up near the tree line for sure. It's an interesting situation and counter intuitive that the southern exposure areas are not necessarily good places to plant maple trees because of the winter warming effects you mention. I just bought some Siberian pine tree seeds with a mind to use them as wind breaks on the North side. They might be a good candidate for your South side.

We're apparently experiencing warmer weather here these days too, and less snow. But I'm now convinced this is a temporary counter trend that is near done. You're in a different part of the country, but we both experience the arctic jet stream and how that behaves is dependent on a lot of variables, which are not steady state by any means.

I don't know about being close to the tree line, I think Ft. Nelson and Watson Lake are in trees, and they are considerably further north than I am.  But with our long days in the summer, we have lots of flowers up here.  I think your SIberian pine is closely related to the Swiss stone pine?  I put a bunch of Korean pine and Swiss stone pine in an ultrasound machine for a while, and they have been warm stratifying for about a month now.  I am just about to switch them over to cold stratification for as many months as I have until it is time to plant them.  The rule of thumb for up here has been, to not plant anything outdoors until the end of the May long weekend.

All the other tree seeds for this year I think only need cold stratification, so they will go at the time the pines go.  Well, except I decided to take a chance on hop hornbeam.  I gave it the ultrasound treatment, and I am about to soak them for a while in GA-3, and then we'll try cold stratifying them.

Open Source, Open Hardware, Open Engineering, ...


(These are my words, and I take credit for all the mistakes you find.)

There are lots of Open things in the world.  Open Source is probably the first, which is a take on "Free Software" for which Richard M. Stallman (RMS) is known.  RMS is known for a few things: the Free Software Foundation, the GNU Project and the world's best text editor (and also an operating system) emacs.  

By and large, Free Software was not about being without cost (also called "free" as in "free beer").  Although much Free Software was also free of cost.  Rather, Free Software was about being without licenses or encumbrances (also called "free" as in "free speech").  That there was lots of "freeware" of dubious quality as well as a lot of "Free Software" which could be quite high in quality was also a problem.  Marketers have pounded for centuries, that things are worth what you pay for them.  If something is without cost (free), it is worthless.

Feelings like those got involved in the creation of a concept called Open Software.  And to talk about either/both at the same time, one often seen FOSS (Free and Open Source Software).  As FOSS started to accumulate a LOT of really good software, lots of businesses started putting the word "Open" on things which were never "free" and never would be "free" (as in free speech).  Something about riding on coat-tails.

In terms of software, software which is free (speech) is protected by a concept called copyright (and a license).  There are things called Copyleft, which is sort of a Copyright.  But Copyright is largely there to stop you from making a copy of something for yourself, or from you making a copy and giving it to someone else (legally).  Free software does not want to stop that, commercial software has to stop that.  Since the Free Software movement started, there have been times where there were few licenses and times where there were many.  There were often arguments about which were true in name, and which in spirit.  Or which were good and which were bad.  A problem for many, is that there were no legal challenges to show that courts agreed with opinions people had.

There have been some legal challenges, I have not memorized the challenges or the results.

----

Some people started talking about it being nice if other things were "free" (speech).  In computing, the flip side of software is hardware.  And Open Hardware does seem to exist, and perhaps the original standard bearer in this is Arduino (sort of known as Genuino in some circumstances).  Most often called a microcontroller (or too small to be a real computer).  Arduino is no longer the only technology considered in this category, there are also various things prepended with Beagle and probably a gazillion other names).  But for the user, the microcontroller board is typically cheap, the programming environment required to program it is "free" (speech) and there are copies of programs written to run on these microcontrollers available in lots of place that are also "free" (speech).

For example, if you want to measure temperature, you can find lots of Arduino projects which talk about interfacing this or that (cheap) temperature sensor to the ADC built into the Arduino and measuring temperature.

Today, you can find lots of "cheap" projects which you can do with Arduino (Beagle, ...).  I have also run across very expensive sensors, for which the manufacturer has set up a source code repository (probably github, could be other places) to let that sensor work with Arduino and the code at the repository is free (speech).

And I expect all/mostly of the above, is also free (beer).

But there is a limit to how complex a thing a microcontroller can do.  And because it is limited, and is somewhat inaccessible as a learning device.

This created a vacuum, and Raspberry Pi stepped in to fill that vacuum.  This is a microcomputer.  The first ones were credit card sized computers for about $35 (USD).  (Arduino were cheaper, so it really is affordable for most people.)  There are lots of things like Raspberry Pi out there (Banana Pi, Orange Pi, Odroid, and a gzaillion more).  The hardware typically is not as free (speech) as Arduino.  A long standing "problem" in RPi is the binary blob to run the GPU on these little computers.

But RPi are more or less powerful enough now, that some people use them for portable computing devices.  A smartphone they build for themself, so to speak.

----

There are lots of other Opens out there, some of which are open and some of which are closed.

Arduino, and Similar

(Again, all errors here are mine.)

These chips which will run a program you can write, don't really have an operating system.  Or it isn't something you need even consider.

There are lots of different Arduino (and Beagle* and ...).  One thing that differentiated the older ones from newer ones, was the kind of "logic" they used.  Transistor-Transistor Logic (TTL) typically operates on 5V.  Many newer Arduino operate with Complementary Metal Oxide Semiconductor (CMOS) logic, which typically operates on 3.3V.  If you are building a board and run out of TTL parts and you have a CMOS part, you probably shouldn't try to use that.  It probably won't work.  You can get "level shifters" to let TTL and CMOS talk to each other, but in terms of building something a board should pretty much be TTL or CMOS, not a mixture of both.

Another thing which differentiates different Arduino is "capacity".  How many programming steps can they hold, how many digital I/O pins do they have, that sort of thing.  A third major way to differentiate amongst the various Arduino is how fast the clock is.

When Arduino first came out, it was very common to see a project board powered by a transistor battery (9V).  The boards were TTL, so they needed 5V.  So on the wire coming from the hot side of the battery, you typically had a L7805 voltage regulator which output 5V (to run the board) by turning all the voltage above 5 into heat.  That energy was wasted.

At some level of complexity, people started using power supplies, and computer power supplies are fairly common.  We have also become more conscious about wasting energy.  Wasting 44% of the power out of a transistor battery is dumb.  Today, we can find or make more efficient power supplies for Arduino projects.

A common sort of project to learn from, is to power a fan.  In particular, a DC fan.

It is possible to 5 or 6V fans, 12V fans or other voltages.  In general, you can drive a 12V DC fan with 5V, it just won't spin as fast.  You cannot "under drive" a fan endlessly, at some voltage there isn't enough power to turn the fan over.  Maybe it is 3V?  Depends on the fan.

There are 2 wire fans, a positive voltage (power) and ground (or return).  One thing a microcontroller can do, is turn a fan on and off.  Let's say we have a a big case fan from a computer, and a couple of feet of say 12 inch ID plastic pipe.  We get a piece of plywood that can cover one end, we cut a hole in it for our fan, and we set the assembly on a couple of blocks so that air can get to the supply side of the fan.  If we connect the fan to power, it blows air into the pipe.  So the first project, is to be able to turn the fan on and off, with an Arduino.

If instead of an Arduino, we connect the fan to a variable voltage power supply, we can try to find a voltage to run the fan at, so that a feather placed inside the pipe, stays in the pipe.  The air flow is not so low that the fan falls to the bottom, and it isn't so high that the feather gets blown out the end.  And maybe we have a tachometer and we measure the speed of the fan when the feather is doing what we want.

It would be very unusual if our really simple Arduino controller has just the right setup to keep the feather "forever falling".

If we go looking at case fans, some of them are 2 wire, but a lot of them aren't.  There are 3 wire fans.  The third wire, is a fan speed indicator.  But, unless our voltage is just right, all we can do to control our setup is to turn the power to the fan on and off.  We could try.

Another kind of fan has 4 wires, the fourth wire is a PWM (or control) wire.  What we can control is "duty cycle" for the power, such as on for 10% of the time and off for 90% of the time.

Implicit in the above statement, is that we are parcelling out time into packages.  A PWM cycle may be 0.2 seconds long, or it could be 20 seconds long or ?  If a cycle is 0.2 seconds and the duty cycle is 10%, then the power is on for 0.02 seconds and off for 0.18 seconds.  It may be easy to find a duty cycle value, which keeps the feather falling where we want it in the pipe.  If our duty cycle is 20 second, our 10% on cycle is 2 seconds and our 90% off is 18 second.  It is very possible that it takes less than 18 seconds for the feather to fall from the top of the pipe to the bottom.  In this case, we may not be able to keep our feather forever falling.

One thing that goes into designing fans, is how much noise they make.  And human hearing can hear things up to about 20 kHz.  So, if our PWM cycle is such that we can put more than 20,000 PWM cycles in each second, it is very likely we won't hear the fan being controlled.  In general, an Arduino is not "fast enough" to control a fan at 20 kHz.  It may be that control at a slower speed still works.  Some things when controlled at lower speeds, break down more often.

An Arduino can run over a range of PWM speeds.  Some PWM speeds have an effect on how well you can set the duty cycle.  Higher speeds probably have coarser control.

In 2018, the common Arduino have a clock speed of 16 MHz.  I think one of the current pack, and a couple of the old ones ran slower (probably 8MHz).  There are a couple of higher speed Arduino, but really the speedy one is the Arduino Due (84 MHz).

I haven't looked at all the other (sort of) open hardware microcontrollers, but it would not surprise me to find they had more ability to run faster than Arduino typically can.

----

But regardless of clock speed, an Arduino does one thing at a time.  It has a single program and it steps through that program one step at a time.

If you have looked at Arduino Due, and all the other similar Open Hardware type microcontrollers and none of them are fast enough, you have a couple of options.  You can switch over to a microcomputer (Raspberry Pi) which might have a clock speed approaching (or exceeding) 1 GHz.  But microcomputers are designed to juggle a variety of tasks, and if you have it doing too many things, its effective speed might still not be fast enough.

An alternative is the Field Programmable Gate Array (FPGA).  While these devices have clocks, when you set them up they have N paths active.  All N paths advance a step at a time by the clock, as opposed to one step on one path, which is the microcontroller way.  FPGA have not really been considered low cost.  But, not too long ago someone reverse engineered a FPGA hardware language, and Open Sourced it.  It works for a particular variety of FPGA.  And then a group of guys decided to see if they could do a Raspberry Pi type thing, and make a cheap FPGA based on Ice Storm.  They were successful, and that board first came out (less than 1 year ago?) as MyStorm.  They are on the 4th iteration now, and I think it is called BlackIce.

Seed Germinating

A couple of messages ago, I mentioned some seed germination work I've been doing.  Perhaps it needs explanation.  Gives you a break from endless geek speak.

GA-3 was mentioned, as was ultrasound.

----

A while ago, someone in Japan was investigating some odd behavior in rice, and stumbled across a family of compounds which are called Gibberellins.  Gibberellic Acid is the 3rd member of this family (hence called GA-3).  Gibberellins are plant hormones that can be involved in a number of different processes, and if they do one job in one plant they might do something else in another plant.  A fairly common thing for GA-3, is to be involved in seed dormancy.  If we add just the right amount of GA-3 to a seed (how do you do that?), it may make a seed germinate faster.

Well, seeds typically have a hard outer covering, how do you get the GA-3 (regardless of how much) into the seed?

One idea is to expose the seed to ultrasound.  One of the things ultrasound does, is create bubbles which can then collapse.  This process is called cavitation.  When a bubble collapses, it can send a jet of liquid in some direction, and this jet me be strong enough to crack on a microscopic basis, a section of seed coat.  So putting seed in an ultrasound unit can damage the coating on a microscopic basis, making it more permeable to water.  And things dissolved in water, like GA-3.  Ultrasound also creates localised regions at very high temperatures, which only last a short period of time.  So ultrasound can drive some chemical reactions that otherwise wouldn't happen.

The research papers I've read on ultrasound treatment of seeds talk about adjusting the power of the ultrasound unit.  I just have a cheap one, it only has one power.  So, I can adjust how long the seed is exposed to the power I have.

Quite often, the ultrasound work is carried out at elevated temperatures.  The ultrasound unit I have, can be run at room temperature (for whatever temperature your room is), or at 65C.  So, I did my ultrasound exposures at 65C.  I put my seed in small polyethylene bags, and the seed quite often was touching the polyethylene on one or two sides.  Does the ultrasound get through the polyethylene without being attenuated?  Does the polyethylene touching the seed have any affect?

The temperature, 65C, while warm is not terribly high.  I wasn't sure what the presence of bacteria or fungi on the seed might have.  So, I went down to my local liquor store.  They didn't have Everclear (190 proof - 95%), but they did have Smirnoff Blue (100 proof - 50%).  So, I soaked my seeds in vodka for about 5 minutes.

Alcohol is actually a pretty reasonable thing for sterilization in biology.  If you want to work with the nitrogen fixing bacteria in legume root nodules, alcohol is a preferred sterilization agent.

But, it turns out that seeds quite often contain abscissic acid (ABA), which promotes dormancy by its presence.  It turns out that abscissic acid is soluble in alcohol.  So, my seeds not only are getting what is likely a germination promoter (GA-3), they are having a dormancy promoter removed.

But, I've no idea if any of this works.  At the moment, it is all just theory.

I'm surprised they focused on power and not frequency.

The last time I was looking around BangGood, I remember seeing driver boards and transducers there.  A typical one is 35W @40kHz.  From what little I've read about a bunch of ultrasound stuff (fish monitors, depth meters, ...) it seems that 40kHz is going to be fairly common.  Why should these biology people have been looking at frequency?

Raspberry Pi and Similar

As with any computer technology, the bleeding edge is a little unstable.  I think the bleeding edge for Raspberry Pi is pretty close to the Odroid-XU4.  This comes with an Exynos 5422 Octa big.LITTLE ARM Cortex-A15 processor @ 2GHz and a quad Cortex-A7.  For a GPU, it has the Mali-T628 MP6.  2GB of DDR3 RAM, with a microSD/eMMC5 "disk", USB, Gigbit ethernet and some other stuff.  But, my data is 6 months old or so, maybe something twice as fast is available now?  

I have a slightly less powerful Odroid-C2 (C1?) here (still in shipping), which I mean to use with a GPS chip to set up to be a Stratum 1 time server and to serve corrections (differential or kinematic) for GPS on the farm.  I had run across someone who had done something very similar, and the suggestion was that a timeserver like that could serve thousands of requests per second for time.  So probably enough for a small city, let alone a farm.  Whether it can calculate corrections, or that has to be offloaded to something with more horsepower is to be determined.

At Hardkernel.com in South Korea, the 201802 price is listed at $46 (USD) plus shipping, quantity 1.

The lower end of the current Raspberry Pi spectrum is probably the RPi Zero.  Adafruit (in New York City) has them listed at $5.  This is roughly comparable to the original RPi.

A problem I think shared by all of the RPi and clones, is the binary blob needed for their GPUs to function.

I've recently built a couple of mini-ITX board PCs, and with a reasonable (non-gaming) GPU they have an estimated peak energy usage of about 200W.  I haven't attempted to measure an idle power draw, but maybe 10% of that?  I think I've seen people talk about RPi-2 drawing 3W.  For projects running off battery, or a solar cell, RPi can be quite reasonable.

While you can apparently install some version of M$-Windows on a RPi, I know nothing about that.  There are several versions of Linux you can run on these single board computers that are roughly the size of a credit card.  It is quite common for people working with the MyStorm FPGA board (which I think is almost $100 usd?  Far less than the FPGAs you could use for calculating bitcoin) to work from a RPi.

Being in a sense a real computer running a real operating system, it has a lot of capabilities way beyond an Arduino.  If your needs for automation, instrumentation and so on require something to oversee or manage a bunch of microcontrollers, a RPi is probably capable of this.  There are likely a bunch of other OpenHardware single board computers which can also do this.

I often wonder why people compare Arduinos to 'Pis, that is a lot like comparing a desktop computer to a super computer...literally.  Most Arduinos are roughly equivalent to a 1980's desktop computer, while many of the PI clones are more powerful than a 1980's Cray-2 supercomputer.

I like the Arduino mini-pros, I use those a lot for controlling different devices.  Depending on what you're using it for, it's possible to power one of those of a coin cell battery for several months.

As for Pis, the ODoids are definitely cutting edge, probably the most powerful Pis available, but they are also relatively large and fairly expensive.  Right now I'm partial to the NanoPi line from FreindlyArm.com.  I'm using a NanoPi Neo2 for my email/web/dns/etc. server


What I really like about the Neo is the heatsink they use, especially when you have it mounted in the aluminum enclosure shown above.  The heatsink presses on the enclosure, basically making the entire enclouser a heatsink.  Because of this the CPU temperature runs less than 5C above ambient temperature.  Power consumption is extremely low, even with the OLED display running it draws around 2/3 of a watt.  I'm guessing without the display somewhere around 1/3-1/2 watt.
The other thing I really like is the price.  The above setup, Neo2 with aluminum case and OLED display, was only $35 + shipping. With the NanoPi Neo (slightly less powerful than the Neo2) it's less than $30.

Monitoring Crops

If a person buys vegetable seed (such as carrots), you might find that it suggests some number of frost free days.  By and large, the customers of these seeds are about latitude 40, and here I am at 56 North.  We get a lot of sun in the summer, and a lot of night in the winter.

The "Peace Country" (watershed of the Peace River in NE British Columbia and NW Alberta) is a huge agricultural area.  About the size of Germany.  And our long summer days have among other things, made us well known worldwide for honey and crops that need bees (or honeybees) for pollination.

If I look up "full sun", it says something like: must get 8 hours of sunlight per day.  On the summer solstice, we are closing in on 19 hours?  We are on the eastern slopes of the Rocky Mountains, so we get a fairly well defined and sharp sunset in the summer.  But there are on a few local hills which perturb sunrise, and so we get twilight well before sunrise.

The limiting factor for growing pretty much anything on the Great Plains is water.  We don't get a lot.  As you get closer to the eastern slopes of the Rocky Mountains, you get less.  And even though the "Peace Country" is not part of the Great Plains, we continue this in spirit.

----

Monitoring rainfall is pretty easy.  Monitoring snowfall is a Pain in the  * (we'll just call it PITA).    If you live in a place where snow and freezing happens, it probably isn't worth worrying about how much snow has fallen.  What you want to know, is that after the snow has gone, what contribution did it make to ground moisture levels.  There are subsidiary worries about erosion from spring run-off.

We can install soil moisture probes to sample our landscape.  At some point, we might want to reduce how many sample points we have, because some are dependent on others (highly correlated), and we may want to move probes due to microclimate concerns.  

Quite a while ago, I read a non-technical article (it pointed to a technical article behind a paywall) about the movement of energy in chlorophyll from absorbing a photon to producing what the plant wants.  But for us, the absorption of a photon of that energy results in a definite change.  It the cell with that chlorophyll molecule had the water and other stuff it needed to prepare for the absorption event.  If the cell wasn't ready, I suspect the absorption event was wasted.

In any event, counting photons (or measuring light intensity) is probably something we should do to predict when the crop is harvestable.

That particular chemical event involving chlorophyll is based on quantum mechanics.  There is some ordinary chemistry needed before that, and there is ordinary chemistry after that.  There are a gazillion reactions, all of which have an activation energy, and through that a temperature dependence on kinetics.  In materials science, there is a rule of thumb that for reactions which happen in water, a change of 10 degrees Celsius (in the vicinity of room temperature) results in a factor of 2 in kinetics.  Increase the temperature by 10C, we double the kinetics.  So keeping track of temperature over the year is useful.

But there are lots of different chemical reactions happening in any plant, all with slightly different activation energies.  What people seem to find that works, is to keep track of "degree days" with respect to some base temperature.  I will suggest degree hours or even degree minutes, but that is a detail.  But different plants had different baseline temperatures.

So, if we are working in days for the time unit, we take the average temperature for the day, subtract off the base temperature (for the plant of interest), and add that to the plant degree days for the plant we are interested in.

Most of these kinds of "hacks", come out of the continental USA and have implicit assumptions.  Assumptions like: it doesn't snow in June, July or August.  I'm sorry, where I live we can get snow (even a lot of snow) in any month.  So, we need to find rules for what happens if we find a particular day gives rise to a negative degree day value (assuming we are working in degree days).

Salt depresses the freezing point, which highways departments often take advantage of.  In the plant world, sugar is often used to depress freezing.  So plants with a lot of sugar are (probably) less susceptible to slightly colder temperatures than plants that don't have a lot of sugar?

In any event, for each crop we are working with, we want to track temperature, light (possibly at some number of energies) and water.

If you have ponds on your property, keeping track of water levels can be correlated against soil moisture levels, at least for land close to the pond.  The correlation problem depends on what you are growing, and how much fungi is in your soil.

But at the end of the day, you should be able to predict (with better and better accuracy as the season progresses) when the crop is minimally and maximally harvestable, and guess at yields.

Predicting Crops

Monitoring crops involved some predictions.  We are monitoring what we planted this year, looking for information as to when we can harvest our crop (either to avoid a disaster, or to maximize yields).

Predicting crops, is a "winter" project.  Let's find predictions for the weather where we are located, for next year.

If you use the Farmer's Almanac for this, by all means continue to do so.  I've no idea what goes into their predictions or what their accuracy is.

We don't have a problem predicting how much solar radiation reaches the upper atmosphere from the Sun in any given year.  Or not much.  Figuring how much of that light gets to the ground is more of a problem.  Some of that light is absorbed, and some of it is scattered.  Light that is absorbed, either leads to heating or chemical reactions (in the atmosphere).  Light that is scattered is still potentially useful.

There may be some ability to forecast how much light is going to be absorbed in the next time period, and how much is scattered.  But things like volcanic eruptions can have profound affects on both absorption and scattering, and are still not really predictable.

Climate models are trying to produce information which leads to predicting temperature and precipitation.  In the daily forecast we see things like cloudy, partially sunny or sunny and those are useful.  But for growing things the predictions of absorption and scattering are probably more useful (and give related information).

Even on a clear sky day, absorption due to water vapour, salts present in the atmosphere and pollution can lead to something like 30% difference in light getting to the ground (if I understand a recent paper for the UN written by people in Australia properly).  Are climate models going to include pollution information?  Do we need to add that in?

----

Maybe we can start to use the information alluded to above, for predicting what might happen next year.  There are lots of big holes in it..

Where I am, is just barely in British Columbia.  Last year was a record year for forest fires in British Columbia.  And reading a recent article at NPR, it was an exceptional year for Montana.  Health effects from the smoke of forest fires last year is going to produce future knowledge, because it was almost a delta function of significant magnitude.  Extensive forest fires do affect growing crops.  Precipitation from clouds full of smoke from such fires, also have an affect.  This complicates predicting how our crop will grow next year, but big chunks of this are similar to predicting what happens from volcanic eruptions.

----

We are getting better at predicting.  Do we have the answers?  Can I sell you a bridge?

My hope is that as we continue to try to predict different things which have an influence on our ability to grow crops, that our predictions will have more value and rely less on just guessing.  Which at least to me, is what personal experience is.

None of us have experienced the future, we have only experienced the past.

Welcome to Monte Carlo

In Monte Carlo, we think about throwing dice, or having balls fall into slots.  Both of which are random events, at least according to most people.

To predict things in the manner of outlined above, may not be easy.  Some of the required data is not of a format I've ever seen the weather office produce.

But, if we are keeping our own data and we measure all the required data, we can use our own data to feed predictions.  Two names associated with this are bootstrap and jack knife.  Or perhaps we can get data of the same nature from other people, and build up a synthetic data set for ourselves?

Lots of people plant crops according to rules like: wait until the ground is 50F.  This kind of statement is often neglecting information, such as at what depth is this temperature taken from?  If nature is planting seeds, the wind typically plants seeds maybe one seed diameter into the ground.  Often just laying on top.  Animals adjust the seed distribution depth by stepping on seeds, and pushing them a bit deeper.  How deep depends on the soil and the animal.  Farm equipment typically plants seeds deeper than nature (in my humble opinion).  I think a contributing factor to the depth of planting by farm equipment, is a larger desire for consistency.  It is easier to consistently plant deeper, than it is to plant shallow.  Also, we really don't want animals coming along and eating the seeds we've planted.

Up here, where we do get winter and cold, there are other reasons for specifying soil temperatures for seeding conditions.  I don't think many seeds germinate, when water is frozen.  But, if we are  told to seed at a ground temperature of 50F, we probably are not looking for the first occurrence of 50F in the seeding season.  Which is perhaps better worded as the first time the soil temperature exceeds 50F.  It is more likely that what we mean is the latest temperature at which the soil temperature is as low as 50F.  The first circumstance can be found by inspection of daily data, the second requires judgment, as we don't know until many days pass that the condition happened.

----

So, let's work out a Monte Carlo type model for spring planting.

We start with some date before we have ever seeded before.  For me, a time when the ground is covered in snow.  The snow can disappear due to two reasons:
1. It is windy, and the snow is blown off.
2. It is warm, and the snow melts.
In any event, snow covered ground does not absorb much sunlight, and snow is frozen water with a temperature of about 0C (32F).  So we have to write some code which for any given day, predicts what could happen to the snow based on the conditions from yesterday.  What we want on any given day, is that we produce a list of possibilities (possibly controlled by conditions found yesterday) and we throw a dice (generate a random number) to pick which condition happens.  Then we advance the day counter and do it over again.

At some point, the snow is at least partially gone, and now we can see faster melting of the remaining snow, as the ground is uncovered.  But we can also get spring snow storms, and cover everything in snow again.  

The snow is gone and has been gone for a while.  Every day, we generate a list of how much the soil temperature will change that day, possibly based on what the temperature was yesterday.  And just like snow, we generate a random number to pick which situation is to happen today.

At some point, the ground is definitely above 50F, so we might want to go back and look at the data, to see if we can come up with a rule which says "this is planting day".

In any event, a decision on planting is made.  Perhaps our model then goes on to grow the crop to harvest, under some different kinds of Monte Carlo type control.

But all of that is 1 trial.  With Monte Carlo what we then want to do, is run through the entire sequence again.  And as all of this are based on drawing random numbers, we see different results for every trial.

I've been looking at the English Premier League (football, you might call it soccer) for a couple of years, and when I do a Monte Carlo study of something, I might generate millions of trials.  I have gotten into the billions of trials.  There are few theoretical distributions that I can draw random samples from, many of these studies involve looking at the just over 20 years of data present in the EPL for data, from which to derive an experimental data distribution.

It may take me a day or even a week to run a large enough number of trials to get a feel for the data.  Sometimes I make an error, and the model is doing something different than it should, so you need to recognize that and correct the model.  At the beginning, I may only do 100 trials, to see what is happening.

The weather office may have 20 or 100 years of data.  That is for one location (often at the local airport).  Maybe we have friends who  have been collecting data, one person for 3 years another for 7 years, and so on.  We live in a time of changing climate.  Using climate data that is before things stared changing isn't very useful.  So using small amounts of data from lots of different locations that are not too far away, nor too different from conditions "here" often lets us get started with better data.  If we plot the data, we may get a histogram with some "holes" in it, and we are allowed to "smooth out" the holes to generate an experimental data distribution.

It is entirely possible that we get different results if we look at predicting the probability of something being true today, ignoring what happened yesterday; as compared to predicting the difference between today and yesterday based on what yesterday was.   That latter data set is not likely something you'll find at the weather office.  You'll need to make it up yourself (or get a geek like me to do so).

Hey y'all, I did not read any of this, been in on the previous discussions enough to know I don't have brain RAM enough to learn this right now. (I am learning WAY MORE building codes than any sane person should ever have to consider.) But I did think of something the other day after we were talking...

A Jack Spirko podcast was talking about businesses you can do that work with permaculture style lives. One of them was to make plug and play systems to automate things and sell it to people who don't find it easy (or have the brain RAM to learn right now!) Some of you with good skills might consider something along the lines of getting the parameters people would like monitored, assemble all the hardware, tell the Pi or Arduino what to do, and ship them a box with good instructions where to install the sensors. Might be a neat niche market for you who are wise in the ways of both dirt and computers.

One of the things I will probably end up doing is the layout drawings like I did on my stuff for others, early on I saw a video where the guy who was doing it said "There are people who know computer graphics, and people who know plants, and there's a need for those of us who know both." I looked at how he was doing things, piece of cake for me, I find graphic layouts and layers easy. You may not realize that your casual ability to say "just make the Pi do this" is a sellable skill, especially if you understand why they want what they do. Another visual here, I had to hire someone to cut some grass here, he's SO not a Permie type, went in with a brushcutter and hacked down volunteer fruit and nut trees etc. There are lots of people who have tractors, but the ones who know what trees are valuable are few and far between. You know what's valuable to permie types, AND you know how to make it happen.

Worth considering. You probably have a niche market here.

Not sure where that find work message comes in?

While I have about 60 programming languages under my belt (and things like Statistical Mechanics), I am a reasonable rough and finish carpenter, mechanic, electrician, plumber, painter, drywall installer and other stuff.  I have an Austrian scythe, but I am still learning how to sharpen it properly.  I've also taught weight lifting, and not too long ago I benched 400, dead-lifted 500 and leg press 1200 pounds (I was always leg strength dominated from too many years of soccer).

----

But, if a person has a way to predict weather (such as from global circulation models?), you could run such a weather prediction against a Monte Carlo type setup as above, to look at the feasibility of growing new crops.  Any crop has "classical" needs in terms of water and heat units, but crops also have quantum needs in terms of how many photons they absorb to drive their reactions with chlorophyll.

I suspect most crops show some correlation to frost free days for growing, but there must be some that have looser correlations.  I would guess plants known to produce high amounts of sugar.

Gordon Haverland wrote:The last time I was looking around BangGood, I remember seeing driver boards and transducers there.  A typical one is 35W @40kHz.  From what little I've read about a bunch of ultrasound stuff (fish monitors, depth meters, ...) it seems that 40kHz is going to be fairly common.  Why should these biology people have been looking at frequency?

The picture is from a biomechanics or medical physics textbook?

I'm still lost on why you would want to vary frequency (of ultrasound, beyond 20 kHz) used in treating seed.  One of the objectives of the seed treatment is to cause the formation and collapse (cavitation) of bubbles near the seed surface.  This can produce a "jet" of liquid which impacts the seed surface, and causes microdamage and localised heating.

It's to do with interference patterns and harmonics.
Here's an example

I looked briefly at the topic of networking on the farm, possibly not with the right set of search terms.  There was one hit largely from a company, which sells networking on the farm.  Or rather, a few hits which sort of revolved around this one.  And really nothing else.

Company president/CEO?  Marketing droid?  I don't know.  Anyway, the one note started out with the apparent desire for someone on the farm to need to be watching video at any place on the farm.  Video?  Really?  At least it was wanting to build off of wifi.  Although I think older technologies (such as 900 MHz, or even lower in frequency) are going to be useful.

Wifi tends to be line of sight (5 GHz more so than 2.4 GHz).  If a farmer is practices cereal (or hay) monoculture, the tallest things on the landscape could easily be the fenceposts, the house and the barn or grain storage.  That assumes a landscape that is basically level.  There are lots of places where the landscape is not flat, but is composed of small peaks and valleys.  One idea might be to put a wifi antenna on the top of each hill.  But some hills are big enough and have enough curvature, that unless you use a very tall tower on the top of the hill, there are still radio shadows on the hill (more so for a 3 foot tall robot as compared to a 15 foot tall tractor).  And then you have this tall tower and possibly guy wires on the top of the hill, interfering with your farming plans.

What makes more sense is to look at the farm with the point of view of putting antennae on the periphery, which can view the landscape within the range of the antenna.  Wireless n (802.11n) is listed as 820 feet (250m) by Wikipedia.  A problem with 802.11n, is that it may use more energy than you can supply.  I was sizing some pumps for solar use, and a 5W pump almost needs a 25W solar panel (fixed orientation) in order to have a reasonable chance of operating enough over 3 days to fill the watering barrel for the next time I have to water seedlings.  One router I looked at, was listed as needing 6W, so about the same situation.  A difference is that the router needs to operate all the time, and so you are going to need a battery (the solar pump was batteryless).  But you still need to account for cloudy days, and if you want to track the sun, you need to power the tracking apparatus.

Most permaculture ideas involve trees, often a lot of trees.

The common rural solution is to "get a tower".  Okay, once I have the CN Tower on my farm, I still wonder if it is going to be high enough to see all of the farm?  In a forest, you may find that there are branches at all heights, although some trees show a tendency to only have branches near the crown.  There are farm "forests" which were meant to be used by animals as a location of shelter from the sun, and later on to allow lawnmowers to cut underneath them, that have few branches below 15 feet or so..  Is that enough gap to send your wifi under the branches?

The spaces between trees in the forest revolves around every plant wanting to try and exploit any amount of light that hits the canopy.  A common theme of permaculture is to space trees further apart than that, so that sunlight (and wifi radio) can see the ground.  People also tend to think in a Cartesian grid, so you may find alleys that car reach deep into a group of trees.

To me, there is some visual information which is useful, but I wouldn't think it is video.  I am thinking more of multispectral imaging.  Just because humans see light over a range of energies, doesn't mean we should limit our use of cameras to that range.  Visible light for humans is something like 1.65 to 3.26 eV per photon.  How far into the infra red do we go?  Near IR is probably down to 1.2 eV.  But there can be lots of organic groups which strongly interact down to 0.2 or 0.1eV.  On the ultraviolet side, we might venture at least part of the way to 12 eV.  Bees and high energy animals like hummingbirds tend to see UV, because they are looking for high energy food.

So we might have 3 or 4 cameras at any given imaging location: visible, UV, near IR, farer IR.  What frame rate do we use?  I suspect something like 1 frame per day is fine.Probably the camera is on a perch that we can rotate, so the cameras gather a panorama every day at some opportune time (local noon?).  The IR cameras might be useful for night imaging, and then we might also wonder about using IR  emitters to illuminate IR images further away from the camera.

Do we send the raw images to the server, for processing there, or do we process at the camera?  If each camera has 10 degrees of view and is a megapixel, that is 144 megapixels of information (potentially) at noon, and however many at night.  For each camera site.  Well, maybe not quite that much.  Like with wifi antenna, we might want to put them on the perimeter.  But if the perimeter is publicly accessible (next to a county road), you might find the vapour pressure of instrumentation too high.

You do want a GPS ground station at your farm, and you want that ground station to be running NTP server software so that it can "serve time" to all the network locations on your farm.  This will probably keep all your instrumentation synchronized to something better than 1 millisecond (possibly much better than that).

I'm still lost Nick.

I understand some of the math of resonances, and I've used Cauchy distributions before to add noise to some processes.

Having that GPS ground station gives you what is technically referred to as a Stratum 1 time server.  With another set of software, this GPS ground station can serve "corrections" to "rovers" on your land (or nearby).  There are a couple of different possibilities in terms of what kind of corrections.  But it is possible to get much better than 1m precisions.  In 1997 with differential corrections, we were getting 90% of our data to better than 20cm (horizontal).  Unless you were in the middle of a blunder or something.

You probably should try to set up a set of soil moisture monitors.  Ideally in representative places, but that can only be found through examining the data and testing other samples.  Having a series of barometric pressure, relative humidity, temperature wind speed and direction setups is probably useful, on the windward and leeward sides.  I'm talking quick and dirty instruments, not formal weather stations.  You probably should go to the trouble of setting up a formal weather station as well (with a wind speed device elevated sufficiently to be useful).  These quick and drity wind speed/direction measurements close to ground level need not have much correlation with general wind patterns in the area.  The proper weather station will have a Stevenson screen (or similar) way to get accurate air temperature data.  For the curious, having instrumentation to detect lightning or measure surface fields (having to do with spaceweather) is nice.  Having cameras that can watch the sky on clear nights can pick up meteors.

If you have a dugout, you probably want to monitor water levels, surface winds, surface temperature, and maybe water temperature at a number of depths and locations.  If you have small ponds for holding water (swales), having some indication of when they have water and how much is probably useful.

Having cameras in known locations taking pictures (especially multispectral) should be useful.  Having cameras on roving robots, that can take close-up pictures, possibly with special illumination is probably also useful.

It looks like Raspberry Pi can run OpenVPN without too many problems.  I ran across one article which recommended going with 1024 bit keys, because of performance.  It may be that if you are running RPi-1, if you do VPN it has to be 1024 bit keys.  RPi-2 and RPi-3 seem to be able to work with 2048 bit keys.

It is possible that for farm-LAN purposes, you want to split your data into 2 segments:
1. data that you are willing to be freely available to anyone.
2. data that is yours.

For sensor stations which warrant a Raspberry-Pi type supervisor, encrypting SOME data is reasonable.  For stations which lack the ability to encrypt, the only solution I have is a data card, and at some point you will need to remove the data card to retrieve the data (possibly inserting a new data card for new data).

In terms of public information, my guess from years ago would have been to "wrap" it in XML.  Today, it would seem that JSON is preferred.

If you have many sensor locations on the farm, it is not inconceivable that all of them produce temperature, relative humidity and barometric pressure.

You could broadcast all of that data "in the clear" (wrapped in XML or JSON), but neighbours snooping on your transmissions aren't going to know that station 17 is the west most Evans cherry next to the road.

If you have many sensors reporting "public" data, it is probably best to send all of that data as encrypted (if you can), and then send something like: minimum, average, maximum as XML and JSON to a "broadcast address".  For people who care about what IPv4 addresses are about, this at least lets your neighbours know what is happening on your farm (in some way).

I think min/average/max is probably the best set of data to be broadcast.  

The best example I have for this, is neighbours who grow vegetables.  In the fall, they are worried about frost damage to plants like tomatoes.  If you broadcast minimum temperature on your farm wifi, and they can receive it, that gives them a heads up.

If you have a website for your farm, I think you should have a section (page) devoted to what information is publicly available.  If you transmit some (or all) information from sensors over an encrypted VPN, do not describe that information in any way.

The website is permies.  It is meant to be viewed by anyone, from anywhere.  If you have a page that talks about publicly available information, that is telling them that if they "snoop" on that IP/port number, what information is there, is information that you mean to be permies.

At the same time, it tells everyone that you have other data on other places.  Which for some people is an invitation to break in.  And so you are likely to find people trying to abuse what you do.

What are people wanting to circumvent protections looking for:
1. Free access to a high bandwidth connection to the Internet (solved in part by QOS?).
2, Proprietary information (why to use a VPN, what kind of key to use).
3. Porn and similar content (don't have porn on your server).
4, Other stuff?

I haven't read all of this (the tech stuff is over my head, sorry), but I grew up on a homestead in the Interior of Alaska, so I am familiar with a similar climate to what you are in.  I was wondering if you had checked with the University of Alaska Extension office for any information they might have.  I don't know what they are doing now at the Experimental Farm, since the powers that be seem to have given up on all possibility of Alaska doing much with agriculture (their own fault, since they passed a law that made it cheaper to import food than to grow it).  But they used to do a lot of work trying to find out what would work best for crops and livestock in a sub-Arctic climate, and should still have that information available, most of it for free.  

Kathleen

Hi Kathleen.

You haven't read all this geeky stuff?  Well, maybe some night if you can't get to sleep, it will put you to sleep.  

I've rummaged around various sites in Alaska for quite a few things, but not agriculture.  The weather here in the summer comes off the Gulf of Alaska (occasionally bringing volcanic ash from the Alaskan volcanos), and in the winter it comes from a couple of different places.

I used to do GPS and geostatics applications in agriculture, which touches on some of this.  But that was 15 or so years ago.

I will see what I can find.

LOL!  Honestly, I started on the geeky stuff but didn't even have any idea what you all were talking about!  I can talk about plants, crops, livestock, and soils all day long, but get into the geeky stuff and I need a dunce cap!

I hope you find something helpful in the Alaska information -- I know there's lots of info out there, it's just finding it that's the problem.

Kathleen

I am not a farmer, but I seem to be good at learning some things (mostly things that aren't social).  So, among other things I screw up that bees, cows, sheep, ... don't like being by themselves.  Most of my "experience" with farming is with cereal crops, oil seed crops and hay.

If I have a parcel of pasture, and I put animals on it, different animals will eat different things.  Cows (maybe I should call them cattle?) have preferences.  Sheep will eat different things.  Goats and pigs are also different.  If I have horses, inevitably they seem to eat only the stuff that is bad for them.

All animals develop habits, and this includes not eating things that are good for them.  Sort of like people and brussel sprouts or brocoli.

If I ask the question, what is the favorite food of cows, Google has no answer.  But if I knew what the favorite flavour of a cow was, I could think about extracting that flavour (ethanol or supercritical carbon dioxide) and then going around the pasture and spraying foods that cows should eat (but for some reason don't) with that flavouring agent.  Maybe that gets cows eating that stuff.  Chances are, this only works with calves.

----

That doesn't obviously lead into some computational aspect.  One trace element I know a fair amount about, is selenium.  There are plants which will concentrate selenium.  If a cow eats plants that are high in selenium, it will show signs of being loco.  And so there is at least one weed called locoweed, which is a plant which concentrates selenium.

There are places where selenium is high, and places where it is low.  And this happens for most of the trace elements in the diet.  If your soil is deficient in boron, sunflowers are among the first plants to suffer from deficiency.

We can look up nutritional data on white clover.  The nutritional tables will present some value for the average content of some nutrient.  Better tables should also provide some measure of dispersion in that value (standard deviation probably).  I doubt they present guesses for probability density functions - they probably just assume a Gaussian is what one should use.

National bodies such as USDA or Agriculture Canada have a reason to present nutrient data values that are averaged over the nation.  Subcountry jurisdictions (state, province, prefecture, ...) similarly have reasons to present nutrient data for the same crops, but averaged only over that subcountry jurisdiction.  Regions like the "Peace Country" (where I live), which are substantially smaller than both Alberta and British Columbia, but still large (the Peace Country is nominally the size of Germany) may have reasons to produce nutritional averages for crops as well.

How do these bodies average the data?  Is it by the area of the region?  Is it by the area the crop is grown in, in that region?  Is it averaged over the population of animals eating that food (in a region)?  By mass?  By how many people live in the region?

In any event, if I go looking for selenium content in white clover, it is probably different for Canada, Alberta, British Columbia and the Peace Country.  It is possible it is different for the BC Peace and the Alberta Peace.  Presumably all these nutrient levels measured and averaged in different places are consistent with each other.

As one looks at smaller regions, you can see where some regions seem to be lower in a nutrient than perhaps they should be.  This could be a deficiency of the nutrient in the soil, or a bioavailability issue.  To look at bioavailability, I think you probably compare monoculture to permaculture (one has more fungi and bacteria in the soil).

Anyway, too much introduction.  To responsibly feed animals, we need to provide them the nutrients they need.  Presumably we want to minimize how much it costs us to do this, cost being money or other things.

I think it could be possible to do some kind of Monte Carlo type work, which could be used to plan feeding based on all the nutrient data available.  Oh, we bought X amount of alfalfa from some guy 500 miles away.

-----

My immediate interest in this?  I would like to get some chickens.  And I mostly want them eating weeds and bugs and the things they ate historically.  But winter is typically 4-5 months here (it has been 7), so it is kind of hard for a chicken to go find weeds to eat in winter.  Even tough birds like Chantecler chickens.

I think I am losing people.  Or putting them to sleep, just as bad.

Our last "caller" was interested in livestock, and they eat feed.  So, lets do a feed thing.

Something that may be interesting, is to generate the histogram for all the data.  Then go through  all the data "in order", generating 2 point averages (points 1 and 2, points 3 and 4, ...).  Now shuffle your data and generate another set of 2 point averages.  Repeat a few times.

One report has clover having about 3% nitrogen in 2-3 year white clover when the crop is at 3000 kg digestible matter per hectare.  Feedipedia has this to say about white clover:
> 24.9 2.7 19.5 29.6 51

The average protein content (fraction) was 0.249, with a standard deviation of 0.027.  The minimum amount of protein observed was 0.195, and the maximum is 0.296.  The number of samples taken was 51.    The minimum fraction observed is about as much below the average, as the maximum is above the average.  The range in values is about 3.75 times the standard deviation observed.  If we were counting things, we would expect the fraction standard deviation (from a Poisson distribution) to be 0.14.  The observed fractional standard deviation is smaller than this at 0.108.

That is really a brief summary on 51 data points.  And we don't even have the data points to work from.

By using a "standard deviation" describe the dispersion of the data about the measure of central tendency (average or mean or expected value), they are suggesting that the Guassian is an appropriate distribution for this data.  There are situations where the Poisson can be an approximation to a Gaussian.  The binomial distribution can be an approximation to a Poisson distribution, except it is under-dispersed compared to a Poisson.  A negative-binomial can also be an approximation to a Poisson, except it is over-dispersed with respect to a Poisson.

So, I wrote a quick and dirt program in Perl, which generates sets of 51 data points from the 4 distributions.  It then proceeds to fudge the data.  It linearly rescales the data so that the fudged data set has the same average and standard deviation as the desired (Feedipedia) data.  What is left is to see if the minimum and maximum of each trial distribution is close enough to the observed minimum and maximum.

On a particular run making 1000 outer loops, I generated 4000 synthetic datasets, fudged them, and then looked to see which fit the minimum and maximum.  I got 27 datasets, or 1377 values.

There are ways to do better.  One thing you may want, is that between two values within the minimum/maximum range, that the fractional histogram count error be less than 10%.  A rough guess at what that means, is a minimum number of counts in a bin of 100.  So instead of looping over trials, you are binning your data as you go along, and you stop when your target bin(s) obtain a count of 100.  And then you plot the histogram to see what it looks like.

----

#!/usr/bin/perl -w
use strict;
use diagnostics;

# Generate synthetic data for %crude protein in white clover for the
# values found in Feedipedia using 4 different random number distributions
# and a bit of fudging.

# Copyright 2018 by Gordon Haverland
# License is typical of code written in Perl - it is your choice of the
# Perl Artistic License or The Gnu General Public License.
# https://dev.perl.org/licenses/

# There is no guarantee or warranty associated with this program.  It
# is just a quick and dirty solution with hardwired data to demonstrate
# a computation intensive approach to generate synthetic data meeting
# requirements.

# There are lines of code which could be removed.  This chunk of comments
# is all the comments that are present.

use Math::Random::MT::Auto qw(rand);
use Math::Random qw(random_poisson random_normal
                   random_binomial random_negative_binomial
                   random_permutation
);

# 24.9 2.7 19.5 29.6 51
my $minimum = 0.195;
my $maximum = 0.296;
my $average = 0.249;
my $sd      = 0.027;
my $nb      = 51;

my $cf = 2;

my $scale = int( ($average / $sd)**2 / $average);   # =341;

my @dists;

my $trials = 1000;
for( my $i = 0; $i < $trials; $i++ ) {
   my @tmp1      = random_binomial( $nb, $nb, $average );
   my @tmp2      = random_poisson( $nb, $average * $scale );
   my @rnormal   = random_normal( $nb, $average, $sd );
   my @tmp3      = random_negative_binomial( $nb, $nb, $average );

   my @rbinomial;
   foreach my $d (@tmp1) {
push @rbinomial, $d / $nb;
   }
   my @rpoisson;
   foreach my $d (@tmp2) {
push @rpoisson, $d / $scale;
   }
   my @rnbinomial;
   foreach my $d (@tmp3) {
push @rnbinomial, $d / $scale;
   }

   print "\n\nPoisson:  ";
   fudgeDist( \@rpoisson, $average, $sd );
   if( verifyDist( \@rpoisson, $average, $minimum, $maximum, $sd ) ) {
print "\tworked";
push @dists, \@rpoisson;
   }
   print "\nBinomial: ";
   fudgeDist( \@rbinomial, $average, $sd );
   if( verifyDist( \@rbinomial, $average, $minimum, $maximum, $sd ) ) {
print "\tworked";
push @dists, \@rbinomial;
   }
   print "\nNormal:   ";
   fudgeDist( \@rnormal, $average, $sd );
   if( verifyDist( \@rnormal, $average, $minimum, $maximum, $sd ) ) {
print "\tworked";
push @dists, \@rnormal;
   }
   print "\nN-Binomial: ";
   fudgeDist( \@rnbinomial, $average, $sd );
   if( verifyDist( \@rnbinomial, $average, $minimum, $maximum, $sd ) ) {
print "\tworked";
push @dists, \@rnbinomial;
   }
   print "\n";
}

print "done\n";

sub fudgeDist {
   my $list    = shift;
   my $average = shift;
   my $sd      = shift;

   my $n = scalar( @$list );
   my $sum = 0;
   foreach my $d (@$list) {
$sum += $d;
   }
   $sum /= $n;
   my $sumsq = 0;
   foreach my $d (@$list) {
$sumsq += ($d - $sum)**2;
   }
   $sumsq /= ($n-1);
   $sumsq = sqrt( $sumsq );

   foreach my $d (@$list) {
$d = ($d - $sum) * $sd / $sumsq + $average;
   }
}

sub verifyDist {
   my $list    = shift;
   my $average = shift;
   my $minimum = shift;
   my $maximum = shift;
   my $sd      = shift;

   my $min = $list->[0];
   my $max = $list->[0];
   my $n = scalar( @$list );
   my $sum = 0;
   foreach my $d (@$list) {
$sum += $d;
$min = $d < $min ? $d : $min;
$max = $d > $max ? $d : $max;
   }
   $sum /= $n;
   my $sumsq = 0;
   foreach my $d (@$list) {
$sumsq += ($d - $sum)**2;
   }
   $sumsq /= ($n-1);
   $sumsq = sqrt( $sumsq );

   my $reject = 0;
   
   if( $min < ($minimum - $cf * 0.001) ) {
print " < min  ";
$reject++;
   } elsif( $min > ($minimum + $cf * 0.001) ) {
print "   min >";
$reject++;
   } else {
print "        ";
   }
   if( $max < ($maximum - $cf * 0.001) ) {
print " < max  ";
$reject++;
   } elsif( $max > ($maximum + $cf * 0.001) ) {
print "   max >";
$reject++;
   } else {
print "        ";
   }
   if( $sum < ($average - $cf * 0.001) ) {
print " < avg  ";
$reject++;
   } elsif( $sum > ($average + $cf * 0.001) ) {
print "   avg >";
$reject++;
   } else {
print "        ";
   }
   if( $sumsq < ($sd - $cf * 0.001) ) {
print " < sd  ";
$reject++;
   } elsif( $sumsq > ($sd + $cf * 0.001) ) {
print "   sd >";
$reject++;
   } else {
print "        ";
   }

   return 0 if $reject > 0;
#    return 0 if( $min < ($minimum - 0.001) || $min > ($minimum + 0.001) );
#    return 0 if( $max < ($maximum - 0.001) || $max > ($maximum + 0.001) );
#    return 0 if( $sum < ($average - 0.001) || $sum > ($average + 0.001) );
#    return 0 if( $sumsq < ($sd - 0.001) || $sumsq > ($sd + 0.001) );
   return 1;
}

They say a picture is worth 1000 words, so I am working on a picture.  But for more data.  If I do 40,000 trials, I am getting a bit over 100,000 data points.  That is starting to be enough data.  

But, before I do that (I am hoping I can attach a 400x600 PNG?), I should make some comments which are not obvious.

By default, lots of programs use /dev/random (or its equivalent on non-UN*X operating systems).  Or possibly the non-blocking /dev/urandom.  This is DUMB!  If you want to simulate something in a "random" manner, you only need  to waste a tiny bit of /dev/random to "seed" a long period random number generator of your choosing.

For work in Perl, my GOTO is the Mersenne Twister (MT).  At the beginning of my Perl script, I have a call to the Mersenne Twister library for Perl:

use Random::Math::MT::Auto qw(rand);

This replaces calls to the standard Perl rand() system call, with calls to a Mersenne Twister random number generator.  Now when your operating system needs to send a packet over the Internet, it doesn't have to worry about being out of entropy because some idiot was running a stochastic simulation.

On any machines where I run code like this, I try to help out the kernel entropy generator.  On Linux, there is a package to draw randomness out of sound in the ALSA sound system.  I don't know if country is more or less random than rock and roll.  Another package which can help is havge, which looks at timing information on interrupts.  There used to be a fairly portable facility called egd, but I am not sure if it is still around.

If you have a computer (RPi ?) that has a GPS chip on it, it may be able to provide some entropy for your LAN.  Arduino or RPi running things like Geiger counters can be source of entropy.

Why is it DUMB?  If your computer runs out of entropy, it becomes much easier for someone to break in.  It is too easy to run things like the Mersenne Twister for simulation work, and it leaves your system with just as much entropy as it normally has.

If you're interested in growing fenceposts there, you might consider Siberian larch rather than the native tamarack. It's much faster growing and more robust, and just as hardy and supposedly similarly rot-resistant. They're popular as ornamentals here in Fairbanks at nearly 65 degrees north, and I've never seen one with winter damage even at zone 1 temps. Two or three feet of growth a year isn't unusual. The native tamaracks here are stunted by comparison. I collected a bunch of seed in the fall from big local specimens that I'm getting ready to plant. Like you said, none of the native trees are suitable beside larch. Even the Siberians will probably take 20 years to get fencepost sized though, so get them in the ground!

I hadn't thought about Siberian larch.  And (so far, with no time spent) you seem to be the only known source of seed.    I will look to see what is around.

Someone mentioned some kind of tree being harvested near Ft.Nelson, BC that was useful for some purpose in building houses.  The only thing that came to mind was the native tamarack.

Thanks for the info.

Gordon Haverland wrote:I hadn't thought about Siberian larch.  And (so far, with no time spent) you seem to be the only known source of seed.  :-)  I will look to see what is around.

Someone mentioned some kind of tree being harvested near Ft.Nelson, BC that was useful for some purpose in building houses.  The only thing that came to mind was the native tamarack.

Thanks for the info.

Might have been white spruce in Ft. Nelson; it's what grows here. I think they also have lodgepole and jack pines, and also balsam and subalpine firs.

I'll save you some time :-) . Here are some L. sibirica seed sources:

https://www.amazon.com/gp/offer-listing/B01BYG28W6/
https://sheffields.com/seeds/Larix/sibirica

Or you might just get seedlings:

http://www.ncrtrees.com/main/containerized_seedlings/

They're really nice looking trees too; the needles are a lot longer than the tamarack's.

I didn't read through all of the discussion here, so unfortunately this is a bit of a drive-by post.

There's a device that is being developed that hopes to track data for farms - using lower tech and more affordable tech.

It's in a blog post by Bill Gates (???) here:  Can the Wi-Fi chip in your phone help feed the world?, which includes this video, too:

Surveying

I have gathered data by taking a handheld GPS on a walk (or ride on the back of the mower to do the lawn).  But the convergence is not so great, and if you are not using differentials or real time kinematic, it is hard to get good results in a reasonable time frame.

Image procesing can be a path to faster data, you still want the GPS with differentials or RTK.  First off is to find out how non-linear your camera is.  In principle you probably want to take pictures of a regular grid in the colours of your RGB camera, so that you can derive corrections for each of the colours.

Sure, some people can fly airplanes and take pictures.  It's not in my budget.  The next thing that came to my mind were drones (quadcopters and related), but after Gatwick who knows what will happen.  This summer when I thought of it, the Canadian regulations more or less say if you are with 3.5 km of an airport, nyet.  It's a little more complicated than that.  Well, maybe a slow flying model plane (RC or autonomous) could do the trick?  I've always like the Martin Mars, from its work in firefighting.  It seems as a model, you probably have room for a camera and GPS in the plane.  Do you have to fill the cargo hold with helium to get a little "lift"?

There are autonomous robots rolling around farms doing things.  So, I "floated" the idea of a a tethered balloon (with a gimbal for the camera and GPS) attached to an autonomous robot past some people regulating "drones", and that seems to be okay.

You probably need a fairly high speed winch on the robot, in case you need to pull the balloon down quickly.  The robot has to weigh enough, that the robot doesn't carry it away.  You probably want to have a balloon destructor in case the tether breaks.

I would imagine you want to be able to take pictures that are very nearly horizontal, and pictures that are very nearly vertical.  You might want to have two nearly identical cameras, one for RGB in the visible and the other altered to be sensitive in near IR.  You should be able to do surveying with that, as well as look for things like stress due to drought.  You probably could have a robot wander your property every year, measuring the height of all your trees, even if some of them are coastal Douglas fir or something.  

You could have another robot set up to take pictures of "a plant" on the ground, with a fancy multispectral camera (possibly with special LED lighting?)

Cartesian grids vs aperiodic tilings

We tend to live on farms that have boundaries that are very nearly E-W or N-S, and if the farm is divided into fields, it is likely those fields have a similar orientation.

And then we come along to plant tress (or whatever), and we end up planting E-W or N-S.

Prevailing winds are often from the west, and so we can have alleys of wind in our plantings.

We could go over to an equilateral triangle instead of a square for the basis of "tiling" the land, this can make it easier to avoid having "alleys" of wind.

But perhaps another idea is to apply an aperiodic tiling to the land?

To tile the land with squares, we have translational symmetry in 2 directions (X and Y), and because they are squares we have a 4 fold rotational symmetry.  To go to equilateral triangles or hexagons (those two tilings are related) we can have 6  or 3 fold rotational symmetry.

In an aperiodic tiling, if you look very closely at the local surroundings of any point, and then look in any direction, the pattern _never_ exactly repeats.  You could get almost a repeat, except that the pattern at the second point is rotated with respect to the first location.

If a person was going to make something out of concrete with aggregate, we would want the smallest dimension of what we are making to be about 5 times the size as the largest aggregate in the concrete.  So if you were to take a section of a tiling and scale it so that on the smallest dimension of the field in question, you got something like 5 tiles across that dimension, then you  might have a reasonable start for planting things like trees.  You are unlikely to have an alleys, but you would need to look for that.  You could rotate the tiling to minimize a single prevailing wind, or possibly two different prevailing winds.

As our farms typically are rectangular, and you will never find an aperiodic tiling fitting in a rectangular area (by logic, the rectangular area is periodic, you can't have something aperiodic inside it where the borders now become periodic).  So you adjust the scale so that the "clipping" to a rectangular area suits your needs the best, and now you have all these strange fill areas on the border.

If you live on a flat plain, an aperiodic tiling might be easier to work with, than if you live on a slope.  I have two different sources of prevailing winds, one is from the SW and is typically warmer, and the other is from somewhat N of W and is typically colder.  The N of W dominates in summer, but the SW can come any time of year.  We just notice it more in the winter.  I live on a north facing slope.  So in the event of frost, I get cold flows across the land that are basically from the south.  I guess you don't want to set up "pockets" with trees, such that they gather a cold flow.  You would prefer that they disperse a cold flow.

Very interesting thread. Regarding surveying, I think drones would be your best and easiest option. I have devised a fairly easy approach using a DJI Phantom 4 and the Litchi app (for Android and iphone) to fly a grid over my farm and take photos at regular intervals. Those photos can then be stitched together using either FOSS or relatively cheap software (Agisoft Photoscan). The photos are geotagged with the drone's location (onboard GPS) and after you stitch the photos together you get a georeferenced orthomosaic or you can even create a 3D textured mesh or digital elevation model (DEM). I haven't yet tried to attach a multispectral camera to the drone but it's definitely possible to use one to monitor crop growth, nutrient/water deficiency etc.

Thanks for stopping by.

In general, I think a "slow" fixed wing vehicle will be better suited to surveying.  For people like me close to airports, the tethered balloon makes sense.

Where the drone seems to make sense to me, is to "trail" something.  Maybe you are looking to track down a source of pollution.  The drone can (possibly) analyze the signal in near real-time, and change course to follow changes in the signal on the ground.

Jetson Nano in the drone?

Where I will find time, I don't know.  But, for this fall (when I start to have frost problems), I would like to have a tethered balloon fitted with visible, NoIR and LWIR cameras.