[NNagain] the real state of "smart agriculture"?

[dan]

I have a business that does various sensing including AG market as well as
bar/restaurant and produce.  We use LoRaWAN because all other techs were
far too costly and/or low performing.  I'll comment in-line.

On Mon, Nov 13, 2023 at 5:44 AM Dave Taht <dave.taht at gmail.com> wrote:

> (I am hoping others on this list with real-world AG experience can
> chime in? I enjoy realworld stories about present solutions and pain
> points[2])
>
> I have often been dubious of the 5g hope to dominate any major
> component of a smart ag architecture except perhaps FWA, (where
> starlink is poised and people also want to run fiber) to give it a
> good run for the money- 5g chips are too big, too hard to power, and
> too complex, and come with a monthly billing model and other
> centralized requirements that make organic evolution and solid support
> in remote environments dicy and expensive.
>

5G is FAR too costly for this.  The AG market and many markets that could
benefit from sensors are far to price conscious.  5G as well as catm and
nb-iot are great if you have a very small number of highly mobile sensors,
but if you need a high number of sensors it's far far far too costly.  And
it's very difficult to run private networks so it's essentially stuck in
the hands of major carriers.  Just look at the catm/nb-iot market, it's
barely alive.

lorawan sensors can be extremely cheap, just a few dollars, and run for
months to years on a battery.  I've placed lorawan asset trackers in
packages and tracked them across country accurately and cheaply.  A $15
sensor's chirps can be extrapolated into location tracking as well as
identification of impact and temps from the sensor.  We currently track a
bait (as in fishing bait) company's cartons in a few hundred mile radius as
well as their coolers and freezers.  We get temps, humidity, and pressure
and can extract door opens from the pressure and a trigger we have built on
the sensors (sharp increase is a door close, sharp decrease is a door
open).  We triangulate location from gateway locations and wifi beacons
much like you get reasonably accurate locations on your PC w/o GPS using
semtek's location services.

I have a small number of catm devices, including catm on my victron global
relays and a few GPS sensors which work great, but I only use them because
I need long distance roaming.



>
> I freely concede that I may be wrong, that with sufficient subsidies,
> we will end up hanging the equivalent of a cellphone off of every
> suitably large piece of gear and ship all the data up to the cloud,
> rather than pre-process locally. Certainly the benefits of gps and
> drones are being shown every day, along with satellite weather and
> other forms of satellite analysis. [1]
>
> But the 5g sensor market? No. Nowadays smart sensors are easily
> constructed out of wifi devices such as these which cost 5 dollars or
> less:


>
> https://www.amazon.com/DORHEA-Development-Microcontroller-NodeMCU-32S-ESP-WROOM-32/dp/B086MJGFVV/ref=asc_df_B086MJGFVV/
>
> And the more meshy LoRA stuff now has much better range (4 miles), at
> low complexity and power also.
>

LoRa isn't actually meshy, you can run some simblance of a mesh on top of
LoRa radios but this is really not necessary.  We have lorawan GPS sensors
that have pinged at 110km away in clear line of site.  We have refrigerator
lorawan sensors that have been read 2km away in a city at other client's
locations.   Lorawan has very cheap gateways that can easily be installed
at client locations for under $100 that can forward to a number of
'national' services (aws iot, helium, the things network) as well as your
own lora stack such as chirpstack.



>
> then there are things like amazon sidewalk:
> https://www.amazon.com/Amazon-Sidewalk/b?ie=UTF8&node=21328123011

Sidewalk is a hybrid not a wireless tech per se, but includes lora (not
lorawan) and is very well distributed.  I have a few test kits for this and
have been very very impressed by coverage.


>
> And airtags.
>
airtags suck.  Slow chirpers, only really useful for tracking with apple's
kit.  I wouldn't consider this a player in the sensor market.


>
> [1] On the other hand rigorous analysis of the food we produce has
> recently discovered a marked decline in the percentage of nutritious
> minerals over the past 100 years. Please see:
>
> https://www.tandfonline.com/doi/full/10.1080/09637486.2021.1981831
>
> How smart is that?
>
monocrops I would assume.  Plus longer transit times, earlier harvests and
'truck ripening'.  I would imagine flash freezing of produce as well.

>
> [2] Massive subsidy and diversion of river resources to the water
> hungry california almond industry during the last 7 years of drought
> led to the cancellation of the salmon fishing season last year.
>
Are you coming for my Almond milk?!?!


>
> You should hear some of the invective that I used to hear aimed at
> "the f-ing vegetarians" along the docks I frequent in half moon bay.
> That I used to hear, anyway, The docks are eerily silent, the workers
> at other jobs, the boats not going out for anything except crab and
> squid.
>
> How smart is that? The California water table is a disaster, too. I
> vastly prefer salmon to almonds personally....
>
> I guess a meta point is easily gathering tactical data is one thing,
> sharing it sanely another, deciding on how to use it strategically,
> another.
>

There are real dangers in collecting and publishing data unfortunately.  I
have a few sort of a creepy anecdotes from beta testing sensors at a pizza
place.  This is based around 1, 5, 10, 15 minute sensor readings from
dragion temp, humidity, and pressure sensors with triggors on rapid changes
to any reading.

We were able to predict freezure failure 3 weeks in advance on 15 minute
reads by analysing the condensor pump runtimes.
We were able to identify which freezers were the oldest or last refurbished
a couple of ways.  The condensor cycle times compared to the decrease in
temps show how long it takes to cool the box which accurately described the
age of the unit, and the time it took the temp to rise accurately
determined the state of the door seals.  between the two we could identify
which coolers were new, which were refurbished, and which were old and
needed a refurb.  This was over a number of stores in the chain.
that's not so creepy, but it's data extracted from 15 minute intervals that
didn't directly measure the condesor or doors.

However, where it gets a bit more creepy is that we were able to extract
when workers went on break.  accurately.   No door opens, no temp drops, no
changes in pressure meant no workers working, they were out back smoking.
We could identify the smoke breaks PERFECTLY.  That means low pressence in
the front of the store and a back door propped open.

We could also identify the food delivery by changes in the walk-in cooler
pressure, and rise in temps, and very slow drop in temps when freezer was
running.  That means a back door propped open.

We could identify if someone was sitting in the office, or if there were
more that 1 person in the office.  pressure, temp, and humidity all altered
from people being in the room and by a predictable amount.


This seems pretty begning data and private data that the public wouldnt
see, but that we could extrapolate this very accurately from sensors in the
walk-in and reach in coolers should give a little pause about massive
sensor networks and publicly accessible data.  You don't know what you
might expose and what security conserns might pop out from data
'innocently' collected.  Big data is very dangerous.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.bufferbloat.net/pipermail/nnagain/attachments/20231113/317e4a25/attachment-0001.html>