Pete and Randy

> >These excerpts show how confidence in correlating internal hive
> conditions with colony survival has changed over time.
With the correct set of sensors, one can tell as much or more about colonies than visual inspection.  However, the majority of the recent startups tend to throw sensors into hives, but fail to address two important issues:
1) Which of an almost infinite array of sensors actually provides useful information for management, and2) How to best assess and analyze streaming data from multiple sensors.
We spent years working on these  two issues and still have more to learn.  However, we have learned a few things.  Probably the most important sensor for distance monitoring of colony condition is hive weight - not to identify health and survival, but to know whether distant colonies are starving or making honey and maybe, detect a severe pesticide exposure incident.
Temperature monitoring has to  be done with more than one sensor and these need to be be in or near the brood nest.   More than any other sensor, representative temperatures of the colony core will flag the loss of a laying queen and track her replacement.
RH - so far we see little value.  It's a difficult sensor to keep functioning, since most need a porous membrane cover, and the bees tend to coat them with propolis.  Except at extreme levels of RH or low RH, such as in some semi-arid and arid areas, where brood dessication is a threat, we see little evidence that colonies regulate RH.  It tends to follow the RH of ambient air.   This is contrary to what has been oft reported, but we've had RH sensors in hives in humid/wet coastal locations and in hot, dry areas like Yuma.  If bees pay much attention to RH, we've yet to figure out when and why.
Acoustic - lots of interest, and we're all learning what can be derived from that data.  

Bi-directional bee counters - by far the best way to assess forager activity, overall  condition of hive, detect both acute and slow/chronic bee loss.  But these are expensive, if properly constructed for accuracy and  reliability.   EPA is partially correct in stating that bee counters affect foraging - that's  because of several inexpensive knock-offs that  use only a few entry/exit portals.  That causes congestion at the entrance, just like too many planes at an airport.  But adding monitored portals, going from the cheap 1-3 to 32 or more, greatly adds to cost.
From here the list goes on to sensors that may be possible, e.g., at least two companies are working on detection and killing of individual mites with lasers, to improbable (in my opinion) - predicting where bees are foraging based on geomagnetics.  I say improbable because I am well aware of a Ph.D. thesis aimed at proving or disproving the ability of bees to detect geomagnetic fields.   His first results indicated that they couldn't, his final results indicated that they could.   He used a reward incentive in his first trials (e.g., syrup available if various strength geomagnetic fields were present).  The results, nada.   But he wondered, so he used negative conditioning.  Bees got a hot-foot if a geomagnetic field was present.  Turns out, they did know, just weren't much interest in positive conditioning and a reward, but were motivated to avoid a hot foot.
Biggest and most expensive lesson we've learned about sensors in bee hives.  What sensors, what to analyze, and how to analyze are critical to success.   One of the first issues is TOO MUCH DATA.  It's like going from a drip of a faucet to the full  blast of a fire hose.  That introduces possible need for tools like wavelet analysis used for things like finding a prisoner hidden in the van part of a truck.  The monitoring done as the truck leaves through a, with the engine running.  Lot of sounds, but the key one is: how many heart beats?  

In addition, one often needs more than one indicator (sensor type) of colony condition, and the analysis has to  be some form of multi-variate analysis.  Statistical approaches may work, but machine learning learning that focuses on patterns may be more useful.  Think about Siri, Cortana, Alexi, Google voice recognition.  It took decades to figure out voice recognition - Dragon Speaking Naturally changed the analysis paradigm.
More than anything else, we  find that the key to is in the analysis.

             ***********************************************
The BEE-L mailing list is powered by L-Soft's renowned
LISTSERV(R) list management software.  For more information, go to:
http://www.lsoft.com/LISTSERV-powered.html