Forgot your password? Please contact us!
Pt1000 measuring resistors of accuracy class AA are used for measuring temperatures inside the eHives. They are calibrated to an accuracy of ± (0.1 °C + 0.0017 ∙ T) with T being the temperature in °C. Because the sensors are connected by a two wire circuit to the electronics for further signal evaluation, the line resistance also causes an error. Since the length of this circuit is small, the associated error should be comparatively small as well. On the circuit board, the sensors are connected in series with a 4.7 kΩ precision resistor each and a reference voltage of 10 V, c.f. Fig. 1. The precision resistors are exact with a tolerance of 0.1 %. The reference voltage is created by an ADR01ARZ with a tolerance of 0.14 %.
Now, the voltage difference between the temperature sensors and the reference voltage is digitised using an AD7789 analog-digital-converter (ADC). The ADC's ± 0,3 μV offset can be neglected because the change in voltage for a 1 °C temperature difference amounts to 5.7 mV. The ADC's resolution of 19 bit with a reference voltage of 2.5 V yields an effective resolution of 4.8 μV. In consequence, the ADC's contribution to the measurement error can be neglected as well. The 2.5 V reference voltage is created by an ADR03ARZ with a tolerance of 0.14 %, being equal to the tolerance of the ADR01ARZ mentioned earlier.
The sensors' measuring resistor itself is shielded by a shell made of stainless steel which reaches up to 20 cm inside the hive and is located at around two-thirds of the honeycombs' height. Usually, a sensor should be in between two honeycombs. However, sometimes the bee colony integrated one or more sensors in the combs. This is especially the case for top-bar-hives. Since different types of hives are used, the sensors' distribution differs and may not even be symmetric. Usually, a sensor is placed in every second comb alley. The exact arrangement is shown in the following table, where each column represents a comb alley according to its number in the first row of the table. The maximum number of comb alleys is 14, while most of the eHives have less. (Dark grey cells represent non-existing alleys.) The numbers 1-6 (respectively 1-5 for eHives with only 5 temperature sensors) in the remaining rows mark the position of the temperature sensors inside the hive according to their label in the database.
The air humidity inside the hive is measured using a SHT21 humidity sensor. This component is sold in a system that converts the digital signal into an analog signal using a PIC32MX micro controller. In consequence, the signal must be digitised again on the main circuit board. The tolerance of the sensor component is 2 % for humidity values between 20 % and 80 %. It increases linearly to 3 % for humidity values approaching 0 % or 100 % respectively. We guess the convertion to an analog signal increases the measurement error by approximantely 0.5 %.
Because the humidity sensor is located in the lower part of the hive below the combs, the measured value correlates heavily with the outside humidity. Since the sensor is relatively huge in dimensions, it was not possible to place it in a higher location of the hive where the colony's micro climate can be monitored better.
The eHive measures the hive's weight using load cells of type SEB46B, c.f. Fig. 2. On a spring torso made of stainless steel, attached distention straps change their electrical resistance depending on the load acted on the torso. Combining multiple distention straps to a Wheatstone bridge, the bridge's output voltage is amplified using an AD623 instumentation amplifier and then digitised using an AD7789 ADC (c.f. temperature section above). The voltage source for the Wheatstone bridge is identical to the one for the temperature sensors. The amplifier increases the signal strength by a factor of 99 with a deviation of 0.35 %. The amplification factor is regulated by a resistor and inherits its 0.1 % tolerance. The load cell's resolution amounts to 0.2 μV/g prior respectively 19.8 μV/g after amplification.
The load cell's smallest possible resolution is 6.6 g according to its spec sheet. At constant temperature this resolution can be reached by our weight measuring system. However, temperature changes can result in weight changes of some one hundred gram. Rainfall can lead to additional deviations because not all eHives are equipped with a roof (c.f. blog post The eHives' locations and surroundings). The sensors' drift over long periods of time is not known.
All weather data, except air pressure, is measured by the weather station Vantage Pro 2 from supplier company Davis. Since the measured values are read out digitally all tolerances can be taken from the weather station's spec sheet. In the following paragraphs additional error sources and external influences are listed.
For measuring air temperature and humidity in a comparable manner, the weather stations must be installed on a open area with defined surface and in equal height. However, reality is complicated and in consequence some weather stations are located on grassland, some on roofs, etc. which may influence the measured values. Nonetheless, it was tried to install the station around 2 m above the hive to monitor the relevant local micro climate at the hives' site.
Rainfall is measured with a seesaw below a funnel. As soon as a well-defined amount of water accumulated in one of the seesaw's two shovels, the seesaw dips over and activates a Reed relay using a magnet attached to the seesaw below its turning axis. One dip translates to 0.2 mm of rainfall. As a consequence of this technique, snow can only be measured in the process of melting, supposing it was not blown away from the funnel due to high wind speeds and/or the funnel's overflow prior to melting. Though electrical heating components are installed inside the rain meter at some of the eHives' locations, those heating components are not in use in most places due to low effectiveness. On top of that, the funnel sometimes gets blocked and may not be cleaned regularly, resulting in measuring zero rainfall over long periods of time.
Equally to air temperature and humidity the weather stations installation site may influence the wind values. The height of the wind sensor is supposed to be around 5 m above ground but can deviate heavily at some locations. For measuring the wind direction, the sensor must be oriented exactly northwards. However, that is not always possible to a high precision, so the measured absolute values may deviate by some degree. The listed measurement error is a relative error.
At some locations the two irradiation sensors are shadow-casted by trees of buildings during parts of the day. Usually, once a day the shadow of the installation mast casts the sensors. This happens because the two sensors are attached to the weather station's base component and in consequence to the fact that in most cases both the base component and the wind sensors component of the weather station are installed to the same mast.
For measuring air pressure a BMP280 sensor is used which is located on the main circuit board. The sensor's output is digital and therefore the measurement error in the spec sheet can be adopted.
For measuring currents ACS712 sensors are used which supply a voltage proportinal to the current. This voltage is digitised using the Arduino's internal 12 bit ADC that works with a maximum voltage of 3.3 V and therefore theoretically exhibits a resolution of 0.81 mV. Since the current sensor outputs 0.185 V/A, one step of 0.81 mV translates to a change in current of 4.4 mA. According to the sensor's spec sheet its resolution amounts to 75 mA, which was rounded to 0.1 A due to the small effective resolution of the Arduino.
The sensor for the radiator current can be allocated freely. For example, it would be possible to measure the charging current if an eHive is powered by battery.
The charging voltage is the voltage at which the electrical system is operated. Since as of today this voltage is supplied by a power-supply unit at all eHives and therefore is almost constant, the name is a bit confusing. The voltage is measured using the Arduino's ADC after being reduced by a voltage divider using a 1 kΩ and a 4.7 kΩ precision resistor. The resisitors' contribution to the total measurement error can be neglected. The theoretical resolution therefore amounts to 4.4 mV, however it was rounded up due to the small effective resolution.
The Arduino's mirco controller features an internal temperature sensor which absolute value must be calibrated individually for every single sensor. Since this was not done, the absolute values of this sensor exhibit an offset of ± 45 °C. The relative error amounts to ± 3 °C according to the spec sheet.
(for details and comments c.f. sections above)
|Inside humidity||2.5||%||below 20 % and above 80 %: increase to 3.5 %|
|Weight||0.01||kg||higher deviations at changing temperatures|
|Outside humidity||3||%||above 90 %: increase to 4 %|
|Rainfall||0.2||mm/h||above 4 mm/h: 5 % of the measured value|
|Wind speed||3||km/h||above 60 km/h: 5 % of the measured value|
|Wind direction||7.5||°||relative error|
|Microchip temperature||3||°C||relative error, offset ± 45 °C|
The diagram viewer of BeeBIT's website has a function to export raw data. That offers the possibility for a deeper analysis of the collected data. In the following article, temperature and weight data of eHive AUT-GSC-1 is studied to exemplify this statement. For better traceability the data export and the subsequent analysing attempts are structured in steps. Data processing is done using Python. The complete Python script as well as an explanatory PDF document are linked at the end of this article. The purpose of this text is to give inspiration and some examples of approaching the investigation of BeeBIT's data sets. As we will see: Many things can be noticed on second sight only.
In the diagram viewer data sets that will be examined afterwards can be selected and viewed intuitively. As an example we will take a look at the temperature data inside and outside of the eHive (6+1 sensors, the sensors inside are placed at different locations between the honeycombs) as well as the weight data. As a time frame we select June 10th until (and including) June 12th. Doing this, we must be aware that all times are given in the standard coordinated universal time UTC+0. As a consequence, the time in the local time zone (that is central European summer time UTC+2) will be shifted by 2h. So for looking at the three day timespan from midnight to midnight in local time UTC+2, we want to export data from June 9th 10 pm until June 12th 10 pm in the diagram viewer's default time UTC+0.
By clicking the Download button in the upper right corner a csv data file containing the selected sensor data sets is generated by the server and provided for download. This file can be opened using text editors or spreadsheet programs. However, since we also want to perform computational heavy tasks on the data, a script in the high-level programming language Python is used to read, analyse and illustrate the data set.
Nevertheless, we first open the csv file with a text editor to grasp the data set's structure. The default name of the downloaded file is data.csv. In the first line the names of the archived data sets are saved. The raw data is stored in the following lines where a line break divides two consecutive time steps. Inside a line the semicolon is used to separate two data points. Hereby we can structure the file into rows and columns. The first two columns of each downloaded csv file contain time information: in the first column as Unix timestamp in decimal representation, in the second column in a human-readable format. In the other columns the beforehand selected sensor data is stored, each column corresponding to one sensor. If data from before or after the selected 72h timespan is still stored in the file, it could now be deleted manually.
The first and last three lines with raw data from the downloaded file data.csv are printed here. The names/identifiers of the data sets are not shown.1560117600;Sun, 09 Jun 2019 22:00:00 GMT;28,5;28;30;34;34,3;34,9;52,61;22
Since we have understood the structure of the data we now want the computer to read it using a script. After that we try to visualise the raw data unaltered before we go on with further analysing steps. In the paragraphs below we use the free programming language Python 3 in combination with its external function packages NumPy and Matplotlib. Python is platform-independent and relatively easy to learn. Thus, the presented analysing methods could be applied and expanded by students with interest in computer technology, e.g. within the scope of a school project.
First of all, the data is read and saved in NumPy-arrays for plotting it over time using Matplotlib, cf. Fig. 2. Since only the relative time information is relevant to us, we can organize the horizontal time axis in units of hours with 0h representing June 09th 10 pm (UTC+0, June 10th 00 am in local time). Because both temperature and weight data is stored in the data file we will need two vertical axis: one in units of °C, the other one in units of kg. Different colours and a legend will be used to distinguish the sensors. As a guide to the eye two vertical lines with a time spacing of 24h are inserted to mark the day change at midnight in local time.
We have already accomplished the hardest task! Now, that we understood the data set's structure and managed to visualise the exported data, we can perform some investigations on it. Before doing this, we want to point out that just by looking at the temperature data one can already get an insight on the position of the bee colony inside the hive. As it can be seen in the plotted temperature curves, the inside temperature sensors 4 to 6 show little change in the measured values and fluctuate by approximately 3 °C around a mean value of 35 °C. This exactly meets the target temperature the bee colony needs for the larvae brood. In contrast, the fluctuation amplitude of the sensors 1 to 3 is nearly 10 °C. As a consequence, we can conclude that the colony mainly lives near sensors 4 to 6.
We now would like to deeper investigate the observed temperature regulation as this would be possible with the diagram viewer on the BeeBIT website. Especially the differences between the temperatures inside and outside of the hive depending on the position of the inside sensor are of relevance to us. Therefore, we exemplary select sensors 2 and 5 and plot the inside-outside difference. Since we want to make a more general statement which should not be dependent on single events during data capture, we take the daily mean over the three day timespan, that is we calculate one mean temperature value out of three values measured on three days at the same time. NumPy's fast and intuitive array handling makes this an easy task. The complete 72h data set as well as the 24h mean are shown in Fig. 3. For both a shared temperature axis is used.
Looking at the inside-outside difference we can conclude that the bee colony regulates the hive temperature. We already have stated that the colony is located near sensors 4 to 6. During night time the hive is heated there so that the temperature does not fall too low for the larvae brood. Near sensor 2 the temperature follows the outside temperature curve. Between 2 am and 7 am the inside-outside temperature difference at sensor 5 reaches a maximum of approximately 14 °C in the three day average. We get a completely other picture if we look at the warm afternoon and evening hours: Here the temperature difference at sensor 5 falls below the values at sensor 2. If we assume that the curve at temperature 2 follows the outside temperature nearly unperturbed (with a constant offset value), we can conclude that the colony actively cools the surrounding brood-/honeycombs. Indeed this behaviour was observed by beekeepers. Worker bees carry water into the hive which evaporates and dissipated heat so that the hive stays cool. Thus, we have reason to believe that we were able to observe this interesting phenomenon in the analysed data!
After investigating the temperature regulation inside the eHive we want to conclude this article by analysing the weight readings. As one could already have noticed by using the website's diagram viewer, the hive weight increases from 52.5 kg to approximately 58.0 kg within the three day timespan. However, the absolute weight of the hive contains the weight of all associated components (wooden frame, sensors, IT components, biomass). Another quantity that can describe the relative weight change is of more interest to us: the change rate, that is the time derivative of the weight readings. This quantity can be calculated by subtracting two neighbouring data points and divide the result by the length of the corresponding time step. If we visualise the calculated change rate (Fig. 4, upper half) we notice that the measurement signal is very noisy. The noise reduces if we again calculate the three days mean. However, we still cannot make a reasonable statement about the weight curve using this data.
It would be desirable to process the data in a way that removes the noise from the signal without changing the underlying meaning of the measured values too much. One known method of noise suppression in signal processing is the use of signal filters. We will apply the so-called Gaussian filter to the signal. For this purpose, interpolation points are calculated from the data points. Here, the individual data points that are used to calculate an interpolation point are weighted differently. A data point that is close in time to the interpolation point is weighted more heavily than a data point many minutes or even hours away. The weighting corresponds to a Gaussian bell curve, which is centred over the respective interpolation point. Details on the interpolation and its algorithmic implementation are linked at the end of the article in the form of an explanatory PDF file.
In the smoothed curves (Fig. 4, lower half) we can see that especially in the evening between 4 pm and 8 pm the hive's weight increases strongly with up to 0.3 kg/h. Overall, a positive balance of the change rate is observed, that is in daily average it is greater than zero. This effectively leads to an increase in the hive's weight, which we already noticed at the beginning of the analysis. Surprising is the strong rash in the early morning: After the weight first decreases sharply with up to -0.4 kg/h, it rises up to +0.5 kg/h again shortly afterwards. At 9 am in the morning, the rate of change then has dropped back to about 0.0 kg/h. This effect can be seen on all three days of the observation period and is also reflected accordingly in the three-day average. Meikle et al. interpreted the sharp decline at dawn as a wake up phase of the bee colony. While the workers prepare for the first flight of the day (heavily consuming food) and then gradually leave the hive, the weight drops. The subsequent sharp increase could not be found in the publication by Meikle et al.. It may be possible, for example, that the bees carry water into the hive. However, the change in weight may have no biological cause. One can think of it being caused by the morning dew precipitating on the outside of the hive. A quick look on the website's diagram viewer could reinforce this second guess: just at the time that the weight change rate rises sharply after the wake up process, the outside humidity reaches a maximum of up to 80 % on all three days of the observation period.
We exported data from the BeeBIT diagram viewer and processed and visualised it with a Python script. The temperature regulation in the inside of the hive and the weight readings were examined in more detail. The difference between two internal temperature data sets and the time derivative of the weight curve was calculated numerically. On the latter, a Gaussian filter was applied to suppress the noise of the measurement signal. The methods shown are intended as examples of reasonable analytical approaches. Such approaches are used in scientific research, but can also be developed by students or interested adults. The eHives are providing interesting and interpretable data sets. The shown data analysis methods can be varied in complexity and their degree of difficulty. Working with just a small set of data in an easy-to-use spreadsheet program can also generate new insights. (An example suitable for school lessons can be found on the BeeBIT website in the learning materials section under the title Data analysis with simple functions and charts. There, data on outside temperature and precipitation are organised and visualized in an Excel spreadsheet.) In-depth numerical analyses, as they could be carried out in school or university projects by older pupils or students, allow a thorough study of the biological aspects of the collected data and the technical methodologies.
If this article, dear reader, has piqued your interest, do not hesitate to try it out for yourself. Download a data file from the diagram viewer, open it with a program of your choice, and take a closer look than you could before on the website. As promised at the beginning of the article: Often a second look is worthwhile.
If you need any kind of help, have questions concerning this article or want to tell us about your work with the eHives' data, feel free to contact us.
Since many month, we thought about a new design for our website and tried to expand and renew the published information. With the new website, a blog system was implemented in which small or big news can be released easily. At https://beebit.de/en/blog you find worth reading articles from the up-to-now spreaded newsletters. In future, the new written and published articles will be summed up to big newsletters in regular time periods. This provides the chance for you to access news on the blog directly without having to wait for a newsletter. In consequence, you can read the articles shortly after they are written.
Furthermore, the start page contains a new feature: a table with informations about the technical state of the eHives. As you may have noticed, not all eHives send data. We tried to make the reasons for this transparent in the table. It will be updated in regular time periods.
On the old website all up-to-now released newsletters could be accessed. The new blog system makes this largely unnecessary. However, if you wish to read the old newsletters, they can be downloaded by clicking on the following links. You will be redirected to the corresponding PDF file on our server.
Please keep in mind that the old newsletters may contain outdated information. Up-to-date informations about the eHive project can be found for example on the FAQ page of the website.
The first eHive with version number 0.1 was built by a school seminar of the "Deutschhaus-Gymnasium" in Würzburg back in 2014. This prototype was stocked with bees and operated for one year. During operation, data was collected on location. In 2015 it was replaced with the eHive 0.2 that was built in cooperation with the "Franz-Oberthür-Berufsschule". This version was shipped to our partners from the EU project. At first, it did not work properly, but until April 2016 amendments were applied. Since then the eHives work comparatively stable.
Since 2017 eHives are distibuted by the BeeBIT association. After the first two eHives, the design was slightly changed, but the functionality stayed the same. Hence, the eHive 1.2 is the only one that can be purchased at the moment. The weather station at all different versions of the eHive is the "Davis Vantage Pro 2 Plus Cable".
The eHive 0.2 exists in two models: As "Zander" hive (eHive 0.2.1) and as "Top-Bar-Hive" (eHive 0.2.2). At the beginning all eHives were Top-Bar-Hives, because these allow the bees to live closer to nature. A small hive with 13 not-preshaped honeycombs was used. In it six temperature sensors were obstucted and one outer honeycomb was replaced by a wooden separation board in order to distribute the sensors symmetrically. A humidity sensor is installed in the lower part of the hive.
The sensors' data go through a junction box on the side of the hive and are guided from there by a water-resistent RJ-45-cable with screw connectors until the data ends up in the actual junction box that contains all the electronic componentes. At some eHives this box is mounted on the other side of the eHive, at some it is attached to different support structures or is laying on the ground.
To offer compatibility with other hives some of the eHives were converted to other hive-types (other than the Top-Bar-Hive). During conversion only five temperature sensors were installed due to spatial limits. The weather station was attached to an external support. This produced different hights of the weather stations and their wind detectors and different distances to the corresponding eHive. For weather protection a roof had to be provided by the local institution. It was not possible to enforce this on all location.
The two hives of version 1.1 were built in early 2017 with the most strinking change being the new wooden outer case. It contains the eHive and all electronic components and serves as a weather protection. At this outer case the attachment for the weather station is mounted. The height of the weather station was defined to be 2 m and the height of the wind sensor 10 m. Both eHives are built as Top-Bar-Hives and have the dimensions of a Zander hive for 12 honeycombs at the top part. Between the combs six temperature sensors are installed. Both the temperature sensors' connection cables as well as the cables from the humidity sensor, the scale and the weather station are guided directly to the central junction box. This avoids unnecessary causes of defect. Because no more eHives of version 1.1 are planned, this version does not exist in other hive types than the Top-Bar-Hive.
The eHive 1.2 is almost identical to version 1.1, only some changes at the outer case were made. It is enlarged by a removable heightening in order to support hive types with multiple honeycomb boxes. Furthermore, the roof is divided into two parts. This simplifies the usage and allows a waterproof lead through of the weather station's mounting rod though the smaller non-removable roof. The first eHive of version 1.2 was shipped in early 2019. Since then, it can be purchased with different hive types (almost all popular types as well as Top-Bar-Hive).
During a didactic teaching seminar in the summer term (2017 and 2018) with focus on "Innovations in the didactic of natural sciences - gain of scientific knowledge using the honeybee model organism" Christoph Bauer (vice chairman of BeeBIT e.V.) presented the possible use cases of eHives in school classes and in small research projects in front of university students.
In November 2018, Dr. Monika Fröhlich and Christoph Bauer were invited to the national "Science on Stage" festival in Berlin. There they presented innovative teaching methods in front of German teachers. These methods are related to the eHive project. On top of that, experiments about conditioning and the making of durable biological preparations were shown.
The article "Varrose online erkennen" ("Detect the varroa desease online") was written by Isabelle and Christoph Bauer and published in the journal "Biologie 5-10" of the Friedrich Verlag. In the article a lesson is presented, that is based on the eHives' data. By comparing the weight diagrams of two eHives in the same time period, conclusions about the health of the corresponding bee colonies can be deduced. The article can be downloaded here (only available in German).
The eTwinning platform of the EU allows European students from different countries to work together on educational projects. It will be possible to use the eHives' data for establishing a collective "bee project".
BeeBIT already received the first requests concering this topic. If you are interested in such a project, we would be happy to offer you support.