Two personal hobbies collide in this post; vivarium temperature and humidity management for the home of our two little dart frogs, and data analysis of large volumes of data from multiple sources. What fun!
There were two main phases to this project. The first was to locate and analyze temperature and humidity data from the natural habitat of our dart frogs to determine what our target environmental conditions within our vivarium should be. The second phase was to buy a temperature and humidity data logger that could be used in and around the vivarium to profile the temperature and humidty patterns each day for initial baseline readings and subsequent readings following corrective actions.
This is a fairly large, long running project. I’ll continue to update this page as I move this project forward.
Disclaimer – I’m a dart frog hobbyist – I do not claim to be an expert in the field, nor am I directly familiar with this region or the specific weather station being studied. I enjoy data analysis and this project was developed to aid my understanding of azureus climate requirements an as a vehicle for exploring interesting data visualization techniques. YMMV
So what was I after?
Why wicked cool visualizations that could be applied to the hobby of course! Here’s what I ended up with:
Read on if you want to understand what you are looking at, and how I went about building them.
Profiling dart frog locale specific weather
We have a post about our dart frogs discussing their origins. They hail from Suriname, South America, specifically from the rain forests surrounding the Sipaliwini Savannah. The first phase of this project was to come up with a data source detailing weather conditions near this locale. Lucky for me, there’s a weather station in the preserve that protects the savannah.
Sipaliwini Savannah Data
After a good bit of searching the Internet I came across a site that contained weather data for the weather station of interest within the Sipaliwini Savannah Preserve.
Here’s a good source of data that I will be tapping in the coming months:
Automating data prep
I ended up skipping the automation of the data collection and manually pulled the data for this analysis. It took a few nights of screen scraping. I may explore automating this moving forward if I decide to expand the analysis to other weather stations, but for now, I wanted to focus on capturing the data for this project.
The weather station data
There are two main graphics summarizing the available data. Click on the plot to the right for a full screen view. Depending on your browser you may need to click on the graphic again to fully expand it.First up is coverage. This is a time series plot illustrating the available data points over time, with a “dot” representing a daily reading. Transparency of the point was set to 50% allowing for a gradient to display the relative density of samples.
What you will notice is a dramatic break in sampling for nearly a decade between 1987 and 1996. Coverage was typically erratic especially from 2000 to 2011. The gradient plot below the scatter plot illustrates the monthly totals for each year progressing down the chart, one year per row. The gradient illustrates no data (white) to full data (dark gray) for each month. Coverage was best from 1984 through 1986. In all, about 4,500 daily values were available.
The next set of plots summarize the monthly statistics compiled across the entire data set. Three sets of data have been presented – temperature in Fahrenheit, temperature in Celsius, and Relative Humidity. It should be noted that the statistics presented are slightly different for temperature versus RH data. The average daily temperature, average high and average low are presented, along with the record high and record low for each month. I was able to present the data in this manner because daily high, low and average temperatures were compiled. RH data is represented with median, 25th percentile and 75th percentile values as well as the record high and record low for each month. The switch in statistical presentations was made because only a single daily RH value was captured.
Temperature plots are presented to the right in the upper two plots. Click on the plot for a full screen view. Depending on your browser you may need to click on the graphic again to fully expand it. Temperatures in the region were fairly consistent, ranging from 77 to 80 degrees Fahrenheit over the course of the year. The range of daily highs and lows increased in August through December with average highs peaking in October and November.
Relative Humidity was specifically of interest due to the importance mapping back to the vivarium conditions. Questions around optimal vivarium humidity seem to be more prevalent in the hobby then temperature. The data (final plot in the compound plot to the right).
Median RH values ranged from daily lows of 70% in October and November (periods of highest temperatures) to 83% in May. This was interesting data, as common advice within the PDF hobby is to target 90%+ RH in the vivarium. This data however indicates the macro-climate is less humid. This data does not provide any insight into the micro-climates within the island forests themselves, but overall, this forested region showed lower RH values.
Further reading within the PDF forums indicated more advanced keepers targeted 80% RH for their vivariums which aligns with the data collected. Further these keepers also “cycle” their frogs, often dropping the temperature and humidity to simulate a winter for some species. This may not necessarily mimic the azureus habitat based on the data collected, but the general lower RH range that I have encountered anecdotally seems to align better with the field data then most hobbyists target ranges.
A second, parallel phase of the project was to profile our vivarium to determine what the internal temperature and humidity variation looks like in several locations of the vivarium. Prior to this project we monitored our vivarium with entry level analog gauges from Zoomed (dual units – temperature and humidty) placed in the lower front left and upper rear right. These gauges did not show much fluctuation in their readings. The accuracy of this class of gauge is often called into question within the hobby, so an alternative gauge will be required for this study. Acquiring a more reliable data logger will also provide a comparison to these “entry level” devices.
A data logger was required to collect continuous temperature and humidity data within the vivarium. After some research, the EL-USB-2 form Lascar was selected. This unit can collect 11 days of data with a 1 minute sample interval. The logger can only be configured through the included Windows based software. Installation and setup of the software was straight forward, and the interface was simple to use. Within minutes the device was configured and ready for use. Here are a few screen shots from the user interface:
|data logger interface||sample output||built-in plotting tool|
A 7 day sampling period was selected for this project. It should provide a suitable duration for tracking any variations that may occur by typical weekly activities like cleaning the viv once a week on the weekend. The approximate schedule for data collection was:
|ambient room (6 ft elevation)||7|
|lower front left||7|
|upper rear right||11|
|lower front right||7|
|upper front left||ongoing|
Baseline room data
Initial data collection occurred next to the vivarium at an elevation of approximately 6 feet. This elevation is above the top of the vivarium. Data collection occurred for just over 7 days. The intent was to capture a prolonged run to provide a sizable number of data points for a stable hourly temperature and humidity profile for an area just outside the vivarium. The following plot summarizes the hourly averages for the seven day period:
You can see the stretches where the recorded temperatures are spot on with the programmed targets. The variability during the adjustment periods is expected, and the variability at night can be attributed to the layout of the house; during the day the main zone and 2nd floor zones of our house (open floor plan) are set to the same temperature – at night the upstairs (viv location) drops while the downstairs increases, so the temps upstairs fluctuate more yielding more variable logging results. Overall I feel comfortable with its accuracy at this point. I do have a voltagemonitor with temp/humidity logging capabilities that I can fire up, run side by side and compare the differences, but I’ll run that in a few weeks after I am done profiling the viv.
So far, the data I have collected is summarized in the plot to the right- the average values for each location for the 7 days of collection is on the left. Full hourly profiles are the majority of the plot moving from left to right by date of collection. Temperature readings are on top, relative humidity is below. Gradient shading was used for the min/max of the full range extending from room temperature (first block of data in the middle) to the first viv location (partial data on right).
After initial profiling external to the vivarium, I placed the data logger in the front lower left of the vivarium. This is the most humid area of the viv based on a visual assessment. The logger was placed in this location for 7 days using a similar method to what was outlined with the baseline profile. The data was plotted using the same method as the baseline data, aligned in a time series view with the baseline heat map.
Overall there was a dramatic increase in humidity relative to the ambient readings. Temperature data was relatively consistent between the ambient temperatures and vivarium temperatures. Further ambient readings should be collected at a lower elevation to determine if the warmer ambient temperatures that were recorded for the initial sample interval were more a function of the vertical temperature gradient in the room rather than a difference in vivarium conditions (the logger was about 18″ higher than the top of the viv for the initial sample period – had to keep it out of the kiddo’s hands…).
Changing the graphic
After several months of collection, a horizontal view proved to be challenging. Though this seemed like a more logical way to display the data, I found it easier to view the data when I transposed the view and plotted days as rows and hours as columns. The new views with adjusted gradient scales are (click on each graphic for a full screen image):
Thought the yellow/red color scale seems to match up better conceptually with temperature, I found sticking with my original preference for more muted tones to be more appealing. I did modify the blue ranges in use for the relative humidity data, but I did not shift to a gray or earthtone here to maintain a greater visual distinction between the two plots. We’ll see how this continues to evolve.