Today I launch Jungle Rhythms, an online citizen-science project that aims to digitize thousands of pages of detailed observations of the life cycle of trees in Africa.

Belgian scientists were stationed at the Yangambi Research Station in what is now the Democratic Republic of the Congo from 1938 until 1958 as part of an agriculture-based research project. During that time, the scientists – for reasons unknown– also began collecting detailed observations on the life cycle of trees in the local forest. Those observations were kept in a series of notebooks, and later summarized in large tables, which were discovered, nearly 80 years later, stored in an archive under less-than-ideal conditions.

To avoid losing the data as the pages crumbled, I digitized the tables in the hopes of using computers to automatically capture the data, but quickly realized the marks were simply too faint.

While the project’s ultimate goal is to preserve the data for future study, it also gives the public an up-close-and-personal view on how scientific research is conducted. To do so, I’ll be blogging about the project to keep users up to date on new exciting results, and any discoveries I make about the history of the data itself.

A new paper I co-authored in PLOS One just came online discussing divergence of above- and belowground carbon stocks in different forest types. In short, the study shows that despite similar vegetation, soil and climatic conditions, soil organic carbon stocks in an area with greater tree height (= larger aboveground carbon stock) were only half compared to an area with lower tree height (= smaller aboveground carbon stock).

This suggests that substantial variability in the aboveground vs. belowground C allocation strategy and/or C turnover in two similar tropical forest systems can lead to significant differences in total soil organic C content and C fractions with important consequences for the assessment of the total C stock of the system, with nutrient limitation, especially potassium, as the driver for divergent C allocation.

Congrats to Sebastian Doetterl and Elizabeth Kearsley for pushing this effort.

The ORNL DAAC which hosts the Daymet data is shifting to a HTTPS only policy. As such I updated my DaymetR R package to reflect these changes. Users running an old version will have to reinstall the new version to reflect the migration to secure data connections.

You can find the updated package on my github page (see link on the software page). Please reinstall the package to guarantee continued functionality. Similar corrections have been made to the DaymetPy code.

Richardson Lab members almost unanimously stated that this year was the brightest and most spectacular autumn display of colours in a long time. However, people are notoriously bad at accurately quantifying pretty much anything they see, hear, smell or feel around us. We often perceive the world differently from it’s physical (absolute) reality. This is easily illustrated by various types of optical illustions or the fact that we do not feel absolute temperatures, only change.

Not only does our brain play tricks on our senses, it also fails us when it comes to interpreting long term (environmental) change, as climate change is slow and our cognitive bias leans towards instant action and gratification.

Begs the questions, can we tease apart the intensity of this years autumn colours, and how they compare to previous years using PhenoCam based data using a more quantitative approach? More so, was this autumn really as spectacular as most people in the lab seem to agree upon?

In order to answer these questions we need to look at some PhenoCam data. Normally we look at the greenness signal, which is mostly driven by the presence of healthy green leaves. In this case however, I focused on how red the leaves were during this autumn. In the figure below you see a time series of canopy redness at the Bartlett research forest in the White Mountains, NH. In this graph you see two high peaks which are caused by leaves turning colour, from green to yellow / red, in autumn in 2014 and 2015.

The yearly maximum value in this graph tells us how intense the colours were. A metric for the duration of the colour intensity can be calculate as the sum of a set number of values before and after this peak value, where a higher value indicates a longer duration of peak colour. Below I show peak values and the summed values for a week before and after the maximum Rcc value.

When looking at absolute values one can see that 2014 was actually the better (left figure, dashed line), brighter year. However, if considering the duration of peak colour 2015 wins by far (right figure, dashed line). At this location autumn colours were especially intense and long lasting (which increases the odds for leaving lasting impressions on lab members going hiking and leaf peepers in general). So in this case, our senses weren’t fooled - and we have the data to prove it!

In order to use the raspberry pi cameras within a more rigorous framework I needed the full spectral response of the chipset used in these cameras, the OV5647 by OmniVision. I set out to do acquire the response curves using a DIY diffraction grating approach.

During this process I was contacted by Howard Shapiro who volunteered his old spectrofluorometer to accomplish this task faster and more precisely. Although, initial tests together with Howard showed promise, it would remain an arduous task to quantify the whole spectral response. Recently, Howard managed to dig up some documentation on the chipset which displayed the spectral response (quantum efficiency, QE) in a graph (given that the OmniBSI chipset as displayed is the one residing in the OV5647). This has made things considerably easier. I used these slides to digitize the quantum efficiency of the OV5647 between 400 and 700 nm. A physical measurement will still be needed to quantify the remaining near infrared spectrum (> 700 nm, data can be found on the OV5647 spectral response github page).

Sadly, the original graph only covers wavelenghts between 400 and 700 nm, leaving out the near infrared (NIR) part of the spectrum. The red edge, located at 680-730nm, is a key part of the spectrum key in vegetation remote sensing applications. At the red edge vegetation reflectance changes from low to higher values. The magnitude of this differences is an indication of plant health.

Although I got most of the picture, due to good detecitve work by Howard, I still don’t have the complete picture. Some physical measurements will still be necessary to get the complete spectral response / QE of the chip but I’m at least halfway there.