A cup of deep learning coffee

Recently I started experimenting with deep learning, a set of algorithms that attempt to model high-level abstractions in data by using multiple processing layers. The reason for this excursion in a more exotic classifier framework is the tricky issue of snow.

Snow on evergreen canopies artificially decreases the greenness value of an image, corrupting an otherwise rather smooth PhenoCam time series of greenness. Below you see a split image of a normal snow free canopy, and a snow covered canopy, visually showing this decrease in greenness.

These snowy days result in the dips in Gcc (greenness) as seen in the time series of image greenness below.

Within the lab we tried various techniques to spot snow on these canopies. These techniques were mostly colour metrics, based upon the distances in colour space (from either white or grey) of a particular pixel or region of interest. However, all efforts failed or were not generalizable across all sites, meaning that every site would need to be parameterized independently (which is a processing headache).

In an effort to address this classification problem I installed the SegNet variety of the Caffe Deep Learning Framework (hence the blog post title). The SegNet framework allows for pixel based classification based upon a deep learning approach, originally designed to quickly (matter of millisecons) classify street images to assist autonomous vehicles. However, I hope this approach might help solve the issue of classifying these snowy days, recognizing snowy canopies instead of pedestrians. Results will follow in the coming weeks.





Dotted lines and asynchronous behaviour in tropical trees

A lot of the Jungle Rhythms project is shrouded in secrets. Some of them more elusive then others. One of these secrets is the reason for the dashed appearance of some of the life cycle events in the hand written notes, instead of the usual full lines or cross hatched lines.

Dotted lines in a Jungle Rhythms yearly section[/caption]

These dashed patterns are very difficult to transcribe. The question therefore remains, should it be transcribed to begin with? From a data retention point of view the answer is simple, yes. Any data which is not marked, as written in the original, is data lost. However, this might not be a convincing argument for most citizen scientists, as they are a nuisance to mark, nor does it explain the underlying nature of this signal.

Browsing through some of the data I found evidence for mixed use of dashed / alternating patterns and full lines (marking a continuous process, see right side of the example above). This suggest that the alternating pattern is true, and not a different style of marking continuous life cycle events.

Tree displaying multiple life cycles[/caption]

Thinking about the dotted line problem, I realized that this might have been the way to mark the occurrence of multiple asynchronous growth phases on the same tree. Partial blooming, fruit and leaf development are common among tropical species and obviously hard to classify as a continuous and discrete process, the tree doesn’t behave as one. A picture I took near the Congo river in Yangambi shows a tree displaying three different leaf development stages is shown above. Here, the dull green leaves are the old ones, the bright green leaves are new ones, while the yellow / red ‘leaves’ are either very young leaves or fruit

Until I find the protocol used in creating these tables I will not know for sure. However, cross referencing some species with known life cycle behaviour in existing databases could confirm that dotted lines in the markings are those with partial blooming / leaf out. My search for answers continues.

Why Jungle Rhythms’ pages are yellow

One of the main reasons to digitize the large summary tables used within the Jungle Rhythms project was to save a digital copy of the physical original. The original copy as stored in the Yangambi herbarium was in a dire state, and preservation was a primary concern. Certain paper, when not stored properly, has the tendency to yellow and slowly become brittle and decay as time passes.

Mainly, newsprint and some other papers are made from ground wood or straw contain more non-cellulose compounds (lignin) compared to cellulose in European paper (pure cellulose) or grass fibers in Asian style paper.  These non-cellulose compounds and lignin in particular are subject to oxidation, which alters their molecular structure by turning them into amongst others phenolic acids, and changing the colors we perceive to yellow and brown.

These phenolic acids not only make the paper yellow, they also make it brittle. These days paper will have either all lignin removed, or will contain an alkaline substance (calcium bicarbonate) to neutralize any lignin left, this kind of paper is called acid-free paper.

Luckily the paper used for the summary tables was sufficiently thick to keep it from falling apart due to oxidation. However, even so it was not safe from water damage or the occasional rodent looking for nesting material. The current state of the paper is therefor a mix of internal paper chemistry, adsorption of dust and dirt, water damage and mechanical damage / stress (rodents, manipulation,…).

(sections copy edited from the Smithsonian Institution Archives. For more information see this research abstract and the series of articles.)


Warm winter weather confuses plants

Warm December weather dominates in both the north eastern part of the US as well as large parts of Europe. At both sides of the ocean this warm weather makes plants equally confused.

Many plants are blooming in New York’s botanical gardens and Boston parks. It’s unlikely that most perennial plants will suffer irreparable damage. However, Belgian fruit farmers fear that an untimely frost on the unhardened fruit trees might have serious consequences for both the tree vigour and fruit yields (see movie below).

Plants can withstand frost, however, the continuous warm weather in Belgium has left many trees not acclimated to true winter conditions. A sudden return to normal, freezing, winter conditions could cause frost damage to tissues otherwise protected by a tree’s natural anti-freeze, e.g. sugars in living tissue.

With no real frost days in Belgium the effect of this mild winter might even extend into next year’s spring. Many trees need a certain amount of chilling days, or days which are sufficiently cold / freezing, to trigger a proper leaf-out response next spring. High spring temperatures should move the development of leaves towards earlier start dates. However, a lack of chilling days has shown to delay this expected response to warmer spring temperatures. Warm winter temperatures therefore not only pose an immediate risk, due to sudden freezing of tissue, but have delayed consequences which extend into the next growing season.

(Header image: early leafing Sambucus nigra)

PhenoCam grassland paper accepted

I haven’t made much noise about this but the grassland study I submitted to Nature Climate Change was accepted. I’m working on the final edits to resubmit after Christmas. I’m rather happy with this outcome. For more details I suggest to keep an eye on the Nature Climate Change portal. I will discuss the work in detail after the press moratorium.

With the upcoming data paper, presenting an extensive dataset of curated PhenoCam data, there might be enough data to do more grassland related work as many grassland sites were added, where I was rather data limited before.


© 2018. All rights reserved.

Powered by Hydejack v7.5.1