I really like the idea of making useful data by pointing a camera out my window. I have a Raspberry Pi lying around (no camera yet) and lots of windows with trees outside. I guess the window should be one that is not going to be opened all summer so the equipment is not disturbed. Also, changes other than vegetative should not be a big part of the scene (no cars parking intermittently, or new structures being built). Does it make any difference whether half of the scene is white pine trees (the other half is Vermont deciduous forest or lawn). How will the Pi do if installed on a window in an unheated garage (drops to -20 F in winter)? There are also lots of bugs in the garage. This could be a fun project.

Chris

Hi Chris,

With respect to vegetation types, everything goes. As long as there is vegetation in the field of view this should work. If you check the PhenoCam webpage you see a lot of diverse sites (but mostly still research locations, but some schools as well). I'm currently working on an evergreen study which involves white pines, so yes those would work. Actually the more vegetation types in the field of view the better as this provides a way to assess the response of different vegetation types to varying climatological conditions e.g. the evergreens might not care so much about temperature as the deciduous trees do in spring.

I'm not sure about the temperature limits on the electronics. Most of electronics generate some heat which prevents them from freezing (but -20F is low). We have StarDot cameras in Alaska working the whole year round, but they run rather hot. Conceptually I did design the housing to be simple and for indoor (room temperature) use. So time will tell if it will keep functioning at such extreme temperatures. The bugs won't be a problem I think, and if not sure I might suggest to cover the gpio pins on the pi with some insulating material so they can't get shorted by a bugs goo.

Cheers, Koen

So it could be useful to point the camera at a scene with solid white pine trees on the left and mixed deciduous on the right and lawn at the bottom and then the images could be batch cropped for finer scale analysis.

How about opening the window in summer? Finding a window that stays in the same place all year and never has a screen in front of it might be a problem here.

Are the photos saved to SD card? So anytime you want you can steal the SD card and dump the photos and return the card before the next scheduled photo? There is no live feed of the photos?

So anybody with a Pi and a Pi cam can grab your software and start collecting?

I guess I should make a 3D printed housing for the Pi. I have a couple of suction cups. Do you have the hardware plans in STL format (or Sketchup)?

I'm still working on the software part to add functionality, as suggested in the note.

As it is currently set up, the script now uploads images to my ftp server, unless there is not connection. The no connection situation is not ideal, and would require the addition of a hardware clock to keep accurate time.

There is a live feed of the images, better the last image is saved in the home directory. In my latest install script (untested) I will install a http server on the pi. This would allow you to browse to your pi and look at what the camera was seeing during the last snapshot.

There is also no reason why concurrent services couldn't be run if the handoff of the camera is done properly, hence my suggestion to stream the data as a security service in addition to having it record at an interval.

An STL file can be found here, but I'm not sure if this will hold up. The 3mm plywood is rigid enough but I'm not sure how the plastics compare.

Ah, uploading each image sounds like the way to go. Will the default connection be wired or wireless? Thanks for the stl, it looks good.

Hi Chris,

I think I didn't answer your questions fully. With some tweaking of the script, yes you can grab the upload_image.sh and get_weather.sh scripts to start logging data.

The upload_image.sh script needs to be edited. You have to set a proper site name, and you might want to check the orientation of the image before you upload. I'll put some more hours in development to make everything easier to deal with.

As I mentioned in my post, the software is rough around the edges and is currently what needs the most attention, mainly making the install as easy as possible (if not make it so that you can do it remotely). I'm aiming for similar functionality as described in my PhenoCam installation tool. This little script takes care of setting up the camera on commercial (expensive) StarDot cameras. Unlike the pi it relies on telnet, which is easier to script I feel.

Also, the window blinds as well as moving are kind of an issue. I hope people would be able to put it up somewhere on a lost (not often used) window (think attic windows or as you mentioned garage windows).

The connection doesn't matter, however a wireless connection will need some more attention during setup (you have to input the password to your router). A wired connection is easier and probably more reliable, but wifi has the advantage of being relatively location independent.

You obviously need a wifi usb stick, as wifi is not integrated into the raspberry pi. If you have one you can just follow the normal installation instructions.

This is a really elegant project! How do you address the differences in light temperature (RGB) and intensity throughout the day? I wasn't sure if you would include a separate light sensor to adjust for that or if that's not necessary. Is there any value in the hourly resolution data as compared to the yearly stuff? I'm wondering if someone like Planet Labs (https://www.planet.com/) might be able to get daily level satellite data, but they definitely will not get hourly (plus of course that data will always be expensive to access).

I also wonder if you've considered anything else you could measure if you're going to have a camera set up besides just greenness? Not sure of ideas, just seems like if you've got the hardware sitting there may as well get the most out of it!

Hi gbathree,

we did a study on the "behaviour" of the greenness (Gcc) values across a day and across the year. In general the best values (greenest) are acquired on sunny days with a camera facing north (little shading as possible).

Below you can see the daily patterns of Gcc for sunny and and cloudy days for cameras pointing north. The pattern on sunny days is fairly symmetrical around solar noon. This changes when not pointing north. However, if you acquire enough images than using the 90th percentile of 3 consecutive days guarantees more or less that you will approximate a "good" Gcc value. Streaks of bad weather can pull the values down however. A few of the cameras take far less pictures (e.g. just around noon) and still deliver reasonable time series. However given that you rely on the distribution of good to bad images having more images makes for better final data.

You can download the full study here

My setup relies on the cameras being indoors, so additional measurements should be derived from image properties other than colour. I've worked on using texture instead of colour to look at seasonality. The figure below illustrates that there is a discrepancy between what you see using colour (red line) and what you see using texture (black dots). The "proof of concept" texture index is variable as I didn't smooth across days, due to the calculation intensive nature of the analysis. The black dots fall of way later in autumn than the Gcc (red line) values. This is due to the fact that a changes in the physical structure of the canopy in autumn do not align with changes in colour. In spring leaf development coincides with an increase in greenness, in autumn leaf drop isn't necessarily synchronous with leaf coloration (most of the time far from it actually).

Additional measurements, using other sensors will be difficult I fear due to the fact that you are indoors and mostly orthogonal to the sky. You could measure broadband radiation using a simple LED sensor, but this would add little data but would increase complexity a lot. The link to the full article is here. A more complex setup can be found in my other remote sensing notes, the multispectral camera. This opens up some opportunities but again increases complexity and cost (beyond a citizen science / schoolyard budget).

The planet labs data is interesting for sure but as with most high resolution data it is prohibitively expensive (especially if you want to do continuous monitoring).