Members: Tommaso Dognini, Stefano Fanciosti, Pietro Donghi, Lorenzo Fumagalli
Tutor: Emanuela Magni
School: Liceo Scientifico e Musicale G.B. Grassi (http://www.grassilecco.edu.it/)
Teams of young people design and program a scientific experiment to run on board the International Space Station.
With the experiment the team wants to observe from a very different perspective (from which we are not used to seeing things) the Earth vegetation. They want to investigate how this vegetation has changed in time, what were the causes of this change, and what consequences it will have. With the data collected from the Astro Pi and the ability to know where the camera was pointed at the time it shot our images, students will be able to compare the conditions of the field at the time of the experiment with the condition of the field in the past, thanks to old satellite imagery. The Earth's vegetation is the fertile ground on which animals and humans live and depend on, that’s why a deep understanding of its changes can help us understand how life on Earth will change.
My team will collect data using the “near-infrared camera” sensor mounted on the Astro Pi, which will let us take infrared images of the Earth surface. These images can be used to determine the presence of green vegetation in that area using the Normalized Difference Vegetation Index (NDVI) index.
The NDVI index values range between -1 and 1. They derive from the ratio between the difference and the sum of the radiations reflected in the near infrared and in the red: NDVI = (NIR -RED) / (NIR + RED). For this reason the negative values will be obtained in areas where there is water, snow or clouds. Around 0 there will be areas of rocky or sandy expanses, around 0.5 shrubs and pastures while around 1 rainforests. For this purpose we will use the near-infrared camera to obtain images of the Earth's surface. Each image will be treated as a matrix of pixels. The astro Pi computer will then calculate the NDVI index for each pixel. At the end of this process we will obtain a matrix of NDVI index values for every image shoted. From these data, a false color image will be created, depicting the same area identified by the photograph. This will allow us to visually understand the value of the NDVI index.
From these modified images it will be possible to determine the presence and density of green vegetation, and therefore the type of biome. Comparing the result of this analysis with known data we will be able to determine the changes that happened in that area.
Here is a video of the unboxing experience of our astroPi kit.
The code we have created has the purpose to capture the images of daylight landscapes on the earth and calculate the NDVI index in order to compare it to samples calculated with our code using public photos and samples data that are available for all teams .
We have paid attention to small details, such as highlighting the place and the time the photo has been made in order to make also a geographical comparison and justify the expected data.
The color mapping has been projected in order to make the data reading easier and more intuitive, also thanks to the legend with the reference color.
Here is a list of the implemented features:
- The program shoots 300 images thanks to the astroPi IR camera and processes them producing a color mapped image, for each raw image, based on NDVI index values.
- The program croppes the raw image in order to avoid distortion in the NDVI index calculation and color mapping due to the ISS Columbus Module window frame.
- The program keeps track of the ISS position and its location is added to every raw image (IR) as exif data; this will help us in the phase 3 to analyse and locate images.
- Thanks to the ability of tracking the ISS the programm is able to shoot images only in daylight period, this helps to avoid shooting images that are not relevant and useful for this experiment.
- The program keeps track of time, the experiment is ment to run for 3 hours (2h and 57 minutes to be sure everything has been shut down at the end time) and then it automatically stops.
The available disk space is 3GB. By testing our code, we understood the average dimension of a photo is about 5MByte. So we calculated that, in order to stay in the 3GB limit, we can shoot 300 images with the max resolution (4060x3040). For every image we save the original IR image and a copy of the photo with the calculation of the NDVI index (this image is cropped so the resolution is lower), so in total the code saves 600 images. In this way we should end up occupating no more than 2.8 GB of space.
If we consider that our program shoots only in daylight and we approximate the night time equal to day time we can conclude that our experiment will have 90 minutes for shooting 300 images so we have to shoot 1 image every 18 seconds.
This are some example of images
| Raw image | NDVI cropped |
|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
We have receaved the images taken by the AstroPi on the ISS running our program. Check them out in the phase3 folder.
PDF report: F2D2_Report_AstroPiProject.pdf
Our team has created a code in order to shoot images of our planet’s landscapes, store the useful ones and calculate for them the NDVI index. To gauge the process, we compared our images with NDVI images we obtained from standard samples available for all teams. Our aim was to have a good understanding of the photographed subjects and how they have been changing over time. We paid attention to details, such as where and when pictures were taken, to also make a comparison and link our data to events that have affected those particular geographical areas. The colour mapping has been set to make dissemination and data reading and understanding easier. The expected rate of images good enough for the described work was at least 10%.
Our program had the aim to shoot photographs of geographical areas and calculate the NDVI index. To make this possible we used the infrared camera provided with the astroPi kit and we obtained jpg images and .log files of the program logs. During the testing phase, we understood that the ideal number of images to be taken was 300 (eventually 600, because we had chosen to create an NDVI image from each IR image) to respect the 3.0 GB data limit. Originally, we had thought that having the NDVI image would have made the analysis phase easier but, after having received the data, we acknowledged that the results had some problems. This happened because the NDVI images were cropped in a square that was not centred, due to the camera position on the ISS which was different from what we expected. Therefore, black portions in the image altered both contrast and colour mapping process, as we had predicted and tried to avoid. However, we believe that our original idea still works perfectly with the changes we made during the analysis phase on the Earth.
From phase 3 we received 300 images. 80 (26.7%) were black or very dark because we used the .is_sunlit() method of the skyfield library to shoot during light time, but we could not avoid getting images during twilight. Of the 220 photos left, only 40 (13.4%) were distinguishable and had a clear and cloudless subject.
To perform the analysis we cropped the chosen images in a square to avoid black margins as we originally aimed to.
We then wrote a program that calculates the NDVI and creates the colour-mapped image. For this purpose, we looked for a better colour map that could enhance the variation of the NDVI index values. After some testing, we found that YlGn from matplotlib library suited our needs the best.
The legend shown in every image goes from -1 to +1, which are the min and max values of the index, where +1 represents very flourishing vegetation.
Here is what we obtained:
The dark-green area in picture number one is due to the presence of the Yumori National Park, a wooded area. Here the coloration is uniform, while the third picture presents different shades of green because the forest is rich in swamps and lakes. The second and third NDVI images are very defined, so it is possible to recognise the lakes in the forest and the islands in the Volga river. The second image is also interesting because the dark-green area in the upper part is due to the presence of flourishing fields while in the down part the colouration is lighter for the presence of cities, as along the Japanese coast. Using the NDVI index in the Gargano image we can understand the wood is now completely reborn after the wildfire happened in 2007 which destroyed 4.500ha of Mediterranean bush.
While working together on this project, we developed skills and learned much more than we thought. In the first phases of the project, we had to reach the same Python coding level and learn how to collaborate on the same code project using GIT and Github. We also had to learn Markdown and, obviously, how to use RaspberryPi OS. To overcome the numerous challenges, we learnt to split them into smaller tasks according to our points of strength. Having to face the same problems altogether led us to share our knowledge and learn something new from each other. For example, we tried to cancel lens reflection rings from pictures. We found out that one possible solution was to operate a subtraction of a chosen image and a white image with the reflection blemish. Despite our efforts, we could not solve this problem due to the lack of a blemished white image. Looking back we should have focused more on the raw IR images, leaving the NDVI calculation and colour mapping to the phase4 analysis on Earth. We also strengthened our communication skills to describe our work in an efficient way.
Our aim was to analyse the effects of climate change using the IR camera and NDVI index. In particular, we expected to photograph some natural disasters such as wildfires, floods or other issues caused by human activity, such as deforestation. The biggest problem with our experiment is the great variety of libraries with detailed NDVI images useful for comparison with the past to understand how vegetation health changes year by year, but despite this issue we are very satisfied with the final result. By using international libraries and code standards, we are now able to share our project and discoveries, and make the data collection easier to anybody who would like to use our work. We would like to thank the ESA and the Raspberry Pi Foundation for giving us the possibility to work on such a unique project which led us to develop new skills and knowledge. This project has been a great opportunity for our career as students and worked for us as a compass in the process of choosing our future studies.