My name is Achraf. I am a land surveying and geomatics student and I’m currently preparing for my end-of-studies project. The project concerns the study of rangeland degradation by estimating biomass of a study area using time-series imagery (vegetation indices) and other data (sample terrain measurements of biomass, rainfall, temperature and DTM). I have learned that regression through machine learning or deep learning techniques can be great ways to estimate biomass. I have recently heard about KNIME and found the software to be really interesting since it’s an alternative to traditional coding. Is there a way to combine these data in KNIME and obtain biomass of the entire study area as the final output? Do I need to convert tiff images into another format for processing? I would really appreciate it if you could help me resolve this problem.
This sounds like an interesting project for sure. In this case it sounds like you need to combine elements of image processing, deep learning (probably with convolutional neural networks), and maybe even a bit of time series and/or spatial correlation for a regression estimation. Certainly not a trivial effort.
I’m fairly certain we don’t have ready-made example workflows for something this specific, but there are several workflows available on the KNIME Hub centered around deep learning with CNNs. Most of these, unfortunately, deal with classification problems, so you would need to adapt the network structure for your purposes (also including your other numeric predictors as well).
Speaking of those other predictors, it occurs to me there are some additional issues to consider: for example, how do you spatially correlate meteorological data with the images, if you have multiple measurement sites and a large area of interest?
Anyway, if you have additional information you can provide, maybe someone can point you in a more specific direction.
Thank you for your reply and information about the subject.
When it comes to data, I was hoping to use images of soil temperature and precipitation as well as DTM images. This way I don’t have to worry about correlation since each pixel will have 5 values : vegetation index, precipitation, elevation, temperature and biomass measurement (for each site). I haven’t received biomass ground measurements yet, but I believe I will obtain surfacic ground measurements, which means I am gonna have to calculate the mean value of the vegetation index, precipitation, temperature and elevation for each surface.
In other words, each surface (or site) of biomass measurement will be represented with 5 values that I can put in a csv file for processing. My goal is to obtain a regression equation that is precise and that I can use to predict biomass in other parts of the study area that were not subject to ground measurements.
I hope I was clear in explaining the problematic and the data that I want to use. I would really appreciate it if I could have your opinion, or other knimers’ opinions, on the matter. Thank you for your help and consideration.