Remote sensing and Daisy

[Sarah Vecchietti]

Hi everyone, I am working on a thesis on winter wheat and I have experience using machine learning models with remote sensing data (multispectral satellite data and RGB drone imagery). I would like to explore whether it is possible to integrate remote sensing-derived variables such as NDVI, NDRE, or canopy cover into Daisy, either for calibration or to improve yield prediction. Has anyone used remote sensing data together with Daisy (e.g. for calibration, validation, or data assimilation)?

Any suggestions or references would be greatly appreciated.

Thanks in advance!

[Per Abrahamsen]

Hi Sarah,

LAI can be estimated from remote sensing Data, and can be fed into Daisy.  This should improve yield prediction. Here is one publication:

https://doi.org/10.1016/j.jhydrol.2003.10.018

Unfortunately. I can't find it in the documentation.  Maybe Simon can help, he is the last (I know of) who has done it.

Kindly,

Per A.

--

---
You received this message because you are subscribed to the Google Groups "Daisy soil plant atmosphere system model" group.
To unsubscribe from this group and stop receiving emails from it, send an email to daisy-model...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/daisy-model/02a70b89-c715-4b99-9c69-72e00d9b9a6en%40googlegroups.com.

[Kiril Manevski]

Hi Sara, Per and all,

Neural networks algorithm invented at MIT in the USA in 1944 (you can imagine at what response). I saw recently online the (now old) guy saying on the deep learning euphoria "People fall in love with an idea, get excited about it, hammer it to the end, get tired of it, until next newbies”.

Similarly, forward-feeding a process-based model is dated, please check the literature, especially on yield prediction. Unfortunately, the added value of remote sensing in that regard is often very limited, and bottom line whether it improves yield prediction depends on the study objectives, i.e. whether observed/simulated yields vary in relation to LAI. 

What we recently did is to investigate the added value of forward-feed for simulating belowground processes, and this is very interesting (let me know if you need the codes, I can ask Magdalena- she did the coupling, but it's not difficult): https://doi.org/10.3390/rs17243965 

Because our previous study pointed in this direction: https://doi.org/10.1016/j.agsy.2024.104149 

The paper of Boegh et al (2004) Per shared kinda showed abit of this 20 years ago, we now try to catch up... 

You can find more details on advanced use of remote sensing with process-based models in our last book chapter (frontiers section):  

https://www.researchgate.net/publication/400977347_Remote_sensing_of_soil_moisture_in_agroecosystems 

And the latest detailed overview of the Daisy model (what it can do):

https://www.researchgate.net/publication/400524040_The_Daisy_crop_model

 

Imagine if you inverse-feed model like Daisy, that is, estimate evapotranspiration from thermal imagery using TSEB (https://www.doi.org/10.3390/rs13152998) and calibrate the model on this; it actually opens a door for near-real-time irrigation, very interesting stuff.

Many greetings,

K.

To view this discussion visit https://groups.google.com/d/msgid/daisy-model/CAGBK-rfniHndFSKJGfxhk8F%2B_eTjzE3W2i%2BqnB3brSEVNbgQxg%40mail.gmail.com.

[Per Abrahamsen]

Hi Sarah.

Using Daisy for this purpose would definitely make sense, but  you would need to take the Daisy course 

https://daisy.ku.dk/courses/

which may or may not fit into the time for your master project.

Have you considered CoupModel or USSF, both of which are local to SLU?

Kindly,

Per A.

On Mon, Mar 23, 2026 at 9:20 AM Sarah Vecchietti <sarol...@gmail.com> wrote:

Dear Maja, Per and Kiril,

Thank you again for your thoughtful replies.

I am currently a master student in Agronomy at SLU in Uppsala, working on my thesis at RISE. The project is based on a long-term testbed (7 years) with detailed data on soil properties, in situ sensors (soil moisture, temperature, salinity), meteorological data, management practices, and crop yields. I am mainly focusing on winter wheat within the crop rotation.

My initial idea was to explore yield prediction using a machine learning approach, potentially integrating remote sensing data (satellite and drone imagery). However, after learning more about Daisy, I became interested  how this approach could offer additional insight, especially if combined with observational data.

I understand that approaches such as using “forced LAI” are possible but not widely documented, and I am currently reflecting on how far it would be meaningful (and realistic) to go in that direction within the scope of a thesis.

From your perspective, I would be very interested in your thoughts on this:

·       whether integrating satellite-derived LAI into Daisy is something you would consider worthwhile in a case like this,

·       and more generally, how you would approach this wether it would be more meaningful to (1) compare a standard Daisy simulation with a machine learning approach, or (2) explore a hybrid approach by integrating satellite-derived data into Daisy.

I am not necessarily looking for a definitive answer, but I would really value your perspective given your experience with the model.

Thank you again for your time. This exchange is very helpful.

Kind regards,
Sarah 

[Sarah Vecchietti]

Dear Maja, Per and Kiril,

Thank you again for your thoughtful replies.

I am currently a master student in Agronomy at SLU in Uppsala, working on my thesis at RISE. The project is based on a long-term testbed (7 years) with detailed data on soil properties, in situ sensors (soil moisture, temperature, salinity), meteorological data, management practices, and crop yields. I am mainly focusing on winter wheat within the crop rotation.

My initial idea was to explore yield prediction using a machine learning approach, potentially integrating remote sensing data (satellite and drone imagery). However, after learning more about Daisy, I became interested  how this approach could offer additional insight, especially if combined with observational data.

I understand that approaches such as using “forced LAI” are possible but not widely documented, and I am currently reflecting on how far it would be meaningful (and realistic) to go in that direction within the scope of a thesis.

From your perspective, I would be very interested in your thoughts on this:

·       whether integrating satellite-derived LAI into Daisy is something you would consider worthwhile in a case like this,

·       and more generally, how you would approach this wether it would be more meaningful to (1) compare a standard Daisy simulation with a machine learning approach, or (2) explore a hybrid approach by integrating satellite-derived data into Daisy.

I am not necessarily looking for a definitive answer, but I would really value your perspective given your experience with the model.

Thank you again for your time. This exchange is very helpful.

Kind regards,
Sarah 

On 20 Mar 2026, at 19:07, Kiril Manevski <kiril.m...@yahoo.com> wrote: