Soil-crop models are worthwhile to better understand processes

Soil-crop models are worthwhile to better understand processes. These models are used to evaluate agricultural system management scenarios, forecasting climate change impacts, proposing new crop types and more. Crop models are widely used, but using them effectively requires an understanding of the major methods and main principles for practical application. Young professionals and PhD candidates came together from 17th to 22th September 2023 at the University of Kassel in Witzenhausen for the summer school ‘Working with Dynamic Soil-Crop Models’ to dive deeper into the aspects of modelling.

Young professionals and PhD candidates participating in the summer school ‘Working with Dynamic Soil-Crop Models’.

Is a model the reality?

A model is per definition a simplification of the real world. However, I am convinced that the combination of models and field experiments could give a good indication of reality. A model that describes the system behavior in the best way could be achieved by calibrating the model to field measurements. The choice of parameters to estimate combines expertise and field data, but also knowledge on avoiding numerical problems. Examples of calibration methods we discussed were Ordinary Least Squares (OLS), maximum likelihood principle, and Bayesian calibration.

Data of field experiments are required to understand the real-world situation better, leading to better models in terms of schematization, processes modelled, and model parameter values. We discussed to have the best model to have with the dataset we have, not the best parameters. Model evaluation could summarize the goodness of the fit in a number. During the course, we considered the model as an engineering tool: we evaluated how well the model fulfilled its objectives instead of considering the model as a scientific hypotheses (evaluate whether or not the model is an accurate description of the way the world works; answer could be true or false).

Sensitivity and uncertainty in models

Another aspect of modelling is the sensitivity and uncertainty in models. It is important to understand how do uncertainties in model input translate into uncertainties in model output. Examples to explore this with generating samples is to i) explore all possible combinations, ii) use (random) sampling like Monte Carlo, or iii) use Latin hypercube sampling. Besides uncertainty plays the sensitivity of a model also a role: what are the contributions of each source of uncertainty to model output uncertainty. Sensitivity analysis can be done in many different ways. An useful scientific paper about sensitivity analysis is written by Hamby et al. (1994): a review of techniques for parameter sensitivity analysis of environmental models. During the course we applied the Morris and FAST methods.

The city Witzenhausen with the Werra river.

Applications of models worldwide

Models are worldwide applied in the agricultural field. Several lectures, given by professors around the world, were organized throughout the week called ‘Food-for-thought-lectures’.

• Prof. Dr. Martine van der Ploeg (WUR) started with a presentation of the community International Soil Modeling Consortium (ISMC).
• Dr. Peter Thorburn (CSIRO, Australia) gave insight into how insurance could help farmers to mitigate nitrogen pollution from intensive cropping.
• In America are farms almost digital twins as shown by Prof. Dr. Bruno Basso (MSU, United States); the agricultural fields are designed and scaled with climate-smart practices like drones. The drones create GPS cards (like the ‘TAAK’ cards in the Netherlands) based on pictures to indicate the amount of water every plant receives.
• Prof. Dr. Nicholas Jarvis (SLU, Sweden) explained some theories behind water flow models like the Richard’s equation and the tipping bucket model. It is all written in the paper ‘Improved descriptions of soil hydrology in crop models: the elephant in the room?’
• Many scientific papers are written, but not all of them are open access. Nowadays it becomes more and more that papers (and data) is open access. Dr. Stanislav Schymanski (FNR, Luxemburg) gave a presentation about ‘Good science is open science’.
• Another topic that becomes more common nowadays is machine learning. Prof. Dr. Ioannis Athanasiadis (WUR, The Netherlands) gave a presentation about mixing machine learning and crop models – experiences for yield forecasting.
• The final presentation was about the impacts of climate change in global agriculture given by Dr. Jonas Jägermeyer (NASA, United States).

The side experiences of a summer school

The side advantage experiences of a summer school are the amazing participants and professors to have a good talk with during breakfast, lunches and dinners. I met people from of course Germany, but also Poland, Denmark, China, Mali, Burkina Faso, Italy and more. For example, I learned to knock on the table to give an applause, but also the differences in PhD system between Europe and Africa. Best of all, I met someone with whom I did research in the desert in Israel 5 years ago.

On Wednesday evening we had a nice pizza evening where each presented his or her own (PhD) research. On Thursday we had a nice diner in Göttingen (place nearby Witzenhausen) where we also learned about the history of this area in the 20th century. It was an intensive week; the inspiring talks, the grown network, the knowledge about international research and some pieces of the bigger model puzzle. For sure, there is much more to learn, but don’t forget: ‘Walk before you run, but start walking’ and that is what I did: start walking.