1000FARMS Webinars - Video of Webinar 08 - Kauê de Sousa

Click on the link below to view the webinar by Kauê de Sousa from the Alliance of Bioversity and CIAT on 15 May 2025, presenting Review of Tricot Principles and Data Analysis Workflows

Presenter Bio: Kauê de Sousa is an applied ecologist at the Alliance of Bioversity and CIAT specializing in agrobiodiversity, climate adaptation, and participatory research. He holds a degree in forestry from the Federal University of Amazonas, Brazil, an MSc in tropical agroforestry from CATIE, Costa Rica, and a PhD in applied ecology and biotechnology from the University of Inland Norway. Kauê’s work bridges ecological science, digital innovation, and local knowledge to support sustainable farming in biodiverse regions. Over the past decade, he has helped scale the tricot approach across Latin America, Africa, and Asia. He leads the capacity building work package in the 1000FARMS project and develops analytical workflows for the ClimMob platform.

For more info, contact Kauê at: k.desousa@cgiar.org

Summary : The presentation offers a comprehensive overview of the tricot approach (Triadic Comparison of Technologies), developed to address critical challenges in on-farm variety testing, especially in the Global South. These challenges include slow variety turnover, climate change uncertainty, limited farmer participation in traditional trials, and weak data capture and standardization. Tricot emphasizes citizen science, on-farm testing, digital tools, and participatory research, enabling farmers to test and rank three randomly assigned, anonymized crop varieties under local conditions. This data, managed through the ClimMob platform, is analyzed using ranking models like Plackett-Luce to derive insights. The approach is supported by a multi-institutional partnership and has evolved to integrate climate, socio-economic, and breeding data for deeper, scalable insights. The speaker highlighted how this inclusive and data-driven model empowers farmers, enhances decision-making, and contributes to better agricultural outcomes across diverse regions.

Link to the presentation slides: Review of tricot principles and data analysis workflows

Question Answer
Can you add an open ended question to the analysis (such as a comment box) using large language models (LLMs)? A key part of the process is having a way to clean and standardize the data. In our work on the Donovan paper, we created a synonym library to standardize terminology by matching different terms that refer to the same concept. This kind of normalization is essential and can be done manually, but large language models (LLMs) can improve this process significantly. We also explored voice recognition, where spoken input is transcribed and then analyzed using the same standardized library. Once you have a consistent set of terms, you can perform more effective analyses. Building this synonym library is relatively quick, and it will soon be integrated into ClimMob for broader use in data analysis.
How do you account for the variation among farmers scores? We use a standardized survey with a consistent set of questions for farmers to gather key information about their field trials—such as previous land use, management practices, and land preparation methods. These questions are based on recommendations from the literature on on-farm trials, which emphasize capturing variability in farmers’ fields. We’ve conducted an extensive literature review to support this approach.
Can the trials be expanded to include four varieties instead of three, so include the local check? While it is technically possible to include a fourth variety in a trial, we do not recommend it. Asking farmers to rank four options is simply more difficult than identifying the best and worst. Human judgment tends to be more accurate when comparing extremes rather than making full rankings. So to reduce complexity and improve the quality of responses, it’s better to stick with three varieties and use best–worst scaling instead of full rankings.
Has any progress been made on giving simplified feedback to farmers after a trial has been completed? A team recently conducted feedback evaluation work in Nigeria with IITA, assessing both the feedback process and related research. I’m looking forward to their report. Similar efforts are also happening in Rwanda. Feedback has been somewhat neglected—especially during COVID—but we’re now working to revive it. It’s easy to overlook, especially after a project ends, since returning results to farmers requires planning and budget. However, it’s crucial to have a good method for effectively disseminating feedback, even if it involves hundreds of farmers.
I’ve used other data collection platforms and noticed some limitations with ClimMob that could be improved. For example, there’s a restriction on how long questions can be, and editing questions after moving them is not easy. Are there any ongoing efforts to improve these aspects? We’re currently developing more detailed standard operating procedures (SOPs) for Tricot, focusing on key crops like sweet potato and maize through major projects like 1000FARMS and BOLD. These SOPs aim to standardize the questions and traits used in trials even more than before. The process is community-driven, with crop experts and the 1000FARMS team collaborating on protocols tailored to specific crops and regions. As Tricot becomes integrated into CGIAR’s broader on-farm testing framework, we welcome feedback and support to improve the system further.
Can you feed large climate data into ClimMob and using an LLM see what in the data is interacting with tricot or do you have select a covariate like precipitation at emergence? While I haven’t tried that specific approach, in our first paper we focused on making agronomic sense of the data by calculating growing degree days and defining crop stage windows—like for beans—to show how Tricot data can align with climate and agronomic reasoning. That was the paper’s main goal. Going forward, Rachel, an experienced ecophysiologist, will be working on integrating ecophysiological analysis into Tricot, particularly for crops like banana and enset. We expect this to bring new insights starting this year and into the next.