This use cases addresses challenges how to deal with incomplete data on the case of data from long-term experiments.

Background

Data from long-term field agricultural field experiments (LTE) are of high interest in agrosystem sciences, especially for use in model validation and simulations. Although numerous LTEs with different factors are conducted both nationally and internationally (e.g., lte.bonares.de), their data have rarely been published and thus are not available for application in crop models. LTE studies often aim to identify the respective major constraints of crop yield in a particular setting, e.g., the effects of pests and diseases, water or nutrient deficiency, or detrimental climate conditions and try to understand long-term effects on soil carbon to support measures for enhancing soil carbon sequestration. However, usually only one or two factors are considered, providing only limited information about the respective setting, which can be a major obstacle in terms of using LTE data for more generic meta studies or the application of biophysical or machine learning yield models. Furthermore, the use of LTE data for crop model calibration and validation requires knowledge about the data’s quality and plausibility, which is often missing in published datasets.

Objectives

This UC aims to improve availability and to complement LTE data, which are of special interest for crop modellers or meta-studies. One activity in this use case will be the support of LTE experimentalists in publishing their datasets in existing FAIRagro infrastructures. Furthermore, to make LTE data applicable for generic studies, additional contextual information will be collected complementing LTE data. An iterative approach will be developed, automated as far as possible and published as a guideline on how to complement existing LTE datasets for further use. Additionally, a service will be developed to analyse and assess data quality and plausibility, which is needed to reuse LTE data in model applications.

Actions

1. Support and standardize new LTE data publication
2. Data curation supplementing already published incomplete LTE data
3. Implement a service to assess data plausibility

Progress & next steps

UC Update: Advancing AI-Supported Metadata Extraction for Long-Term Experiments

The UC team led by Lachmuth et al. has successfully demonstrated how large language models (LLMs) can support metadata and context extraction for agricultural long-term experiments. Their approach shows that AI can efficiently identify and structure metadata across heterogeneous sources, improving data findability and interoperability.