Assessing the environmental impact of crop protection products

Pressure on agriculture to reduce its ecological footprint is growing. Crop protection products play a key role in protecting crops from pests and disease, and are at the same time criticised – for possible effects on non-target organisms and the environment.

Between that criticism and any improvement sits a measurement. As long as environmental impact cannot be quantified comparably, one application cannot be weighed against another and no reduction can be demonstrated. That quantification was the task.

At the core of the project was a flexible infrastructure that assesses environmental impact using recognised scientific models. The decisive part: those models rest on broad scientific consensus – the only way an assessment holds up when third parties examine it.

The infrastructure had to be scalable, flexible and easy to integrate into existing systems, so it could process varied datasets and use cases efficiently. Alongside that stood a second goal beyond technology: encouraging more conscious use of these products and raising awareness of their ecological effects.

Two sources fed the assessment. Application data for crop protection products records use under differing conditions – the basis of any modelling of environmental effects, because the same substance behaves differently depending on how it is applied.

Datasets on Good Agricultural Practices reflect established agricultural practice and help encourage responsible use.

A calculator non-experts understand. The environmental impact calculator had to cover various application scenarios and produce results interpretable without specialist background. Transparency and ease of interpretation took precedence over depth of detail – a figure nobody can place changes no behaviour.

Data quality and integration. Application data and GAP data had to be prepared consistently and to a high standard before any calculation was possible.

Scalability without loss of performance. The infrastructure had to handle large and complex datasets without performance degradation – the precondition for the calculator being usable in an interactive application.

Data processing in Python. Complex datasets were processed and made analysable with Python and libraries such as Pandas and NumPy.

A modular ETL architecture. A scalable, modular ETL architecture brings together data from different sources – and makes it possible to add further sources without rebuilding the existing processing.

Scientific models rather than in-house heuristics. We used recognised scientific methods and worked with domain experts. For an assessment that is publicly cited and examined by third parties, the provenance of the model matters as much as its accuracy.

The infrastructure detects data errors, corrects them and produces quality reports, which makes the reliability of the assessments demonstrable rather than merely asserted.

On that foundation sits a flexible assessment tool that evaluates the environmental impact of crop protection products using scientific models, among them work from the Technical University of Denmark. It can be used standalone or integrated into other applications such as ResiYou.

In collaboration with DTU, an alpha version of a publicly accessible calculator was created as well. Users can compare environmental impacts and make informed decisions – so the assessment moves out of internal use and becomes checkable.

The pattern – bringing scientifically grounded models into a scalable infrastructure, securing data quality automatically, and making the result interpretable for non-experts – applies wherever a regulatorily or publicly relevant figure is calculated. The requirement there is always the same: traceable model provenance, verifiable data quality, comprehensible output.

The project is ongoing. The obvious next steps: moving the public calculator from alpha into stable operation, extending the ResiYou integration so that residue compliance and environmental impact become visible within a single decision, and connecting further data sources through the existing ETL architecture.