xLandscape
Slide deck: SETAC xLandscape Training — (2) xLandscape Intro
Why a Modular Landscape Modelling Framework?
The demand for landscape-scale modelling in pesticide risk assessment is growing, driven by:
- More realistic, higher-tier risk assessment — EFSA increasingly calls for spatially explicit reference scenarios in tiered schemes.
- Improved process integration — linking hydrology, fate, exposure, individual effects, and population dynamics in a single modelling chain.
- Holistic view to pesticide risk — covering multiple stressors, recovery, biodiversity, and indirect effects of weed control.
- Integration of monitoring and modelling — to continuously improve knowledge and validate predictions.
- Digital Agriculture and Digital Twins — enabling what-if analyses for integrated pest control and environmental risk management (e.g., Destination Earth).
Why do we develop a modular landscape modelling framework?
Simply because there is a demand — and there is nothing like this available.
Conceptual Drivers for xLandscape
The xLandscape framework was designed around three key concepts:
| Concept | Description |
|---|---|
| Tiered approach | Flexibility and adaptivity — "get only as complex as necessary" |
| Specific Protection Goals (SPGs) | Models operate in explicit spatial, temporal, and structural dimensions aligned to SPGs |
| Systems Approaches & Digital Twins | Foundation for future real-world digital twins and DestinE-compatible scenarios |
The Vision
Imagine… you are facing one of these challenges:
- You are developing a new species effect model (e.g. amphibians, wild pollinators, beetles) for pesticide risk assessment and population protection.
- You are interested in multiple pesticide exposure, toxicity, and effects across large regions.
- You want to compare risks of pest control strategies under different baseline assumptions.
- You study real-world effects of different risk management decisions and landscape design options.
- You want to build a regional scenario (or digital twin) to study a range of protection goals and ecosystem services.
- You want all this work to be consistent with related projects, fully versioned, and reproducible in a regulatory context.
If you start from scratch, it's quite an effort. But if you can focus on your focal expertise and build on open, shared work from your colleagues — that changes everything.
You have a tool at hand enabling you to build and run processes and models at landscape level, using real-world data. You can adapt this tool to your problem. It is open source. You can run landscape models on your laptop or large cloud systems. The conceptual foundation is embedded in regulatory-scientific frameworks, with tiered scenario development that lets you start quickly at a screening level and get more detailed as you proceed. Scenario services support you on request. Versioning ensures full long-term availability and reproducibility within a regulatory context.
xLandscape aims to make this possible: an open, AI-supportable platform where researchers can build on what others have already done rather than starting from scratch.
Wherever your expertise lies — toxicology, hydrology, ecology, agronomy — the modular architecture lets you focus on your core competence while benefiting from a consistent, validated framework.
Modularity as a Principle
Modularity is central to xLandscape because it simultaneously enables:
- Focus — developers and users work within their domain expertise.
- Stepwise validation — each component can be tested independently.
- Collaboration — teams can contribute without needing to understand the entire system.
- Flexibility — model complexity is adapted to the problem, not forced by the platform.
- Harmonisation — consistent interfaces and semantics across all models.
- Transparency — similar solutions to similar problems; modules can become authorised APIs.
The framework integrates diverse domains: hydrology, land use, habitats, pesticide use, exposure, environmental fate, individual effects, and population effects.
Model Composition: Components
The xLandscape microkernel provides the foundational services:
Interfaces · Control · Monte Carlo · Dimensions · Scales · Semantics · Data Store
Individual components wrap existing models (TOXSWA, PRZM5, GUTS, StreamCom, …) and are integrated at four levels via standardised interfaces. Models written in Python, R, C++, Fortran, Java, or Smalltalk can be incorporated.
xAquaticRisk as an example composition
| Layer | Components |
|---|---|
| Agriculture / Land management | Crop cultivation, PPP use, land use/cover |
| Exposure / Fate | xDrift, xRunoff, xDrainage, Toxswa, Steps1234, CmfContinuous |
| Effects | GUTS (CVASI), StreamCom, MASTEP |
| Analysis | LP50, population exposure, reporting |
| Environment | Weather, soil, elevation, hydrographic network |
Selected components
| Component | Domain | Notes |
|---|---|---|
| Crop cultivation | Agriculture | Land use / crop calendar |
| PPP use (xCropProtection) | Agriculture | Pesticide application scheduling |
| xDrift | Exposure | In-field spray drift deposition |
| xRunoff / xDrainage | Exposure | Surface runoff and drain flow |
| PRZM5 / PrzmEU / PrzmUS | Exposure/Fate | Pesticide fate in soil and runoff |
| Drainage-PEARL / Drainage-MACRO | Exposure/Fate | Drainage leaching models |
| TOXSWA | Exposure/Fate | Pesticide fate in surface water |
| Steps1234 | Exposure | Step 1–4 exposure modelling |
| RunOffFilter / PECsoil / PlantResidue | Exposure | Off-field and terrestrial PECs |
| GUTS (CVASI) | Effects | Individual survival under toxicant stress |
| StreamCom / MASTEP | Effects | Stream community and population effects |
| BeeHave / BeeForage | Effects | Honeybee colony and foraging models |
| Land use/cover | Environment | Spatial landscape structure |
| Weather, soil, elevation | Environment | Abiotic boundary conditions |
| Hydrographic / topographic network | Environment | Stream and catchment routing |
The xLandscape model ecosystem
| Model | Focus |
|---|---|
| xAquaticRisk (invertebrates) | Aquatic invertebrate population risk |
| xAquaticRisk (water plants) | Aquatic macrophytes and algae |
| xPollinator | Honeybee and wild bee population risk |
| xNTTP-EU / xNTP-US | Non-target terrestrial plants |
| xOffFieldSoil | Off-field soil exposure |
| xWeedIndirectEffects | Indirect effects of weed control |
| xResidues | Crop residue exposure |
All models are available at https://github.com/xlandscape.
Propagating Real-World Variability
xLandscape handles input variability (agriculture, environment, biology) and propagates it to output variability (PECs, effects, population sizes) using a hybrid explicit–Monte Carlo approach:
- Discretisation in space and time captures landscape heterogeneity.
- Monte Carlo simulation samples from probability density functions for uncertain parameters.
- Multidimensional HDF5 data arrays store and exchange all intermediate and final outputs in a semantically consistent model space, accessible from R, Python, Jupyter, Matlab, Tableau, KNIME, and more.
Technology: Multidimensional Data Arrays
All intermediate and final model outputs are stored in HDF5 — a format originally developed at the U.S. National Center for Supercomputing Applications and maintained by The HDF Group as a non-profit standard:
| Data class | Examples |
|---|---|
| Hydrology | water depth, flow rate |
| Exposure | PPP use, PEC~sw~, PEC~sed~ |
| Effects | individual mortality (SD/IT), population size |
| Environment | bee forage availability |
These arrays are accessible via any HDF5-compatible tool — enabling flexible, user-driven post-processing and analysis without vendor lock-in.
User Levels and Roles
xLandscape supports different levels of engagement — from model users applying pre-built scenarios to developers building new components and contributing to the framework. The training focuses on building familiarity across these roles.
Take-Home Messages
- Landscape modelling = improved realism and integration — it places risk assessment in the real world.
- Modularisation manages complexity and enables effective collaboration across disciplines.
- xLandscape = an open-source modular landscape modelling approach — components, models, and scenarios can be built for a wide range of problems.
- The goal is not to establish fixed, authoritative landscape models, but to enable flexible, adaptive research and higher-tier models of the highest possible harmonisation, continuity, and transparency.
Our Values
Openness & Collaboration — open-source principles foster innovation, participation, and real-world relevance.
FAIR Principles — Findable, Accessible, Interoperable, and Reusable. Modular frameworks and transparent sharing let researchers build on existing work.
Invitation to Participate — xLandscape is a mindset. Everyone is welcome to contribute ideas, share expertise, and help shape the future of landscape modelling.
What's Next
- Growing AI support for model composition and analysis
- Community building — connecting research and development teams
- xLandscape Version 4: microkernel and component-based clean architecture, improved component autonomy, Python packaging, digital-twin readiness
- Stepwise introduction of formal semantics up to established ontologies
- Continued development, testing, documentation, and publication of components