🌍 IMPETUS Climate Embedding Model

This is an embedding model trained using public data from the Horizon 2020 project IMPETUS Climate (Grant Agreement No. 101037084). The model has been fine-tuned to capture domain-specific semantic relationships related to climate change, urban resilience, and adaptation strategies.

📁 Data Sources

The training data for this model was exclusively derived from public deliverables of the IMPETUS Climate H2020 project. These deliverables are available through the official European Commission portal and the project’s website.

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("Applied-Artificial-Intelligence-Eurecat/IMPETUS-Climate-bge-small")
# Run inference
sentences = [
    'What specific interactions are part of the strategic approach?',
    'The IMPETUS project employs a structured Communication, Collaboration, and Dissemination (CCD) framework within its WP7 activities, divided into three phases: \n1. **Phase I (M1-M18)**: Focuses on content creation and collaboration planning. \n2. **Phase II (M19-M29)**: Aims to maximize results and monitor impact. \n3. **Phase III (M30-M48)**: Engages stakeholders and promotes results. \n\nThe project targets three audience categories: Policy Makers, Future Adopters, and Sectoral Specialists, emphasizing effective stakeholder engagement through tailored communication and outreach events. Guidelines for learning activities include technical guidance for video production, course structure recommendations, and support for speakers. \n\nCollaboration with other projects enhances knowledge sharing, and the development of a Massive Open Online Course (MOOC) involved extensive planning, expert selection, and content creation, culminating in a launch campaign coordinated by multiple partners.',
    '1.2.1 Advanced tools for regional water management\n\nObjective and challenge\n\nThe state of Berlin is required to map and justify the water demand from the catchment areas of the\nSpree and Havel rivers from considerations of the Berlin water cycle through resilience studies. The\ninvestigations within IMPETUS case study 1 (Berlin/Brandenburg) are intended to serve this purpose\nby integrating sectoral data and analytical tools. The main focus is on the following questions:\n\n9\n\nD4.1\n10/10//2022\n• Which surface water inflows to Berlin are absolutely necessary in order to cover the current and\nfuture raw water requirements while complying with the requirements of the Drinking Water\nRegulation?\n• What are the maximum permissible proportions of treated wastewater in the respective\ncatchment areas in order to ensure compliance with the drinking water limit values for drinking\nwater abstraction from groundwater replenished with riverbank filtrate?\n• What are the minimum surface water inflows to Berlin that are needed to ensure that the\nminimum ecological and chemical requirements for water bodies can be met in accordance with\nlegal requirements and other usage demands?\n• What is the relevance of stormwater runoff (including combined sewer overflows) and the\npotential of stormwater management measures in regard to the required surface water\nqualities?\n• What are the current and future shares of groundwater recharge, riverbank filtrate and (old)\ngroundwater in Berlins drinking water supply and what qualitative changes are to be expected?\nUsage scenarios will include drinking water supply from groundwater, partially augmented by river water\nvia managed aquifer recharge and bank filtration schemes; usage of rivers as receiving water bodies\nfor treated wastewater; increasing water demand for agriculture; water demand for cooling; large\nindustrial users; and other scenarios identified in co-designing and co-development workshops with local\nstakeholders.\n\nTechnical description\n\nFulfilling task T4.7.1 requires a modelling cascade capable of representing surface water hydraulic\nconditions, discharge of treated wastewater and other impacts to surface water, mixing as well as\nsurface water-groundwater interaction with all its in- and outflows and representative boundary\nconditions.\nClimate change will be accounted for by simulating stepwise impacts from\nreduced surface water inflow to the Berlin water cycle\nincrease in wastewater volumes and treated wastewater share in Berlins surface water bodies\n•\n•\n• decrease in natural groundwater recharge\n\n10\n\nD4.1\n10/10//2022\nSurface water balance and tipping points in drinking water quality abstracted from groundwater as a\nmixture of natural groundwater, riverbank filtrate and potentially artificial groundwater recharge are the\ntwo end members considered in system analysis. The results will feed into T4.18 decision theatre.\nThe single elements of the model cascade and required data are summarized in Table 2 below.\nElement Models & tools Data requirements\nWater level (selected locations)\ndischarge (selected River\nlocations)\nInput:\ntreatment\n(water works,\nplants,\nWater users\nsewage\nthermal power plants)\nPrecipitation\nSurface water\n(Spree, Havel)\nbodies\nHYDRAX model:\nHydrological balance\nmodel for surface water\nincluding inflow/outflow\nand resulting water levels Lake Evaporation\nOutput:\nRiver level; Water\ndischarge; Flow velocity; Cross-\nsectional area of the stream;\nVolume\nInput:\nprecipitation, evapotranspiration,\nnatural groundwater recharge\nClimate change\nscenarios\nOutput:\nReferences from literature\nand analysis of historic\ndata to be used, e.g.\npercentual reduction or\nhistoric minima/maxima\nsurface water volume;\ninfiltration/exfiltration volume in\nsurface water-groundwater\ninteraction; mixing/ shares of\nwater by source\nflow rates of surface waters and\nsewage effluents as averages,\nmanual data input\nInput:\nWastewater',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

🧠 Model License

While the training data was publicly available, the resulting embedding model is distributed under a different license to reflect the additional work and fine-tuning performed:

📄 License: Creative Commons Attribution-NonCommercial 4.0 (CC BY-NC 4.0)

This license allows:

• ✅ Attribution: You must give appropriate credit when using the model.
• 🔄 Reuse and adaptation: You can share and adapt the model.
• 🚫 Non-commercial use only: You may not use the model for commercial purposes without prior written permission.

If you are unsure whether your intended use qualifies as commercial, or if you wish to obtain a commercial license, please contact us.

📌 Required Citation

If you use this model in a publication, software, or presentation, please cite the following:

Ian Palacín, Oriol Alàs, IMPETUS Climate Embedding Model, Eurecat – Technology Centre of Catalonia, 2025.
Trained using public deliverables from the IMPETUS Climate project, funded by the European Union’s Horizon 2020 programme (Grant Agreement No. 101037084)