Waterbase serves as the EEA’s central database for managing and disseminating data regarding the status and quality of Europe's rivers, lakes, groundwater bodies, transitional, coastal, and marine waters. It also includes information on the quantity of Europe’s water resources and the emissions from point and diffuse sources of pollution into surface waters. Specifically, Waterbase - Biology focuses on biology data from rivers, lakes, transitional and coastal waters collected annually through the Water Information System for Europe (WISE) – State of Environment (SoE) reporting framework. The data are expected to be collected within monitoring programs defined under the Water Framework Directive (WFD) and used in the classification of the ecological status or potential of rivers, lakes, transitional and coastal water bodies. These datasets provide harmonised, quality-assured biological monitoring data reported by EEA member and cooperating countries, as Ecological Quality Ratios (EQRs) from all surface water categories (rivers, lakes, transitional and coastal waters).
This metadata refers to the geospatial dataset representing the status of the EEA Industrial Reporting database as of 15 December 2025 (version 15). The release and emissions data cover the period 2007-2024 as result of the data reported under the E-PRTR facilities, 2017-2024 for IED installations and WI/co-WIs, and 2016-2024 for LCPs. These data are reported to EEA under Industrial Emissions Directive (IED) 2010/75/EU Commission Implementing Decision 2018/1135 and the European Pollutant Release and Transfer Register (E-PRTR) Regulation (EC) No 166/2006 Commission Implementing Decision 2019/1741. The dataset brings together data formerly reported separately under E-PRTR Regulation Art.7 and under IED Art.72. Additional reporting requirements under the IED are also included.
The datasets includes 1) the noise exposure data, 2) the noise contours data, 3) razterized noise contours data and 4) potential quiet areas all under the terms of the Environmental Noise Directive (END). Data covers the EEA32 member countries and the United Kingdom (excluding Turkey for the third round of noise mapping in 2017).
Estuaries and coasts are characterized by ecological dynamics that bridge the boundary between habitats, such as fresh and marine water bodies or the open sea and the land. Because of this, these ecosystems harbor ecosystem functions that shaped human history. At the same time, they display distinct dynamics on large and small temporal and spatial scales, impeding their study. Within the framework of the OTC-Genomics project, we compiled a data set describing the community composition as well as abiotic state of an estuary and the coastal region close to it with unprecedented spatio-temporal resolution. We sampled fifteen locations in a weekly to twice weekly rhythm for a year across the Warnow river estuary and the Baltic Sea coast. From those samples, we measured temperature, salinity, and the concentrations of Chlorophyll a, phosphate, nitrate, and nitrite (physico-chemical data); we sequenced the 16S and 18S rRNA gene to explore taxonomic community composition (sequencing data and bioinformatic processing workflow); we quantified cell abundances via flow cytometry (flow cytometry data); and we measured organic trace substances in the water (organic pollutants data). Processed data products are further available on figshare.
In Phase 1 lieferte "Operation Hydrometeors" für das Schwerpunktprogramm einen effizienten polarimetrischen Vorwärtsoperator als wichtige Voraussetzung der Verschmelzung von Radarpolarimetrie und numerischer Atmophärenmodellierung. Der nicht-polarimetrische “Efficient Modular VOlume RADar Operator” (EMVORADO) wurde unter Annahme von ellipsenförmigen Hydrometeoren um die Berechnung synthetischer Polarisationsparameter bei Wetterradarwellenlängen erweitert und ist sowohl in Form von Online-Implementierungen in den Wettervorhersagemodellen COSMO und ICON als auch als eigenständiges Programm verfügbar. Des Weiteren haben wir, basierend auf einem Clustering der polarimetrischen Variablen, einen ausgefeilten und robusten Algorithmus zur Hydrometeor-Klassifikation (HMC) entwickelt. Neben der Klassifizierung des dominierenden Hydrometeortyps bietet er zusätzlich die Möglichkeit, die anteiligen Massenverhältnisse der Hydrometeortypen an einem Ort zu schätzen, und zwar sowohl auf Basis polarimetrischer Radarbeobachtungen als auch auf Basis synthetischer Variablen, welche durch die Anwendung von EMVORADO auf ICON-Simulationen erhalten werden. Mit den entwickelten Werkzeugen haben wir polarimetrische Signaturen und Hydrometeore in stratiformen Regen unter Verwendung einer dualen Strategie ausgewertet, die synergistisch die hochentwickelte Hydrometeorklassifizierung und -Quantifizierung mit dem direkten Vergleich von multivariaten beobachteten und synthetischen polarimetrischen Radarparametern kombiniert. Die Analysen lieferten bereits einige Verbesserungen der Zwei-Momenten-Wolkenmikrophysik-Parametrisierung in ICON.In Phase 2 werden die entwickelten Werkzeuge weiter verfeinert, um ihre Anwendbarkeit zu erweitern. Wir verbessern das Schmelzschema in EMVORADO, koppeln möglicherweise auch eine Streudatenbank an und verfeinern unsere neue HMC weiter, um Hydrometeor-Cluster über die 0°C-Grenze hinweg zu ermöglichen. Mit diesen Entwicklungen und der Anwendung unserer dualen Strategie evaluieren und verbessern wir die Darstellung von drei weiteren wichtigen polarimetrischen Signaturen, die auf verschiedene Modellschwächen hinweisen: (1) Säulen mit erhöhter differentieller Reflektivität ZDR weisen den Weg zu einer besseren Darstellung des Gefrierprozesses von Regentropfen, (2) das ZDR in Bodennähe liefert Informationen zur besseren Parametrisierung der Regentropfengrößenverteilung, und (3) polarimetrische Variablen in der Zone bevorzugten Dendritenwachstums liefern die Informationen zur Verbesserung und Erweiterung der Darstellung von mikrophysikalischen Eisprozessen wie z.B. die sekundäre Eisproduktion im Modell. Schließlich werden wir zwei von der Polarimetrie abgeleitete Informationen für die indirekte Datenassimilation in ICON ausnutzen: wir assimilieren ZDR-Säulenobjekte und/oder die aus unserer neuen HMC abgeleiteten Massenverhältnisse der Hydrometeortypen in geeigneter Form, höchstwahrscheinlich als kategorische Ranginformationen.
# robbenblick A Computer Vision project for object detection and annotation management using YOLOv8, SAHI, and FiftyOne, with the primary aim of counting objects (Robben) in large aerial images. ## Overview This repository provides a complete MLOps pipeline for: * **Data Preparation:** Converting raw CVAT annotations (XML) and large images into a tiled, YOLO-compatible dataset. * **Automated Experiments:** Systematically training and tuning YOLOv8 models. * **Tiled Inference:** Running optimized inference (SAHI) on large, high-resolution images for object counting. * **Evaluation:** Assessing model performance for both detection (mAP) and counting (MAE, RMSE, R²). * **Visualization:** Analyzing datasets and model predictions interactively with FiftyOne. ## Pretrained Model Weights Pretrained model weights are available on Hugging Face: https://huggingface.co/ki-ideenwerkstatt-23/robbenblick/ ## Project Workflow The project is designed to follow a clear, sequential workflow: 1. **Prepare Data (`create_dataset.py`):** Organize your raw images and CVAT `annotations.xml` in `data/raw/` as shown below. ```text data/raw/ ├── dataset_01/ │ ├── annotations.xml │ └── images/ └── dataset_02/ ... ``` Run the script to generate a tiled, YOLO-formatted dataset in `data/processed/` and ground truth count CSVs. 2. **Tune Model (`run_experiments.py`):** Define a set of hyperparameters (e.g., models, freeze layers, augmentation) in `configs/base_iter_config.yaml`. Run the script to train a model for every combination and find the best performer. 3. **Validate Model (`yolo.py`):** Take the `run_id` of your best experiment and run validation on the hold-out `test` set to get **detection metrics (mAP)**. 4. **Infer & Count (`predict_tiled.py`):** Use the best `run_id` to run sliced inference on new, large images. This script generates final counts and visual outputs. 5. **Evaluate Counts (`evaluate_counts.py`):** Compare the `detection_counts.csv` from inference against the `ground_truth_counts.csv` to get **counting metrics (MAE, RMSE)**. 6. **Visualize (`run_fiftyone.py`):** Visually inspect your ground truth dataset or your model's predictions at any stage. ## Configuration This project uses two separate configuration files, managed by `robbenblick.utils.load_config`. * **`configs/base_config.yaml`** * **Purpose:** The single source of truth for **single runs**. * **Used By:** `create_dataset.py`, `predict_tiled.py`, `run_fiftyone.py`, and `yolo.py` (for validation/single-predict). * **Content:** Defines static parameters like data paths (`dataset_output_dir`), model (`model`), and inference settings (`confidence_thresh`). * **`configs/base_iter_config.yaml`** * **Purpose:** The configuration file for **experiments and tuning**. * **Used By:** `run_experiments.py`. * **Content:** Any parameter defined as a **YAML list** (e.g., `model: [yolov8n.pt, yolov8s.pt]`) will be iterated over. `run_experiments.py` will test every possible combination of all lists. ## Environment Setup 1. Clone the repository: ```sh git clone git@github.com:ki-iw/robbenblick.git cd robbenblick ``` 2. Create the Conda environment: ```sh conda env create --file environment.yml conda activate RobbenBlick ``` 3. (Optional) Install pre-commit hooks: ```sh pre-commit install ``` ## Core Scripts & Usage ### `create_dataset.py` * **Purpose:** Converts raw CVAT-annotated images and XML files into a YOLO-compatible dataset, including tiling and label conversion. * **How it works:** * Loads configuration from a config file. * Scans `data/raw/` for dataset subfolders. * Parses CVAT XML annotations and extracts polygons. * Tiles large images into smaller crops based on `imgsz` and `tile_overlap` from the config. * Converts polygon annotations to YOLO bounding box format for each tile. * Splits data into `train`, `val`, and `test` sets and writes them to `data/processed/dataset_yolo`. * Saves a `ground_truth_counts.csv` file in each raw dataset subfolder, providing a baseline for counting evaluation. * **Run:** ```sh # Do a 'dry run' to see statistics without writing files python -m robbenblick.create_dataset --dry-run --config configs/base_config.yaml # Create the dataset, holding out dataset #4 as the test set python -m robbenblick.create_dataset --config configs/base_config.yaml --test-dir-index 4 ``` * **Key Arguments:** * `--config`: Path to the `base_config.yaml` file. * `--dry-run`: Run in statistics-only mode. * `--test-dir-index`: 1-based index of the dataset subfolder to use as a hold-out test set. * `--val-ratio`: Ratio of the remaining data to use for validation. ### `run_experiments.py` * **Purpose:** **This is the main training script.** It automates hyperparameter tuning by iterating over parameters defined in `base_iter_config.yaml`. * **How it works:** * Finds all parameters in the config file that are lists (e.g., `freeze: [None, 10]`). * Generates a "variant" for every possible combination of these parameters. * For each variant, it calls `yolo.py --mode train` as a subprocess with a unique `run_id`. * After all runs are complete, it reads the `results.csv` from each run directory, sorts them by `mAP50`, and prints a final ranking table. * **Run:** ```sh # Start the experiment run defined in the iteration config python -m robbenblick.run_experiments --config configs/base_iter_config.yaml # Run experiments and only show the top 5 results python -m robbenblick.run_experiments --config configs/base_iter_config.yaml --top-n 5 ``` ### `predict_tiled.py` * **Purpose:** **This is the main inference script.** It runs a trained YOLOv8 model on new, full-sized images using Sliced Aided Hyper Inference (SAHI). * **How it works:** * Loads a trained `best.pt` model specified by the `--run_id` argument. * Loads inference parameters (like `confidence_thresh`, `tile_overlap`) from the `base_config.yaml`. * Uses `get_sliced_prediction` from SAHI to perform tiled inference on each image. * Saves outputs, including visualized images (if `--save-visuals`), YOLO `.txt` labels (if `--save-yolo`), and a `detection_counts.csv` file. * **Run:** ```sh # Run inference on a folder of new images and save the visual results python -m robbenblick.predict_tiled \ --config configs/base_config.yaml \ --run_id "best_run_from_experiments" \ --source "data/new_images_to_count/" \ --output-dir "data/inference_results/" \ --save-visuals ``` ### `evaluate_counts.py` * **Purpose:** Evaluates the *counting* performance of a model by comparing its predicted counts against the ground truth counts. * **How it works:** * Loads the `ground_truth_counts.csv` generated by `create_dataset.py`. * Loads the `detection_counts.csv` generated by `predict_tiled.py`. * Merges them by `image_name`. * Calculates and prints key regression metrics (MAE, RMSE, R²) to assess the accuracy of the object counting. * **Run:** ```sh # Evaluate the counts from a specific run python -m robbenblick.evaluate_counts \ --gt-csv "data/raw/dataset_02/ground_truth_counts.csv" \ --pred-csv "data/inference_results/detection_counts.csv" ``` ### `yolo.py` * **Purpose:** The core engine for training, validation, and standard prediction. This script is called by `run_experiments.py` for training. You can use it directly for validation. * **How it works:** * `--mode train`: Loads a base model (`yolov8s.pt`) and trains it on the dataset specified in the config. * `--mode validate`: Loads a *trained* model (`best.pt` from a run directory) and validates it against the `test` split defined in `dataset.yaml`. This provides **detection metrics (mAP)**. * `--mode predict`: Runs standard (non-tiled) YOLO prediction on a folder. * **Run:** ```sh # Validate the 'test' set performance of a completed run python -m robbenblick.yolo \ --config configs/base_config.yaml \ --mode validate \ --run_id "best_run_from_experiments" ``` ### `run_fiftyone.py` * **Purpose:** Visualizes datasets and predictions using FiftyOne. * **How it works:** * `--dataset groundtruth`: Loads the processed YOLO dataset (images and ground truth labels) from `data/processed/`. * `--dataset predictions`: Loads images, runs a specified model (`--run_id`) on them, and displays the model's predictions. * **Run:** ```sh # View the ground truth annotations for the 'val' split python -m robbenblick.run_fiftyone \ --config configs/base_config.yaml \ --dataset groundtruth \ --split val \ --recreate # View the predictions from 'my_best_run' on the 'test' split python -m robbenblick.run_fiftyone \ --config configs/base_config.yaml \ --dataset predictions \ --split test \ --run_id "my_best_run" \ --recreate ``` ### `streamlit_app.py` * **Purpose:** Quick test runs with the trained model of your choice for counting the seals in the image(s) and visualization. * **How it works:** * Loads the selected YOLO model from `runs/detect/`. * Upload images, run model, then displays the counts and model's predictions as image visualization. * **Run:** ```sh # View the ground truth annotations for the 'val' split export PYTHONPATH=$PWD && streamlit run robbenblick/streamlit_app.py ``` ## Recommended Full Workflow 1. **Add Raw Data:** * Place your first set of images and annotations in `data/raw/dataset_01/images/` and `data/raw/dataset_01/annotations.xml`. * Place your second set (e.g., from a different location) in `data/raw/dataset_02/images/` and `data/raw/dataset_02/annotations.xml`. 2. **Create Dataset:** * Run `python -m robbenblick.create_dataset --dry-run` to see your dataset statistics. Note the indices of your datasets. * Let's say `dataset_02` is a good hold-out set. Run: `python -m robbenblick.create_dataset --config configs/base_config.yaml --test-dir-index 2` * This creates `data/raw/dataset_02/ground_truth_counts.csv` for later. 3. **Find Best Model:** * Edit `configs/base_iter_config.yaml`. Define your experiments. ```yaml # Example: Test two models and two freeze strategies model: ['yolov8s.pt', 'yolov8m.pt'] freeze: [None, 10] yolo_hyperparams: scale: [0.3, 0.5] ``` * Run the experiments: `python -m robbenblick.run_experiments`. * Note the `run_id` of the top-ranked model, e.g., `iter_run_model_yolov8m.pt_freeze_10_scale_0.3`. 4. **Validate on Test Set (Detection mAP):** * Check your best model's performance on the unseen test data: `python -m robbenblick.yolo --mode validate --run_id "iter_run_model_yolov8m.pt_freeze_10_scale_0.3" --config configs/base_config.yaml` * This tells you how well it *detects* objects (mAP). 5. **Apply Model for Counting:** * Get a new folder of large, un-annotated images (e.g., `data/to_be_counted/`). * Run `predict_tiled.py`: `python -m robbenblick.predict_tiled --run_id "iter_run_model_yolov8m.pt_freeze_10_scale_0.3" --source "data/to_be_counted/" --output-dir "data/final_counts/" --save-visuals` * This creates `data/final_counts/detection_counts.csv`. 6. **Evaluate Counting Performance (MAE, RMSE):** * Now, compare the predicted counts (Step 5) with the ground truth counts (Step 2). Let's assume your "to_be_counted" folder *was* your `dataset_02`. `python -m robbenblick.evaluate_counts --gt-csv "data/raw/dataset_02/ground_truth_counts.csv" --pred-csv "data/final_counts/detection_counts.csv"` * This gives you the final MAE, RMSE, and R² metrics for your **counting task**. ## Additional Notes This repository contains only the source code of the project. The training data and the fine-tuned model weights are not included or published. The repository is currently not being actively maintained. Future updates are not planned at this time. For transparency, please note that the underlying model used throughout this project is based on **YOLOv8 by Ultralytics**. ## License Copyright (c) 2025 **Birds on Mars**. This project is licensed under the **GNU Affero General Public License v3.0 (AGPL-3.0)**. This aligns with the license of the underlying **YOLOv8** model architecture used in this project. Please note: **Training data and fine-tuned model weights are not part of the licensed materials** and are not included in this repository. For full details, see the LICENSE file. ## Troubleshooting ### FiftyOne: images (partially) not visible Try using `--recreate` flag to force FiftyOne to reload the dataset: ```sh python robbenblick/run_fiftyone.py --dataset groundtruth --split val --recreate ``` ### FiftyOne: failed to bind port If you get: ``` fiftyone.core.service.ServiceListenTimeout: fiftyone.core.service.DatabaseService failed to bind to port ``` Try killing any lingering `fiftyone` or `mongod` processes: ```sh pkill -f fiftyone pkill -f mongod Then rerun your script. ``` # Collaborators The code for this project has been developed through a collaborative effort between [WWF Büro Ostsee](https://www.wwf.de/themen-projekte/projektregionen/ostsee) and [KI-Ideenwerkstatt](https://www.ki-ideenwerkstatt.de), technical implementation by [Birds on Mars](https://birdsonmars.com). <p></p> <a href="https://ki-ideenwerkstatt.de" target="_blank" rel="noopener noreferrer"> <img src="assets/kiiw.jpg" alt="KI Ideenwerkstatt" height="100"> </a> <p></p> Technical realization <br> <a href="https://birdsonmars.com" target="_blank" rel="noopener noreferrer"> <img src="assets/bom.jpg" alt="Birds On Mars" height="100"> </a> <p></p> An AI initiative by <br> <a href="https://www.bundesumweltministerium.de/" target="_blank" rel="noopener noreferrer"> <img src="assets/bmukn.svg" alt="Bundesministerium für Umwelt, Klimaschutz, Naturschutz und nukleare Sicherheit" height="100"> </a> <p></p> In the context of <br> <a href="https://civic-coding.de" target="_blank" rel="noopener noreferrer"> <img src="assets/civic.svg" alt="Civic Coding" height="100"> </a>
The KOR250 (INSPIRE) in the scale of 1:250,000 shows occurrences and deposits of mineral resources in Germany, which lie close to the Earth’s surface, i.e. can be mined in open-pits, quarries or near-surface mines. These mineral resources include industrial minerals, aggregates, peat, lignite, oil shales, and natural brines. The map is derived from the KOR250, the digital successor of the map series KOR200 „Map of Near-Surface Deposits of the Federal Republic of Germany 1:200,000”, which has been published since 1984. The KOR200 and KOR250 have been published by the Federal Institute for Geosciences and Natural Resources together with the State Geological Surveys of the federal states on behalf of the Federal Ministry for Economic Affairs and Energy. Primary purpose of the KOR250 is to display Germany’s potential of domestic raw materials in a comparable way. The explanations given in the printed booklets accompanying the KOR200 are not available in the digital KOR250. In the KOR250 besides the defined deposits and differently coloured areas of raw materials, "active mines" (= operations) at time of publication or "focal points of several active mines" are marked with one symbol each. These mines are not included in the KOR250 (INSPIRE) as often the headquarters of the mining company and not the mining site itself is displayed as well as in many regions the dataset is outdated. As the map sheets of the KOR200 have been generated over more than three decades the timeliness of data is extremely different. For more detail, the current large-scale raw material maps of the Federal State Geological Surveys should always be consulted. The point data displayed in KOR250 (INSPIRE) indicate very small, but worth mentioning prospects of certain raw materials. According to the Data Specification on Mineral Resources (D2.8.III.21) the content of the map is stored in two INSPIRE-compliant GML files: KOR250_EarthResource_polygon.gml comprises the mineral resources as polygons. KOR250_EarthResource_point.gml comprises the mineral resources as points. The GML files together with a Readme.txt file are provided in ZIP format (KOR250-INSPIRE.zip). The Readme.text file (German/English) contains detailed information on the GML files content. Data transformation was proceeded by using the INSPIRE Solution Pack for FME according to the INSPIRE requirements. Notes: It should be noted that according to the INSPIRE commodity code list, most magmatites and metamorphites were assigned to the two values "granite" and "basalt". From a geological point of view and with regard to its origin, this assignment is often misleading. For more information on the outcropping rock of a specific raw material occurrence, the German name from the original KOR250 was mapped to the attribute name of the class GeologicFeature.
Soil is the first component of the environment that can be effected by GM plants, because they do not only consume the nutritive substances from the soil, but also release there different compounds during a growing period, and leave in the soil their remains. If the plants are modified to increase their resistance to plant pathogens, particularly bacteria, they can also affect the other microorganisms important for plant development. Also there are no considerable data about possible effect of GM plants on soil organic matter and chemical processes in soil. For the experiment it is planned to use transgenic potato plants (Solanum tuberosum L. cv. Desiree) expressing a chimerical gene for T4 lysozyme for protection against bacterial infections; - obtaining and short-term growing of GM plants in laboratory conditions; - extraction and collection of root exudates and microbial metabolites from rhizosphere; - analysis of these exudates by Pyrolysis-Field Ionisation Mass Spectrometry (Py-FIMS) in comparison with the exudates of wild-type plants and transgenic controls not harbouring the lysozyme gene, and with dissolved organic matter from non-cropped soil; - creation of 'fingerprints' for each new transgenic line in combination with certain soil on the basis of marker signals. Expected impacts: - New highly cost-effective express testing system for the risk assessment of genetically modified plants at the earliest stages of their introduction; - The conclusion about safety/danger of GM plants for the soil ecosystems; - Model for prediction of possible risk caused by GM plants.
Ultrafeine Partikel (UFP) mit einem aerodynamischen Durchmesser kleiner als 100 nm stehen unter dem Verdacht die menschliche Gesundheit zu schädigen, allerdings fehlt bisher die abschließende wissenschaftliche Evidenz aus epidemiologischen Studien. Zur Herleitung von Expositionskonzentrationen gegenüber UFP wurden zum Teil statistische Modellierungsverfahren genutzt um UFP-Anzahlkonzentrationen vorherzusagen. Ein häufig genutztes Verfahren ist eine auf Flächennutzung basierte lineare Regression („land-use regression“, LUR). Allerdings wurden in luftqualitativen Studien auch andere, ausgefeiltere Modellansätze benutzt, z.B. „machine learning“ (ML) oder „deep learning“ (DL), die eine bessere Vorhersagegenauigkeit versprechen. Das Ziel des Projekts ist die Modellierung von UFP-Anzahlkonzentration in urbanen Räumen basierend auf ML- und DL-Algorithmen. Diese Algorithmen versprechen eine bessere Vorhersagegenauigkeit gegenüber linearen Modellansätzen. Mit unserem Modellansatz wollen wir sowohl räumliche als auch zeitliche Variabilität der UFP-Anzahlkonzentrationen abbilden. In einem ersten Schritt werden die Messergebnisse aus mobilen Messkampagnen genutzt um ein ML-basiertes LUR Modell zu kalibrieren. Zusätzlich werden urbane Emissionen aus lokalen Quellen, abseits vom Straßenverkehr, identifiziert und explizit in das Modell einbezogen. In einem zweiten Schritt wird ein DL-Modellansatz basierend auf Langzeit-UFP-Messungen mit dem ML-Modell gekoppelt um die Repräsentierung der zeitlichen Variabilität zu verbessern. Unser vorgeschlagenes Arbeitsprogramm besteht aus fünf Arbeitspaketen (WP): WP 1 beinhaltet mobile Messungen mittels eines mobilen Labors und eines Messfahrads. WP 2 besteht aus stationären Messungen, die an Stationen des German Ultrafine Aerosol Network durchgeführt werden. In WP 3 werden wichtige UFP-Emissionsquellen, insbesondere Nicht-Verkehrsemissionen, mit Hilfe von zusätzlichen kurzzeitigen stationären Messungen identifiziert und quantifiziert. In WP 4 werden ML-Algorithmen genutzt um ein statistisches Modell aufzubauen. Als Kalibrierungsdatensatz werden die Messungen aus WP 1 benutzt. Das Modell wird UFP-Anzahlkonzentrationen mit Hilfe eines Datensatzes aus erklärenden Variablen, u.a. meteorologische Größen, Flächennutzung, urbaner Morphologie, Verkehrsmengen und zusätzlichen Informationen zu UFP-Quellen nach WP 3, vorhersagen. In WP 5 werden die UFP-Anzahlkonzentrationen aus WP 2 für einen DL-Modellansatz genutzt, der die zeitliche Variabilität repräsentieren wird. Dieser wird dann mit dem ML-Modell aus WP 4 gekoppelt. Der Nutzen der Modellkopplung wird mit dem Datensatz aus WP 3 validiert. Aus unserem Projekt wird ein Modell hervorgehen, das in der Lage ist die räumliche und zeitliche Variabilität urbaner UFP-Anzahlkonzentrationen in einer hohen Genauigkeit zu repräsentieren. Damit wird unsere Studie einen Beitrag zur Quantifizierung von Expositionskonzentrationen gegenüber UFP z.B. in epidemiologischen Studien leisten.
The formation of biogeochemical interfaces in soils is controlled, among other factors, by the type of particle surfaces present and the assemblage of organic matter and mineral particles. Therefore, the formation and maturation of interfaces is studied with artificial soils which are produced in long-term biogeochemical laboratory incubation experiments (3, 6, 12, 18 months. Clay minerals, iron oxides and charcoal are used as major model components controlling the formation of interfaces because they exhibit high surface area and microporosity. Soil interface characteristics have been analyzed by several groups involved in the priority program for formation of organo-mineral interfaces, sorptive and thermal interface properties, microbial community structure and function. Already after 6 months of incubation, the artificial soils exhibited different properties in relation to their composition. A unique dataset evolves on the development and the dynamics of interfaces in soil in the different projects contributing to this experiment. An integrated analysis based on a conceptual model and multivariate statistics will help to understand overall processes leading to the biogeochemical properties of interfaces in soil, that are the basis for their functions in ecosystems. Therefore, we propose to establish an integrative project for the evaluation of data obtained and for publication of synergistic work, which will bring the results to a higher level of understanding.
| Origin | Count |
|---|---|
| Bund | 7222 |
| Europa | 660 |
| Global | 546 |
| Kommune | 2 |
| Land | 3285 |
| Wirtschaft | 6 |
| Wissenschaft | 40820 |
| Zivilgesellschaft | 6 |
| Type | Count |
|---|---|
| Daten und Messstellen | 5118 |
| Ereignis | 2 |
| Förderprogramm | 2648 |
| Hochwertiger Datensatz | 41 |
| Kartendienst | 2 |
| Repositorium | 35 |
| Sammlung | 2 |
| Software | 1 |
| Taxon | 113 |
| Text | 391 |
| unbekannt | 40106 |
| License | Count |
|---|---|
| geschlossen | 1513 |
| offen | 8471 |
| unbekannt | 38373 |
| Language | Count |
|---|---|
| Deutsch | 814 |
| Englisch | 47677 |
| andere | 1 |
| Resource type | Count |
|---|---|
| Archiv | 1397 |
| Bild | 10372 |
| Datei | 13398 |
| Dokument | 859 |
| Keine | 4618 |
| Unbekannt | 346 |
| Webdienst | 155 |
| Webseite | 38827 |
| Topic | Count |
|---|---|
| Boden | 8082 |
| Lebewesen und Lebensräume | 43234 |
| Luft | 7058 |
| Mensch und Umwelt | 47984 |
| Wasser | 40314 |
| Weitere | 48357 |