Environmental monitoring of the future

The Baltic Sea, including the Swedish coast, is one of the worlds most polluted seas. In addition, several areas along the Swedish west coast are also affected by chemical pollution. A good framework is necessary to track the consequences of the contamination on the environment closely using reliable strategies. European requirements for monitoring are outlined in the EU Water Framework Directive (WFD) that regulates monitoring in fresh water and coastal areas and the Marine Strategy Framework Directive (MSFD) that regulates marine monitoring. To live up to these requirements, most countries within the European Union have annual standardized campaigns that often includes chemical analysis and biological effect screenings. Swedish coastal monitoring has focused on both chemical analysis and analysis of effects of chemicals on biological systems using biological effect markers. While these two approaches have been running in parallel for decades, very few sampling sites overlap and there have been few attempts at integrating the monitoring data. The Swedish coastal effect monitoring and the contaminants monitoring program have accumulated large amount of data that can be used to test novel methods for data integration and generation of assessment criteria. This, with the addition of novel biological effects markers, including assessment of meiofauna biodiversity, will be a cornerstone to outline new strategies for future strategies for environment risk assessments. Risk assessments that are necessary to evaluate coastal marine areas for good environmental status (GES). This is of great value for Swedish monitoring but also for harmonizing monitoring activities within the EU and MSFD. The main aim of this project is to use available data, and new data from case studies, to develop new data integration strategies and assessment criteria’s necessary for data integration. In addition, the project will optimize new biomarkers for monitoring. 

Project leader: Joachim Sturve, University of Gothenburg

Amount: 5 000 000 SEK

We currently lack cost-effective sampling for multiple environmental pollutants and guidelines for how exposures and their determinants should be studied. Likewise, there are shortcomings in how existing data within the Swedish health-related environmental monitoring (HÄMI) is made useful for all potential stakeholders. With the HÄMImpact study, we aim to improve and develop the monitoring of environmental chemicals, evaluate and recommend state-of-the-art analysis of multiple exposures, and develop data processing of all human biomonitoring data collected within HÄMI. This will be accomplished by 1) evaluating new methods for self-sampling of small amounts of blood for measurements of organic and inorganic environmental chemicals and implementing these in ongoing studies for biological monitoring within HÄMI; 2) evaluating new statistical tools for processing and presenting the individual exposures, mixed exposures and potential determinants; and 3) developing new methods for data processing by HÄMI on new and existing data via making reports and data available. In the end, the project will develop and strengthen the human biomonitoring of toxic chemicals, which in turn will lead to improved data for following up the Swedish environmental goals, the drinking water directive, official statistics, and the EU's zero pollution strategy. 

Project leader: Karin Broberg, Karolinska Institutet

Amount: 4 947 060 SEK

Electrofishing data form the basis for the monitoring of numerous fish populations across the Nordic countries and play a key role in the status assessment of our rivers under the Water Framework Directive. However, the methods used to analyze electrofishing data in Sweden are currently not able to handle the data in a consistent way, so that catchability is fixed to an average value for large proportion of the data (over 40% in 2023), potentially creating significant biases. In addition, the current experimental design of the Swedish electrofishing program may lack statistical power to detect trends in fish population abundance.

We will develop a hierarchical Bayesian model (HBM) for the electrofishing data, that will combine data from single- and multiple-pass fishing and allow flow of information between sites within a river, and between similar rivers, allowing efficient usage of all the available data. We will create a user-friendly interactive web interface (Shiny app) for the model, to promote its use by practitioners.

We will also use the HBM to provide new recommendations on optimizing the experimental design of electrofishing programmes to ensure that the data collected are suited to answer the questions of interest in the most cost-effective way. 

Overall, this project is expected to i) substantially improve the utility of the data collected in both historical and future electrofishing and ii) update analytical methods to minimize bias in fish density estimates. This will lead to an improved basis for decision-making for local, national and international assessments of riverine fish. Quality-assured fish density estimates from electrofishing also have an important role to play in the calibration of estimates from emerging technologies like eDNA. 

Project leader: Rebecca Whitlock, Swedish University of Agricultural Sciences

Amount: 2 474 543 SEK

National environmental monitoring programs, such as Sweden's National Forest Inventory (NFI) and the National Inventories of Landscapes in Sweden (NILS), provide important data on habitats such as wetlands, grasslands, and forests. The data helps identify trends and support monitoring for environmental goals and the EU Habitat Directive, but the programs are under constant pressure to be cost-effective. The effectiveness of these programs depends on factors such as sample size, data quality, and the ability to detect early changes in vegetation related to research questions. Although NFI and NILS provide reliable national estimates, local estimates are difficult due to limited sample sizes and costs. Maps based on models offer a solution by providing localized information using satellite and LIDAR data, but the accuracy depends on the quality of the models, the training and remote sensing data.

Despite advancements in remote sensing, aerial and drone images are rarely used in Sweden's monitoring programs. To address this, NILS developed a two-step method in 2020 that uses model-based maps to improve the selection of sampling sites in alpine regions. Drone images, with their high resolution, can enhance data collection and analysis by serving as a bridge between field, aerial and satellite data. They also have the potential to reduce fieldwork by automating estimates of coverage for various species, currently done visually.

The project's goal is to improve the efficiency of monitoring alpine vegetation in Sweden by integrating Deep Learning image analysis with data from drones, aerial images, LIDAR, and satellites. This will improve data collection and produce better model-based maps for vegetation monitoring and change detection.

The project is divided into three work areas: improving the accuracy of field data using drones, improving modelbased maps with aerial and LIDAR data, and combining time series satellite data and aerial images for better vegetation mapping. 

Project leader: Sven Adler, Swedish University of Agricultural Sciences

Amount: 4 991 442 SEK

Estimating the impact of fisheries on protected, endangered, and threatened species (Pets) is essential for managing unintentional bycatch. However, Pets bycatch events are sporadic, requiring large sample sizes and extended monitoring periods to generate reliable estimates. Electronic monitoring (EM) offers a way to gather the necessary data. In Sweden, an EM monitoring programme has been implemented, where large quantities of video data are collected and manually screened. However, this process is time-consuming, costly and raises privacy concerns. Machine learning (ML) has the potential to automate EM bycatch detection, significantly reducing the need for manual analysis and data handling.

A pilot study carried out by Slu demonstrated the feasibility of ML for detecting bycatch in film data from gillnet fisheries. However, the models require further refinement and larger training datasets to achieve the accuracy needed for reliable Pets identification. A major challenge is the insufficient availability of rare bycatch footage within any single fishery, a global issue that limits model performance. Our approach addresses these challenges through a multi-pronged strategy. First, we will augment training datasets and optimise our bycatch detection models by collaborating with other institutions to collect additional bycatch data. Second, we will develop a complementary ML model capable of identifying hauling operations in video footage, automatically discarding irrelevant footage early in the analysis process, thereby significantly reducing the volume of data to be processed and stored. Third, we will implement a novel federated machine learning approach. By setting up decentralised nodes at collaborating institutes to train a shared model without transferring raw video data to a central server, we can preserve data privacy while leveraging larger, combined datasets to improve model accuracy. Finally, we will develop methods for implementing ML models in practice. 

Project leader: Sara Königson, Swedish University of Agricultural Sciences

Amount: 4 982 017 SEK

The overall aim of this project is to facilitate the introduction of cellular effect-based methods into Swedish environmental toxicant monitoring and risk assessment of aquatic systems. Current monitoring mainly focuses on known chemicals with known toxicity, without considering chemical mixtures. Given the vast number of chemicals released into the environment, it's impractical to measure each one and the toxicity of chemicals is in many cases unknown or uncertain. Moreover, these chemicals form complex mixtures in the environment. Although cellular effect-based methods that account for these unknowns have been developed, they have mostly been applied in research, with limited uptake by the Swedish water sector and regulators. The challenge in interpreting results and establishing action-triggering effect levels hinders their use in environmental monitoring.

Through three Swedish case studies with different pollution sources and involving a network of water management representatives from each case site, the project will collect data using accessible standardized cellular effect-based assays. Parallel chemical analyses will provide a more complete assessment of chemical risk, showing how much risk traditional analyses miss compared to effect-based methods. Continuous dialogue with water management representatives will help develop, test, and share concrete guidelines for using cellular effect-based methods in Sweden, addressing regulatory needs. The collected data will serve as a reference for future interpretation. The guidelines will cover methods, effect-based thresholds, and follow-up actions, helping facilitate the adoption of these methods in Sweden.

Project leader: Gunnar Thorsén, IVL Swedish Environmental Research Institute Ltd.

Amount: 4 992 615 SEK

Sweden reports emissions and uptake of GHG in forests and grassland based on repeated inventories in the Swedish Forest Soil Inventory. Emissions and uptake of CO2 in soils are quantitatively important in Swedens national GHG inventory but the statistical uncertainty in the estimates are high. In connection to an increased burden for Sweden in EU’s ambitious plan to increase uptake in the LULUCF sector these statistical uncertainties are a problem.

The purpose of this project is to develop a sampling and analysis model for carbon stock determinations in forest and grasslands based on taking more samples over a larger sample plot (300 - 500 m²) rather than a single sample. This reduces the statistical uncertainty between inventory rounds for each surveyed sample area because the local representativeness of the carbon stock estimate increases. By introducing analyses with infrared spectroscopy, an analytical technique that allows a large number of samples to be analyzed at a low cost per sample, the analysis costs for the increasing number of samples can be kept lower.

The project will conduct fieldwork over two seasons. In the first year, 10 large square sample plots will be sampled in different types of forest and grassland in three different horizons: litter and two mineral soil horizons (0-10 cm, 10-20 cm). Once the soil samples are analysed, the data will be analyzed using geostatistical methods and by experimenting with sample spacing and the number of samples, both the sample plot size and sampling design can be optimized. In the second year, a larger number of sample plots, 40 in total, will be sampled but with an optimized sample plot size and sample spacing. With this data, the reduction in statistical uncertainty can be determined, and we can predict how much we can reduce uncertainties in calculations of carbon stock changes.

The project will present a proposal for a new model for sampling and soil analysis within the soil inventory in the final report.

Project leader: Erik Karltun, Swedish University of Agricultural Sciences

Amount: 4 519 000 SEK