Machine learning-based global maps of ecological variables and the challenge of assessing them

How can we quantify, explain, and apply the uncertainty of complex soil maps predicted with neural networks?

Artificial neural networks (ANNs) have proven to be a useful tool for complex questions that involve large amounts of data. Our use case of predicting soil maps with ANNs is in high demand by government agencies, construction companies, or farmers, given cost and time intensive field work. However, there are two main challenges when applying ANNs. In their most common form, deep learning algorithms do not provide interpretable predictive uncertainty. This means that properties of an ANN such as the certainty and plausibility of the predicted variables, rely on the interpretation by experts rather than being quantified by evaluation metrics validating the ANNs. Further, these algorithms have shown a high confidence in their predictions in areas geographically distant from the training area or areas sparsely covered by training data. To tackle these challenges, we use the Bayesian deep learning approach "last-layer Laplace approximation", which is specifically designed to quantify uncertainty into deep networks, in our explorative study on soil classification. It corrects the overconfident areas without reducing the accuracy of the predictions, giving us a more realistic uncertainty expression of the model's prediction. In our study area in southern Germany, we subdivide the soils into soil regions and as a test case we explicitly exclude two soil regions in the training area but include these regions in the prediction. Our results emphasize the need for uncertainty measurement to obtain more reliable and interpretable results of ANNs, especially for regions far away from the training area. Moreover, the knowledge gained from this research addresses the problem of overconfidence of ANNs and provides valuable information on the predictability of soil types and the identification of knowledge gaps. By analyzing regions where the model has limited data support and, consequently, high uncertainty, stakeholders can recognize the areas that require more data collection efforts.

A multi-resolution ensemble model of three decision-tree-based algorithms to predict daily NO2 concentration in France 2005-2022

Understanding and managing the health effects of Nitrogen Dioxide (NO2) requires high resolution spatiotemporal exposure maps. Here, we developed a multi-stage multi-resolution ensemble model that predicts daily NO2 concentration across continental France from 2005 to 2022. Innovations of this work include the computation of daily predictions at a 200 m resolution in large urban areas and the use of a spatio-temporal blocking procedure to avoid data leakage and ensure fair performance estimation. Predictions were obtained after three cascading stages of modeling: (1) predicting NO2 total column density from Ozone Monitoring Instrument satellite; (2) predicting daily NO2 concentrations at a 1 km spatial resolution using a large set of potential predictors such as predictions obtained from stage 1, land-cover and road traffic data; and (3) predicting residuals from stage 2 models at a 200 m resolution in large urban areas. The latter two stages used a generalized additive model to ensemble predictions of three decision-tree algorithms (random forest, extreme gradient boosting and categorical boosting). Cross-validated performances of our ensemble models were overall very good, with a ten-fold cross-validated R2 for the 1 km model of 0.83, and of 0.69 for the 200 m model. All three basis learners participated in the ensemble predictions to various degrees depending on time and space. In sum, our multi-stage approach was able to predict daily NO2 concentrations with a relatively low error. Ensembling the predictions maximizes the chance of obtaining accurate values if one basis learner fails in a specific area or at a particular time, by relying on the other learners. To the best of our knowledge, this is the first study aiming to predict NO2 concentrations in France with such a high spatiotemporal resolution, large spatial extent, and long temporal coverage. Exposure estimates are available to investigate NO2 health effects in epidemiological studies.

High-resolution global maps of yield potential with local relevance for targeted crop production improvement

Identifying untapped opportunities for crop production improvement in current cropland is crucial to guide food availability interventions. Here we integrated an agronomically robust bottom-up approach with machine learning to generate global maps of yield potential of high resolution (ca. 1 km2 at the Equator) and accuracy for maize, wheat and rice. These maps serve as a robust reference to benchmark farmers' yields in the context of current cropping systems and water regimes and can help to identify areas with large room to increase crop yields.

Innovative graph neural network approach for predicting soil heavy metal pollution in the Pearl River Basin, China

Predicting soil heavy metal (HM) content is crucial for monitoring soil quality and ensuring ecological health. However, existing methods often neglect the spatial dependency of data. To address this gap, our study introduces a novel graph neural network (GNN) model, Multi-Scale Attention-based Graph Neural Network for Heavy Metal Prediction (MSA-GNN-HMP). The model integrates multi-scale graph convolutional network (MS-GCN) and attention-based GNN (AGNN) to capture spatial relationships. Using surface soil samples from the Pearl River Basin, we evaluate the MSA-GNN-HMP model against four other models. The experimental results show that the MSA-GNN-HMP model has the best predictive performance for Cd and Pb, with a coefficient of determination (R2) of 0.841 for Cd and 0.886 for Pb, and the lowest mean absolute error (MAE) of 0.403 mg kg-1 for Cd and 0.670 mg kg-1 for Pb, as well as the lowest root mean square error (RMSE) of 0.563 mg kg-1for Cd and 0.898 mg kg-1 for Pb. In feature importance analysis, latitude and longitude emerged as key factors influencing the heavy metal content. The spatial distribution prediction trend of heavy metal elements by different prediction methods is basically consistent, with the high-value areas of Cd and Pb respectively distributed in the northwest and northeast of the basin center. However, the MSA-GNN-HMP model demonstrates superior detail representation in spatial prediction. MSA-GNN-HMP model has excellent spatial information representation capabilities and can more accurately predict heavy metal content and spatial distribution, providing a new theoretical basis for monitoring, assessing, and managing soil pollution.

Global insect herbivory and its response to climate change

Herbivorous insects consume a large proportion of the energy flow in terrestrial ecosystems and play a major role in the dynamics of plant populations and communities. However, high-resolution, quantitative predictions of the global patterns of insect herbivory and their potential underlying drivers remain elusive. Here, we compiled and analyzed a dataset consisting of 9,682 records of the severity of insect herbivory from across natural communities worldwide to quantify its global patterns and environmental determinants. Global mapping revealed strong spatial variation in insect herbivory at the global scale, showing that insect herbivory did not significantly vary with latitude for herbaceous plants but increased with latitude for woody plants. We found that the cation-exchange capacity in soil was a main predictor of levels of herbivory on herbaceous plants, while climate largely determined herbivory on woody plants. We next used well-established scenarios for future climate change to forecast how spatial patterns of insect herbivory may be expected to change with climate change across the world. We project that herbivore pressure will intensify on herbaceous plants worldwide but would likely only increase in certain biomes (e.g., northern coniferous forests) for woody plants. Our assessment provides quantitative evidence of how environmental conditions shape the spatial pattern of insect herbivory, which enables a more accurate prediction of the vulnerabilities of plant communities and ecosystem functions in the Anthropocene.

Modelling Europe-wide fine resolution daily ambient temperature for 2003-2020 using machine learning

To improve our understanding of the health impacts of high and low temperatures, epidemiological studies require spatiotemporally resolved ambient temperature (Ta) surfaces. Exposure assessment over various European cities for multi-cohort studies requires high resolution and harmonized exposures over larger spatiotemporal extents. Our aim was to develop daily mean, minimum and maximum ambient temperature surfaces with a 1 × 1 km resolution for Europe for the 2003-2020 period. We used a two-stage random forest modelling approach. Random forest was used to (1) impute missing satellite derived Land Surface Temperature (LST) using vegetation and weather variables and to (2) use the gap-filled LST together with land use and meteorological variables to model spatial and temporal variation in Ta measured at weather stations. To assess performance, we validated these models using random and block validation. In addition to global performance, and to assess model stability, we reported model performance at a higher granularity (local). Globally, our models explained on average more than 81 % and 93 % of the variability in the block validation sets for LST and Ta respectively. Average RMSE was 1.3, 1.9 and 1.7 °C for mean, min and max ambient temperature respectively, indicating a generally good performance. For Ta models, local performance was stable across most of the spatiotemporal extent, but showed lower performance in areas with low observation density. Overall, model stability and performance were lower when using block validation compared to random validation. The presented models will facilitate harmonized high-resolution exposure assignment for multi-cohort studies at a European scale.

A new ChatGPT-empowered, easy-to-use machine learning paradigm for environmental science

The quantity and complexity of environmental data show exponential growth in recent years. High-quality big data analysis is critical for performing a sophisticated characterization of the complex network of environmental pollution. Machine learning (ML) has been employed as a powerful tool for decoupling the complexities of environmental big data based on its remarkable fitting ability. Yet, due to the knowledge gap across different subjects, ML concepts and algorithms have not been well-popularized among researchers in environmental sustainability. In this context, we introduce a new research paradigm-"ChatGPT + ML + Environment", providing an unprecedented chance for environmental researchers to reduce the difficulty of using ML models. For instance, each step involved in applying ML models to environmental sustainability, including data preparation, model selection and construction, model training and evaluation, and hyper-parameter optimization, can be easily performed with guidance from ChatGPT. We also discuss the challenges and limitations of using this research paradigm in the field of environmental sustainability. Furthermore, we highlight the importance of "secondary training" for future application of "ChatGPT + ML + Environment".

Explanation of the influence of geomorphometric variables on the landform classification based on selected areas in Poland

In recent years, automatic image classification methods have significantly progressed, notably black box algorithms such as machine learning and deep learning. Unfortunately, such efforts only focused on improving performance, rather than attempting to explain and interpret how classification models actually operate. This article compares three state-of-the-art algorithms incorporating random forests, gradient boosting and convolutional neural networks for geomorphological mapping. It also attempts to explain how the most effective classifier makes decisions by evaluating which of the geomorphometric variables are most important for automatic mapping and how they affect the classification results using one of the explainable artificial intelligence techniques, namely accumulated local effects (ALE). This method allows us to understand the relationship between predictors and the model's outcome. For these purposes, eight sheets of the digital geomorphological map of Poland on the scale of 1:100,000 were used as the reference material. The classification results were validated using the holdout method and cross-validation for individual sheets representing different morphogenetic zones. The terrain elevation entropy, absolute elevation, aggregated median elevation and standard deviation of elevation had the greatest impact on the classification results among the 15 geomorphometric variables considered. The ALE analysis was conducted for the XGBoost classifier, which achieved the highest accuracy of 92.8%, ahead of Random Forests at 84% and LightGBM at 73.7% and U-Net at 59.8%. We conclude that automatic classification can support geomorphological mapping only if the geomorphological characteristics in the predicted area are similar to those in the training dataset. The ALE plots allow us to analyze the relationship between geomorphometric variables and landform membership, which helps clarify their role in the classification process.

Convergence and divergence in science and practice of urban and rural forest restoration

Forest restoration has never been higher on policymakers' agendas. Complex and multi-dimensional arrangements across the urban-rural continuum challenge restorationists and require integrative approaches to strengthen environmental protection and increase restoration outcomes. It remains unclear if urban and rural forest restoration are moving towards or away from each other in practice and research, and whether comparing research outcomes can help stakeholders to gain a clearer understanding of the interconnectedness between the two fields. This study aims to identify the challenges and opportunities for enhancing forest restoration in both urban and rural systems by reviewing the scientific evidence, engaging with key stakeholders and using an urban-rural forest restoration framework. Using the Society for Ecological Restoration's International Principles as discussion topics, we highlight aspects of convergence and divergence between the two fields to broaden our understanding of forest restoration and promote integrative management approaches to address future forest conditions. Our findings reveal that urban and rural forest restoration have convergent and divergent aspects. We emphasise the importance of tailoring goals and objectives to specific contexts and the need to design different institutions and incentives based on the social and ecological needs and goals of stakeholders in different regions. Additionally, we discuss the challenges of achieving high levels of ecological restoration and the need to go beyond traditional ecology to plan, implement, monitor, and adaptively manage restored forests. We suggest that rivers and watersheds could serve as a common ground linking rural and urban landscapes and that forest restoration could interact with other environmental protection measures. We note the potential for expanding the creative vision associated with increasing tree-containing environments in cities to generate more diverse and resilient forest restoration outcomes in rural settings. This study underscores the value of integrative management approaches in addressing future forest conditions across the urban-rural continuum. Our framework provides valuable insights for policymakers, researchers, and decision-makers to advance the field of forest restoration and address the challenges of restoration across the urban-rural continuum. The rural-urban interface serves as a convergence point for forest restoration, and both urban and rural fields can benefit from each other's expertise.

Estimating daily air temperature and pollution in Catalonia: A comprehensive spatiotemporal modelling of multiple exposures

Environmental epidemiology studies require models of multiple exposures to adjust for co-exposure and explore interactions. We estimated spatiotemporal exposure to surface air temperature and pollution (PM2.5, PM10, NO2, O3) at high spatiotemporal resolution (daily, 250 m) for 2018-2020 in Catalonia. Innovations include the use of TROPOMI products, a data split for remote sensing gap-filling evaluation, estimation of prediction uncertainty, and use of explainable machine learning. We compiled meteorological and air quality station measurements, climate and atmospheric composition reanalyses, remote sensing products, and other spatiotemporal data. We performed gap-filling of remotely-sensed products using Random Forest (RF) models and validated them using Out-Of-Bag (OOB) samples and a structured data split. The exposure modelling workflow consisted of: 1) PM2.5 station imputation with PM10 data; 2) quantile RF (QRF) model fitting; and 3) geostatistical residual spatial interpolation. Prediction uncertainty was estimated using QRF. SHAP values were used to examine variable importance and the fitted relationships. Model performance was assessed via nested CV at the station level. Evaluation of the gap-filling models using the structured split showed error underestimation when using OOB. Temperature models had the best performance (R2 =0.98) followed by the gaseous air pollutants (R2 =0.81 for NO2 and 0.86 for O3), while the performance of the PM2.5 and PM10 models was lower (R2 =0.57 and 0.63 respectively). Predicted exposure patterns captured urban heat island effects, dust advection events, and NO2 hotspots. SHAP values estimated a high importance of TROPOMI tropospheric NO2 columns in PM and NO2 models, and confirmed that the fitted associations conformed to prior knowledge. Our work highlights the importance of correctly validating gap-filling models and the potential of TROPOMI measurements. Moderate performance in PM models can be partly explained by the poor station coverage. Our exposure estimates can be used in epidemiological studies potentially accounting for exposure uncertainty.

The neglected role of abandoned cropland in supporting both food security and climate change mitigation

Despite the looming land scarcity for agriculture, cropland abandonment is widespread globally. Abandoned cropland can be reused to support food security and climate change mitigation. Here, we investigate the potentials and trade-offs of using global abandoned cropland for recultivation and restoring forests by natural regrowth, with spatially-explicit modelling and scenario analysis. We identify 101 Mha of abandoned cropland between 1992 and 2020, with a capability of concurrently delivering 29 to 363 Peta-calories yr-1 of food production potential and 290 to 1,066 MtCO2 yr-1 of net climate change mitigation potential, depending on land-use suitability and land allocation strategies. We also show that applying spatial prioritization is key to maximizing the achievable potentials of abandoned cropland and demonstrate other possible approaches to further increase these potentials. Our findings offer timely insights into the potentials of abandoned cropland and can inform sustainable land management to buttress food security and climate goals.

Earthworms contribute significantly to global food production

Earthworms are critical soil ecosystem engineers that support plant growth in numerous ways; however, their contribution to global agricultural production has not been quantified. We estimate the impacts of earthworms on global production of key crops by analyzing maps of earthworm abundance, soil properties, and crop yields together with earthworm-yield responses from the literature. Our findings indicate that earthworms contribute to roughly 6.5% of global grain (maize, rice, wheat, barley) production and 2.3% of legume production, equivalent to over 140 million metric tons annually. The earthworm contribution is especially notable in the global South, where earthworms contribute 10% of total grain production in Sub-Saharan Africa and 8% in Latin America and the Caribbean. Our findings suggest that earthworms are important drivers of global food production and that investment in agroecological policies and practices to support earthworm populations and overall soil biodiversity could contribute greatly to sustainable agricultural goals.

Global determinants of insect mitochondrial genetic diversity

Understanding global patterns of genetic diversity is essential for describing, monitoring, and preserving life on Earth. To date, efforts to map macrogenetic patterns have been restricted to vertebrates, which comprise only a small fraction of Earth's biodiversity. Here, we construct a global map of predicted insect mitochondrial genetic diversity from cytochrome c oxidase subunit 1 sequences, derived from open data. We calculate the mitochondrial genetic diversity mean and genetic diversity evenness of insect assemblages across the globe, identify their environmental correlates, and make predictions of mitochondrial genetic diversity levels in unsampled areas based on environmental data. Using a large single-locus genetic dataset of over 2 million globally distributed and georeferenced mtDNA sequences, we find that mitochondrial genetic diversity evenness follows a quadratic latitudinal gradient peaking in the subtropics. Both mitochondrial genetic diversity mean and evenness positively correlate with seasonally hot temperatures, as well as climate stability since the last glacial maximum. Our models explain 27.9% and 24.0% of the observed variation in mitochondrial genetic diversity mean and evenness in insects, respectively, making an important step towards understanding global biodiversity patterns in the most diverse animal taxon.

A spatio-temporal analysis investigating completeness and inequalities of global urban building data in OpenStreetMap

OpenStreetMap (OSM) has evolved as a popular dataset for global urban analyses, such as assessing progress towards the Sustainable Development Goals. However, many analyses do not account for the uneven spatial coverage of existing data. We employ a machine-learning model to infer the completeness of OSM building stock data for 13,189 urban agglomerations worldwide. For 1,848 urban centres (16% of the urban population), OSM building footprint data exceeds 80% completeness, but completeness remains lower than 20% for 9,163 cities (48% of the urban population). Although OSM data inequalities have recently receded, partially as a result of humanitarian mapping efforts, a complex unequal pattern of spatial biases remains, which vary across various human development index groups, population sizes and geographic regions. Based on these results, we provide recommendations for data producers and urban analysts to manage the uneven coverage of OSM data, as well as a framework to support the assessment of completeness biases.

Integrating large-scale stationary and local mobile measurements to estimate hyperlocal long-term air pollution using transfer learning methods

Mobile air quality measurements are collected typically for several seconds per road segment and in specific timeslots (e.g., working hours). These short-term and on-road characteristics of mobile measurements become the ubiquitous shortcomings of applying land use regression (LUR) models to estimate long-term concentrations at residential addresses. This issue was previously found to be mitigated by transferring LUR models to the long-term residential domain using routine long-term measurements in the studied region as the transfer target (local scale). However, long-term measurements are generally sparse in individual cities. For this scenario, we propose an alternative by taking long-term measurements collected over a larger geographical area (global scale) as the transfer target and local mobile measurements as the source (Global2Local model). We empirically tested national, airshed countries (i.e., national plus neighboring countries) and Europe as the global scale in developing Global2Local models to map nitrogen dioxide (NO₂) concentrations in Amsterdam. The airshed countries scale provided the lowest absolute errors, and the Europe-wide scale had the highest R². Compared to a "global" LUR model (trained exclusively with European-wide long-term measurements), and a local mobile LUR model (using mobile data from Amsterdam only), the Global2Local model significantly reduced the absolute error of the local mobile LUR model (root-mean-square error, 6.9 vs 12.6 μg/m³) and improved the percentage explained variances compared to the global model (R², 0.43 vs 0.28, assessed by independent long-term NO₂ measurements in Amsterdam, n = 90). The Global2Local method improves the generalizability of mobile measurements in mapping long-term residential concentrations with a fine spatial resolution, which is preferred in environmental epidemiological studies.

Projected landscape-scale repercussions of global action for climate and biodiversity protection

Land conservation and increased carbon uptake on land are fundamental to achieving the ambitious targets of the climate and biodiversity conventions. Yet, it remains largely unknown how such ambitions, along with an increasing demand for agricultural products, could drive landscape-scale changes and affect other key regulating nature's contributions to people (NCP) that sustain land productivity outside conservation priority areas. By using an integrated, globally consistent modelling approach, we show that ambitious carbon-focused land restoration action and the enlargement of protected areas alone may be insufficient to reverse negative trends in landscape heterogeneity, pollination supply, and soil loss. However, we also find that these actions could be combined with dedicated interventions that support critical NCP and biodiversity conservation outside of protected areas. In particular, our models indicate that conserving at least 20% semi-natural habitat within farmed landscapes could primarily be achieved by spatially relocating cropland outside conservation priority areas, without additional carbon losses from land-use change, primary land conversion or reductions in agricultural productivity.

A quixotic view of spatial bias in modelling the distribution of species and their diversity

Ecological processes are often spatially and temporally structured, potentially leading to autocorrelation either in environmental variables or species distribution data. Because of that, spatially-biased in-situ samples or predictors might affect the outcomes of ecological models used to infer the geographic distribution of species and diversity. There is a vast heterogeneity of methods and approaches to assess and measure spatial bias; this paper aims at addressing the spatial component of data-driven biases in species distribution modelling, and to propose potential solutions to explicitly test and account for them. Our major goal is not to propose methods to remove spatial bias from the modelling procedure, which would be impossible without proper knowledge of all the processes generating it, but rather to propose alternatives to explore and handle it. In particular, we propose and describe three main strategies that may provide a fair account of spatial bias, namely: (i) how to represent spatial bias; (ii) how to simulate null models based on virtual species for testing biogeographical and species distribution hypotheses; and (iii) how to make use of spatial bias - in particular related to sampling effort - as a leverage instead of a hindrance in species distribution modelling. We link these strategies with good practice in accounting for spatial bias in species distribution modelling.

Oceanic differences in coral-bleaching responses to marine heatwaves

Anomalously high ocean temperatures have increased in frequency, intensity, and duration over the last several decades because of greenhouse gas emissions that cause global warming and marine heatwaves. Reef-building corals are sensitive to such temperature anomalies that commonly lead to coral bleaching, mortality, and changes in community structure. Yet, despite these overarching effects, there are geographical differences in thermal regimes, evolutionary histories, and past disturbances that may lead to different bleaching responses of corals within and among oceans. Here we examined the overall bleaching responses of corals in the Atlantic, Indian, and Pacific Oceans, using both a spatially explicit Bayesian mixed-effects model and a deep-learning neural-network model. We used a 40-year global dataset encompassing 23,288 coral-reef surveys at 11,058 sites in 88 countries, from 1980 to 2020. Focusing on ocean-wide differences we assessed the relationships between the percentage of bleached corals and different temperature-related metrics alongside a suite of environmental variables. We found that while high sea-surface temperatures were consistently, and strongly, related to coral bleaching within all oceans, there were clear geographical differences in the relationships between coral bleaching and most environmental variables. For instance, there was an increase in coral bleaching with depth in the Atlantic Ocean whereas the opposite was observed in the Indian Ocean, and no clear trend could be seen in the Pacific Ocean. The standard deviation of thermal-stress anomalies was negatively related to coral bleaching in the Atlantic and Pacific Oceans, but not in the Indian Ocean. Globally, coral bleaching has progressively occurred at higher temperatures over the last four decades within the Atlantic, Indian, and Pacific Oceans, although, again, there were differences among the three oceans. Together, such patterns highlight that historical circumstances and geographical differences in oceanographic conditions play a central role in contemporary coral-bleaching responses.

Integrating gradient with scale in ecological and evolutionary studies

Gradient and scale are two key concepts in ecology and evolution that are closely related but inherently distinct. While scale commonly refers to the dimensional space of a specific ecological/evolutionary (eco-evo) issue, gradient measures the range of a given variable. Gradient and scale can jointly and interactively influence eco-evo patterns. Extensive previous research investigated how changing scales may affect the observation and interpretation of eco-evo patterns; however, relatively little attention has been paid to the role of changing gradients. Here, synthesizing recent research progress, we suggest that the role of scale in the emergence of ecological patterns should be evaluated in conjunction with considering the underlying environmental gradients. This is important because, in most studies, the range of the gradient is often part of its full potential range. The difference between sampled (partial) versus potential (full) environmental gradients may profoundly impact observed eco-evo patterns and alter scale-gradient relationships. Based on observations from both field and experimental studies, we illustrate the underlying features of gradients and how they may affect observed patterns, along with the linkages of these features to scales. Since sampled gradients often do not cover their full potential ranges, we discuss how the breadth and the starting and ending positions of key gradients may affect research design and data interpretation. We then outline potential approaches and related perspectives to better integrate gradient with scale in future studies.

Random forest-based modeling of stream nutrients at national level in a data-scarce region

Nutrient runoff from agricultural production is one of the main causes of water quality deterioration in river systems and coastal waters. Water quality modeling can be used for gaining insight into water quality issues in order to implement effective mitigation efforts. Process-based nutrient models are very complex, requiring a lot of input parameters and computationally expensive calibration. Recently, ML approaches have shown to achieve an accuracy comparable to the process-based models and even outperform them when describing nonlinear relationships. We used observations from 242 Estonian catchments, amounting to 469 yearly TN and 470 TP measurements covering the period 2016-2020 to train random forest (RF) models for predicting annual N and P concentrations. We used a total of 82 predictor variables, including land cover, soil, climate and topography parameters and applied a feature selection strategy to reduce the number of dependent features in the models. The SHAP method was used for deriving the most relevant predictors. The performance of our models is comparable to previous process-based models used in the Baltic region with the TN and TP model having an R² score of 0.83 and 0.52, respectively. However, as input data used in our models is easier to obtain, the models offer superior applicability in areas, where data availability is insufficient for process-based approaches. Therefore, the models enable to give a robust estimation for nutrient losses at national level and allows to capture the spatial variability of the nutrient runoff which in turn enables to provide decision-making support for regional water management plans.