Search

To properly evaluate the quantity of mineral resources in a deposit, modeling the rock units of the deposit is essential since it informs the deposit's mineralization controls. The deterministic methods of modeling include geological assumptions, especially between drill holes, and therefore, might predict the deposit's complexity biasedly. Additionally, the uncertainty of rock units at unsampled locations cannot be quantified. Truncated/plurigaussian Gaussian and sequential indicator simulations are two most frequent stochastic approaches better suited to address these shortcomings. The ability to precisely depict the spatial correlation between rock units and quantify their uncertainties has made the truncated Gaussian simulation an essential technique for delineating rock units. However, using this method, long-range geological features like veins, faults, and fractures are often loosely modeled thus necessitating the use of soft data such as deterministic interpretive geological model to avert this inaccuracy. In this paper, the truncated Gaussian simulation is used to assess the spatial variability of rock units and quantify the uncertainty in their occurrence in a vein-dominated gold deposit. A novel approach is presented to model the rock units, particularly the veins, by incorporating a collocated co-simulation algorithm via a global correlation coefficient parameter. The realizations are conditioned to drill hole data and a collocated interpretive geological model created by a machine learning algorithm. The outcome shows realistic reproduction of the veins and other rock units, as representing the drill hole data and assessing their uncertainties thereof. [ABSTRACT FROM AUTHOR]

Acquiring comprehensive insights into soil properties at various spatial scales is paramount for effective land management, especially within small catchment areas that often serve as vital pastured landscapes. These regions, characterized by the intricate interplay of agroforestry systems and livestock grazing, face a pressing challenge: mitigating soil degradation while optimizing land productivity. This study aimed to analyze the spatial distribution of eight topsoil (0–5 cm) properties (clay, silt, sand, pH, cation exchange capacity, available potassium, total nitrogen, and soil organic matter) in a small grazed catchment. Four machine learning algorithms—Random Forest (RF), Support Vector Machines (SVM), Cubist, and K-Nearest Neighbors (kNN)—were used. The Boruta algorithm was employed to reduce the dimensionality of environmental covariates. The model's accuracy was assessed using the Concordance Correlation Coefficient (CCC) and Root Mean Square Error (RMSE). Additionally, uncertainty in predicted maps was quantified and assessed. The results revealed variations in predictive model performance for soil properties. Specifically, kNN excelled for clay, silt, and sand content, while RF performed well for soil pH, CEC, and TN. Cubist and SVM achieved accuracy in predicting AK and SOM, respectively. Clay, silt, CEC, and TN yielded favourable predictions, closely aligning with observations. Conversely, sand content, soil pH, AK, and SOM predictions were slightly less accurate, highlighting areas for improvement. Boruta algorithm streamlined covariate selection, reducing 23 covariates to 10 for clay and 4 for soil pH and AK prediction, enhancing model efficiency. Our study revealed spatial uncertainty patterns mirroring property distributions, with higher uncertainty in areas with elevated content. Model accuracy varied by confidence levels, performing best at intermediate levels and showing increased uncertainty at extremes. These findings offer insights into model capabilities and guide future research in soil property prediction. In conclusion, these results urge more research in small watersheds for soil and territorial management. [ABSTRACT FROM AUTHOR]

Abstract Background The Qinghai-Tibet Plateau is the “sensitive area” of climate change, and also the “driver” and “amplifier” of global change. The response and feedback of its carbon dynamics to climate change will significantly affect the content of greenhouse gases in the atmosphere. However, due to the unique geographical environment characteristics of the Qinghai-Tibet Plateau, there is still much controversy about its carbon source and sink estimation results. This study designed a new algorithm based on machine learning to improve the accuracy of carbon source and sink estimation by integrating multiple scale carbon input (net primary productivity, NPP) and output (soil heterotrophic respiration, Rh) information from remote sensing and ground observations. Then, we compared spatial patterns of NPP and Rh derived from the fusion of multiple scale data with other widely used products and tried to quantify the differences and uncertainties of carbon sink simulation at a regional scale. Results Our results indicate that although global warming has potentially increased the Rh of the Qinghai-Tibet Plateau, it will also increase its NPP, and its current performance is a net carbon sink area (carbon sink amount is 22.3 Tg C/year). Comparative analysis with other data products shows that CASA, GLOPEM, and MODIS products based on remote sensing underestimate the carbon input of the Qinghai-Tibet Plateau (30–70%), which is the main reason for the severe underestimation of the carbon sink level of the Qinghai-Tibet Plateau (even considered as a carbon source). Conclusions The estimation of the carbon sink in the Qinghai-Tibet Plateau is of great significance for ensuring its ecological barrier function. It can deepen the community’s understanding of the response to climate change in sensitive areas of the plateau. This study can provide an essential basis for assessing the uncertainty of carbon sources and sinks in the Qinghai-Tibet Plateau, and also provide a scientific reference for helping China achieve “carbon neutrality” by 2060.

Background: The Qinghai-Tibet Plateau is the "sensitive area" of climate change, and also the "driver" and "amplifier" of global change. The response and feedback of its carbon dynamics to climate change will significantly affect the content of greenhouse gases in the atmosphere. However, due to the unique geographical environment characteristics of the Qinghai-Tibet Plateau, there is still much controversy about its carbon source and sink estimation results. This study designed a new algorithm based on machine learning to improve the accuracy of carbon source and sink estimation by integrating multiple scale carbon input (net primary productivity, NPP) and output (soil heterotrophic respiration, Rh) information from remote sensing and ground observations. Then, we compared spatial patterns of NPP and Rh derived from the fusion of multiple scale data with other widely used products and tried to quantify the differences and uncertainties of carbon sink simulation at a regional scale. Results: Our results indicate that although global warming has potentially increased the Rh of the Qinghai-Tibet Plateau, it will also increase its NPP, and its current performance is a net carbon sink area (carbon sink amount is 22.3 Tg C/year). Comparative analysis with other data products shows that CASA, GLOPEM, and MODIS products based on remote sensing underestimate the carbon input of the Qinghai-Tibet Plateau (30–70%), which is the main reason for the severe underestimation of the carbon sink level of the Qinghai-Tibet Plateau (even considered as a carbon source). Conclusions: The estimation of the carbon sink in the Qinghai-Tibet Plateau is of great significance for ensuring its ecological barrier function. It can deepen the community's understanding of the response to climate change in sensitive areas of the plateau. This study can provide an essential basis for assessing the uncertainty of carbon sources and sinks in the Qinghai-Tibet Plateau, and also provide a scientific reference for helping China achieve "carbon neutrality" by 2060. Highlights: This study designed a carbon sink estimation method for Qinghai-Tibet Plateau by integrating machine learning and multiple scale ground- and remote sensed-based data. The estimated total carbon sink of the Qinghai-Tibet Plateau is 22.3 Tg C/year, accounting for about 10% of China's total carbon sink. The carbon sink of former estimation maybe greatly underestimated due to the underestimation of carbon input in the Qinghai-Tibet Plateau. [ABSTRACT FROM AUTHOR]

Real-time error correction is an effective measure to improve forecast accuracy. This paper develops a real-time error correction model based on machine learning (ML) ensemble method, including the establishment of base learners, heterogeneity test, and Stacking combination. Then the Copula-based Bayesian processor of forecast (BPF) is adopted for probabilistic forecasts, which quantitatively describes the uncertainty of the correction results. Finally, the model performance is comprehensively evaluated from accuracy and stability, with deterministic metrics used to evaluate the correction effect and uncertain metrics for assessing probabilistic forecasts. The proposed model is applied to the multireservoir system in the Huai River Basin, and the results reveal the following: (1) Unlike the single ML algorithm with the performance of oscillations, the Stacking ensemble method can aggregate the advantages of multiple learners, showing robust correction effect and high adaptability across all data samples. (2) The forecasting error is significantly reduced by the Stacking method, with the average Nash–Sutcliffe efficiency coefficient ( NSE ) value increasing above 0.9, which is 4.93% higher than that of the autoregressive (AR) method. The greater superiority is also shown in the remaining evaluation metric values. Moreover, as the lead time increases, the performance of the stacking method tends to have a slower decline trend than the AR method. (3) The changes in the structure of the Stacking method have a relatively small influence on the forecast uncertainty, with all the Containing ratio ( CR ) values over 80% for different samples. The flexible combination of ML algorithms in the ensemble method will not add additional uncertainty factors and ensure the stability of the correction performance. The framework for real-time error correction and its uncertainty assessment has overall optimal correction performance with less observed data and lower computational cost required, which is promising for further improving the accuracy and reliability of real-time flood forecasting. [ABSTRACT FROM AUTHOR]

• SOC stock was predicted between 1992 and 2016 at various aggregation levels. • Both static and dynamic environmental covariates were extensively used. • Crucial to consider spatiotemporal correlation in upscaling with quantified uncertainty. • Larger support reduces uncertainty, aids in identifying significant SOC stock changes. • Exhaustive SOC stock predictions support environmental and agricultural challenges. The role of soil organic carbon (SOC) is crucial not only for numerous soil functions and processes but also for addressing various environmental crises and challenges we face. Consequently, the demand for information on the spatiotemporal variability of SOC is increasing, posing new methodological challenges, such as the need for information on SOC and SOC changes with quantified uncertainty across a wide variety of spatial scales and temporal periods. Our objective was to present a methodology based on a combination of machine learning and space–time geostatistics to predict the spatiotemporal variability of SOC stock with quantified uncertainty at various spatial supports (i.e., point support, 1 × 1 km, 5 × 5 km, 10 × 10 km, 25 × 25 km, counties, and the entire country) for Hungary, between 1992 and 2016. The role of geostatistics is pivotal, as it accounts for the spatiotemporal correlation of the interpolation errors, which is essential for reliably quantifying the uncertainty associated with spatially aggregated SOC stock and SOC stock change predictions. Five times repeated 10-fold leave-location-out cross-validation was used to evaluate the point support predictions and uncertainty quantifications, yielding acceptable results for both SOC stock (ME = −0.897, RMSE = 19.358, MEC = 0.321, and G = 0.912) and SOC stock change (ME = 0.414, RMSE = 16.626, MEC = 0.160, and G = 0.952). We compiled a series of maps of SOC stock predictions between 1992 and 2016 for each support, along with the quantified uncertainty, which is unprecedented in Hungary. It was demonstrated that the larger the support, the smaller the prediction uncertainty, which facilitates the identification and delineation of larger areas with statistically significant SOC stock changes. Moreover, the methodology can overcome the limitations of recent approaches in the spatiotemporal modelling of SOC, allowing the prediction of SOC and SOC changes, with quantified uncertainty, for any year, time period, and spatial scale. This methodology is capable of meeting the current and anticipated demands for dynamic information on SOC at both national and international levels. [ABSTRACT FROM AUTHOR]

Sensor uncertainty significantly affects the performance of autonomous vehicles (AVs). Sensor uncertainty is predominantly linked to sensor specifications, and because sensor behaviors change dynamically, the machine learning approach is not suitable for learning them. This paper presents a novel learning approach for predicting sensor performance in challenging environments. The design of our approach incorporates both epistemic uncertainties, which are related to the lack of knowledge, and aleatoric uncertainties, which are related to the stochastic nature of the data acquisition process. The proposed approach combines a state-based model with a predictive model, where the former estimates the uncertainty in the current environment and the latter finds the correlations between the source of the uncertainty and its environmental characteristics. The proposed approach has been evaluated on real data to predict the uncertainties associated with global navigation satellite systems (GNSSs), showing that our approach can predict sensor uncertainty with high confidence.

Sensor uncertainty significantly affects the performance of autonomous vehicles (AVs). Sensor uncertainty is predominantly linked to sensor specifications, and because sensor behaviors change dynamically, the machine learning approach is not suitable for learning them. This paper presents a novel learning approach for predicting sensor performance in challenging environments. The design of our approach incorporates both epistemic uncertainties, which are related to the lack of knowledge, and aleatoric uncertainties, which are related to the stochastic nature of the data acquisition process. The proposed approach combines a state-based model with a predictive model, where the former estimates the uncertainty in the current environment and the latter finds the correlations between the source of the uncertainty and its environmental characteristics. The proposed approach has been evaluated on real data to predict the uncertainties associated with global navigation satellite systems (GNSSs), showing that our approach can predict sensor uncertainty with high confidence. [ABSTRACT FROM AUTHOR]

Process‐based hydrological models seek to represent the dominant hydrological processes in a catchment. However, due to unavoidable incompleteness of knowledge, the construction of "fidelius" process‐based models depends largely on expert judgment. We present a systematic approach that treats models as hierarchical assemblages of hypotheses (conservation principles, system architecture, process parameterization equations, and parameter specification), which enables investigating how the hierarchy of model development decisions impacts model fidelity. Each model development step provides information that progressively changes our uncertainty (increases, decreases, or alters) regarding the input‐state‐output behavior of the system. Following the principle of maximum entropy, we introduce the concept of "minimally restrictive process parameterization equations—MR‐PPEs," which enables us to enhance the flexibility with which system processes can be represented, and to thereby investigate the important role that the system architectural hypothesis (discretization of the system into subsystem elements) plays in determining model behavior. We illustrate and explore these concepts with synthetic and real‐data studies, using models constructed from simple generic buckets as building blocks, thereby paving the way for more‐detailed investigations using sophisticated process‐based hydrological models. We also discuss how proposed MR‐PPEs can bridge the gap between current process‐based modeling and machine learning. Finally, we suggest the need for model calibration to evolve from a search over "parameter spaces" to a search over "function spaces." Plain Language Summary: Modelers make many decisions in their quest to formulate a working model. It is important to evaluate the impact of each modeling decision, and to assess the extent to which different decisions improve the representation of the actual system. Building upon past work, we present a framework that enables an improved assessment of individual modeling decisions. Specifically, we suggest that modelers should pay more attention to the hierarchical structure of model building decisions, and to the impact that each such decision can have on the fidelity of resulting process representation. Key Points: We present a strategy for characterizing and quantifying the information added at each model building stepModel building steps are interdependent in a hierarchical mannerWe call for the focus of model calibration to shift from "parameter spaces" to "function spaces" [ABSTRACT FROM AUTHOR]

Accurate forest growing stock (GSV) mapping is crucial for informed land management decisions, climate change mitigation, and sustainable resource utilization. This study aimed to digital mapping and uncertainty assessment of GSV levels in the Bashkiriya Nature Reserve, Russia. A random forest approach was used to predict GSV at a spatial resolution of 10 m (10 ×10 m) using 8395 plots and a set of 36 environmental covariates representing remote sensing data, relief and climate variables. The results revealed that the GSV levels ranged from 1 to 439 m3/ha, with a median of 200 m3/ha. Our model was evaluated using cross-validation and three performance metrics: root mean square error (RMSE=76 m3/ha), coefficient of determination (R2 =0.55) and Nash-Sutcliffe efficiency coefficient (NSE=0.55). In addition, we assessed the uncertainty of our predictions using a 90% prediction interval. According to the assessment of covariate importance, the significance of explanatory variables decreased in the following order: remote sensing > terrain > climate. We found, that Sentinel-2A satellite data, along with spectral indices, played a pivotal role in our spatial prediction model. Among these variables, the SWIR 12 band emerged as a standout contributor, highlighting the importance of this spectral information in GSV mapping. According to the generated map, the highest GSV values were identified primarily in regions characterized by low-lying terrain with mature trees of the reserve, while the lowest levels were distinctly evident on treeless mountain peaks. While it is widely acknowledged that remote sensing constitutes a primary source for modelling forest properties, our research has underscored the significance of the SWIR spectrum and elevation in future forest investigations. An advantage of the produced map of GSV and their associated uncertainties is that the end users can incorporate these uncertainties into decision-making processes involved in forest and environmental planning. • Random forest (RF) was applied to model forest growing volume (GSV). • A set of 36 covariates and 8395 ground-based measurements were incorporated. • We conducted an uncertainty analysis of GSV predictions. • Sentinel-2A SWIR band 12 was identified as a key variable in the RF model. [ABSTRACT FROM AUTHOR]

• The effect of sampling densities on accuracy of prediction model was explored. • Geostatistical models without environmental variables have higher prediction accuracy in plain areas when the sampling size is large. • Machine learning models had an advantage of prediction accuracy when sampling densities were below 5.31 samples/103 km2. • Quantitatively identified the driving factors of the spatial distribution of soil sand content in eastern China. Accurate prediction of the spatial distribution of soil sand content is a pre-requisite for land use management, soil quality evaluation and erosion control, as it determines the transport and movement of soil water, fertilizer, air and heat. Digital soil mapping (DSM) is extensively employed for predicting soil properties. However, practical research is required to address the challenge of selecting an optimal prediction model that is both cost-effective and accurate at a specific sampling density. In this study, topsoil samples were collected from 2,848 sampling points in the eastern plains of China (107,200 km2). The performance of different prediction models for mapping soil sand content was compared at 12 levels of sampling density. Moreover, the geographical detector, a statistical method used to assess the spatial stratified heterogeneity of variables, was adopted to determine the major drivers of spatial variation in soil sand content. The results indicated that climate factors are the major drivers of the spatial variability in soil sand content. For the 100% sample size (26.57 samples/103 km2), the geostatistical models that did not depend on environmental variables (ordinary kriging, sequential Gaussian simulation) performed best, followed by the machine learning models (random forest, cubist and support vector machine) and the geostatistical model with environmental variables (co-kriging). Sampling density had a considerable impact on model accuracy, and the advantages of machine learning models became apparent when sampling densities were below 20% (5.31 samples/103 km2). Therefore, the best combination of prediction model and sampling density should be selected to obtain maps of soil sand content economically and accurately. This study provides a valuable reference for the selection of prediction methods in the practical application of DSM. [ABSTRACT FROM AUTHOR]

Although estimating the uncertainty of models used for modelling nitrate contamination of groundwater is essential in groundwater management, it has been generally ignored. This issue motivates this research to explore the predictive uncertainty of machine-learning (ML) models in this field of study using two different residuals uncertainty methods: quantile regression (QR) and uncertainty estimation based on local errors and clustering (UNEEC). Prediction-interval coverage probability (PICP), the most important of the statistical measures of uncertainty, was used to evaluate uncertainty. Additionally, three state-of-the-art ML models including support vector machine (SVM), random forest (RF), and k -nearest neighbor (k NN) were selected to spatially model groundwater nitrate concentrations. The models were calibrated with nitrate concentrations from 80 wells (70% of the data) and then validated with nitrate concentrations from 34 wells (30% of the data). Both uncertainty and predictive performance criteria should be considered when comparing and selecting the best model. Results highlight that the k NN model is the best model because not only did it have the lowest uncertainty based on the PICP statistic in both the QR (0.94) and the UNEEC (in all clusters, 0.85–0.91) methods, but it also had predictive performance statistics (RMSE = 10.63, R2 = 0.71) that were relatively similar to RF (RMSE = 10.41, R2 = 0.72) and higher than SVM (RMSE = 13.28, R2 = 0.58). Determining the uncertainty of ML models used for spatially modelling groundwater-nitrate pollution enables managers to achieve better risk-based decision making and consequently increases the reliability and credibility of groundwater-nitrate predictions. Unlabelled Image • Predictive uncertainty of models was estimated using the QR and UNEEC methods. • Random Forest model had the lower uncertainty band width based on the both methods. • Groundwater nitrate (NO 3) concentrations were predicted using RF, SVM, and K NN. • Random Forest model outperformed other models in terms of predictive performance. • Hydraulic conductivity and elevation had the highest contribution to the modelling. [ABSTRACT FROM AUTHOR]

Considering the nature of soils distribution, an important modeling issue in soil class mapping is imbalanced class observations. Imbalanced number of data in observed soil classes in an area can result in the underestimation or loss of minority classes and an overestimation of the majority classes in predictive modeling. The effect of this phenomenon is that an area of land with comparatively fewer soil profile observations could be unmapped in the digital maps. To address this problem, this paper investigated the usefulness of data pretreatment techniques called over- and under-sampling of data applied on three predictive models including decision trees (DT), random forest (RF), and multinomial logistic regression (MNLR). The study area is situated in the northwest of Iran with 452 profiles observations on a regular grid covering about 12,000 ha. This area has 8 USDA soil great groups with an imbalanced frequency distribution. Results showed that modeling using imbalanced distribution of class observation caused uncertain maps with minority classes being lost and relatively poor accuracies. After data treatment, with over- and under-sampling, all models showed significant improvement in maintaining the minority classes, in both calibration and validation evaluations. Balancing the classes led to a notable decrease in uncertainty of all 3 models by decreasing the confusion index and raising the probability of occurrence for the soil classes in the final maps. Comparing the 3 models, decision trees showed the largest calibration and validation accuracies with and without data treatment. RF has an issue of overestimation of some of the majority classes. Data resampling technique can be a useful solution for dealing with imbalanced class observations to produce more certain digital soil maps. • Imbalanced distribution of soil classes resulted in losing some classes in the map. • Imbalanced classes were treated by data resampling techniques before modeling. • Data resampling helped in maintaining minority classes and decreasing uncertainty. • Decision tree model performed well in response to balancing the classes. [ABSTRACT FROM AUTHOR]

Recently, the Global Map of Salt-affected Soils (GSSmap) was launched, which pursued a country-driven approach and aimed to update the global and country-level information on salt-affected soils (SAS). The aim of this paper was to present how Hungary contributed to GSSmap by preparing its own SAS maps using advanced digital soil mapping techniques. We used not just a combination of random forest and multivariate geostatistical techniques for predicting the spatial distribution of SAS indicators (i.e., pH, electrical conductivity and exchangeable sodium percentage) for the topsoil (0–30 cm) and subsoil (30–100 cm), but also a number of indices derived from Sentinel-2 satellite images as environmental covariates. The importance plots of random forests showed that in addition to climatic, geomorphometric parameters and legacy soil information, image indices were the most important covariates. The performance of spatial modelling was checked by 10-fold cross validation showing that the accuracy of the SAS maps was acceptable. By this study and by the resulting maps of it, we not just contributed to GSSmap, but also renewed the SAS mapping methodology in Hungary, where we paid special attention to modelling and quantifying the prediction uncertainty that had not been quantified or even taken into consideration earlier.

Zinc (Zn) is a vital element required by all living creatures for optimal health and ecosystem functioning. Therefore, several researchers have modeled and mapped its occurrence and distribution in soils. Nonetheless, leveraging model predictive performances while coupling information derived from visible near-infrared (Vis-NIR) and soils (i.e. chemical properties) to estimate potential toxic elements (PTEs) like Zn in agricultural soils is largely untapped. This study applies two methods to rapidly monitor Zn concentration in agricultural soil. Firstly, employing Vis-NIR and machine learning algorithms (MLAs) (Context 1) and secondly, applying Vis-NIR, soil chemical properties (SCP), and MLAs (Context 2). For the Vis-NIR information, single and combined pretreatment methods were applied. The following MLAs were used: conditional inference forest (CIF), partial least squares regression (PLSR), M5 tree model (M5), extreme gradient boosting (EGB), and support vector machine regression (SVMR) respectively. For context 1, the results indicated that M5-MSC (M5 tree model-multiplicative scatter correction) with coefficient of determination (R2) = 0.72, root mean square error (RMSE) = 21.08 (mg/kg), median absolute error (MdAE) = 13.69 and ratio of performance to interquartile range (RPIQ) = 1.63 was promising. Regarding context 2, CIF with spectral pretreatment and soil properties [CIF-DWTLOGMSC + SCP (conditional inference forest-discrete wavelet transformation-logarithmic transformation-multiplicative scatter correction-soil chemical properties)] yielded the best performance of R2 = 0.86, RMSE = 14.52 (mg/kg), MdAE = 6.25 and RPIQ = 1.78. Altogether, for contexts 1 and 2, the CIF-DWTLOGMSC + SCP approach (context 2) was the best Zn model outcome for the agricultural soil. The uncertainty map revealed a low to high error distribution in context 1, and a low to moderate distribution in context 2 for all models except CIF, which had some patches with high uncertainty. We conclude that a multiple optimization approach for modeling Zn levels in agricultural soils is invaluable and may provide fast and reliable information needed for area-specific decision-making. • Compared to the other algorithms, conditional Inference Forest performed better. • Prediction accuracy can be increased by recognizing the most effective pre-treatment methods. • Combining several pre-treatments is preferable to employing just one treatment algorithm. • Combination of auxiliary datasets reduces the level of uncertainty. [ABSTRACT FROM AUTHOR]

Coastal catchments in British Columbia, Canada, experience a complex mixture of rainfall- and snowmelt-driven contributions to flood events. Few operational flood-forecast models are available in the region. Here, we integrated a number of proven technologies in a novel way to produce a super-ensemble forecast system for the Englishman River, a flood-prone stream on Vancouver Island. This three-day-ahead modeling system utilizes up to 42 numerical weather prediction model outputs from the North American Ensemble Forecast System, combined with six artificial neural network-based streamflow models representing various slightly different system conceptualizations, all of which were trained exclusively on historical high-flow data. As such, the system combines relatively low model development times and costs with the generation of fully probabilistic forecasts reflecting uncertainty in the simulation of both atmospheric and terrestrial hydrologic dynamics. Results from operational testing by British Columbia's flood forecasting agency during the 2013-2014 storm season suggest that the prediction system is operationally useful and robust. [ABSTRACT FROM AUTHOR]

• Use of geostatistics is indispensable when spatial aggregation with quantified uncertainty is targeted. • Spatial aggregation decreases uncertainty and supports detection of statistically significant SOC stock change. • Many counties in Hungary had a statistically significant topsoil SOC stock increase between 1992 and 2010. • The total topsoil SOC stock increase in Hungary between 1992 and 2010 was predicted as 14.79 Tg C. Many national and international initiatives rely on spatially explicit information on soil organic carbon (SOC) stock change at multiple scales to support policies aiming at land degradation neutrality and climate change mitigation. In this study, we used regression cokriging with random forest and spatial stochastic cosimulation to predict the SOC stock change between two years (i.e. 1992 and 2010) in Hungary at multiple aggregation levels (i.e. point support, 1 × 1 km, 10 × 10 km square blocks, Hungarian counties and entire Hungary). We also quantified the uncertainty associated with these predictions in order to identify and delimit areas with statistically significant SOC stock change. Our study highlighted that prediction of spatial totals and averages with quantified uncertainty requires a geostatistical approach and cannot be solved by machine learning alone, because it does not account for spatial correlation in prediction errors. The total topsoil SOC stock for Hungary was predicted to increase between 1992 and 2010 with 14.9 Tg, with lower and upper limits of a 90% prediction interval equal to 11.2 Tg and 18.2 Tg, respectively. Results also showed that both the predictions and uncertainties of the average SOC stock change were smaller for larger spatial supports, while spatial aggregation also made it easier to obtain statistically significant SOC stock changes. The latter is important for carbon accounting studies that need to prove in Measurement, Reporting and Verification protocols that observed SOC stock changes are not only practically but also statistically significant. [ABSTRACT FROM AUTHOR]

Recently, the Global Map of Salt-affected Soils (GSSmap) was launched, which pursued a country-driven approach and aimed to update the global and country-level information on salt-affected soils (SAS). The aim of this paper was to present how Hungary contributed to GSSmap by preparing its own SAS maps using advanced digital soil mapping techniques. We used not just a combination of random forest and multivariate geostatistical techniques for predicting the spatial distribution of SAS indicators (i.e., pH, electrical conductivity and exchangeable sodium percentage) for the topsoil (0–30 cm) and subsoil (30–100 cm), but also a number of indices derived from Sentinel-2 satellite images as environmental covariates. The importance plots of random forests showed that in addition to climatic, geomorphometric parameters and legacy soil information, image indices were the most important covariates. The performance of spatial modelling was checked by 10-fold cross validation showing that the accuracy of the SAS maps was acceptable. By this study and by the resulting maps of it, we not just contributed to GSSmap, but also renewed the SAS mapping methodology in Hungary, where we paid special attention to modelling and quantifying the prediction uncertainty that had not been quantified or even taken into consideration earlier. [ABSTRACT FROM AUTHOR]