K-nearest Neighbor Regression Research Articles

Advancing Iron Ore Grade Estimation: A Comparative Study of Machine Learning and Ordinary Kriging

Mineral grade estimation is a vital phase in mine planning and design, as well as in the mining project’s economic assessment. In mining, commonly accepted methods of ore grade estimation include geometrical approaches and geostatistical techniques such as kriging, which effectively capture the spatial grade variation within a deposit. The application of machine-learning (ML) techniques has been explored in the estimation of mineral resources, where complex correlations need to be captured. In this paper, the authors developed four machine-learning regression models, i.e., support vector regression (SVR), random forest regression (RFR), k-nearest neighbour (KNN) regression, and extreme gradient boost (XGBoost) regression, using a geological database to predict the grade in an Indian iron ore deposit. When compared with ordinary kriging (R2 = 0.74; RMSE = 2.09), the RFR (R2 = 0.74; RMSE = 2.06), XGBoost (R2 = 0.73; RMSE = 2.12), and KNN (R2 = 0.73; RMSE = 2.11) regression models produced similar results. The block model predictions generated using the RFR, XGBoost, and KNN models show comparable accuracy and spatial trends to those of ordinary kriging, whereas SVR was identified as less effective. When integrated with geological methods, these models demonstrate significant potential for enhancing and optimizing mine planning and design processes in similar iron ore deposits.

Insights into the specific capacitance of CNT-based supercapacitor electrodes using artificial intelligence.

In this study, the specific capacitance characteristics of a carbon nanotube (CNT) supercapacitor was predicted using different machine learning algorithms, such as artificial neural network (ANN), random forest regression (RFR), k-nearest neighbors regression (KNN), and decision tree regression (DTR), based on experimental studies. The results of the simulation verified the accuracy of the ANN algorithm with respect to the data derived from the specific capacitance of the supercapacitor module. It was observed that there was a strong correlation between the experimental results and the predictions made by the ANN algorithm. Comparative analysis showed that the developed ANN algorithm was consistently superior over other algorithms in terms of different metrics, as indicated by the lowest root mean square error (RMSE) value of roughly 26.24 and the highest R 2 value of approximately 0.91. In contrast, the DTR model recorded the least reliable results in the accuracy analysis, as indicated by the highest RMSE value of about 53.46 and the lowest R 2 value of roughly 0.63. To further explore the impact of independent input parameters including pore structure, specific surface area, and I D/I G ratio on a single output parameter (particularly, the specific capacitance) the sensitivity analysis was also conducted using the SHapley Additive exPlanations (SHAP) framework. This investigation sheds light on the relative significance and effects of different input variables on the specific capacitance of supercapacitors based on CNTs. The results indicated that the ANN algorithm accurately predicted the capacitance of the CNT-based supercapacitor, demonstrating the feasibility and significance of neural network algorithms in the design of energy storage devices.

The instability prediction of non-resonant energetic particle modes based on machine learning algorithms

Abstract Non-resonant energetic particle modes (EPMs) are extremely common in tokamak experimental phenomenon, which can disrupt the plasma balance, subsequently reducing the device confinement performance. Nevertheless, it should be stressed that the hybrid simulation of EPMs requires considerable time and computational resources in the study of such phenomena. To solve this issue, this research proposes the Machine Learning (ML) approaches to predict the linear instability of non-resonant EPMs. Here, the current study compares four single model algorithms in order to select the optimal one for predicting the non-resonant EPMs instability, including the Decision Tree Regression (DT), K-Nearest Neighbor (KNN) regression, Support vector Regression (SVR) and Multilayer perceptron (MLP). By optimizing the hyper-parameters for these models, the final evaluation results reveal quite high accuracy and R2 values, all around 0.9. Moreover, this work examines the influence of missing different feature variables on the model evaluation. Through this comparison, the study analyzes the degree of importance of various factors in forecasting the instability of high-energy particle modes.

Machine learning models for predicting interaction affinity energy between human serum proteins and hemodialysis membrane materials

Membrane incompatibility poses significant health risks, including severe complications and potential fatality. Surface modification of membranes has emerged as a pivotal technology in the membrane industry, aiming to improve the hemocompatibility and performance of dialysis membranes by mitigating undesired membrane-protein interactions, which can lead to fouling and subsequent protein adsorption. Affinity energy, defined as the strength of interaction between membranes and human serum proteins, plays a crucial role in assessing membrane-protein interactions. These interactions may trigger adverse reactions, potentially harmful to patients. Researchers often rely on trial-and-error approaches to enhance membrane hemocompatibility by reducing these interactions. This study focuses on developing machine learning algorithms that accurately and rapidly predict affinity energy between novel chemical structures of membrane materials and human serum proteins, based on a molecular docking dataset. Various membrane materials with distinct characteristics, chemistry, and orientation are considered in conjunction with different proteins. A comparative analysis of linear regression, K-nearest neighbors regression, decision tree regression, random forest regression, XGBoost regression, lasso regression, and support vector regression is conducted to predict affinity energy. The dataset, comprising 916 records for both training and test segments, incorporates 12 parameters extracted from data points and involves six different proteins. Results indicate that random forest (R² = 0.8987, MSE = 0.36, MAE = 0.45) and XGBoost (R² = 0.83, MSE = 0.49, MAE = 0.49) exhibit comparable predictive performance on the training dataset. However, random forest outperforms XGBoost on the testing dataset. Seven machine learning algorithms for predicting affinity energy are analyzed and compared, with random forest demonstrating superior predictive accuracy. The application of machine learning in predicting affinity energy holds significant promise for researchers and professionals in hemodialysis. These models, by enabling early interventions in hemodialysis membranes, could enhance patient safety and optimize the care of hemodialysis patients.

Imputation of missing land carbon sequestration data in the AR6 Scenarios Database

Abstract. The AR6 Scenarios Database is a vital repository of climate change mitigation pathways used in the latest Intergovernmental Panel on Climate Change (IPCC) assessment cycle. In its current version, many scenarios in the database lack information about the level of anthropogenic carbon dioxide (CO2) removal via land sinks, as net-negative CO2 emissions and gross removals on land are not always separated and are not consistently reported across models. This makes scenario analyses focusing on CO2 removal challenging. We test and compare the performance of different regression models to impute missing data on land carbon sequestration for the global level and for several sub-global macro-regions from available data on net CO2 emissions from agriculture, forestry, and other land uses. We find that a k-nearest neighbors regression performs best among the tested regression models and use it to impute and provide two publicly available imputation datasets (https://doi.org/10.5281/zenodo.13373539, Prütz et al., 2024) on CO2 removal via land sinks for incomplete global scenarios (n=404) and incomplete regional R10 scenario variants (n=2358) of the AR6 Scenarios Database. We discuss the limitations of our approach, the use of our datasets for secondary assessments of AR6 scenario ensembles, and how this approach compares to other recent AR6 data reanalyses.

Uncertainty Analysis of Remote Sensing Estimation of Chinese Fir (Cunninghamia lanceolata) Aboveground Biomass in Southern China

Forest aboveground biomass (AGB) is not only the basis for forest carbon stock research, but also an important parameter for assessing the forest carbon cycle and ecological functions of forests. However, there are various uncertainties in the estimation process, limiting the accuracy of AGB estimation. Therefore, we extracted the spectral features, vegetation indices and texture factors from remote sensing images based on the field data and Landsat 8 OLI remote sensing images in Southern China to quantify the uncertainties. Then, we established three AGB estimation models, including K Nearest Neighbor Regression (KNN), Gradient Boosted Regression Tree (GBRT) and Random Forest (RF). Uncertainties at the plot scale and models were measured by using error equations to analyze the influences of uncertainties at different scales on AGB estimation. Results were as follows: (1) The R2 of the per-tree biomass model for Cunninghamia lanceolata was 0.970, while the uncertainty of the residual and parameters for per-tree biomass model was 4.62% and 4.81%, respectively; and the uncertainty transferred to the plot scale was 3.23%. (2) The estimation methods had the most significant effects on the remote sensing models. RF was more accurate than other two methods, and had the highest accuracy (R2 = 0.867, RMSE = 19.325 t/ha) and lowest uncertainty (5.93%), which outperformed both the KNN and GBRT models (KNN: R2 = 0.368, RMSE = 42.314 t/ha, uncertainty = 14.88%; GBRT: R2 = 0.636, RMSE = 32.056 t/ha, uncertainty = 6.3%). Compared to KNN and GBRT, the R2 of RF was enhanced by 0.499 and 0.231, while the uncertainty was decreased by 8.95% and 0.37%, respectively. The uncertainty associated with the scale of remote sensing models remains the primary source of uncertainty when compared to the plot scale. On the remote sensing scale, RF is the model with the best estimation effect. This study examines the impact of both plot-scale and remote sensing model-scale methodologies on the estimation of AGB for Cunninghamia lanceolata. The findings aim to offer valuable insights and considerations for enhancing the accuracy of AGB estimations.

Hyperspectral estimation of chlorophyll density in winter wheat using fractional-order derivative combined with machine learning.

Chlorophyll density (ChD) can reflect the photosynthetic capacity of the winter wheat population, therefore achieving real-time non-destructive monitoring of ChD in winter wheat is of great significance for evaluating the growth status of winter wheat. Derivative preprocessing has a wide range of applications in the hyperspectral monitoring of winter wheat chlorophyll. In order to research the role of fractional-order derivative (FOD) in the hyperspectral monitoring model of ChD, this study based on an irrigation experiment of winter wheat to obtain ChD and canopy hyperspectral reflectance. The original spectral reflectance curves were preprocessed using 3 FOD methods: Grünwald-Letnikov (GL), Riemann-Liouville (RL), and Caputo. Hyperspectral monitoring models for winter wheat ChD were constructed using 8 machine learning algorithms, including partial least squares regression, support vector regression, multi-layer perceptron regression, random forest regression, extra-trees regression (ETsR), decision tree regression, K-nearest neighbors regression, and gaussian process regression, based on the full spectrum band and the band selected by competitive adaptive reweighted sampling (CARS). The main results were as follows: For the 3 types of FOD, GL-FOD was suitable for analyzing the change process of the original spectral curve towards the integer-order derivative spectral curve. RL-FOD was suitable for constructing the hyperspectral monitoring model of winter wheat ChD. Caputo-FOD was not suitable for hyperspectral research due to its insensitivity to changes in order. The 3 FOD calculation methods could all improve the correlation between the original spectral curve and Log(ChD) to varying degrees, but only the GL method and RL method could observe the change process of correlation with order changes, and the shorter the wavelength, the smaller the order, and the higher the correlation. The bands screened by CARS were distributed throughout the entire spectral range, but there was a relatively concentrated distribution in thevisible light region. Among all models, CARS was used to screen bands based on the 0.3-order RL-FOD spectrum, and the model constructed using ETsR reached the best accuracy and stability. Its R2c, RMSEc, R2v, RMSEv, and RPD were 1.0000, 0.0000, 0.8667, 0.1732, and 2.6660, respectively. In conclusion, based on the winter wheat ChD data set and the corresponding canopy hyperspectral data set, combined with 3 FOD calculation methods, 1 band screening method, and 8 modeling algorithms, this study constructed hyperspectral monitoring models for winter wheat ChD. The results showed that based on the 0.3-order RL-FOD, combined with the CARS screening band, ETsR modeling has the highest accuracy, and hyperspectral estimation of winter wheat ChD can be realized. The results of this study can provide some reference for the rapid and nondestructive estimation of ChD in winter wheat.

Machine Learning Models in Predicting Failure Times Data Using a Novel Version of the Maxwell Model

This work aims to introduce a novel statistical distribution based on Maxwell distribution that can handle both positive and negative data sets with varying failure rates, including decreasing and bathtub-shaped distributions. The novel statistical distribution can be derived via the log transformation approach with an additional exponent parameter, defining the Transformed Log Maxwell (TLMax) distribution. The numerical investigation reveals that the developed TLMax distribution can effectively fit negative and positive data sets. A data set containing failure times for Kevlar 49/epoxy at a pressure of approximately 90% was employed to compare the proposed model against the traditional Maxwell model, and the results obtained indicated that the novel distribution outperformed the comparator. Finally, for the prediction of failure times in the dataset, we employed a machine learning model, including support vector regression (SVR), K-nearest neighbors’ regression (KNN), linear regression (LR), and gradient boosting regression (XGBoost). The findings indicate that the KNN model demonstrates greater prediction robustness than the other models. Beyond practitioners and researchers, this research holds relevance for professionals in physics and chemistry, where the Maxwell distribution is commonly employed.

Research on carbonation percentage of carbonated recycled concrete fine aggregate: experimental investigation and machine learning prediction

The carbonation percentage (CP) is an essential carbonation degree evaluation index that determines the physicochemical properties of carbonated recycled concrete fine aggregate (CRFA), and its accurate prediction is necessary. Therefore, this paper first uses six machine learning algorithms to build predictive models, aiming to forecast the CP of CRFA under various carbonation control parameters and analyze their advantages and shortcomings. Six models including k-nearest neighbor regression, back propagation neural network, decision tree (DT), random forest, gradient boosting decision trees, and extreme gradient boosting (XGB) are trained, tested, and validated by employing a comprehensive dataset collected from our 100 mix design samples. As expected, the XGB model shows a fascinating prediction ability superior to other proposed models, owing to the largest coefficient of determination being 0.974 and the lowest root mean squared of 1.521 in the testing set. Moreover, the sensitive analysis of input feature parameters is conducted using Shapley additive explanations (SHAPs) according to the XGB-based prediction model, illustrating C-Con (average SHAP value = 0.72) shows the largest impact on the development of CP. This study promotes effective forecasting and modifies the suitable design of carbonation control parameters, providing a theoretical foundation for utilizing building waste following low-carbon and ecologically friendly development concepts.

A Novel Hybrid Machine Learning Framework for Wind Speed Prediction

The growing urgency of environmental challenges and the depletion of fossil fuels have accelerated the search for sustainable and renewable energy sources. Wind energy, for example, is an important source of green electricity. However, using wind power is challenging due to the variability and unpredictability of wind patterns. Consequently, the ability to predict wind power in advance is crucial. The integration of artificial intelligence within the renewable energy sector could provide a viable solution to this challenge. In this study, we investigate the potential of machine learning to improve wind power forecasting by conducting a comparison of three regression models: K-Nearest Neighbor regression, Random Forest regression, and Support Vector regression. These models are combined with a feature selection technique to forecast wind power. Additionally, we propose a novel hybrid approach that combines these machine learning models with Multiple Linear Regression to address the complexities of wind energy forecasting. The performance of the models is evaluated using the R² score, Mean Absolute Error, and Root Mean Squared Error. The dataset for this study was generated from a numerical simulation conducted at a location with a latitude of 22.55° N and a longitude of -14.33° E. The findings demonstrate that the proposed hybrid model outperforms the individual machine learning models in terms of prediction accuracy. This study provides a solid foundation for future research and development in wind energy forecasting.

Estimating punching performance in fiber-reinforced polymer concrete slabs utilizing machine learning and gradient-boosted regression techniques

ABSTRACT The study explores the perforating shear performance of Fiber-Reinforced Polymer (FRP) concrete blocks using machine learning techniques like Gradient-Boosted Regression Trees (GBRT), k-nearest Neighbours (KNN), and Lasso Regression. It aims to predict the structural integrity of FRP blocks under shear conditions based on experimental data. The models were assessed using Coefficient of Determination (R2), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). GBRT demonstrated superior performance during training with an R2 of 0.9786, RMSE of 52.75, and MAE of 34.12, indicating strong predictive accuracy and minimal error. It outperformed KNN (R2 = 0.92, RMSE = 83.91, MAE = 45.71) and Lasso Regression (R2 = 0.71, RMSE = 162.45, MAE = 115.83). In validation, GBRT again excelled with an R2 of 0.93, RMSE of 76.23, and MAE of 58.46, confirming its robustness in generalizing unseen data. KNN showed lower performance in validation (R2 = 0.86), with increased error values, while Lasso lagged further behind (R2 = 0.681, RMSE = 185.23, MAE = 138.34). GBRT consistently outperformed traditional regression methods, highlighting its potential for more accurate and reliable structural analysis in FRP concrete slabs.

Developing a New Feature for Vulnerability Risk Scoring Model for Enhanced Cybersecurity

As organisations increasingly rely on technology, the risk of cyber threats to data integrity and security grows significantly. Traditional vulnerability risk-scoring models may not adequately address the rapidly evolving nature of cyber threats, necessitating the development of more adaptable and context-specific models. This research aims to achieve two primary objectives: developing a flexible risk-scoring model that can be customised for different industries, companies, or situations, and creating a new feature that accurately reflects the risk score based on the current dataset. The study employs correlation analysis and machine learning-based regression modelling, utilising appropriate evaluation metrics to assess model performance. Results indicate that the K-Nearest Neighbors regression model performs particularly well, offering precise risk assessments. The risk score produced by the model serves as a critical tool for prioritising cybersecurity efforts, where higher scores denote a greater need for immediate action. This research contributes to the field by providing a scalable and customisable framework for developing specialised risk-scoring models, enhancing the effectiveness of cybersecurity strategies across diverse contexts.

Monitoring a Coffee Roasting Process Based on Near‐Infrared and Raman Spectroscopy Coupled With Chemometrics

ABSTRACTRoasting is a fundamental step in coffee processing, where complex reactions form chemical compounds related to the coffee flavor and its health‐beneficial effects. These reactions occur on various time scales depending on the roasting conditions. To monitor the process and ensure reproducibility, the study proposes simple and fast techniques based on spectroscopy. This work uses analytical tools based on near‐infrared (NIR) and Raman spectroscopy to monitor the coffee roasting process by predicting chemical changes in coffee beans during roasting. Green coffee beans of Robusta and Arabica species were roasted at 240°C for different roasting times. The spectra of the samples were taken using the spectrometers and modeled by the k‐nearest neighbor regression (KNR), partial least squares regression (PLSR), and multiple linear regression (MLR) to predict concentrations from the spectral data sets. For NIR spectra, all the models provided satisfactory results for the prediction of chlorogenic acid, trigonelline, and DPPH radical scavenging activity with low relative root mean square error of prediction (pRMSEP < 9.649%) and high coefficient of determination (R2 > 0.915). The predictions for ABTS radical scavenging activity were reasonably good. On the contrary, the models poorly predicted the caffeine and total phenolic content (TPC). Similarly, all the models based on the Raman spectra provided good prediction accuracies for monitoring the dynamics of chlorogenic acid, trigonelline, and DPPH radical scavenging activity (pRMSEP < 7.849% and R2 > 0.944). The results for ABTS radical scavenging activity, caffeine, and TPC were similar to those of NIR spectra. These findings demonstrate the potential of Raman and NIR spectroscopy methods in tracking chemical changes in coffee during roasting. By doing so, it may be possible to control the quality of coffee in terms of its aroma, flavor, and roast level.

Graphene Metasurface Based Biosensor for COVID-19 Detection in the Terahertz Regime with Machine Learning Optimization using K-Nearest Neighbours Regression

Graphene Metasurface Based Biosensor for COVID-19 Detection in the Terahertz Regime with Machine Learning Optimization using K-Nearest Neighbours Regression

Machine learning and regression in the management of runoff in bauxite mines under rehabilitation.

Accurate and reliable forecasting of monthly runoff considering several years of rehabilitation helps in planning and managing the water resources system of bauxite mining areas. A combination of linear regression models and artificial intelligence was proposed and applied to predict runoff from mined areas. The ideal number of input variables was chosen considering a correction threshold (correlation above 95%), stepwise selection, and according to the recursive feature elimination (RFE). The linear regression equations were divided into explanatory and predictive, with the predictive ones being trained and validated with all possible combinations of annual data periods. Five machine learning models (linear model (LM), K-nearest neighbor regression (KNN), support vector machine (SVM), random forest (RF), and Cubist) were trained and used as a prediction instrument on the test set of each subsequent year. Precipitation, accumulated precipitation duration, maximum precipitation intensity, and number of events were the most explanatory variables in most linear regression and machine learning equations. The linear regression model approach presented satisfactory performance indices in predicting surface runoff in a mined bauxite area. The Cubist and RF models were the best predictors for the greatest number of years of rehabilitation, with a coefficient of determination varying with maximum values between 0.52 and 0.88 in validation. There was no best model that consistently presented the best result in all years of rehabilitation, and it is recommended to use the annual model for the respective year of rehabilitation. The linear regression equations and algorithms were useful tools in predicting runoff series and, therefore, promising monthly runoff prediction. The presented approach becomes a starting point in the search to optimize the hydrological planning of surface bauxite mines.

Unveiling predictive factors for household-level stunting in India: A machine learning approach using NFHS-5 and satellite-driven data.

Childhood stunting remains a significant public health issue in India, affecting approximately 35% of children under 5. Despite extensive research, existing prediction models often fail to incorporate diverse data sources and address the complex interplay of socioeconomic, demographic, and environmental factors. This study bridges this gap by employing machine learning methods to predict stunting at the household level, using data from the National Family Health Survey combined with satellite-driven datasets. We used four machine learning models-random forest regression, support vector machine regression, K-nearest neighbors regression, and regularized linear regression-to examine the impact of various factors on stunting. The random forest regression model demonstrated the highest predictive accuracy and robustness. The proportion of households below the poverty line and the dependency ratio consistently predicted stunting across all models, underscoring the importance of economic status and household structure. Moreover, the educational level of the household head and environmental variables such as average temperature and leaf area index were significant contributors. Spatial analysis revealed significant geographic clustering of high-stunting districts, notably in central and eastern India, further emphasizing the role of regional socioeconomic and environmental factors. Notably, environmental variables like average temperature and leaf area index emerged as strong predictors of stunting, highlighting how regional climate and vegetation conditions shape nutritional outcomes. These findings underline the importance of comprehensive interventions that not only address socioeconomic inequities but also consider environmental factors, such as climate and vegetation, to effectively combat childhood stunting in India.

Empirical Likelihood Ratio Based K-Nearest Neighbours Regression

Regression models play a pivotal role in real-life applications by enabling the analysis and prediction of continuous outcomes. Among these, the k-Nearest Neighbours (KNN) model stands out as a significant advancement in machine learning. KNN's ability to make predictions based on the proximity of data points has found wide-ranging applications in various fields. However, the traditional KNN regression model has its limitations, including sensitivity to noise and an uneven distribution of neighbours. In response, this paper introduces a novel approach: an empirical likelihood ratio (ELR) based regression algorithm. The ELR technique offers distinct advantages over distance-based nearest neighbour computations, particularly in handling skewed data distributions and minimizing the impact of outliers. The proposed ELRbased KNN regression model is rigorously assessed through both simulation studies and real-life scenarios. The results explicitly demonstrate the enhanced performance of the ELR-based approach over the conventional KNN model. This research contributes to a deeper understanding of regression techniques and underscores the practical significance of leveraging empirical likelihood ratios in refining predictive models for real-world applications.. KEYWORDS :Regression model, k-Nearest neighbours, Empirical likelihood ratio, Distance measures, Data distributions.

Regression and Ensemble Models for Analyzing Students Key Performance Factors

Predicting student performance can help identify when students need extra support and tailor interventions to improve outcomes. In this study, we use machine learning to explore the factors influencing student success and build the models that can predict. We apply several algorithms , including Linear Regression , “Random Forest Regression” , “ Gradient Boosting Regression”, “ Support Vector Regression”, “ K-Nearest Neighbors Regression” , each chosen for their ability to capture different patterns in data , whether linear or complex . The models are trained on factors such as attendance, socioeconomic status, parental education , and academic history to understand how these impact student achievement and learning . Our findings show that ensemble techniques like Gradient Boosting and Random Forest tend to provide more accurate predictions than traditional regression models. By comparing these models, we aim to offer insights into how machine learning can help predict and support student performance, helping educators make more informed decisions about learning. This also helps personalized decision-making to support learners effectively.

Comparative Analysis of Machine Learning Algorithms: KNN, SVM, Decision Tree and Logistic Regression for Efficiency and Performance

This paper presents a comparative study of machine learning models namely K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Decision Tree, and Logistic Regression on two datasets: 20Newsgroups and Wine datasets. Based on basic accuracy, and precision, and recall of the models, F1 measures were assessed. When compared with other classifiers such as KNN and Decisions Trees, SVM and Logistic regression gave better results especially in the case of the 20Newsgroups dataset dominated by textual high dimensional data. KNN had poor recall results and Decision Tree was moderate. In the Wine dataset, since the structure of data is comparatively less complex in our context then all the models yielded almost similar results with accuracy and precision factors very close to 1.0 which of course manifested that the choice of the model does not affect much on simple data. These results stress the importance of the choice of the appropriate model for tasks with a certain level of data complexity; detailed models show the greatest efficiency at the accomplishment of complicated tasks, though at the same time, they are not required for simple, structured data.

Optimizing hybrid models for canopy nitrogen mapping from Sentinel-2 in Google Earth Engine

Canopy nitrogen content (CNC) is a crucial variable for plant health, influencing photosynthesis and growth. An optimized, scalable approach for spatially explicit CNC quantification using Sentinel-2 (S2) data is presented, integrating PROSAIL-PRO simulations with Gaussian Process Regression (GPR) and an Active Learning technique, specifically the Euclidean distance-based diversity (EBD) approach for selective sampling. This hybrid method enhances training dataset efficiency and optimizes CNC models for practical applications. Two GPR models based on PROSAIL-PRO variables were evaluated: a protein-based model (Cprot-LAI) and a chlorophyll-based model (Cab-LAI). Both models, implemented in Google Earth Engine (GEE), demonstrated strong performance and outperformed other machine learning methods, including kernel ridge regression, principal component regression, neural network, weighted k-nearest neighbors regression, partial least squares regression and least squares linear regression. Validation results showed moderate to good accuracies: NRMSECprot−LAI = 16.76%, RCprot−LAI2 = 0.47; NRMSECab−LAI = 18.74%, RCab−LAI2 = 0.51. The models revealed high consistency for an independent validation dataset of the Munich-North-Isar (Germany) test site, with R2 values of 0.58 and 0.71 and NRMSEs of 21.47% and 20.17% for the Cprot-LAI model and Cab-LAI model, respectively. The models also demonstrated high consistency across growing seasons, indicating their potential for time series analysis of CNC dynamics. Application of the S2-based mapping workflow across the Iberian Peninsula, with estimates showing relative uncertainty below 30%, highlights the model’s broad applicability and portability. The optimized EBD-GPR-CNC approach within GEE supports scalable CNC estimation and offers a robust tool for monitoring nitrogen dynamics.