Deep learning-based national scale soil organic carbon mapping with Sentinel-3 data

High-precision maps of soil organic carbon (SOC) are beneficial for managing soil fertility and understanding the global carbon cycle. Digital soil mapping plays an important role in efficiently obtaining the spatial distribution of SOC, which contributes to precision agriculture. However, traditional soil-forming factors (i.e., terrain or climatic factors) have weak variability in low-relief areas, such as plains, and cannot reflect the spatial variation of soil attributes. Meanwhile, vegetation cover hinders the acquisition of the direct information of farmland soil. Thus, useful environmental variables should be utilized for SOC prediction and the digital mapping of such areas. SOC has an important effect on crop growth status, and remote sensing data can record the apparent spectral characteristics of crops. The normalized difference vegetation index (NDVI) is an important index reflecting crop growth and biomass. This study used NDVI time series data rather than traditional soil-forming factors to map SOC. Honghu City, located in the middle of the Jianghan Plain, was selected as the study region, and the NDVI time series data extracted from Landsat 8 were used as the auxiliary variables. SOC maps were estimated through stepwise linear regression (SLR), partial least squares regression (PLSR), support vector machine (SVM), and artificial neural network (ANN). Ordinary kriging (OK) was used as the reference model, while root mean square error of prediction (RMSEP) and coefficient of determination of prediction (R2P) were used to evaluate the model performance. Results showed that SOC had a significant positive correlation in July and August (0.17, 0.29) and a significant negative correlation in January, April, and December (−0.23, −0.27, and −0.23) with NDVI time series data. The best model for SOC prediction was generated by ANN, with the lowest RMSEP of 3.718 and highest R2P of 0.391, followed by SVM (RMSEP = 3.753, R2P = 0.361) and PLSR (RMSEP = 4.087, R2P = 0.283). The SLR model was the worst model, with the lowest R2P of 0.281 and highest RMSEP of 3.930. ANN and SVM were better than OK (RMSEP = 3.727, R2P = 0.372), whereas PLSR and SLR were worse than OK. Moreover, the prediction results using single-data NDVI or short time series NDVI showed low accuracy. The effect of the terrain factor on SOC prediction represented unsatisfactory results. All these results indicated that the NDVI time series data can be used for SOC mapping in plain areas and that the ANN model can maximally extract additional associated information between NDVI time series data and SOC. This study presented an effective method to overcome the selection of auxiliary variables for digital soil mapping in plain areas when the soil was covered with vegetation. This finding indicated that the time series characteristics of NDVI were conducive for predicting SOC in plains.

High-resolution digital mapping of soil organic carbon and soil total nitrogen using DEM derivatives, Sentinel-1 and Sentinel-2 data based on machine learning algorithms

Soil organic carbon (SOC) dataset augmentation, which enables the comprehensive monitoring of carbon sinks at regional and global scales, is vital for global carbon cycle management and soil fertility. SOC maps built by conventional laboratory or field measurements are time- and cost-consuming and especially difficult in forests. A new approach to build SOC maps with good accuracy and time efficiency and promptly respond to changes in SOC dynamics is, therefore, being identified. This study aimed to evaluate the ability of SOC estimation using a multiple linear regression model (MLR) and four machine learning algorithms: artificial neural networks (ANN), support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGBoost) with satellite data sources and soil nutrient indicator data to find the optimal method. The results indicate that the SVM and XGBoost models demonstrated the best predictive abilities (R2 = 0.70 and 0.74, %RMSE = 8.8 and 8.3, MAE = 0.176 and 0.155) when using remote sensing variables and soil property variables, respectively. Band7_IDM, Band 5, Band4_IDM, RVI, NSMI, NDVI, Band 6, Band 7, and Band 4 were the most valuable variables in the SVM model, while NDVI, DVI, GVMI, NSMI, and Band4_Dive in the XGBoost model. The SVM model using remote sensing data may be applied to build SOC maps in Vietnamese forests instead of using conventional methods with a high accuracy (R2 = 0.74).

Soil organic carbon (SOC) is an important indicator of soil quality and directly determines soil fertility. Hence, understanding its spatial distribution and controlling factors is necessary for efficient and sustainable soil nutrient management. In this study, machine learning algorithms including artificial neural network (ANN), support vector machine (SVM), cubist regression, random forests (RF), and multiple linear regression (MLR) were chosen for advancing the prediction of SOC. A total of sixty (n = 60) soil samples were collected within the research area at 30 cm soil depth and measured for SOC content using the Walkley–Black method. From these samples, 80% were used for model training and 21 auxiliary data were included as predictors. The predictors include effective cation exchange capacity (ECEC), base saturation (BS), calcium to magnesium ratio (Ca_Mg), potassium to magnesium ratio (K_Mg), potassium to calcium ratio (K_Ca), elevation, plan curvature, total catchment area, channel network base level, topographic wetness index, clay index, iron index, normalized difference build-up index (NDBI), ratio vegetation index (RVI), soil adjusted vegetation index (SAVI), normalized difference vegetation index (NDVI), normalized difference moisture index (NDMI) and land surface temperature (LST). Mean absolute error (MAE), root-mean-square error (RMSE) and R2 were used to determine the model performance. The result showed the mean SOC to be 1.62% with a coefficient of variation (CV) of 47%. The best performing model was RF (R2 = 0.68) followed by the cubist model (R2 = 0.51), SVM (R2 = 0.36), ANN (R2 = 0.36) and MLR (R2 = 0.17). The soil nutrient indicators, topographic wetness index and total catchment area were considered an indicator for spatial prediction of SOC in flat homogenous topography. Future studies should include other auxiliary predictors (e.g., soil physical and chemical properties, and lithological data) as well as cover a broader range of soil types to improve model performance.

Since multi-omics data (high-dimensional data, including many variables, like genomics, transcriptomics and proteomics) are high-dimensional (have many variables), heterogeneous (are data of different types), and interdependent (variables can affect each other), there is a significant challenge in integrating various data types in a synergistic manner (without, it may not be feasible). This paper is an attempt to tackle the problem of predicting drug sensitivity by offering a hybrid machine learning (ML, algorithms that learn patterns by analyzing data) and deep learning (DL, a type of ML based on neural network architectures to learn patterns of data) model of not only predicting drug sensitivity, but also using Genomics of Drug Sensitivity in Cancer (GDSC) data. The method involves applying sophisticated preprocessing, strict feature selection techniques (Random Forest, tree-based algorithm; LASSO regression, feature selection that chooses features by pruning irrelevant ones; Mutual Information which is a dependency measure between variables) as well as a hybrid model design favoring classical ML models (Random Forest, SVM--Support Vector Machine) over the deep learning style (LSTM-Long short-term memory networks, a sequence-based version of recurrent neural network). The metrics of the evaluation give a very high R 2 (coefficient of determination, a measure of model fit) of 0.991, and the evaluations are very small MAE ( Mean absolute error: square root of average squared error), RMSE (Root mean squared error: square root of average squared error), and MSE ( Mean squared error: square root of average squared error), resulting in approximately 12 percent higher accuracy than conventional approximations: Ridge Regression, XGBoost (Extreme gradient boosting), and Random Forest alone. This also shows the strength (capacity to work with diverse circumstances), generalizability (capacity to be used in new information), and enhanced predictive capabilities of the hybrid method of modeling nonlinear associations in multi-omics information, thus enhancing precision medicine (health care customized to patients in terms of genetic, environmental, and lifestyle determinants) uses.

Accurate and cost-effective mapping of soil organic carbon (SOC) is critical for understanding carbon dynamics and informing sustainable land management. Although deep learning-based methods have demonstrated strong potential in digital soil mapping, they typically require large amounts of data. However, the availability of site-level SOC observations is often limited, which poses a challenge for model performance. To address this, we propose a novel transfer learning approach based on a Convolutional Neural Network (CNN) model that does not rely on exogenous data. Specifically, when predicting SOC for a given soil layer, the model is first pre-trained on data from all layers and then fine-tuned using data from the target layer. This design enables more efficient use of limited site data. Experimental results show that the proposed transfer model consistently outperforms other machine learning models, including the Random Forest (RF), standard CNN, and Multi-Task CNN (MTCNN) models. The transfer model achieves a coefficient of determination (R2) of 0.374 and a root mean square error (RMSE) of 2.937%, indicating superior performance. These findings highlight the effectiveness of the proposed approach for digital soil mapping under data-scarce conditions and underscore its potential as a robust tool for accurate SOC estimation.

A novel intelligence approach based active and ensemble learning for agricultural soil organic carbon prediction using multispectral and SAR data fusion

Estimation and monitoring of soil organic carbon (SOC) stocks is important for maintaining soil productivity and meeting climate change mitigation targets. Current global SOC maps do not provide enough detail for landscape-scale decision making, and do not allow for tracking carbon sequestration or loss over time. Using an optical satellite-driven machine learning workflow, we mapped SOC stocks (topsoil; 0 to 30 cm) under natural vegetation (86% of land area) over South Africa at 30 m spatial resolution between 1984 and 2019. We estimate a total topsoil SOC stock of 5.6 Pg C with a median SOC density of 6 kg C m−2 (IQR: interquartile range 2.9 kg C m−2). Over 35 years, predicted SOC underwent a net increase of 0.3% (relative to long-term mean) with the greatest net increases (1.7%) and decreases (−0.6%) occurring in the Grassland and Nama Karoo biomes, respectively. At the landscape scale, SOC changes of up to 25% were evident in some locations, as evidenced from fence-line contrasts, and were likely due to local management effects (e.g. woody encroachment associated with increased SOC and overgrazing associated with decreased SOC). Our SOC mapping approach exhibited lower uncertainty (R2 = 0.64; RMSE = 2.5 kg C m−2) and less bias compared to previous low-resolution (250–1000 m) national SOC mapping efforts (average R2 = 0.24; RMSE = 3.7 kg C m−2). Our trend map remains an estimate, pending repeated measures of soil samples in the same location (time-series); a global priority for tracking SOC changes. While high resolution SOC maps can inform land management decisions aimed at climate mitigation (natural climate solutions), potential increases in SOC are likely limited by local climate and soils. It is also important that climate mitigation efforts such as planting trees balance trade-offs between carbon, biodiversity and overall ecosystem function.

본 연구는 Deep Neural Network(DNN)을 이용하여 광주-기아 챔피언스 필드의 일일 관중 수를 예측함으로써 이를 통해 구단과 관련기업의 마케팅 자료제공 및 구장 내 부대시설의 재고관리에 자료로 쓰임을 목적으로 수행 되었다. 본 연구 에서는 Artificial Neural Network(ANN)의 종류인 DNN 모델을 이용하였으며 DNN 모델의 과적합을 막기 위해 Dropout 과 Batch normalization 적용한 모델을 바탕으로 총 4종류를 설계하였다. 각각 10개의 DNN을 만들어 예측값의 Root Mean Square Error(RMSE)와 Mean Absolute Percentage Error(MAPE)의 평균값을 낸 모델과 예측값의 평균으로 RMSE와 MAPE를 평가한 Ensemble 모델을 만들었다. 모델의 학습 데이터는 2008년부터 2017년까지의 관중 수 데이터를 수집하여 수집된 데이터의 80%를 무작위로 선정하였으며, 나머지 20%는 테스트 데이터로 사용하였다. 총 100회의 데이터 선정, 모델구성 그리고 학습 및 예측을 한 결과 Ensemble 모델은 DNN 모델의 예측력이 가장 우수하게 나왔으며, 다중선 형회귀 모델 대비 RMSE는 15.17%, MAPE는 14.34% 높은 예측력을 보이고 있다.In this paper, we used the Deep Neural Network (DNN) to predict the number of daily spectators of Gwangju -KIA Champions Field in order to provide marketing data for the team and related businesses and for managing the inventories of the facilities in the stadium. In this study, the DNN model, which is based on an artificial neural network (ANN), was used, and four kinds of DNN model were designed along with dropout and batch normalization model to prevent overfitting. Each of four models consists of 10 DNNs, and we added extra models with ensemble model. Each model was evaluated by Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). The learning data from the model randomly selected 80% of the collected data from 2008 to 2017, and the other 20% were used as test data. With the result of 100 data selection, model configuration, and learning and prediction, we concluded that the predictive power of the DNN model with ensemble model is the best, and RMSE and MAPE are 15.17% and 14.34% higher, correspondingly, than the prediction value of the multiple linear regression model.

High resolution 3D mapping of soil organic carbon in a heterogeneous agricultural landscape

Adoption of a good estimation model for the prediction of sub soils properties before the commencement of a construction project, or at the preliminary stage of project planning is highly imperative. This will mitigate the most unexpected costs incurred during construction which are mostly geotechnical in nature. This research aims to use Machine Learning ML tools such as Multiple Linear Regression (MLR) Artificial Neural Network(ANN),Support Vector Machine(SVM), Random Forest(RF) andM5 Tree (M5P) in geotechnical Engineering with a view to correlate Optimum Moisture Content(OMC), Maximum Dry Density(MDD) and Soaked California Bearing Ratio(SCBR) and Unsoaked California Bearing Ratio (USCBR) from the measured index properties. The results from index properties classified the soils of the study area as A-2-4, A-2-6, A-2-7 and A-7-5 for Ekiti Central Senatorial Districts (ECSD) and A-2-4, A-2-5, A-2-6, A-2-7, A-4, A-5, A-6 and A-7-5 for Ekiti South Senatorial Districts ( ESSD) while Ekiti Northern Senatorial Districts (ENSD) were classified as A-2-4, A-2-5, A-2-6, A-2-7, A-6 and A-7-6. Conversely. The strengths of the developed Machine Learning models have been examined in terms of regression coefficient (R 2 ) and Root Mean Square Error (RMSE) values. It is found that all the five ML models predict OMC %, MDD, SCBR and USCBR close to the experimental value. However, the prediction of OMC %, MDD, SCBR and USCBR by RF is found better than other ML models deployed in this research. DOI: 10.7176/CER/14-3-04 Publication date: May 31 st 2022

Visible, near-infrared, and shortwave-infrared spectra as an input variable for digital mapping of soil organic carbon

Soil organic carbon (SOC) concentration is the fundamental indicator of soil health, underpinning food production and climate change mitigation. SOC storage is highly sensitive to several dynamic environmental drivers, with approximately one third of soils degraded and losing carbon worldwide. Digital soil mapping illuminates where hotspots of SOC storage occur and where losses to the atmosphere are most likely. Yet, attempts to map SOC often disagree. Here we compare national scale SOC concentration map products to reveal agreement of data in mineral soils, with progressively poorer agreement in organo-mineral and organic soils. Divergences in map predictions from each other and survey data widen in the high SOC content land types we stratified. Given the disparities are highest in carbon rich soils, efforts are required to reduce these uncertainties to increase confidence in mapping SOC storage and predicting where change may be important at national to global scales. Our map comparison results could be used to identify SOC risk where concentrations are high and should be conserved, and where uncertainty is high and further monitoring should be targeted. Reducing inter-map uncertainty will rely on addressing statistical limitations and including covariates that capture convergence of physical factors that produce high SOC contents.

Large topsoil organic carbon variability is controlled by Andisol properties and effectively assessed by VNIR spectroscopy in a coffee agroforestry system of Costa Rica

Soil Organic Carbon (SOC) influences many soil properties including nutrient and water holding capacity, nutrient cycling and stability, improved water infiltration and aeration. It also is an essential parameter in the assessment of soil quality, especially for agricultural production. However, SOC mapping is a complicated process that is costly and time-consuming due to the physical challenges of the natural conditions that is being surveyed. The best model for SOC mapping is still in debate among many researchers. Recently, the development of machine learning and Geographical Information Systems (GIS) has provided the potential for more accurate spatial prediction of SOC content. This research was conducted in a relatively small-scale capacity in the Central Vietnam region. The aim of this study is to compare the accuracy of Inverse Distance Weighting (IDW), Ordinary Kriging (OK), and Random Forest (RF) methods for SOC interpolation, with a dataset of 47 soil samples for an area of 145 hectares. Three environmental variables including elevation, slope, and the Normalized Difference Vegetation Index (NDVI) were used for the RF model. In the RF model, the values of the number of variables randomly sampled as candidates at each split, (mtry), and the number of bootstrap replicates, (ntree), were determined in terms of 1 and 1,000 respectively The results at our research site showed that using IDW is the most accurate method for SOC mapping, followed by the methods of RF and OK respectively. Concerning SOC mapping based-on auxiliary variables, in areas where there is human activity, the selection of auxiliary variables should be carefully considered because the variation in the SOC may not only be due to environmental variables but also by farming technologies.