Spatial and temporal patterns of above- and below- ground biomass over the Tibet Plateau grasslands and their sensitivity to climate change

The sensitivity of grassland above- (AGB, gC m-2) and below-ground biomass (BGB, gC m-2) to climate has been shown to be significant on the Tibetan Plateau, however, the spatial patterns and sensitivity of biomass with altitudinal change needs to be quantitated. In this study, large data sets of AGB and BGB during the peak growth season, and the corresponding geographical and climate conditions in the grasslands of the Tibetan Plateau between 2001 and 2020 were analyzed, and modelled using a Cubist regression trees algorithm. The mean values for AGB and BGB were 61.3 and 1304.3 gC m-2, respectively, for the whole region over the two decades. There was a significant change in spatial AGB of 64.8 % on the Plateau (P < 0.05, with areas where AGB increased being twice as large as areas where AGB decreased), while BGB did not change significantly in majority the of the region (≥ 90.1 %, P > 0.05). In general, the areas where AGB showed positive partial correlations with precipitation were larger than the areas where AGB had positive correlations with temperature (P < 0.05). However, these trends varied depending on the climatic conditions: in the wetter regions, temperature had a greater effect on the size of the areas with positive AGB responses than precipitation (P < 0.05), while precipitation had a greater effect on the size of areas with positive BGB changes than temperature (P < 0.05). In the drier areas, however, precipitation affected the AGB response significantly compared to temperature (P < 0.05), while temperature influenced the BGB response greater than precipitation (P < 0.05). The response and sensitivity of grassland biomass to temperature and precipitation varied according to the altitude of the Plateau: the response and sensitivity were stronger and more sensitive at medium altitudes, and weak at the higher or lower altitudes. Likely, this phenomenon was resulted from the natural selection of plants to maintain the efficient use of resources during un-favourable and stressed conditions for maximum plant development and growth. These findings will help assess the ecological consequences of global climate change for the grasslands of the Tibetan Plateau, particularly in those regions with highly variable altitudes.

An integrated feature selection approach to high water stress yield prediction

The timely and precise prediction of winter wheat yield plays a critical role in understanding food supply dynamics and ensuring global food security. In recent years, the application of unmanned aerial remote sensing has significantly advanced agricultural yield prediction research. This has led to the emergence of numerous vegetation indices that are sensitive to yield variations. However, not all of these vegetation indices are universally suitable for predicting yields across different environments and crop types. Consequently, the process of feature selection for vegetation index sets becomes essential to enhance the performance of yield prediction models. This study aims to develop an integrated feature selection method known as PCRF-RFE, with a focus on vegetation index feature selection. Initially, building upon prior research, we acquired multispectral images during the flowering and grain filling stages and identified 35 yield-sensitive multispectral indices. We then applied the Pearson correlation coefficient (PC) and random forest importance (RF) methods to select relevant features for the vegetation index set. Feature filtering thresholds were set at 0.53 and 1.9 for the respective methods. The union set of features selected by both methods was used for recursive feature elimination (RFE), ultimately yielding the optimal subset of features for constructing Cubist and Recurrent Neural Network (RNN) yield prediction models. The results of this study demonstrate that the Cubist model, constructed using the optimal subset of features obtained through the integrated feature selection method (PCRF-RFE), consistently outperformed the RNN model. It exhibited the highest accuracy during both the flowering and grain filling stages, surpassing models constructed using all features or subsets derived from a single feature selection method. This confirms the efficacy of the PCRF-RFE method and offers valuable insights and references for future research in the realms of feature selection and yield prediction studies.

Digital mapping and spatial modeling of some soil physical and mechanical properties in a semi-arid region of Iran

The soil's physical and mechanical (SPM) properties have significant impacts on soil processes, such as water flow, nutrient movement, aeration, microbial activity, erosion, and root growth. To digitally map some SPM properties at four global standard depths, three machine learning algorithms (MLA), namely, random forest, Cubist, and k-nearest neighbor, were employed. A total of 200-point observation was designed with the aim of a field survey across the Marvdasht Plain in Fars Province, Iran. After sampling from topsoil (0 to 30 cm) and subsoil depths (30 to 60 cm), the samples were transferred to the laboratory to determine the mean weight diameter (MWD) and geometric mean diameter (GMD) of aggregates in the laboratory. In addition, shear strength (SS) and penetration resistance (PR) were measured directly during the field survey. In parallel, 79 environmental factors were prepared from topographic and remote sensing data. Four soil variables were also included in the modeling process, as they were co-located with SPM properties based on expert opinion. For selecting the most influential covariates, the variance inflation factor (VIF) and Boruta methods were employed. Two covariate dataset scenarios were used to assess the impact of soil and environmental factors on the modeling of SPM properties including SPM and environmental covariates (scenario 1) and SPM, environmental covariates, and soil variables (scenario 2). From all covariates, nine soil and environmental factors were selected for modeling the SPM properties, of which four of them were the soil variables, three were related to remote sensing, and two factors had topographic sources. The results indicated that scenario 2 outperformed in all standard depths. The findings suggested that clay and SOM are key factors in predicting SPM, highlighting the importance of considering soil variables in addition to environmental covariates for enhancing the accuracy of machine learning prediction. The k-nearest neighbor algorithm was found to be highly effective in predicting SPM, while the random forest algorithm yielded the highest R2 value (0.92) for penetration resistance properties at 15-30 depth. Overall, the approach used in this research has the potential to be extended beyond the Marvdasht Plain of Fars Province, Iran, as well as to other regions worldwide with comparable soil-forming factors. Moreover, this study provides a valuable framework for the digital mapping of SPM properties, serving as a guide for future studies seeking to predict SPM properties. Globally, the output of this research has important significance for soil management and conservation efforts and can facilitate the development of sustainable agricultural practices.

Assessment of macronutrients status using digital soil mapping techniques: a case study in Maru'ak area in Lorestan Province, Iran

The present study was conducted to compare generalized linear model (GLM), random forest (RF), and Cubist to produce available phosphorus (AP) and potassium (AK) maps and to identify the covariates that control mineral distribution in Lorestan Province, Iran. To this end, the locations for collecting 173 soil samples were determined through the conditioned Latin hypercube sampling (cLHS) method, at four different land-uses (orchards, paddy fields, agricultural, and abandoned fields). The performance of the models was assessed by coefficient of determination (R²), root-mean-square error (RMSE), and mean absolute error (MAE) indices. The results showed that the RF model fitted better than GLM and Cubist models and could explain 40 and 57% of AP and AK distribution, respectively. The R², RMSE, and MAE of the RF model were 0.4, 2.81, and 2.43 for predicting AP and equal to 0.57, 143.77, and 116.61 for predicting AK, respectively. The most important predictors selected by the RF model were valley depth and soil-adjusted vegetation index (SAVI) for AP and AK, respectively. The maps showed higher AP and AK content in apricot orchards compared to other land-uses. No difference was observed between AP and AK content on paddy fields, agricultural, and abandoned areas. The higher AP and AK contents were related to orchard management practices, such as failure to dispose of plant residuals and fertilizer consumption. It can be concluded that the orchards (by increasing soil quality) was the best land-use in line with sustainable management for the study area. However, generalizing the results needs more detailed research.

Machine Learning for In Silico ADMET Prediction

ADMET (absorption, distribution, metabolism, excretion, and toxicity) describes a drug molecule's pharmacokinetics and pharmacodynamics properties. ADMET profile of a bioactive compound can impact its efficacy and safety. Moreover, efficacy and safety are considered some of the major causes of clinical attrition in the development of new chemical entities. In past decades, various machine learning or quantitative structure-activity relationship (QSAR) methods have been successfully integrated in the modeling of ADMET. Recent advances have been made in the collection of data and the development of various in silico methods to assess and predict ADMET of bioactive compounds in the early stages of drug discovery and development process.

Estimation of anthropogenic heat emissions in China using Cubist with points-of-interest and multisource remote sensing data

Rapid urbanization and industrialization in China stimulated the great increase of energy consumption, which leads to drastic rise in the emission of anthropogenic waste heat. Anthropogenic heat emission (AHE) is a crucial component of urban energy budget and has direct implications for investigating urban climate and environment. However, reliable and accurate representation of AHE across China is still lacking. This study presented a new machine learning-based top-down approach to generate a gridded anthropogenic heat flux (AHF) benchmark dataset at 1 km spatial resolution for China in 2010. Cubist models were constructed by fusing points-of-interest (POI) data of varying categories and multisource remote sensing data to explore the nonlinear relationships between various geographic predictors and AHE from different heat sources. The strategy of developing specific models for different components and exploiting the complementary features of POIs and remote sensing data generated a more reasonable distribution of AHF. Results showed that the AHF values in urban centers of metropolises over China range from 60 to 190 W m^-2. The highest AHF values were observed in some heavy industrial zones with value up to 415 W m^-2. Compared with previous studies, the spatial distribution of AHF from different heating components was effectively distinguished, which highlights the potential of POI data in improving the precision of AHF mapping. The gridded AHF dataset can serve as input of urban numerical models and can help decision makers in targeting extreme heat sources and polluters in cities and making differentiated and tailored strategies for emission mitigation.

Improved digital soil mapping with multitemporal remotely sensed satellite data fusion: A case study in Iran

Modeling and mapping of soil properties are critical in many environmental, climatic, ecological and hydrological applications. Digital soil mapping (DSM) techniques are now commonly applied to predict soil properties with limited data by developing predictive relationships with environmental covariates. Most studies derive covariates from a digital elevation model (named static covariates). Many works also include single-day remotely sensed satellite imagery. However, multitemporal satellite images can capture information about soil properties over time and bring additional information in predicting soil properties in DSM. We refer to covariates derived from multitemporal satellite images as dynamic covariates. The objective of this study was to assess the performance of DSM when using terrain derivatives (static covariates), single-date remotely sensed satellite indices (limited dynamic covariates), multitemporal satellite indices (dynamic covariates), and combinations of terrain derivatives and satellite indices (covariate fusion) as covariates in predicting soil properties and estimating uncertainty. Three soil properties are considered in this study: organic carbon (OC), sand content, and calcium carbonate equivalent (CCE). Inclusion of single and/or multitemporal remotely sensed satellite indices improved the prediction of soil properties over traditionally used terrain indices. Significant improvements were observed in the prediction of soil properties using two models, Cubist and random forest (RF). The increase in the R² values for Cubist and RF were 126% and 78% for OC, 110% and 54% for sand, and 87% and 32% for CCE. The RMSE decreased by 34% and 27% for OC, 25% and 12% for sand, and 39% and 19% for CCE, when compared to the terrain indices only model. This also reduced the uncertainty of estimation and mapping. These clearly showed the advantage of using multitemporal satellite data fusion rather than simply using static terrain indices for DSM of soil properties to deliver a great potential in improving soil modeling and mapping for many applications.

Machine learning-based detection of soil salinity in an arid desert region, Northwest China: A comparison between Landsat-8 OLI and Sentinel-2 MSI

Accurate assessment of soil salinization is considered as one of the most important steps in combating global climate change, especially in arid and semi-arid regions. Multi-spectral remote sensing (RS) data including Landsat series provides the potential for frequent surveys for soil salinization at various scales and resolutions. Additionally, the recently launched Sentinel-2 satellite constellation has temporal revisiting frequency of 5 days, which has been proven to be an ideal approach to assess soil salinity. Yet, studies on detailed comparison in soil salinity tracking between Landsat-8 OLI and Sentinel-2 MSI remain limited. For this purpose, we collected a total of 64 topsoil samples in an arid desert region, the Ebinur Lake Wetland National Nature Reserve (ELWNNR) to compare the monitoring accuracy between Landsat-8 OLI and Sentinel-2 MSI. In this study, the Cubist model was trained using RS-derived covariates (spectral bands, Tasseled Cap transformation-derived wetness (TCW), and satellite salinity indices) and laboratory measured electrical conductivity of 1:5 soil:water extract (EC). The results showed that the measured soil salinity had a significant correlation with surface soil moisture (Pearson's r = 0.75). The introduction of TCW generated satisfactory estimating performance. Compared with OLI dataset, the combination of MSI dataset and Cubist model yielded overall better model performance and accuracy measures (R² = 0.912, RMSE = 6.462 dS m^-1, NRMSE = 9.226%, RPD = 3.400 and RPIQ = 6.824, respectively). The differences between Landsat-8 OLI and Sentinel-2 MSI were distinguishable. In conclusion, MSI image with finer spatial resolution performed better than OLI. Combining RS data sets and their derived TCW within a Cubist framework yielded accurate regional salinity map. The increased temporal revisiting frequency and spectral resolution of MSI data are expected to be positive enhancements to the acquisition of high-quality soil salinity information of desert soils.

Improved Estimates of Population Exposure in Low-Elevation Coastal Zones of China

With sea level predicted to rise and the frequency and intensity of coastal flooding expected to increase due to climate change, high-resolution gridded population datasets have been extensively used to estimate the size of vulnerable populations in low-elevation coastal zones (LECZ). China is the most populous country, and populations in its LECZ grew rapidly due to urbanization and remarkable economic growth in coastal areas. In assessing the potential impacts of coastal hazards, the spatial distribution of population exposure in China’s LECZ should be examined. In this study, we propose a combination of multisource remote sensing images, point-of-interest data, and machine learning methods to improve the performance of population disaggregation in coastal China. The resulting population grid map of coastal China for the reference year 2010, with a spatial resolution of 100 × 100 m, is presented and validated. Then, we analyze the distribution of population in LECZ by overlaying the new gridded population data and LECZ footprints. Results showed that the total population exposed in China’s LECZ in 2010 was 158.2 million (random forest prediction) and 160.6 million (Cubist prediction), which account for 12.17% and 12.36% of the national population, respectively. This study also showed the considerable potential in combining geospatial big data for high-resolution population estimation.

Prediction of Left Ventricular Mechanics Using Machine Learning

The goal of this paper was to provide a real-time left ventricular (LV) mechanics simulator using machine learning (ML). Finite element (FE) simulations were conducted for the LV with different material properties to obtain a training set. A hyperelastic fiber-reinforced material model was used to describe the passive behavior of the myocardium during diastole. The active behavior of the heart resulting from myofiber contractions was added to the passive tissue during systole. The active and passive properties govern the LV constitutive equation. These mechanical properties were altered using optimal Latin hypercube design of experiments to obtain training FE models with varied active properties (volume and pressure predictions) and varied passive properties (stress predictions). For prediction of LV pressures, we used eXtreme Gradient Boosting (XGboost) and Cubist, and XGBoost was used for predictions of LV pressures, volumes as well as LV stresses. The LV pressure and volume results obtained from ML were similar to FE computations. The ML results could capture the shape of LV pressure as well as LV pressure-volume loops. The results predicted by Cubist were smoother than those from XGBoost. The mean absolute errors were as follows: XGBoost volume: 1.734 ± 0.584 ml, XGBoost pressure: 1.544 ± 0.298 mmHg, Cubist volume: 1.495 ± 0.260 ml, Cubist pressure: 1.623 ± 0.191 mmHg, myofiber stress: 0.334 ± 0.228 kPa, cross myofiber stress: 0.075 ± 0.024 kPa, and shear stress: 0.050 ± 0.032 kPa. The simulation results show ML can predict LV mechanics much faster than the FE method. The ML model can be used as a tool to predict LV behavior. Training of our ML model based on a large group of subjects can improve its predictability for real world applications.

An extensive experimental survey of regression methods

Regression is a very relevant problem in machine learning, with many different available approaches. The current work presents a comparison of a large collection composed by 77 popular regression models which belong to 19 families: linear and generalized linear models, generalized additive models, least squares, projection methods, LASSO and ridge regression, Bayesian models, Gaussian processes, quantile regression, nearest neighbors, regression trees and rules, random forests, bagging and boosting, neural networks, deep learning and support vector regression. These methods are evaluated using all the regression datasets of the UCI machine learning repository (83 datasets), with some exceptions due to technical reasons. The experimental work identifies several outstanding regression models: the M5 rule-based model with corrections based on nearest neighbors (cubist), the gradient boosted machine (gbm), the boosting ensemble of regression trees (bstTree) and the M5 regression tree. Cubist achieves the best squared correlation ( R²) in 15.7% of datasets being very near to it, with difference below 0.2 for 89.1% of datasets, and the median of these differences over the dataset collection is very low (0.0192), compared e.g. to the classical linear regression (0.150). However, cubist is slow and fails in several large datasets, while other similar regression models as M5 never fail and its difference to the best R² is below 0.2 for 92.8% of datasets. Other well-performing regression models are the committee of neural networks (avNNet), extremely randomized regression trees (extraTrees, which achieves the best R² in 33.7% of datasets), random forest (rf) and ε-support vector regression (svr), but they are slower and fail in several datasets. The fastest regression model is least angle regression lars, which is 70 and 2,115 times faster than M5 and cubist, respectively. The model which requires least memory is non-negative least squares (nnls), about 2 GB, similarly to cubist, while M5 requires about 8 GB. For 97.6% of datasets there is a regression model among the 10 bests which is very near (difference below 0.1) to the best R², which increases to 100% allowing differences of 0.2. Therefore, provided that our dataset and model collection are representative enough, the main conclusion of this study is that, for a new regression problem, some model in our top-10 should achieve R² near to the best attainable for that problem.

Environmental and anthropogenic influences on ambient background concentrations of fluoride in soil

Excess exposure to fluoride causes substantive health burden in humans and livestock globally. However, few studies have assessed the distribution and controls of variability of ambient background concentrations of fluoride in soil. Ambient background concentrations of fluoride in soil were collated for Greater Melbourne, Greater Geelong, Ballarat and Mitchell in Victoria, Australia (n = 1005). Correlation analysis and machine learning techniques were used to identify environmental and anthropogenic influences of fluoride variability in soil. Sub-soils (>0.3 m deep), in some areas overlying siltstone and sandstone, and to a lesser extent, overlying basalt, were naturally enriched with fluoride at concentrations above ecological thresholds for grazing animals. Soil fluoride enrichment was predominantly influenced by parent material (mineralogy), precipitation (illuviation), leaching during palaeoclimates and marine inputs. Industrial air pollution did not significantly influence ambient background concentrations of fluoride at a regional scale. However, agricultural practices (potentially the use of phosphate fertilisers) were indicated to have resulted in added fluoride to surface soils overlying sediments. Geospatial variables alone were not sufficient to accurately model ambient background soil fluoride concentrations. A multiple regression model based on soil chemistry and parent material was shown to accurately predict ambient background fluoride concentrations in soils and support assessment of fluoride enrichment in the environment.

Combining ancillary soil data with VisNIR spectra to improve predictions of organic and inorganic carbon content of soils

While traditional laboratory methods of determining soil organic carbon (SOC) content are generally simple, this becomes more challenging when carbonates are present in the soil; such is commonly found in semi-arid areas. Additionally, soil inorganic carbon (SIC) content itself is difficult to determine. This study uses visible near infrared (VisNIR) spectra to predict SOC and SIC contents of samples, and the impact of including soil pH and soil total carbon (STC) data as predictor variables was evaluated. The results indicated that combining available soil pH and STC content data with VisNIR spectra dramatically improved prediction accuracy of the Cubist models. Using the full suite of predictor variables, Cubist models trained on the calibration dataset (75%) could predict the validation dataset (25%) for SOC content with a Lin’s concordance correlation coefficient (LCCC) of 0.94, and an LCCC of 0.83 for SIC content. This is compared to an LCCC of 0.81 and 0.35 for SOC and SIC content, respectively, when no ancillary soil data was included with VisNIR spectra as predictor variables. These results suggest that there may be promise for using other readily available soil data in combination with VisNIR spectra to improve the predictions of different soil properties.

•

It can be laborious and expensive to measure soil organic and inorganic carbon content with traditional laboratory methods, and there has been recent focus on using spectroscopic techniques to overcome this.

•

This study demonstrates that combining ancillary soil data (pH and total carbon content) with these spectroscopic techniques can considerably improve predictions of SOC and SIC content.

Mapping monthly rainfall erosivity in Europe

Rainfall erosivity as a dynamic factor of soil loss by water erosion is modelled intra-annually for the first time at European scale. The development of Rainfall Erosivity Database at European Scale (REDES) and its 2015 update with the extension to monthly component allowed to develop monthly and seasonal R-factor maps and assess rainfall erosivity both spatially and temporally. During winter months, significant rainfall erosivity is present only in part of the Mediterranean countries. A sudden increase of erosivity occurs in major part of European Union (except Mediterranean basin, western part of Britain and Ireland) in May and the highest values are registered during summer months. Starting from September, R-factor has a decreasing trend. The mean rainfall erosivity in summer is almost 4 times higher (315MJmmha^-1h^-1) compared to winter (87MJmmha^-1h^-1). The Cubist model has been selected among various statistical models to perform the spatial interpolation due to its excellent performance, ability to model non-linearity and interpretability. The monthly prediction is an order more difficult than the annual one as it is limited by the number of covariates and, for consistency, the sum of all months has to be close to annual erosivity. The performance of the Cubist models proved to be generally high, resulting in R² values between 0.40 and 0.64 in cross-validation. The obtained months show an increasing trend of erosivity occurring from winter to summer starting from western to Eastern Europe. The maps also show a clear delineation of areas with different erosivity seasonal patterns, whose spatial outline was evidenced by cluster analysis. The monthly erosivity maps can be used to develop composite indicators that map both intra-annual variability and concentration of erosive events. Consequently, spatio-temporal mapping of rainfall erosivity permits to identify the months and the areas with highest risk of soil loss where conservation measures should be applied in different seasons of the year.

Effect of roll compaction on granule size distribution of microcrystalline cellulose-mannitol mixtures: computational intelligence modeling and parametric analysis

Dry granulation using roll compaction is a typical unit operation for producing solid dosage forms in the pharmaceutical industry. Dry granulation is commonly used if the powder mixture is sensitive to heat and moisture and has poor flow properties. The output of roll compaction is compacted ribbons that exhibit different properties based on the adjusted process parameters. These ribbons are then milled into granules and finally compressed into tablets. The properties of the ribbons directly affect the granule size distribution (GSD) and the quality of final products; thus, it is imperative to study the effect of roll compaction process parameters on GSD. The understanding of how the roll compactor process parameters and material properties interact with each other will allow accurate control of the process, leading to the implementation of quality by design practices. Computational intelligence (CI) methods have a great potential for being used within the scope of quality by design approach. The main objective of this study was to show how the computational intelligence techniques can be useful to predict the GSD by using different process conditions of roll compaction and material properties. Different techniques such as multiple linear regression, artificial neural networks, random forest, Cubist and k-nearest neighbors algorithm assisted by sevenfold cross-validation were used to present generalized models for the prediction of GSD based on roll compaction process setting and material properties. The normalized root-mean-squared error and the coefficient of determination (R²) were used for model assessment. The best fit was obtained by Cubist model (normalized root-mean-squared error =3.22%, R²=0.95). Based on the results, it was confirmed that the material properties (true density) followed by compaction force have the most significant effect on GSD.