Search

This work proposes an approach for establishing coffee management classes mainly supported by pedological information (soil survey) and land parcels, taking into account peculiarities of Brazilian coffee crops (land parcels already implemented with different crop ages, cultivars and density) and inspired by some management zone concepts. Two initial datasets were used based on soil survey and/or coffee crop management information. Eight sequences of tests were developed, involving: ranking of the most important variables for coffee yield modeling by random forest, reduction of data dimensionality through principal component analysis (PCA) or factorial analysis of mixed data (FAMD), generation of clusters with the hierarchical cluster on principal component (HCPC), applying hierarchical tree by using Ward's minimum variance method and improved by k-means classification. Cluster effectiveness was assessed by statistical difference in coffee yield. A total of 3 clusters were considered the most proper number of management classes, composed by the most accurate random forest model (crop age, crop density, silt fraction and soil organic matter content ranked as most important variables) and highest % of variables explanation by PCA. Although not well explored for such a purpose, HCPC applied in this study case was effective on generating homogeneous management classes, differing statistically from each other by means of coffee yield. [ABSTRACT FROM AUTHOR]

Soil organic carbon (SOC) stocks are a remarkable property for soil and environmental monitoring. The understanding of their dynamics in crop soils must go forward. The objective of this study was to determine the impact of temporal environmental controlling factors obtained by satellite images over the SOC stocks along soil depth, using machine learning algorithms. The work was carried out in São Paulo state (Brazil) in an area of 2577 km2. We obtained a dataset of boreholes with soil analyses from topsoil to subsoil (0–100 cm). Additionally, remote sensing covariates (30 years of land use history, vegetation indexes), soil properties (i.e., clay, sand, mineralogy), soil types (classification), geology, climate and relief information were used. All covariates were confronted with SOC stocks contents, to identify their impact. Afterwards, the abilities of the predictive models were tested by splitting soil samples into two random groups (70 for training and 30% for model testing). We observed that the mean values of SOC stocks decreased by increasing the depth in all land use and land cover (LULC) historical classes. The results indicated that the random forest with recursive features elimination (RFE) was an accurate technique for predicting SOC stocks and finding controlling factors. We also found that the soil properties (especially clay and CEC), terrain attributes, geology, bioclimatic parameters and land use history were the most critical factors in controlling the SOC stocks in all LULC history and soil depths. We concluded that random forest coupled with RFE could be a functional approach to detect, map and monitor SOC stocks using environmental and remote sensing data. [ABSTRACT FROM AUTHOR]

This study aimed to i) explore the land-use effect (former lakebed, pasture, and agricultural areas) on elemental concentrations, enrichment factors (EF s), and geo-accumulation indices (Igeo s), ii) provide X-ray diffractograms of the primary minerals and identify the clay minerals dominant in each land-use scenario, iii) map the spatial distribution of elemental concentration, EF s , and Igeo s using Digital Soil Mapping (DSM) approaches. We collected a total of 240 soil samples. Our findings highlight that vermiculite was a common clay-sized mineral, and aragonite, gypsum, and calcite were dominant primary minerals in each land-use. The remote sensing data (Sentinel-2 MSI and Landsat-8 OLI) were used to calculate some environmental covariates useful for digital mapping of elemental concentrations, EF s , and Igeo s. The highest concordance correlation coefficient (CCC) was found in the prediction of Hg (CCC = 0.79) using remote sensing data based on Sentinel-2 with the Random forest (RF) model. There is neither threat in terms of Cu, Zn, and Pb nor the impact of anthropogenic activities in their distribution under each land-use scenario. The study area was enriched with Hg, Cd, and Ni (very high, EF = 20–40), followed by Cr (significant, EF = 5.88) and As (moderate, EF = 2.08). A similar trend was also found for Igeo s across the study area. We conclude that the distribution of elemental concentration, EF s , and Igeo s were significantly affected by the shrinking of the Burdur Lake and the different land-use scenarios in this semi-arid region. • XRD, pXRF, SEM, RS, and DSM were simultaneously used. • Former lakebeds, pasture, and agricultural areas were the land-use scenarios for investigation. • Land-use changes and agricultural activities impacted EF s and Igeo s. • Sentinel-2 MSI data delivered better prediction than Landsat-8 OLI data. • Random forest outperformed cubists in predictions. [ABSTRACT FROM AUTHOR]

Knowledge about parent material (PM) is crucial to understand the properties of overlying soils. Assessing PM of very deep soils, however, is not easy. Previous studies have predicted PM via proximal sensors and machine learning algorithms, but within small areas and with low soil variety. This study evaluates the efficiency of using portable X-ray fluorescence (pXRF) spectrometry together with machine learning algorithms in a large area populated with a wide variety of soil classes and land uses to build accurate PM distribution maps from soil data. Samples from A and B horizons were collected from 117 sites, totaling 234 samples, along with representative PM samples of the predominant PMs in the region. Elemental contents of PM and soil samples were obtained using pXRF. Random Forest (RF), Support Vector Machine (SVM) and Linear Discriminant Analysis (LDA) models built with and without Principal Component Analysis (PCA) were used to predict PMs from A and B horizon samples separately. For validation, PM was identified in 23 different spots and compared with the predicted PM via overall accuracy and Kappa coefficient. Maps built with models excluding PCA had an overall accuracy ranging from 0.87 to 0.96 and kappa coefficient ranging from 0.74 to 0.91. Maps generated with PCA based models reached 100% overall accuracy. Prediction of PM using pXRF with samples from either A or B horizon and machine learning algorithms offers solid results even when applied in large areas with high land use and soil class variability. • Soil parent material can be accurately predicted via pXRF analysis of soils. • Maps from B horizon samples were slightly superior to A horizon samples. • Overall map predictive accuracy was strong. [ABSTRACT FROM AUTHOR]

Proximal sensors have proven capable of predicting multiple soil properties under different conditions. However, doubts remain about which sensor is preferable for delivering optimal prediction models and which preprocessing methods produce the most accurate results. Portable X-ray fluorescence (pXRF) spectrometry and visible near-infrared (Vis-NIR) diffuse reflectance spectroscopy have been widely used, while the NixPro TM color sensor has been explored more recently. This study evaluated the use of pXRF, Vis-NIR, and NixPro TM data to predict soil organic matter content (SOM), pH, base saturation (BS), the sum of bases (SB), cation exchange capacity (CEC) at pH = 7 and effective CEC (eCEC), via each sensor in isolation, and via combined sensors data. Moreover, factors interfering in the prediction models' accuracy (data preprocessing methods, soil horizon, soil class, parent material) were used as auxiliary variables. 604 soil samples were collected in Brazil, encompassing ten soil orders and 19 parent materials. Numerical and categorical prediction models (7,980) were created for six soil properties using a random forest algorithm, totaling 7980 models, delivering almost 24,000 results., Coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), and residual prediction deviation (RPD) were used for validation of numerical predictions., Overall accuracy and kappa coefficient were calcualted for categorical predictions. Although the combination of sensors provided most of the best predictions, pXRF in isolation achieved accuracies close to the three sensors combined. NixPro TM offered superior contributions to SOM and CEC predictions, but pXRF and Vis-NIR were responsible for the best results of most studied variables. On average, by adding pXRF to Vis-NIR data, predictive accuracy improved 32%; while adding Vis-NIR to pXRF data increased accuracy by ca. 6%. Soil-order-specific models improved predictions for Ultisols compared to general models (without soil order distinction), reaching R2 > 0.90. Soil parent material and horizon did not improve models significantly. Categorical predictions improved the accuracy for some properties, reaching an overall accuracy of 100% and kappa index of 1.0 for pH in A horizons of Ultisols via pXRF + Vis-NIR data. Proximal sensor data with no auxiliary variables provided almost all the best results. The fusion of proximal sensors can provide better predictions, but pXRF alone can deliver satisfactory results in most cases for the six soil properties. • Soil pH, SOM, BS, SB, eCEC, CEC can be predicted by pXRF sensor alone. • PXRF provided results similar to the best ones provided by both Vis-NIR and NixPro TM . • Soil horizon, parent material, or class did not significantly improve predictions. • PXRF alone delivers satisfactory results for all the six soil properties evaluated. [ABSTRACT FROM AUTHOR]

• Munsell color was accurately predicted by Nix Pro color sensor. • Mean color difference (ΔE* ab) achieved 2.9, indicating the Nix Pro accuracy. • Results achieved overall accuracy and Kappa index of 0.93. • Using a dirty, old Munsell color chart reduced the accuracy of predictions. • Soil color determined by human eye was compared to Nix Pro color predictions. Soil color has historically drawn humans' attention, although its definition is somewhat subjective. It is correlated with several soil attributes and it allows for inferences about several soil aspects. New proximal sensors, such as the Nix™ Pro color sensor, can determine soil color values, but its correlation with the widely used Munsell soil color chart (MSCC) has yet to be investigated. This work aimed to train machine learning models using the Random Forest (RF) algorithm to predict each notation of MSCC chips from data extracted by the Nix™ Pro sensor, test the model's accuracy by evaluating whether it can identify MSCC chips using a brand-new and a dirty MSCC, and compare model predictions with soil color classifications made by the human eye. Additionally, MSCC data obtained via Nix™ was compared to Munsell renotation data to assess the color detection accuracy of the sensor. Prediction models were calibrated by scanning every MSCC chip (437 in total) in triplicate. All validation samples were excluded from model calibration. Accuracy of the predictions of MSCC notation reached overall accuracy and Kappa index values of 0.93 for the brand-new MSCC and of 0.70 for the dirty MSCC. Soil color classification by human eye had little agreement with the predicted MSCC notation, as expected due to the variable conditions affecting soil color conventional determination in the field. Color difference was calculated by the Euclidian distance (ΔE* ab) between three color stimuli in the CIELAB color space. The mean ΔE* ab between Nix™-provided data and renotation data was 2.9, demonstrating high color detection accuracy. The Nix™ Pro color sensor allows for assessment of accurate color data. When applied together with machine learning algorithms, Nix™ Pro provides a reliable determination of soil color classification equivalent to MSCC in an easily reproducible, rapid, inexpensive and non-subjective way. [ABSTRACT FROM AUTHOR]