Majid Danesh

Dear Corresponding Author, I am grateful to you for the valuable research. Kindly consider and implement the following points: 1. What is the innovation and originality of the research? Please highlight it in the context. 2. What environment was c...

Phosphorus in surface waters accelerate algal growth and eutrophication, considerably influencing water quality. Spatiotemporal changes in phosphorus concentration are crucial for environmental issues. We aimed to study the temporal and spatial changes in water quality in a river and in a drainage water system considering different land uses. To this aim, 15 water samples were collected from the origin of the river to the estuary, in the Bostankar River watershed (N-Iran), during spring and winter. Further samples were collected from agricultural drainage water in rice fields, tea, flower, orange as well as kiwi gardens, and forests during spring and winter. EC, pH, TDS, and three forms of phosphorus (total, particulate, and soluble) were measured in the water samples. The results showed that water quality changes in agricultural drainage water were time-dependent; the average total phosphorus was 0.4 mg l-1 lower in the spring than in the winter. The highest phosphorus concentration (1.29 mg l-1) occurred in the winter in the drainage water of the orange gardens. Temporal and spatial changes of the river showed that water quality reduced from the river upstream (jungles and grasslands) towards the downstream (different agricultural land uses), and the amount of phosphorus increased from 0.25 to 0.5 mg l^− 1. The TDS increased from 60 to 220 mg l^− 1 in the river in the winter. Finally, the results showed that human activities were the main factor in river water quality reduction due to agricultural activities.

Textural components of soil play an essential role in erodibility and should be considered in many projects of conservation and environmental modeling processes. Traditional methods of determining soil texture are usually laborious, expensive and time consuming along with destructive effects on the environment. Meanwhile, spectroscopic technology using the spectral features and signatures from the whole reflected spectra of soil surface promises a competent method to study soil constituents. To investigate this issue, 113 points were selected and sampled randomly from 0-15 cm of soil surface in eastern parts of Mazandaran Province, Iran. Samples were haphazardly divided into 91 for model building and 22 for final verification and accuracy assessment processes. Applying the enhanced PLS-algorithm plus the FLOOCV approach along with spectral transformations and pre-processing, the modeling of each textural components were accomplished. Spectrally, sand and clay fractions were modeled with high accuracy as: R2c= 0.89, RMSEc= 7.42, SEc= 7.46 for sand and R2c= 0.82, RMSEc= 6.88, SEc= 6.92 for the clay content. Whereas, the silt predictive model was slightly weaker than the other constituents. The most effective spectral ranges involved in the modeling process, were also detected and recognized based on beta & spectral weight analyses and Marten’s uncertainty test. Additionally, the most influential spectroscopic ranges included were the visible, NIR and SWIR regions with the specified wavelengths. In general, the efficacy of spectroscopic technology in soil texture studies has been proven by this research. Using the computed spectral models, we are able to study the soil textural components at large scales faster, safer, timelier and also cheaper. That is absolutely true and applicable using the regionalized remotely sensed data but requires further investigation in different geographical regions.

The advent of lab diffuse reflectance spectroscopy (LDRS) that exploits the fundamental vibration, overtones and combination of functional groups of soil components makes the soil study easier. The present research intends to predict sand content utilizing the proximal soil sensing (PSS) tech. Thus, in accord with the supplementary data layers and stratified randomized sampling (SRS) method, eventually, 128 samples were gathered from 20 cm of soil surface of Mazandaran Province, Iran. First of all, the sample set was subdivided into two subsets: calibration (96) and validation (32). Using the multivariate regression analysis-partial least squares regression (PLSR) algorithm with leave-one-out cross-validation (LOOCV) technique and some pre-processing algorithms, such as spectral averaging, smoothing and 1st derivative (1st-D), the definitive calibration model with two & four latent vectors (LVs/LFs) and correlation coefficient (R_P), determination coefficient (R²_P), root mean square error (RMSE_P), ratio of performance to deviation (RPD_P) and ratio of performance to interquartile distance (RPIQ_P) respectively: 0.83&0.82, 0.68&0.67,8.68&8.83%,1.78&1.75,2.45&2.41, were validated and spotted as the most appropriate predictive model for the sand content prediction in the study region. Last, the potentiality of the visible-near infrared diffuse reflectance spectroscopy (VNIR-DRS) for sand content estimation in Mazandaran soils was proved. Also, it is feasible to upscale the sand prediction process utilizing the principal resulted model and the key spectral domains via airborne/satellite hyperspectral data, which emphatically shows the LDRS importance as a commencement point for characterizing the informative optical wavelengths. Likewise, that will be the infrastructure for spaceborne data modeling and upscaling process.

Due to the time and expenses, essential for sampling, and the inaccessibility of different parts of the land in the world, the use of geostatistical methods and artificial neural network (ANN), which can predict the spatial variability of SOM and soil organic carbon (SOC) is of outmost significance. The present research investigated the effects of different land use type including cultivated and uncultivated land, rangeland, forestry, and orchard at different altitudes (-20 to >500 m) on the spatial variability of SOM and SOC using the above-mentioned methods. A total of 249 combined samples from the depth of 0-15 cm (using GPS) on the basis of land use type and topography were collected from different parts of the research area (9545 km²). Using cross validation, the methods were compared and the best fitted one was selected on the basis of mean error (ME) and root mean square error (RMSE). The best-fitted semivariogram models for SOM (R²= 0.894) and SOC (R²= 0.761) were spherical. The cross-validation method indicated ANN as the most accurate method for the prediction of SOM and SOC with the ME and RMSE of 0.37 and 0.36 for SOM, and 0.37 and 0.35 for SOC, respectively. The comparison of different land types indicated that uncultivated land (1.37%) and forestry (3.94%) had the least and the highest SOC, respectively. Altitude also significantly affected SOC, and A500 and A0 resulted in the highest and least SOC, respectively. The use of ANN for the estimation of SOM and SOC is recommendable.

Estimation of the soil organic carbon content is of utmost importance in understanding the chemical, physical, and biological functions of the soil. This study proposes machine learning algorithms of support vector machines, artificial neural networks, regression tree, random forest, extreme gradient boosting, and conventional deep neural network for advancing prediction models of SOC. Models are trained with 1879 composite surface soil samples, and 105 auxiliary data as predictors. The genetic algorithm is used as a feature selection approach to identify effective variables. The results indicate that precipitation is the most important predictor driving 15 percent of SOC spatial variability followed by the normalized difference vegetation index, day temperature index of moderate resolution imaging spectroradiometer, multiresolution valley bottom flatness and land use, respectively. Based on 10 fold cross validation, the DNN model reported as a superior algorithm with the lowest prediction error and uncertainty. In terms of accuracy, DNN yielded a mean absolute error of 59 percent, a root mean squared error of 75 percent, a coefficient of determination of 0.65, and Lins concordance correlation coefficient of 0.83. The SOC content was the highest in udic soil moisture regime class with mean values of 4 percent, followed by the aquic and xeric classes, respectively. Soils in dense forestlands had the highest SOC contents, whereas soils of younger geological age and alluvial fans had lower SOC. The proposed DNN is a promising algorithm for handling large numbers of auxiliary data at a province scale, and due to its flexible structure and the ability to extract more information from the auxiliary data surrounding the sampled observations, it had high accuracy for the prediction of the SOC baseline map and minimal uncertainty.

Estimation of the soil organic carbon (SOC) content is of utmost importance in understanding the chemical, physical, and biological functions of the soil. This study proposes machine learning algorithms of support vector machines (SVM), artificial neural networks (ANN), regression tree, random forest (RF), extreme gradient boosting (XGBoost), and conventional deep neural network (DNN) for advancing prediction models of SOC. Models are trained with 1879 composite surface soil samples, and 105 auxiliary data as predictors. The genetic algorithm is used as a feature selection approach to identify effective variables. The results indicate that precipitation is the most important predictor driving 14.9% of SOC spatial variability followed by the normalized difference vegetation index (12.5%), day temperature index of moderate resolution imaging spectroradiometer (10.6%), multiresolution valley bottom flatness (8.7%) and land use (8.2%), respectively. Based on 10-fold cross-validation, the DNN model reported as a superior algorithm with the lowest prediction error and uncertainty. In terms of accuracy, DNN yielded a mean absolute error of 0.59%, a root mean squared error of 0.75%, a coefficient of determination of 0.65, and Lin’s concordance correlation coefficient of 0.83. The SOC content was the highest in udic soil moisture regime class with mean values of 3.71%, followed by the aquic (2.45%) and xeric (2.10%) classes, respectively. Soils in dense forestlands had the highest SOC contents, whereas soils of younger geological age and alluvial fans had lower SOC. The proposed DNN (hidden layers = 7, and size = 50) is a promising algorithm for handling large numbers of auxiliary data at a province-scale, and due to its flexible structure and the ability to extract more information from the auxiliary data surrounding the sampled observations, it had high accuracy for the prediction of the SOC base-line map and minimal uncertainty.