جستجوی مقالات مرتبط با کلیدواژه « random forest algorithm » در نشریات گروه « کشاورزی »

The genome-wide association study (GWAS) is a powerful approach to identify genomic regions associated with fertility traits that explain a significant portion of the genetic variance associated with these traits and identify the relevant causal mutations. Evaluating the correlation between each genotyped marker and trait is an essential strategy for GWAS studies that examine the effects of all markers by considering their possible interactions, environmental factors, and even mutual effects between markers. Recently, machine learning methods have been introduced to genomic topics, and the basis of these methods is different from the common methods of genomic evaluation. The machine learning method is used to estimate the genomic breeding values of the candidate animals by considering the training data (genotypic and phenotypic information of the reference population). One of the key advantages of this method is the ability to analyze large data. Machine learning is a branch of artificial intelligence whose goal is to achieve machines that can extract knowledge (learning) from the environment. A variety of machine learning methods (random forest, boosting, and deep learning) are used to model genetic variance and environmental factors, study gene networks, GWAS, study epistasis effects, and genomic evaluation. Random forest is one of the machine learning methods that has been successfully used in various fields of science. This research was conducted to identify markers and genes related to reproductive traits such as calving interval (CI), days open (DO), daughter pregnancy rate (DPR), and age at first calving (AFC) in Iranian Holstein dairy cattle. These traits have already been investigated with the ssGBLUP method and using a smaller sample size. However, in the present research, by using more genotyped animals, a random forest algorithm was used to identify markers and genes related to reproductive traits.
The records used in this research were provided by the National Animal Breeding Center and Promotion of Animal Products of Iran and included AFC, DO, CI, and DPR related to the genotyped bulls' daughters. In this research, the pedigree information of 2774183 animals was used. The genotypic information of the markers related to 2419 Holstein bulls was used. Genomic data quality control was performed using factors such as the number of genotyped SNPs per animal (ACR), the number of genotyped animals per SNP (CR), Hardy-Weinberg equilibrium (HWE), and minor allele Frequency (MAF). When filtering genomic data, the markers whose MAF was less than 5% were removed, and then the samples whose genotyped frequency was less than 90% were identified and removed. Then, the markers whose genotyping rate was less than 95% in the samples were identified and removed. Finally, the SNPs that deviated from the HWE test (P<10-6) were excluded from the analysis as a measure of genotyping error. To control the quality of genomic data, PLINK 1.9 software was used. Then Ranfog software was used in the Linux environment to perform analysis through random forest algorithm.
By using the random forest algorithm, a total of 21 important SNPs were observed, then important fertility trait candidate genes were identified by the gene ontology method, and 62 genes were within 250 Kb of these SNPs. The most significant SNP was observed for AFC. The main SNP for AFC is in ARS-BFGL-NGS-22647 BTA3, for CI is in ARS-BFGL-NGS-114194 (BTA11), for DO is in BTA-74076 -no-rs (BTA5), and for DPR is in ARS-BFGL-NGS-32553 (BTA26). The researchers, who studied fertility traits in Nellore cattle using machine learning methods, identified MPZL1 and CD247 genes on chromosome number 3 and this gene was associated with age at first calving. Many pathways of cell biology affect the performance of reproductive traits. Research has reported the relationship between the CD247 gene and pathways of biology, including cell development and function. Research has shown that the IFFO2 gene plays an important role in the molecular structure of cells, as well as in the mechanism of blastocyst formation, embryos, and the length of gestation in cattle. In a study conducted on the mouse population on the structure of the flagellum and the sperm maturation process, the role of the ALDH4A1 gene in the sperm maturation process was reported. The association of the RPS6KC1 gene with pregnancy rate and antral follicle number in Nellore heifers has been reported. The KAT2B gene is a transcriptional activator that plays an essential role in regulating the correction of histone acetylation and plays an important role in improving carcass quality, muscle and fat development, and metabolism in native Chinese cattle. In addition, they play a key role in regulating biological processes and are related to cell growth, metabolism and immune system function.
  According to the objectives of this research, new information on markers and candidate genes related to reproductive traits in Iranian Holstein dairy cattle was reported. The markers and candidate genes identified in the present research can be used in genomic selection to improve the reproductive traits of Holstein dairy cattle.

The occurrence of landslides in mountainous areas may cause serious damage to road infrastructure, and may also lead to human deaths. Therefore, the purpose of this study is to landslide susceptibility mapping using advanced machine learning algorithms in Sarovabad city. In this study, landslide susceptibility was determined using two advanced data mining algorithms including random forest (RF) and decision tree (DT). First, the point file of 166 landslides occurred in Sarovabad city was considered as the landslide inventory map. The landslide points are divided into training data (70%) and validation data (30%). A total of 16 parameters including slope, aspect, elevation, river proximity, road proximity, river density, fault proximity, fault density, road density, precipitation, land use, NDVI, lithology, earthquake, stream power index (SPI) and topographic wetness index (TWI) were used in order to landslide susceptibility mapping. Finally, the performance of the models was evaluated using the ROC curve. The results of the ROC showed that the decision tree and random forest models have AUC values of 0.942 and 0.951, respectively. Therefore, the random forest model has the highest AUC value compared to the decision tree and was the best model for predicting the risk of landslides in the future in the study area. Landslide potential maps are efficient tools; so that they can be used for environmental management, land use planning and infrastructure development.

The negative effects of soil salinity on natural and human environments have turned such phenomenon into one of the serious threats to the sustainable management of arid and semi-arid areas. The present study aims to evaluate the spatial-temporal variations in land use/land cover and soil salinity in a part of Sistan basin located in the arid regions of southeastern Iran, which has been exposed to the phenomenon of soil salinity during the recent years. To this aim, land use/land cover and soil salinity maps were prepared for 1989 and 2019 using the measurements obtained from ground sampling and instruments such as remote sensing (RS) and geographic information system (GIS). Based on the results, the normalized average soil salinity was 0.322 during 1989, reaching 0.52 during 2019 with a growth of 0.188. In addition, comparing the trend of salinity increase and land use/land cover variations in the region indicates that such factors affect each other significantly. Further, the conversion of land use/land cover from irrigated agricultural uses and water bodies to bare lands, rainfed agriculture, and man-made areas has decreased the vegetation cover and water areas, leading to an increase in soil salinity due to the erosion created by the 120- day wind of Sistan and sedimentation of salt particles in the whole region. Chah-nimehs have not been able to fully compensate for the adverse effects generated by the destruction of habitats and drying up of a part of Hamun Lake and Hirmand River, despite their slight success in reducing the environmental obstacles.

The aim of this study was to model the landslide susceptibility using the Random Forest Machine learning technique and prioritization of effective factors on landslide occurrence in Bar watershed in Khorasan Razavi province. The random forest algorithm is based on a bunch of decision trees and is currently one of the best machine learning algorithms. For this purpose, a landslide inventory map was created with 73 historical landslides, which was randomly divided into two datasets for model training (70%) and model testing (30%). A total of 16 landslide-conditioning factors were considered for the susceptibility landslide mapping. The random forest algorithm was run and a landslide susceptibility map was prepared. The RF-based model was validated using the area under the receiver operating characteristic (ROC) curve. The results of evaluation indicated that the success and prediction rates of the model were 90% in training and 89% in validation, respectively. These results confirm the ability of random forest method for prediction of landslide susceptibility models. Also, prioritization of the effective factors showed that the slope length and slope had the highest effect on landslide occurrence. Based on the results of the random forest model, 23.7% of the study area is located in a very high and high sensitivity zone.

Today, reforested stands are one of the most important sources of forest carbon storage and one of the factors that reduce the process of destruction of natural areas. Above-ground biomass (AGB) plays an essential role in sustainable forest management and reducing global warming and is an important source of information. Allometric equations are an important tool for quantifying above-ground biomass in forests. In recent years, remote sensing techniques using non-parametric methods such as the Random Forest algorithm have been widely used to estimate forest tree biomass. In this research, the ability of Sentinel 2 data using the random forest algorithm to estimate the above-ground biomass of Arabdagh reforested stands in Golestan province was evaluated. In this study, 180 circular sample plots with an area of 400 square meters were measured using the cluster sampling method and the diameter at breast height (DBH) and tree height (H) were measured. Also, the exact coordinates of the centers of the sample plots were recorded using DGPS. Then, using the prepared allometric equations, the above-ground biomass of trees was calculated. In this study, Sentinel 2 pre-processed radiometric and geometrical data were used, and based on that, different vegetation indices were prepared. In the implementation of the random forest algorithm, the relationship between the characteristics of biomass as a dependent variable and the spectral values of vegetation indices as independent variables were investigated. Modeling was done using 75% of sample plots (135 sample plots) with random forest algorithm and validation of estimates was done using 25% of sample plots (45 sample plots). The results showed that NDVI and GNDVI indices had the highest correlation in the estimation of above-ground biomass and the random forest algorithm with 310 trees and 5 predictors and the percentage root mean square error of 35.83% and the coefficient of determination 0.51 was able to estimate the above-ground biomass of Arabdagh reforested stands. Also, the results showed that using the data of Sentinel 2, the random forest algorithm has estimated the above-ground biomass of trees more than the actual values. There is no significant difference at the 95% probability level between the estimated and real above-ground biomass values (p-value > 0.05). Also, among the independent variables used. The results of this research showed that Sentinel 2 data has been able to estimate the above-ground biomass of Arabdagh reforested stands with acceptable accuracy. According to the results of this article, it can be said that the information of the main bands and spectral indices played an important role in the estimation of above-ground biomass.

Population growth, along with the development of industry and agriculture, has led to an increase in water consumption. Limited surface water resources have led to over-harvesting of groundwater aquifers and has had irreparable consequences for the country's water resources and environment, including the subsidence phenomenon, which has covered most of the country's plains. The present research aims to identify the effective factors and areas at risk of subsidence in Fadafan village of Kashmar. For risk zoning, during 2019, the lithology, land use, Petrology, aquifer Extraction rate, Distance from the stream, Fault, exploitation wells, springs and aqueducts factors as well as geomorphological factors including slope, direction and height studied and each factor turned into an information layer, then modeling and evaluation were performed using random forest algorithm in R software. Then, to determine the areas prone to subsidence, risk zoning maps in five classes were extracted using two methods of information value and area density in ArcGIS environment. The results showed that in the methods of area density and information value, 97.01 and 91.04% of subsidence were in the very- high and high-risk class, respectively. Therefore, both methods have been successful in risk zoning. Also, the aquifer extraction and land use factors are most important in subsidence. Also based on the ROC curve, random forest algorithm with very high accuracy (93%) has provided good results in prioritizing and the importance of effective factors in subsidence. The southern part of the region with rangeland use, has the highest and irrigated agriculture in the region has the lowest risk in the spatial development of land subsidence.As a result, aquifer recharge management by spreading floods and reducing water extraction in the southern part of the region can be effective and practical in reducing the risk of occurrence and development of subsidence.

Flood damage assessment is often necessary for early flood management. To this end, this paper provides a framework of rapid estimation of flood damage and identification the flooded areas in March 2020 using Sentinel-1 satellite data. To this end, in the present study, after applying the necessary pre-processing in SNAP6 software, the backscattering coefficient, or sigma naught for two images related to before and after the flood occurrence was extracted. The backscattering coefficient histogram was used to separate the image into two classes including water and non-water and the threshold of 0.01 was obtained based on it. Then, by applying mathematical operations on both backscattering images, the binary image of water and non-water was prepared and the flooded areas were determined based on the difference between the two images. After detecting the flooded areas, Sentinel images were classified into three classes including waterbody before flood, flooded area and other lands using supervised classification algorithms. The results indicated the high accuracy of the Random Forest algorithm with kappa of 0.92 compared to other algorithms. By overlaying the land use and flooded areas maps, the inundation percentage for each land use was determined. According to the results, bare lands with 27.9 percent, residential land with 16 percent and rangelands with 12 percent had the highest inundation percentage, respectively.

Land cover map show the spatial distribution of different landscapes such as agricultue, natural resources, water and man-made area. It is a valuable tool to managing and reducing risk in challenging issues such as drought and its effects, food security, flood control, and urban planning. In order to overcome the limitations of field work in the mapping of land cover, the use of satellite images due to the wide, multispectral and update data seems to be suitable. In the study area, the spatially heterogeneous landscapes also makes it difficult to classify features. Therefore, the main purpose of the study is accurate and high resolution land cover mapping using Sentinel-2A images in the Google Earth Engine platform. In this regard, three classification algorithms including RF, SVM and CART were evaluated and compared. Various indices were prepared using ratioing and transformation methods. The accuracy of the classifications was evaluated in comparison with ground reference data. Individual bands evaluation showed that the best overall accuracy (49%) was obtained using the CVI index.The best overall accuracy and kappa coefficient of 86% and 0.82 were obtained by RF algorithm. Therefore, while pointing to the advantages of the GEE including easily accessible data and the ability to process and quickly compare of data, it can be claimed that Sentinel-2A images for land cover mapping in terms of cost, time and accuracy, have high efficiency and the map can be very useful for the management and decision making in different natural and man-made resources for the successful implementation of sustainable development.

Fire severity mapping is very important for managing the fires in forest ecosystems. The extraction of spectral indices from optical sensors is recognized as one of the most effective bands for the classification of vegetation classes. In this study, the ability and sensitivity of some spectral indices extracted from Sentinel-2 and Landsat 8-OLI images with different spatial resolutions have been investigated for fire severity mapping using the Random Forest algorithm in a burned area located in the reforested area of Arabdagh, Golestan province. After necessary preprocessing on the bands, the appropriate mono and bi-temporal spectral vegetation indices were created. The optimal index values for bands in the bi-spectral spaces pre/post-fire were calculated to evaluate the sensitivity of bands to the changes occurring within the fire classes. The best results were obtained for the NIR-SWIR2 bands with an optimal index value of 0.77 for Sentinel-2 and 0.67 for Landsat8-OLI. The best indices were selected based on values of optimality index. The values of these indices were calculated after the fire as well as the differential (pre/post-fire) ones. The ground truth of fire severity classes map was prepared by a selective sampling method through field surveying. The classification was done with different indices by random forest (RF) algorithm and the results were assessed by the ground truth points. The result showed that the best results were obtained for a combination of many differential indices from all bi-bands of Landsat 8-OLI with kappa coefficient (0.96).

Khorasan Razavi province has the potential for growing and producing rapeseed because of favorable environmental conditions, so that the northern and central cities of province have high potential for cultivation of rapeseed. Modeling the correct relationship between environmental conditions and yields is a critical step to find how crop-planting choices in different regions of Iran.Spatial modeling in GIS is one of the most important strategies that can provide a basis for measuring environmental factors and land suitability for the cultivation of a particular product by combining statistical methods and spatial data. In this research, the link between water, soil and meteorological factors and yields modeled during the growing season in sample farms.

In this research, the position of 24 sample fields of rapeseed farming was recorded by Global Positioning System (GPS) and then actual yield was calculated. To explore how the environmental conditions and yields relationship has changed over space, we used ten environmental parameters influencing rapeseed productions yield, including elevation, slope, aspect, EC and pH groundwater resources, mean air temperature, incoming solar radiation, potential evapotranspiration, wind exposition index, Soil texture during the growing season. The values of each independent variables were extracted into samples by nearest neighbor method. Then, after normalizing the variables and taking into account the range of numbers, the samples were divided into two subsets: training (60%, 14 farms) and the testing dataset (40%, 10 farms) randomly. Two methods of nonparametric K of the nearest neighbor and random forest were then used to estimate rapeseed yield over the study area.

The results of mean absolute error percentage in the methods used showed that K is the nearest neighbor with 26% error and random forest with 11% error. The results of Nash–Sutcliffe efficiency index for validation data set represent the value of 0.65 for K nearest neighbor and 0.82 for random forest method. In general, the results indicate that the random forest method has a lesser error than the K nearest neighbor method in estimating the yield of rapeseed productions for the study area.

Based on the results of this research, it can be concluded that among the variables used, two variables of wind supply index and average temperature had the most effect on the yield of rapeseed in comparison with other variables. Also, according to the final map, it was determined that suitable areas for rapeseed cultivation over Sabzevar region are located in the northern and northwestern regions. Low yield in the central regions of this part is mainly due to the excessive salinity of water and gypsum formations. Crop yield is a result of combination of genetic factors and also environmental conditions of the cultivation, which we emphasized on the environmental factors in this study.

Land subsidence occurrence in the Tasuj plane might become more frequent and hazardous in the near future due to its relationship with the water crisis and drought periods. In order to mitigate the damage caused by land subsidence, it is necessary to determine the susceptible or prone areas. The purpose of this study is to produce land subsidence susceptibility map based on the random forest approach to land subsidence occurrence data and eleven environmental variables  that have significant influence on land subsidence occurrences (altitude, slope, aspect, distance to drainage line, drainage density, distance from the fault, topographic wetness index, land cover, lithology, groundwater level and decline in groundwater level) were used as inputs of the random forest model. The random forest approach was applied to produce the land subsidence susceptibility map. The performance of the model was assessed using the receiver operating characteristics (ROC) curve and the area under the curve (AUC). The model results indicate the accuracy of 0.86. Based on the result of the mean decrease accuracy method, the most important conditioning factors were groundwater level, distance from the fault, and a decline in groundwater level, respectively. According to the result, about 18% and 11% of the study area was located within high to very high susceptibility classes. The result of this study can be used by stakeholders and local authorities to mitigate related hazards of land subsidence occurrences in the study area.

Land use/cover maps are the basic inputs for most of the environmental simulation models; hence, the accuracy of the maps derived from the classification of the satellite images reduces the uncertainty in modeling. The aim of this study was to assess the accuracy of the maps produced by machine learning based on classification methods (Random Forest and Support Vector Machine) and to compare them with a common classification method (Maximum Likelihood). For this purpose, the image of the OLI sensor of Landsat 8 for the study area (Sattarkhan Dam’s basin in the Eastern Azerbaijan) was used after the initial corrections. Five land uses including urban, irrigated and rain-fed agriculture, range and water body were considered. For conducting the supervised classification, ground truth data were used in two sets of educational (70% of the total) and test (30%) data. Accuracy indexes were used and the McNemar test was employed to show the significant statistical difference between the performances of the methods. The results indicates that the overall accuracy of Support Vector Machine, Random Forest, and Maximum Likelihood methods was 96.6, 90.8, and 90.8 %, respectively; also the Kappa coefficient for these methods was 0.93, 0.81 and 0.83, respectively. The existence of a significant statistical difference at the 95% confidence between the performances of the Support Vector Machine algorithm and the other two algorithms was confirmed by the McNemar test.