جستجوی مقالات مرتبط با کلیدواژه « مدل ماشین بردار پشتیبان » در نشریات گروه « آبخیزداری، بیابان، محیط زیست، مرتع »

Soil erosion by water and wind is one of the biggest environmental threats worldwide which it has various negative consequences such as soil degradation, mitigation of soil fertility, depletion of nutrients and micro-elements, degradation of soil structure, dust storms, reservoir siltation, and etc. on the natural and anthropogenic ecosystems. Therefore, accurate information of land susceptibility to soil erosion hazard by water and wind, and production of spatial maps of these hazards are necessary for mitigation of their consequences. Therefore, the goal of this research is application of two machine learning models for the spatial modelling of water and wind erosion in the Bakhtegan basin, Fars province.

In order to generate the spatial maps of soil erosion by water and wind, we have used 20 and 16 effective factors on water and wind erosion, respectively. Before the modelling stage, the multivariate adaptive regression spline (MARS) feature selection algorithm and multicollinearity test were used to identify the most important factors controlling water and wind erosion. In next stage, two random forest and support vector learning machine techniques were applied to model soil erosion. Area under curve (AUC) was applied to assess the model performance.

According to results, land use, slope, lithology, roughness, coarse fragment, clay, vegetation cover, aspect, soil bulk density and topographic wetness index were identified as the most important factors controlling water erosion. The most important factors controlling wind erosion are including elevation, land use, wind speed, slope, vegetation cover, soil nitrogen, cation exchange capacity, lithology, aspect, clay and bulk density. The AUC value for wind erosion map generated by random forest model was 99, and for water erosion map produced by support vector machine is 96. Finally, we integrated two maps and results indicate an area of study area about 18% has very high susceptibility to wind erosion and low susceptibility to water erosion.

Due to study area has several lakes consisting of Bakhtegan, Maharlo and Tashk, and also several dams, it is so important to supply water for different purposes such as drinking, agricultural, industry and environment. In case of not paying attention to this challenge especially wind erosion in the study area, this case study especially dried beds for lakes can play as sources for generating dust particles. Therefore, the findings of this research can use by managers to mitigate negative effects of this phenomena.

Rainfall-runoff from urban areas is one of the available water resources, which is wasted due to lack of attention and proper management. Besides, urban runoff excess of drains capacity causing many problems including inundation and urban environmental pollution. Therefore, harvesting this runoff can provide a part of the required water in urban areas, and also reduce flood and urban inundation, thus rainwater harvesting systems can play an important role in this regard. The purpose of this study is identification areas with inundation potential for urban runoff harvesting using the support vector machine model. For this purpose, 46 flood points were collected in Imam Ali (AS) town in Mashhad, and 32 points were randomly selected for the model training plus 14 points for the model validation. The predicting variables such as elevation, slope, drainage density, distance from drainage channels, and land use were used to implement the support vector machine model. The results showed that high and very high potential classes for runoff collection include 11.63% and 4.49% of the total area, respectively. The central and north-eastern parts have more potential for urban runoff harvesting, while the western, eastern, and southern regions have the lowest potential for runoff harvesting. Also, the evaluation performance of the support vector machine model was 87.8% in the training stage and 80% in the validation stage, which indicates that the performance of the model is "very good".

In any water resource management plan, there is a pivotal need to undertake the future conditions to allocate the water resources to different sectors (e.g. drinking-water supply, agriculture sector, etc.) more efficiently. Meanwhile, it is important to forecast water resources inflow for future months. To this aim, it is of prime interest to adopt models that are capable of coping with data scarcity problem and able to forecast the stream flow with the least possible error. The current study was aimed at forecasting the monthly inflow of the Boustan Dam by employing three models namely: time series method, Artificial Neural Network (ANN), Support Vector Machine (SVM), and their ensembles. The hydrometric data was obtained from the Tamar Station. Afterward, the models were compared by using several evaluation criteria. According to the Akaike and Schwarz criteria, the ARIMA (2, 0, 0) (1, 0, 1) was found to be the best time series model with a parsimonious behavior. Moreover, the ANN model with two and four input neurons and the SVM model with three input neurons were the best performing models compared to their other counterparts with different input numbers. Considering the evaluation criteria altogether, the time series method was the best performing model with the RMSE, AARE, MBE, and CE values of 0.88, 4.71, -0.024, and 0.36, respectively. Therefore, the time series method was introduced as the premier model for monthly inflow forecasting in the studied stations.