Application of Data-Mining Algorithms in the Sensitivity Analysis and Zoning of Areas Prone to Gully Erosion in the Indicator Watersheds of Khorasan Razavi Province

Extended abstract 1- Introduction Gully erosion is one of the most important sources of sediment in the watersheds and a common phenomenon in semi-arid climate that affects vast areas with different morphological, soil and climatic conditions. This type of erosion is very dangerous due to the transfer of fertile soil horizons, and the reduction of water holding capacity also is a factor for soil instability and erosion creating problems for the functioning of agricultural machines. Hence, the identification of effective factors in the formation and development of gully, plays an important role in the watershed management and prevention of land degradation in the direction of sustainable development. There are few studies regarding the application of different data mining methods to determine the effective variables in the occurrence of gullies and most studies are based on the other statistical methods. Data mining is called as knowledge discovery in databases and is a way to discover new and potential beneficial information through lots of information. Some of the most important data mining algorithms include the decision tree, random forest, boosting aggregate demand, support vector machine, logistic regression and neural network algorithms. Given that the data mining extracts useful information from large volumes of data and has shown a good performance, therefore, the aim of the present study is the prioritization of environmental factors affecting the occurrence of gully erosion including the altitude, slope, aspect, climate, land use, land capability, lithology, flood zones, evaporation, rainfall, temperature, distance from road, distance from fault, distance from river, topographic wetness index, sediment transport index and stream power index using data mining algorithms and zoning its sensitivity using bivariate statistical models of information value and area density in five Shekarkalat, Qarehtikan, Komajkhoore Dargaz, Sabzevar Kaloots and Mooshak-Kashmar watersheds in the Khorasan Razavi province. 2- Methodology The present study aimed at investigating the factors affecting the occurrence of gully erosion and its zoning in the five Shekarkalat, Qarehtikan, Komajkhoore Dargaz, Sabzevar Kaloots and Mooshak-Kashmar watersheds in the Khorasan-Razavi province. Within these five watersheds, 15 representative gullies were selected and subsequently, the distribution map of gullies in these watersheds were prepared. Then, each of the factors affecting the occurrence of gully erosion in the study areas consisting of 17 information layers including altitude, slope, aspect, climate, land use, land capability, lithology, flood zones, evaporation, rainfall, temperature, distance from road, distance from fault, distance from river, topographic wetness index, sediment transport index and stream power index were prepared and digitized using the ArcGIS10.3 software. By using data mining algorithms in the R software, the factors affecting the gully formation were introduced. Finally, the gully erosion hazard zonation in the GIS was done using bivariate statistical models. 3- Results The results of data mining algorithms showed that the distance from road, lithology, slope, aspect, stream power index, sediment transport index and topographic wetness index variables were the most important factors in the occurrence of gully erosion. According to the obtained zoning map and also the weight of  classes for each parameter, the slope between 0 to 5 degrees, the northeast and southeast directions, a distance more than 2000 m from roads, glauconite sandstone class in the lithology layer and classes with high levels of the secondary topographic indicators had the highest weight and effect on the occurrence of gully erosion in the study areas. The evaluation results of the zoning map using the information value and area density methods based on the weighted quantitative amounts showed that in the information value method, 98.08 % and 1.92% of the gullies were located at the medium and high hazard zones, respectively and in the area density method, 100% of the gullies were located at the very low hazard zone. The results indicated that the areas prone to gully erosion were not detected in high accuracy but the information value method was more favorable than the density area method. 4- Discussion and Conclusions According to the results of this research, the distance from road, lithology, slope, aspect, stream power index, sediment transport index and topographic wetness index variables considered as the most important factors in the occurrence of gully erosion. Although the factors such as lithology, slope, aspect and the secondary topographic indicators (topographic wetness index, sediment transport index and stream power index), which were identified as the most important factors influencing the development of gully in this research, were classified as natural factors, they could be influenced by human factors. Therefore, the sensitivity of the lithological units should well noted. Also, controlling the stream flows, implementing of management plans especially in low slopes and vegetation regeneration, which increase the surface roughness coefficient and decrease the runoff velocity and ultimately control the gully erosion, is essential. In case of zoning with the two methods of information value and area density, it follows that the information value method is more appropriate than the area density method.