ارزیابی حساسیت خاک به فرسایش خندقی با استفاده از مدل CART و سیستم اطلاعات جغرافیایی(GIS) (منطقه موردمطالعه: حوضه آبخیز خسویه استان فارس)

Extended Abstract



Introduction

Soil erosion in any place is affected by various factors, including natural features and human activities. Mediterranean rainfall system and high water erosion, large extent of soils and formations sensitive to erosion, poor natural vegetation in many regions of the country and uneven conditions are some of the important natural factors affecting soil erosion in Iran (Arab Khodri, 2014). So that about 125 million hectares out of 165 million hectares of the country's lands are exposed to water erosiondramatically. Therefore, the incoming blown sand is dropped close to the brink line. Gully erosion as one of type of water erosion is very severe in many areas specially in arid and semi-aird areas. Because in areas with low vegetation cover are mpre prone to the drops of rainfall with high intensity. There are many researches regarding this type of water loss that some of these researches focus on modelling and predicted the prone areas with stochastic model. Rangzen et al. (1401) in their research studied the prone areas to gully erosion using fuzzy membership function in the watershed of Mehr city, in the south of Fars province, they have determined prone areas to gully erosion using fuzzy membership function and hierarchical analysis model (AHP) in Mehr city, in the south of Fars province. The results of their investigation showed that the areas located in the center of the studied area (about 18%) are more sensitive to gullies erosion. They have used the ROC curve to validate the model, the AUC values near 85 indicate the high accuracy of the model for predicting areas prone to gully erosion in the watershed of Mehr city. In this study we have applied the CART model to predict the prone areas in Ljasouh watershed in South of Fasr province in Iran.







Methodology

The study area isKhasuye watershed, with an area of 136,622 hectares, is located 337 km from Shiraz city in Fars province, Zarindasht city. In terms of the geographical location of this watershed between the longitudes of 54 degrees and 9 minutes to 54 degrees and 42 minutes east and the latitudes of 28 degrees and 18 minutes to 28 degrees and 39 minutes north, it occupies the northern strip of Zarindasht city.

The average annual and monthly temperatures of this basin are 21.91 and 19.09 degrees Celsius, respectively. Also, the maximum and minimum average temperature of the basin is 33.09 and 9.78 degrees Celsius, respectively, corresponding to the months of July, the average annual rainfall of Khasouye station is 221.54 mm and its average height above sea level is about 1150 meters.

For the applying the CART model by using the SPM (Salford Peredectie Modeler) software in research, in first step, the locations of the gullies were recorded using the Google Earth software and Global Positioning System (GPS) in the whole study area. The second step, out of 16 influencing variables in gully erosion, including topography-related indicators (height, slope, slope direction, slope length, flow strength, stream depth, topographic wetness and longitudinal curvature), normalized vegetation cover index (NDVI), land use, distance from roads, distance from rivers, geological map, rainfall, soil type and soil erodibility have been used. After assigning the values related to independent indicators to each of gullies area, modeling has been run to predict the prone gullies areas in the study area, in the SPM software environment using the CART model. The applied model has used 70% of the data as training data and 30% of the data as test data in the modeling process.



Results and Discussion



One of the main factors in modeling is checking the accuracy of model results, therefore we have used different stastical index to evaulte the accuracy of our result for each traing and testing data. Table No. 1 shows the CART model accuracy evaluation indices. According to this table, the correlation coefficient or R2 for training data is equal to 0.904 and for test data is equal to 0.845. Considering that (Chain et al., 1998) have defined three values of 0.19, 0.33 and 0.67, respectively, as the criterion value for weak, medium and strong values of the evaluation of the structural parts of the model by means of the R2 index (correlation coefficient). Therfore we can conclude that the model it has high accuracy in predicting the points prone to gully areas height parameter with R square of 0.915 and Std error of estimate of 0.133.



In terms of the relative importance of each of the independent variables participating in the CART modeling process, respectively, the variables of height, amount of precipitation, distance from streams and land use have the highest value with values above 50%, and the variables of slope direction, Longitudinal curvature, topographic wettness have the lowest values.



Conclusion

As we have seen in the predicted map of gully erosion the most of these features occur in the flat area with low slope in the Khasouye basin. Due to rainfall or irregular irrigation of the adjacent lands, the water flows on the surface and finally reaches these flat areas. Usually, the amount of precipitation in high areas is higher than in flat and low-altitude areas, but all the precipitation that occurs in high areas is due to other factors such as topography, slope and the direction of the slope of the area, especially in the Khasouye watershed, where the high areas lack suitable vegetation. By applying the results obtained from the implemented statistical model and the geological maps of the Khosouye watershed, it was determined that the most gullies occurred in the places consisting of the Quaternary formation. Quaternary formation includes alluvial and sedimentary deposits and loess. Due to the high solubility of this type of formation, its sensitivity to water erosion and especially to ditch erosion is very high.



Keywords: Gully erosion, CART model, GIS, Khasouhe watershed.