Sensitivity of Desert Landforms to Trampling Using Digital Photography and XGboost Model: A Case Study of Masile Pit in Qom

Today, even the simplest human activities, such as walking, can have destructive consequences. Human movement on the earth may be done for economic, tourism, military purposes, etc. However, these movements lead to the trampling of the land and have consequences such as changes in the abundance and richness of vegetation, increase in runoff and soil density, and changes in erosion. In addition, the surface of the landforms is disturbed and with a sharp decrease in the threshold of shear speed, their vulnerability to wind erosion increases dramatically. Determining the sensitivity of landforms requires accurate tools, time and a lot of money, while the use of photography in monitoring the amount of changes in landforms can save time and money. The research that has been done so far in relation to trampling is generally based on two approaches. The first approach is the experimental method, that is, small undisturbed parts are selected in the study area and trampled according to the requirements of the experiment. The second method involves making long-term observations of the tracks that have been trodden. Chale Masileh is a part of the desert areas of the country, which is trampled by humans for various reasons. Economic activities such as harvesting potassium, magnesium and other materials from the bed of the salt lake, tourist areas such as Maranjab in the south of the region and military activities and holding large military maneuvers are among them. However, there is still no detailed information about the areas sensitive to trampling. Therefore, the current research tries to investigate the sensitivity of different desert landforms to trampling using a low-cost and fast method because in order to use and manage desert areas, it is necessary to understand the sensitivity of landforms.

The studied area mainly includes Mesila pit. The geographic coordinates of the salt lake as an index point in this hole are 30°34° north latitude and 50°51° east longitude. In this research, various materials, data and tools have been used in different stages of the research. The geomorphological map of the study area, and field data were collected in field operations using camera, tripod, GPS and plot. ArcMap, SNAP software were used to prepare the data, and Python 3.10 programming environment was used to run the model and draw graphs and outputs. 13 landforms were selected in different positions of the region. In the next step, an undisturbed area in the landform was selected and photographed, then trampled with 25 passes and a second photograph was taken. After recording the images by entering fourteen features for each pixel, the data was prepared to participate/to be applied in the model implementation. In the implementation of the model, 75% of the pixels were used as training data and the remaining 25% as test data. The ratio of changed pixels to unchanged pixels was considered as the change rate.

Evaluation of model efficiency using model relative performance characteristic curves (ROC) and area under the curve (AUC) shows that the values of 0.99 to 0.88 indicate the very good efficiency of the model in all samples. The RMSE value for the samples shows that in all the samples the mean square error value is less than 0.5, which confirms the good accuracy of the model in predicting the changed pixels. In addition, the average percentage of accuracy for samples using -k10 shows that the accuracy of the model in each sample is more than 80%. The importance of the image texture features in predicting the changed values in the photographed samples shows that in most samples the feature of the local standard deviation of the image of dissimilarity was the most important factor

Determining the sensitivity of the landforms of desert areas to human trampling is important for the management of these areas. To achieve this purpose, it is very useful to use low-cost, fast and accurate methods such as taking pictures and using machine learning algorithms. This study shows that the XGboost model and the photography method can measure the amount of landform changes with an accuracy of over 90%. These changes are determined according to the texture of the images and do not measure changes in height or soil density. However, it determines the number of changed pixels even when it is barely visible to the eye. On the other hand, the results of this method are the result of the influence of all effective variables such as slope, humidity, roughness, etc. While changes in altitude or density can be affected by one of the mentioned factors. The changes in different landforms based on the results of the model show that the least change is in the plowed land, which consists of very hard mud and salt. The surface of this landform is naturally very messy and uneven. The changes in the comparison of the two photos before and after trampling in this sample are hardly visible. The changed areas mostly correspond to the microslopes and small peaks that have been subjected to the most pressure in each pass. The amount of changes in the landform of sandy surfaces with a small amount of vegetation is 85%, which shows the most/highest changes among the samples. The key feature of this landform is the separation of land surface materials and vegetation.