Introduction

Land subsidence refers to the vertical, downward movement of the Earth's surface. This phenomenon can occur naturally or as a result of human activities. The consequences of subsidence, could damage existing infrastructure such as road networks, facilities, buildings, and even agricultural lands. Land subsidence is not always visible, especially in the early stages. For this reason, it is essential to identify and identify areas prone to subsidence before they endanger lives and the environment.

In Iran, due to the lack of rainfall in the past few years in order to supply the water needed in the agricultural, industrial, and drinking water sectors, water withdrawal from underground aquifers has increased, which has led to the depletion of groundwater resources and soil loosening, resulting in land subsidence. This phenomenon has occurred as a serious crisis in many fertile plains of Iran in recent years. In this study, Shabestar Plain, as one of the high-risk plains of East Azerbaijan Province, has been selected as the study area. Dempster-Shafer evidence theory has been used as a statistical method to determine the sensitivity to land subsidence in the Shabestar Plain. In most relevant research, only the probability of land subsidence occurrence has been obtained. However, appropriate and more accurate decision-makings require that in addition to the estimated probabilities, the uncertainties of the estimates is also provided.

Methodology

Dempster-Shafer evidence theory is a mathematical framework for modeling uncertainty while combining evidence from different sources. It was introduced by Dempster in 1967, and its mathematical structure was further developed by Shafer in 1976. This theory is based on the belief that results from evidence and is useful in situations where data is incomplete or ambiguous helping make better decisions.

Shabestar Plain is one of the important and at risk plains of East Azarbaijan which has been selected as study area. Eight spatial data layers as land subsidence influencing factors were gathered, including groundwater level, geology, slope, elevation, distance from faults, distance from rivers, aspect, and land use. Relevant raster maps of each factor were generated with pixel size of 20×20 meters. In the next step, using the Dempster-Shafer theory, the data were combined and study area susceptibility to land subsidence was prepared. The general steps of conducting this study include:

- selecting a set of influencing factors and preparing related maps

- gathering land subsidence inventory map of the study area

- combining input data using evidence theory to prepare ‘belief’, ‘disbelief’, and ‘uncertainty’ maps

- Producing land subsidence susceptibility map for the study area

- Validating results using independent data

Results and Discussion

The produced maps; belief, disbelief and uncertainty were classified into five classes. The belief map indicates the five degrees of belief to which each region on the map has priority for mitigation and preparedness programs. The disbelief map indicates the five degrees to which each region of the map has not priority for mitigation and prepareness programs. The uncertainty map shows the degree of uncertainty whether each of the regions on the map have priority for mitigation and preparedness programs or not. The final land subsidence susceptibility map was classified into five subsidence susceptibility classes: "very low", "low", "medium", "high", and "very high". It was shown that the southern part of the Shabestar Plain is highly susceptible to land subsidence, which needs to be considered by ahthorities and regional managers.

To examine the accuracy of the prediction results using the Dempster-Shafer method, five piezometric wells of the study area, which field observations reported land subsidence occurrence for them, were selected. The subsidence rate of the selected points were calculated using the Shabestar Plain subsidence inventory map by utilizing GIS analyses tools. We compared subsidence rates of the points with the predicted land subsidence susceptibility classes. The results confirmed a very good agreement between the predictions made by the proposed Dempster-Shafer model and the subsidence rates obtained from the radar interferometry method.

Conclusion

Dempster-Shafer evidence theory presents the degree of belief, disbelief and uncertainty, beside the prediction results. This ability is not supported with other common prediction methods such as machine learning or artificial neural networks. In this paper using Dempster-Shafer theory, we obtained maps of belief, disbelief and uncertainty of land subsidence susceptibility, as well as a land subsidence susceptibility map based on the provided belief, disbelief and uncertainty. The zoning map was classified into five categories using quantile method. The area belonging to the "very low" susceptibility was 41.86 % of the total area of the Shabestar Plain, which has the largest area among other classes of land subsidence susceptibility. The classes of "low", "medium", "high" and "very high" land subsidence susceptibility covered 26.8 %, 24.72 %, 4.6 % and 2.02 % of the whole study area respectively. The results also showed that the southern part of the study area is very sensitive to land subsidence. In this study, an existing land subsidence inventory map was used to validate land subsidence sensitivity modeling. It was shown that the method used for modeling land subsidence susceptibility had good predictive power. Therefore it can be concluded that the introduced model can provide reliable land subsidence predictions of areas at risk and can be utilized in other study areas as well. The results can help urban authorities with proper planning for construction developments in time to prevent and/or reduce damages caused by land subsidence.