Modeling and Prediction of Land use/ Land Cover Changes Based on Spatio-temporal Analysis of Optical and Radar Remotely Sensed Images

The earth and land surface have always changed over time. By identifying and modeling of types of these changes, accurate and up-to-date information will be available from the LULC. It is even possible to predict future land cover changes by considering geographical rules and patterns of change behavior. Satellite sensors are one of the most important remote sensing tools that provide images with different resolutions and regular and extensive data from land surfaces. On the other hand, the combination of these images, especially radar and optical images, according to the specific characteristics of each of them, has had an important impact on the development of methods for detecting and modeling changes. In this research, by using satellite images time series, the pattern of the spatial and temporal behavior of LULC changes in a specific time period was monitored and changes in land surface were modeled. Then LULC changes in the future time are predicted using the spatial-temporal model. The presented model used a spatial autoregressive relationship based on the seasonal and time series trend components. Phenological and temporal changes were modeled by determining the spatial correlation of changes in land cover by the remaining component. Using regression relationships, phenological and temporal changes were modeled according to the spatial correlation of changes in land cover by the component. Finally, considering all changes (gradual, seasonal, and abrupt) according to the designed regression function, a cubic spatial neighborhood window was defined. Array cubes (spatial-temporal) moved on the whole image. In each move, on the one hand, regression parameters were estimated, and by applying these estimates to known data, on the other hand, land cover at the next time was predicted by the spatio-temporal model. The main focus of this research is to recognize and predict the behavior of land cover changes over time by considering the spatial correlation of changes in neighboring pixels within a spatial-temporal model. In this regard, a combination of radar and optical images is used to simultaneously use the capabilities of these images in order to detect the types of land cover and identify their changes. Radar and Optical images are combined with an innovative approach based on a combination of feature-level and decision-level fusion. In the proposed method, the integration is performed based on the definition of a linear relationship between the feature extracted from radar images and optical images in a time series. In the process of combining images, thresholds have been used, which have been determined based on the features extracted from radar and optical images. The threshold for all land cover was determined experimentally by analyzing features extracted during the seasons (rainy and dry). To evaluate the accuracy of the predicted results with the approach presented in this study, the predicted results of land cover changes were compared with the results of land cover classification from the RF algorithm. According to this evaluation, the accuracy of forecasting changes for water bodies, vegetation, barren lands, and salt marsh was estimated to be 82.5, 94.9, 86.8, and 94.7%, respectively.