Iranian Remote Sensing and GIS Society / Shahid Beheshti University Iranian Journal of Remote Sensing and GIS 2008-5966 17 4 2026 01 21 Evaluating the Efficiency of InSAR Coherence in Crop Type Mapping Using Machine Learning Evaluating the Efficiency of InSAR Coherence in Crop Type Mapping Using Machine Learning 133 160 105853 10.48308/gisj.2025.238522.1249 FA Fatemeh Amiri Dep of Remote Sensing and GIS, Faculty of Humanities, Tarbiat Modares University, Tehran, Iran 0009-0006-3591-6934 Ali Shamsoddini Dep of Remote Sensing and GIS, Faculty of Humanities, Tarbiat Modares University, Tehran, Iran 0000-0002-4559-7563 Mohamad Sharifikia Dep of Remote Sensing and GIS, Faculty of Humanities, Tarbiat Modares University, Tehran, Iran Journal Article 2025 01 25 Introduction: The optimal use of agricultural land is a key concern for authorities due to agriculture's significant role in job creation, foreign exchange earnings, ensuring food security, and reducing reliance on imports. Gathering information about the spatial distribution and cultivated areas of various crops can enhance their efficient usage. One effective method for obtaining this information is through satellite imagery. Remote sensing technology, with its ability to provide high-resolution images and extensive spatial and temporal coverage, has become a dominant approach for crop type mapping. One of the remote sensing data that has recently received attention in the field of crop type mapping is the interferometric coherence images of synthetic aperture radar (InSAR). The sensitivity of these images to crop’s structure, making them valuable for monitoring and mapping crop types. In global literature, InSAR coherence images have been widely used in research related to agricultural products. However, in Iran, the use of coherence data for monitoring phenology and distinguishing different crops has not received much attention, despite its unique capabilities. Therefore, evaluating the efficiency of coherence data and its potential for adopting optimal agricultural management policies in Iran can be highly beneficial. Methodology: The main objective of this study is to evaluate the efficiency of machine learning-based InSAR coherence data for crop type mapping. To achieve this, a one-year time series of Synthetic Aperture Radar (SAR) data was compiled from Sentinel-1 phase information for the 2019 crop year, for the Ardabil plain, located to the west and northwest of Ardabil city. A network of SAR image pairs with short spatial and temporal baselines was created to produce coherence data. Field data were collected from 1,358 fields containing various crops. To avoid mixed pixels, a 10-meter buffer was established around the edges of each crop field. A total of 156,026 pixels from the coherence images were sampled and randomly divided into three groups: training (70%), validation (15%), and test (15%). To select the appropriate time interval for using coherence images, the phenological response of the crops to the InSAR coherence was analyzed. During the time interval, the phenological signals of the studied crops were compared with the signals of the built-up areas and bare soil to ensure that they were not mixed. Consequently, the multi-temporal InSAR coherence values in the selected time interval were used as input to the Support Vector Machine (SVM) classifier with different kernels to distinguish and identify the type of crops. Result: The study of the coherence time series values in the selected control areas revealed distinct differences in the coherence behavior of various crops when compared to one another, as well as in comparison to both built-up and bare soil areas. The InSAR coherence data match well with the main phenological stages of the crops. Among the different SVM kernels tested, the radial basis function (RBF) kernel achieved the highest overall accuracy of 59.69% during the validation phase, utilizing various combinations of the parameters c and gamma. In the testing phase, the crop type map produced using the SVM classifier with the RBF kernel reached an overall accuracy of 60.6%. This model performed best in identifying wheat and least effectively in identifying alfalfa. User accuracy was notably higher for wheat and potato plants, while it was lower for corn, broad bean, and alfalfa. Conclusion: Coherence images offer valuable insights for identifying and classifying crops in Iran. Leveraging machine learning techniques can enhance the utility of coherence data in monitoring and categorizing different crop types. Several factors influence the effectiveness of coherence images and the performance of classification algorithms, including the number of training samples available for each crop, the number of coherence features, the use of complementary data, sensor parallax (spatial baseline), topographical features (slope and aspect), the temporal resolution, and the classification algorithm. These characteristics should be carefully considered to optimize the analysis. Introduction: The optimal use of agricultural land is a key concern for authorities due to agriculture's significant role in job creation, foreign exchange earnings, ensuring food security, and reducing reliance on imports. Gathering information about the spatial distribution and cultivated areas of various crops can enhance their efficient usage. One effective method for obtaining this information is through satellite imagery. Remote sensing technology, with its ability to provide high-resolution images and extensive spatial and temporal coverage, has become a dominant approach for crop type mapping. One of the remote sensing data that has recently received attention in the field of crop type mapping is the interferometric coherence images of synthetic aperture radar (InSAR). The sensitivity of these images to crop’s structure, making them valuable for monitoring and mapping crop types. In global literature, InSAR coherence images have been widely used in research related to agricultural products. However, in Iran, the use of coherence data for monitoring phenology and distinguishing different crops has not received much attention, despite its unique capabilities. Therefore, evaluating the efficiency of coherence data and its potential for adopting optimal agricultural management policies in Iran can be highly beneficial. Methodology: The main objective of this study is to evaluate the efficiency of machine learning-based InSAR coherence data for crop type mapping. To achieve this, a one-year time series of Synthetic Aperture Radar (SAR) data was compiled from Sentinel-1 phase information for the 2019 crop year, for the Ardabil plain, located to the west and northwest of Ardabil city. A network of SAR image pairs with short spatial and temporal baselines was created to produce coherence data. Field data were collected from 1,358 fields containing various crops. To avoid mixed pixels, a 10-meter buffer was established around the edges of each crop field. A total of 156,026 pixels from the coherence images were sampled and randomly divided into three groups: training (70%), validation (15%), and test (15%). To select the appropriate time interval for using coherence images, the phenological response of the crops to the InSAR coherence was analyzed. During the time interval, the phenological signals of the studied crops were compared with the signals of the built-up areas and bare soil to ensure that they were not mixed. Consequently, the multi-temporal InSAR coherence values in the selected time interval were used as input to the Support Vector Machine (SVM) classifier with different kernels to distinguish and identify the type of crops. Result: The study of the coherence time series values in the selected control areas revealed distinct differences in the coherence behavior of various crops when compared to one another, as well as in comparison to both built-up and bare soil areas. The InSAR coherence data match well with the main phenological stages of the crops. Among the different SVM kernels tested, the radial basis function (RBF) kernel achieved the highest overall accuracy of 59.69% during the validation phase, utilizing various combinations of the parameters c and gamma. In the testing phase, the crop type map produced using the SVM classifier with the RBF kernel reached an overall accuracy of 60.6%. This model performed best in identifying wheat and least effectively in identifying alfalfa. User accuracy was notably higher for wheat and potato plants, while it was lower for corn, broad bean, and alfalfa. Conclusion: Coherence images offer valuable insights for identifying and classifying crops in Iran. Leveraging machine learning techniques can enhance the utility of coherence data in monitoring and categorizing different crop types. Several factors influence the effectiveness of coherence images and the performance of classification algorithms, including the number of training samples available for each crop, the number of coherence features, the use of complementary data, sensor parallax (spatial baseline), topographical features (slope and aspect), the temporal resolution, and the classification algorithm. These characteristics should be carefully considered to optimize the analysis.

https://gisj.sbu.ac.ir/article_105853_525b84f7df0e8063ec6d9a44d409cad7.pdf