Agricultural Crop Monitoring Using Polarimetric Interferometric Synthetic Aperture Radar Images

Increasing the agricultural crops due to climatic conditions, limitation of water resources, limitation of suitable agricultural lands, as well as financial constraints in the country is faced with a lot of problems. Therefore, in order to provide food, the efficiency of the production factors, especially of the water and soil, should be increased. This requires regular monitoring of crops. Remote sensing is one of the most important techniques used in agricultural crop monitoring. SAR remote sensing can bridge the gap between the need for crop information over large scales and the necessity of frequent observations. SAR observables are sensitive to the various characteristics of crops. Today, developing the monitoring methods in the large scale is an important issue for reasonable management of natural resources, especially for the populous countries. The purpose of this study is to monitor and retrieve some parameters of agricultural crops using time series of polarimetric interferometric synthetic aperture radar (PolInSAR) images. The time series of PolInSAR data include intensity, polarimetric, and interferometric information that reflect a large amount of information on various crops. The information obtained from the optics data, the intensity and the polarimetric data is not suitable for retrieving some of the crop parameters such as the height. However, the interferometric data can play a role in monitoring and retrieving these parameters during the growing season. In the present study, the proposed monitoring method is based on the derived features of a decomposition model and regression based methods. In this method, first, an optimal polarization base with the maximum correlation between the slave and master images is calculated. Then eigenvalue decomposition is applied to the interferometric polarimetric covariance matrix in that optimal base, and the features such as entropy and alpha are calculated. Some of these features have a high linear relationship with height, biomass and phenology, and others provide useful information for improving the estimation performance. Finally, the crop parameters are estimated based on the 13 PolInSAR features and also the artificial neural network and support vector regression. The validation analysis is carried out using the images of E-SAR sensor of the DEMMIN region in Germany. These images are acquired between May and June of 2006 and the ground data during the growth cycle is available. The results for wheat and barley crops indicate the good performance of the proposed method in monitoring and retrieving the parameters. Both methods used for estimation including neural network and support vector regression have the good estimates of crop parameters and can be used to monitor the crops. For example for wheat, the RMSE values were 0.21, 0.59 and 0.21, using neural network and 0.21, 0.52 and 0.46, using support vector regression, in height, biomass and phenology estimation, respectively. The estimation results for height and phenology are better than the biomass. Also, using the neural network in the estimation has a relatively higher computational cost. The proposed method can be an appropriate alternative to the experimental and physical models available in the estimation of parameters such as height, which due to the lack of data with suitable baseline can not be used.