Towards an Improved Algorithm for Estimating Freeze-Thaw Dates of a High Latitude Lake Using Texture Analysis of SAR Images

Analyzing the freeze-thaw dates of high latitude lakes is an important part of climate change studies. Due to the various advantages provided by the use of SAR images, with respect to remote monitoring of small lakes, SAR image analysis is an obvious choice to estimate lake freeze-thaw dates. An important property of SAR images is its texture. The problem of estimating freeze-thaw dates can be restated as a problem of classifying an annual time series of SAR images based on the presence or absence of ice. We analyzed a few algorithms based on texture to improve the estimation of freeze-thaw dates for small lakes using SAR images. We computed the Gray Level Co-occurrence Matrix (GLCM) for each image and extracted ten different texture features from the GLCM. We used these texture features (namely, Energy, Contrast, Correlation, Homogeneity, Entropy, Autocorrelation, Dissimilarity, Cluster Shade, Cluster Prominence and Maximum Probability as previously used in studies related to the texture analysis of SAR sea ice imagery) as input to a group of classification algorithms to find the most accurate classifier and set of texture features that can help to decide the presence or absence of ice on the lake. The accuracy of the estimated freeze-thaw dates is dependent on the accuracy of the classifier. It is considered highly difficult to differentiate between open water (without wind) and the first day of ice formed on the lake (due to the similar mean backscatter values) causing inaccuracy in calculating the freeze date. Similar inaccuracy in calculating the thaw date arise due to the close backscatter values of water (with wind) and later stages of ice on the lake. Our method is promising but requires further research in improving the accuracy of the classifiers and selecting the input features.