Retrieval of eucalyptus planting history and stand age using random localization segmentation and continuous land-cover classification based on Landsat time-series data

Obtaining robust change-detection results and reconstructing planting history are important bases for conducting forest resource monitoring and management. The existence of multiple change points in a very short period can lead to a global segmentation method incorrectly locate the change points, because they could impact each other during model initialization. This is especially true for monitoring plantations such as eucalyptus, which has a unique growth cycle with short rotation periods and frequent disturbances. In this study, we proposed a method to find critical change points in a normalized difference vegetation index (NDVI) time series by combining random localization segmentation and the Chow test. Features of the NDVI time series calculated on the divided segments and change points were used to train a Random Forest classifier for continuous land-cover classification. The proposed method was successfully applied to a eucalyptus plantation for identifying the management history, including harvest time, generation, rotation cycle, and stand age. The results show that our method is robust for different lengths of NDVI time series, and can detect short-interval (cut and stability) change points more accurately than the global segmentation method. The overall accuracy of identification was 80.5%, and successive generations in 2021 were mainly first- and second-generation, accounting for 69.0% and 27.9% of the total eucalyptus area, respectively. The rotation cycle of eucalyptus plantation is usually 5-8 years for 66.9% of the total area. The eucalyptus age was accurately estimated with an R-2 value of 0.91 and RMSE of 13.3 months. One-year-old eucalyptus plantations accounted for the highest percentage of 14.5%, followed by seven-year-old plantations (12.9%). This study provides an important research basis for accurately monitoring the rotation processes of short-period plantations, assessing their timber yield and conducting carbon- and water-cycle research.