Assessing the reforestation effects of plantation plots in the Thai savanna based on 45 cm resolution true-color images and machine learning

Abstract

In forest restoration projects, assessing progress is critically important. This study was conducted to quantify the restoration progress for approximately 30 year old plantation plots of Acacia and Pterocarpus in Sakaerat, Thailand. For 22.5 × 22.5 m areas in the plantation plots, as a continuous measure of forest restoration, dry evergreen forest-likelihood was determined using 45 cm resolution satellite images and an artificial neural network architecture. The Acacia plot achieved a moderate mean evergreen forest likelihood of 0.451 (April 2016) to 0.679 (May 2019), but the values for the Pterocarpus plot were 0.076 or smaller. Two other Pterocarpus plots at cooler and moister sites had mean evergreen forest-likelihood values of up to 0.884 (April 2016), which were significantly greater than that (0.451) for the Acacia plot, but the values dropped in May 2019. Throughout the period of the four image acquisition times from March 2014 to May 2019, the plantation plots had significantly smaller evergreen forest-likelihood than the dry evergreen forest did. The current approach would be a helpful option for stakeholders of forest management by applying sub-m-resolution images and machine learning followed by quantification of the forest restoration effects relying on the appearance and texture of tree canopies at multiple data acquisition times.