Main content area

Optimizing individual tree detection accuracy and measuring forest uniformity in coconut (Cocos nucifera L.) plantations using airborne laser scanning

Mohan, Midhun, Mendonça, Bruno Araujo Furtado de, Silva, Carlos Alberto, Klauberg, Carine, de Saboya Ribeiro, Acauã Santos, Araújo, Emanuel José Gomes de, Monte, Marco Antonio, Cardil, Adrián
Ecological modelling 2019 v.409 pp. 108736
Cocos nucifera, algorithms, canopy, canopy height, coconuts, crop production, forest inventory, forest plantations, forests, fruits, lidar, markets, models, monitoring, remote sensing, tree crown, tree growth, trees, Brazil
Forest inventory and monitoring is indispensable for coconut (Cocos nucifera L.) forest plantation owners as it allows them in assessing tree growth, fruit production rates and vitality of plantations, as well as assists to meet up with the rising market demands by ensuring better yields. Nonetheless, the use of remote sensing techniques for optimizing the management and production of coconut plantations is still at a latent stage. In this paper, we present an airborne laser scanning (lidar) based tree detection method applied for automatically identifying individual coconut trees in a plantation in southeast Brazil. This method locates individual trees by searching treetops on canopy height models (CHM) derived from lidar data through a moving window having fixed treetop window size (TWS). Here, an adaptive TWS approach was implemented as a function of lidar-derived canopy cover (COV, %) along with additional smoothening window sizes (SWS) for enhancing tree detection accuracy. A total of 19 plots characterized by varying levels of canopy cover were used for testing the accuracy of our framework and we were able to obtain an average tree detection accuracy of 86.22%. From a total of 341 trees, 294 trees were detected correctly by the algorithm using adaptive TWS. A low TWS (3x3) value was found to perform best in study plots having COV > 80% and for rest of the cases, a higher TWS (7x7) was perceived suitable. Results were further analyzed and compared, to evaluate the relationship of Individual Tree Detection (ITD) accuracy with varying canopy cover levels, within-plot tree distribution patterns, TWS and SWS values. Proceedings from our work show that appropriate combination of lidar-derived CHMs, local maxima (LM) algorithm and window sizes have the potential to satisfactorily (F-score ˜ 0.90) detect plantation species having irregular canopies such as coconut trees. As an extension of the ITD, estimation of forest uniformity, which gives a measure of the level of homogeneity or heterogeneity of stands, is also performed.