Comparing different k-NN sampling methods for density estimation of wild pistachio (Pistaciaatlantica Desf.) with clustered spatial pattern in a Zagros open stand

Density (i.e. number of trees per unit area) is one of the important structural attributes of forest stands which is appropriate to understand forest dynamics. The k-Nearest Neighbour (k-NN) method is a distance sampling method which is commonly used to estimate quantitative attributes in forest inventories. In this study, the k-NN method with five strategies of Nearest Individual (NI), Nearest Neighbour (NN), Random Pairs (RP), Point-Centered Quarter (PCQ), and Quartered Neighbour (QN) was used to estimate the density of wild pistachio (Pistacia atlantica Desf.) in Zagros woodlands. A natural stand of 45 ha was selected in Bane Research Site, and was fully callipered to derive the true density. The spatial distribution of trees was clustered (R=0.79 and z = -12.38) with a true density of 19.44 trees per ha. While applying the k-NN method, different strategies as well as k ranging between 2 and 10 were tested across 46 sample points in a 100 × 100 m2 sampling grid. The results showed that all strategies except PCQ significantly estimated the density at α=0.05. Furthermore, the number of k and the strategy of k-tree selection affected the accuracy and precision of k-NN results, since NI in k=4, NN in k=7, RP and QN in k=2 estimated the density with the least error (RMSE and Bias). In conclusion, the optimum k-NN method with appropriate k and strategy could estimate the density of wild pistachio trees with clustered spatial distribution in an open stand in Zagros woodlands.