Obtaining tree volume prediction with multivariate adaptive regression splines technique

In this study, single-entry and double-entry tree volume models were developed using the Multivariate Adaptive Regression Curves Technique to estimate the volume of trees in Scots pine (Pinus sylvestris L.) stands located within the boundaries of Sisorta Forest Management Plannig Unit, Koyulhisar Forest Enterprise, Giresun Forest District Directorate. In this study, the "Earth" package, coded with the R programming language (R Core Team 2017), was used to develop and train MARS models for estimating the total stem volumes of the trees. The performance metrics obtained for the single-entry MARS model through the training are as follows: RMSE = 0.06617, MAE = 0.10524, MAE% = 16.93902%, AIC = -181.787, BIC = -128.592, Bias = 5.12x10-9, Bias% = 8.25x10-7%, MAPE = 9.74x10-7%, MSE = 10.6510, and R² = 0.9736. For the double-entry MARS model, the performance metrics are: RMSE = 0.03665, MAE = 0.06932, MAE% = 11.15764%, AIC = -216.238, BIC = -163.043, Bias = 4.28x10-6, Bias% = 0.00069%, MAPE = 0.00099%, MSE = 5.8994, and R² = 0.9885. It has been determined that MARS models, which have been found to provide more successful results than traditional multiple regression models, show a positive increasing trend in both single-entry and double-entry volume estimates. This trend demonstrates an increasingly positive curvilinear development that aligns with expected growth patterns. Therefore, it has been concluded that MARS models not only have a high level of accuracy but also produce estimates that are consistent with growth principles. In MARS models, an important aspect is the identification of knot points. For single-entry predictions, the knot points were determined to be at 23 cm, 31 cm, and 38 cm. For the double-entry MARS model, a height knot point was found at 12.0815 m, and diameter knot points were identified at 19 cm, 32 cm, and 42 cm.