Machine learning applications to reveal the difference in <i>Robinia pseudoacacia</i> growth and its drivers on China’s Loess Plateau

Abstract The decline in tree growth has become a global issue. It is critically important to explore the factors affecting tree growth under the background of global climate change to understand tree growth models. A database was established based on Robinia pseudoacacia growth and its driving factors on China’s Loess Plateau. Linear regression and three machine learning methods, including support vector machine, random forest (RF) and gradient boosting machine were used to develop R. pseudoacacia growth models considering forest age, density, climate factors and topographic factors. The root mean square deviation method was adopted to quantitatively assess the relationship between tree growth and soil properties. The average tree height of R. pseudoacacia on the Loess Plateau was 8.8 ± 0.1 m, the average diameter at breast height (DBH) was 10.4 ± 0.1 cm and the average crown diameter was 3.2 ± 0.1 m. The RF model was a fast and effective machine learning method for predicting R. pseudoacacia growth, which showed the best simulation capability and could account for 67% of tree height variability and 55% of DBH variability. Model importance indicated that forest age and stand density were the main factors predicting R. pseudoacacia growth, followed by climate factors. The trade-off between R. pseudoacacia growth and soil properties revealed that soil texture and soil pH were the primary determinants of R. pseudoacacia growth in this region. Our synthesis provides a good framework for sustainable forest management in vulnerable ecological areas under future climate change.