Modelling and reconstructing tree ring growth index with climate variables through artificial intelligence and statistical methods

Abstract

Climate variables play an important role in the increase of tree ring width (TRW), which is one of the primary paleoclimate signals. In this study, we apply new methods for understanding tree growth responses to climate variables under anthropogenic climate change. This study uses regression and artificial intelligence (AI) modelling techniques to develop a model describing the relationships between climate variables and the tree ring standardised growth index (TRSGI) in northeast South Korea. We examined data from 1901 to 1998, and then used data from 1999 to 2019 to reconstruct the TRSGI. This study comprised four major steps. In the first step, we evaluated the performance of the Climate Reach Unit (CRU) TS4.03 in comparison to synoptic station in situ data by using three well-known bias correction methods, namely delta, quantile mapping, and empirical quantile mapping. CRU TS4.03 obtained the highest cross-correlation coefficients (r2 > 0.90) within the synoptic station data. In the second step, different temperatures, precipitation, vapour pressure, and two drought indices were analysed using TRSGI. In the third step, the Mann-Kendall test was run to detect climate and TRSGI data trends, and it showed that all variables had increasing trends at a significance level of 0.05 during 1901–1998. Finally, in the fourth and final step, by selecting the most significant factors from the statistical tests, four models were developed: multiple linear regression (MLR), stepwise regression (SR), nonlinear autoregressive with exogenous input (NARX), and NARX de-noised wavelet, the last of which was a model we developed by combining NARX and de-noised wavelet models. The results indicated that MLR with r = 0.44 (p < 0.003) and SR with r = 0.27 (p < 0.001) were not strong enough to reconstruct TRSGI. In contrast, the NARX de-noised model (r = 0.80) performed better than the NARX model (r = 0.78), showing highly similar trends to the observed TRSGI in the case study area. This research highlights the high performance of AI approaches for modelling TRW data in comparison to regression models. AI techniques may improve the accuracy of climate variable estimations and reconstructions that use TRW data. The presented artificial neural network method has the potential to be a highly effective method for TRW studies.