Neural Network-Based Climate Prediction for the 21st Century Using the Finnish Multi-Millennial Tree-Ring Chronology

• Published in: Geosciences
• Main Authors: Elena A. Kasatkina, Oleg I. Shumilov, Mauri Timonen
• Format: Article in Journal/Newspaper
• Language: English
• Published: MDPI AG 2024
• Subjects: climate change; climate prediction; super-long tree-ring chronology; neural networks; LSTM; Arctic; Geology; QE1-996.5; Climate change; Fennoscandia
• Online Access: https://doi.org/10.3390/geosciences14080212; https://doaj.org/article/2e359f51293b42e9846e409e4cc6210d

The sun’s activity role in climate change has become a topic of debate. According to data from the IPCC, the global average temperature has shown an increasing trend since 1850, with an average increase of 0.06 °C/decade. Our analysis of summer temperature records from five weather stations in northern Fennoscandia (65°–70.4° N) revealed an increasing trend, with a range of 0.09 °C/decade to 0.15 °C/decade. However, due to the short duration of instrumental records, it is not possible to accurately assess and predict climate changes on centennial and millennial timescales. In this study, we used the Finnish super-long (~7600 years) tree-ring chronology to create a climate prediction for the 21st century. We applied a method that combines a long short-term memory (LSTM) neural network with the continuous wavelet transform and wavelet filtering in order to make climate change predictions. This approach revealed a significant decrease in tree-ring growth over the near term (2063–2073). The predicted decrease in tree-ring growth (and regional temperature) is thought to be a result of a new grand solar minimum, which may lead to Little Ice Age-like climatic conditions. This result is significant for understanding current climate processes and assessing potential environmental and socio-economic risks on a global and regional level, including in the area of the Arctic shipping routes.