A bidirectional long short-term memory (LSTM) neural network was employed for the first time for past temperature field reconstructions. The LSTM method tested in our experiments using a limited calibration and validation dataset shows worse reconstruction skills compared to traditional reconstruction methods. However, a certain degree of reconstruction performance achieved by the nonlinear LSTM method shows that skill can be achieved even when using small samples with limited datasets.

To understand the natural characteristics and future changes of the global extreme daily precipitation, it is necessary to explore the long-term characteristics of extreme daily precipitation. Here, we used climate simulations to analyze the characteristics and long-term changes of extreme precipitation during the past 3351 years. Our findings indicate that extreme precipitation in the past is associated with internal climate variability and regional surface temperatures.

The analysis of the dynamics of western central Mediterranean droughts for 850–2099 CE in the Community Earth System Model indicates that past Mediterranean droughts were driven by the internal variability. This internal variability is more important during the initial years of droughts. During the transition years, the longevity of droughts is defined by the land–atmosphere feedbacks. In the future, this land–atmosphere feedbacks are intensified, causing a constant dryness over the region.