In recent decades, global warming has significantly altered both the spatial and temporal distribution of rainfall patterns. This change has heightened the risk of rainfall-induced landslides, which are the most prevalent natural disasters in the mountainous regions of southwestern China. These events pose unpredictable and severe threats to the region, making it essential to forecast future rainfall trends and assess how landslide susceptibility will respond to these changes. Understanding these dynamics is crucial for developing effective strategies to mitigate and adapt to the changing rainfall patterns that influence landslides. This study focuses on Sichuan Province, China, and uses annual cumulative rainfall (ACR) as a key dynamic variable to create landslide susceptibility maps (LSMs). The goal is to explore the evolving relationship between rainfall and landslide susceptibility and use future rainfall projections to predict these risks. To achieve this, a historical landslide geospatial database was compiled across five temporal categories: 2000, 2001–2005, 2006–2010, 2011–2015, and 2016–2020. The extreme learning machine (ELM) was applied to generate LSMs for the years 2000 to2020, while an elasticity framework was used to assess how sensitive landslide susceptibility is to rainfall variations. To project future scenarios, a long short-term memory (LSTM) model was employed to project the ACR for 2030, using monthly rainfall data from 2000 to 2020. This projected ACR was then used to estimate future landslide susceptibility. Results showed a marked increase in high landslide susceptibility areas: 5.6 % by 2005, 0.3 % by 2010, 0.2 % by 2015, and 12.9 % by 2020, all relative to the year 2000. The elasticity analysis revealed that from 2000 to 2020, a 1 % change in rainfall would cause an average 1.35 % change in landslide susceptibility. Looking forward to 2030, the projected rise in ACR is expected to lead to a 2.44 % increase in areas of high landslide susceptibility. Multiple validation techniques were applied to ensure reliability and robustness of these findings.