AbstractThe discrepancies among the variations in global ice volume, cave stalagmite δ¹⁸O and rainfall reconstructed by cosmogenic ¹⁰Be tremendously restrain our understanding of the evolution of the East Asian summer monsoon (EASM). Here, we present a 430-ka EASM mean annual precipitation record on the Chinese Loess Plateau obtained using branched glycerol dialkyl glycerol tetraethers based on a deep learning neural network; this rainfall record corresponds well with cave-derived δ¹⁸O data from southern China but differs from precipitation reconstructed by ¹⁰Be. Both branched tetraether membrane lipids and cave δ¹⁸O may be affected by soil moisture and atmospheric temperature when glacial and interglacial conditions alternated and were thus decoupled from atmospheric precipitation; instead, they represent variations in the intensity of the EASM. Furthermore, we demonstrate that the brGDGT-DLNN method can significantly extend the temporal scale record of the EASM and is not restricted by geographic location compared with stalagmite records.