5:15 PM - 7:15 PM
[MIS14-P03] Automatic sedimentary organic matter (Palynofacies) analysis using AI to deciphering terrestrial vegetation changes during the Cretaceous Ocean Anoxic Events
Keywords:AI, Vegetation, Cretaceous, OAE
The mid-Cretaceous “hothouse” period is characterized by high atmospheric pCO2 and frequent occurrences of global-scale environmental change such as Ocean Anoxic Events (OAEs). The OAEs are thought to be caused by increased atmospheric pCO2 released by Large Igneous Provinces (LIPs), subsequent enhanced chemical weathering, and expansion of anoxic marine environments due to the increased surface biological production. However, the terrestrial stratigraphic record during the period of OAEs is limited. There have been only a few studies reconstructing the terrestrial environmental changes during the Cretaceous OAEs, such as reconstruction of vegetation change based on the sedimentary organic matter composition (Palynofacies) analysis in marine strata of the OAE2 interval in southern France (Heimhofer et al., 2018) and in lacustrine strata of the Toarcian OAE in northern China (Baker et al., 2017; Li et al., 2023; Huang et al., 2024).
In this study, we performed palynofacies analysis for the Aptian lacustrine strata (Shinekhudag Fm) in Mongolia (Hasegawa et al., 2018, 2022), which has recently been identified as potentially including the OAE1a interval. The samples used in this study are CSH01 core (about 150 m) and CSH02 core (about 190 m) drilled in the Shine Khudag area in 2013 and 2014, respectively. The entire CSH01 core and the lower part of CSH02 core correspond to the Shinekhudag Formation, which consists of alternating layers of shale and dolomite layers, reflecting precipitation/evaporation changes. The upper part of the CSH02 core corresponds to the Khukhteeg Formation, which consists of coal-bearing fluvial strata (Hasegawa et al., 2018).
In order to quantitatively reconstruct the terrestrial vegetation change during OAE1a, we attempted to establish the method of AI-based palynofacies analysis by utilizing the AI-based image recognition system (microfossil Classification and Rapid Accumulation Device: miCRAD) developed by Itaki et al. (2020a,b). First, we used 40 samples of palynofacies slides obtained from the CSH01 core, and compared human observation results (500 particles per slide; Sabine Haase 2020MS) and AI-based analysis results (thousands to tens of thousands of particles per slide). As a result, we found several similarities between human observation and AI results, particularly in the abundance of Phytoclasts (plant fragments) and Botryococcus.
Next, we compared AI-based results of palynofacies abundance with lithology changes. We found abundant pollens and phytoclasts in shale layer, indicating a high lake level environment. On the other hand, Botryococcus tended to increase in the transition from a high lake level (shale layer) to a low lake level (dolomite layer). This is consistent with previous studies that Botryococcus is more tolerant of rapid environmental changes than other algae (Herrmann 2010; Demura et al., 2014). Furthermore, in the case of charcoal, no significant correlation with lithology changes was observed, but a reverse relationship with the AOM abundance was observed. This is consistent with previous studies (Baker et al., 2017) that charcoal is an indicator of wildfire and reflecting higher atmospheric oxygen concentration. In other words, atmospheric oxygen concentrations may have increased in the abundant charcoal layers, suggesting that organic matter decomposition may have increased due to higher oxygen concentrations at the bottom of the lake, resulting in less AOM.
We try to further improve the accuracy of classification by improving the teacher data and conduct 200 samples of AI-based palynofacies analysis covering the period before and after OAE1a using CSH02 cores. We will also compare the results of the palynofasicies analysis with those of carbon and oxygen isotope ratios and elemental and mineral compositions, and attempt to elucidate detailed environmental changes and vegetation evolution during the OAE1a period.
In this study, we performed palynofacies analysis for the Aptian lacustrine strata (Shinekhudag Fm) in Mongolia (Hasegawa et al., 2018, 2022), which has recently been identified as potentially including the OAE1a interval. The samples used in this study are CSH01 core (about 150 m) and CSH02 core (about 190 m) drilled in the Shine Khudag area in 2013 and 2014, respectively. The entire CSH01 core and the lower part of CSH02 core correspond to the Shinekhudag Formation, which consists of alternating layers of shale and dolomite layers, reflecting precipitation/evaporation changes. The upper part of the CSH02 core corresponds to the Khukhteeg Formation, which consists of coal-bearing fluvial strata (Hasegawa et al., 2018).
In order to quantitatively reconstruct the terrestrial vegetation change during OAE1a, we attempted to establish the method of AI-based palynofacies analysis by utilizing the AI-based image recognition system (microfossil Classification and Rapid Accumulation Device: miCRAD) developed by Itaki et al. (2020a,b). First, we used 40 samples of palynofacies slides obtained from the CSH01 core, and compared human observation results (500 particles per slide; Sabine Haase 2020MS) and AI-based analysis results (thousands to tens of thousands of particles per slide). As a result, we found several similarities between human observation and AI results, particularly in the abundance of Phytoclasts (plant fragments) and Botryococcus.
Next, we compared AI-based results of palynofacies abundance with lithology changes. We found abundant pollens and phytoclasts in shale layer, indicating a high lake level environment. On the other hand, Botryococcus tended to increase in the transition from a high lake level (shale layer) to a low lake level (dolomite layer). This is consistent with previous studies that Botryococcus is more tolerant of rapid environmental changes than other algae (Herrmann 2010; Demura et al., 2014). Furthermore, in the case of charcoal, no significant correlation with lithology changes was observed, but a reverse relationship with the AOM abundance was observed. This is consistent with previous studies (Baker et al., 2017) that charcoal is an indicator of wildfire and reflecting higher atmospheric oxygen concentration. In other words, atmospheric oxygen concentrations may have increased in the abundant charcoal layers, suggesting that organic matter decomposition may have increased due to higher oxygen concentrations at the bottom of the lake, resulting in less AOM.
We try to further improve the accuracy of classification by improving the teacher data and conduct 200 samples of AI-based palynofacies analysis covering the period before and after OAE1a using CSH02 cores. We will also compare the results of the palynofasicies analysis with those of carbon and oxygen isotope ratios and elemental and mineral compositions, and attempt to elucidate detailed environmental changes and vegetation evolution during the OAE1a period.