Wed, Aug 17, 2022: 11:00 AM-11:15 AM
515B
Background/Question/MethodsUnderstanding soil carbon (C) and nitrogen (N) cycles and their complicated responses to climate and environmental changes are critical in ecology. Soil moisture change directly regulates microbial activities, thus impacting microbially-mediated soil C-N cycles. Therefore, accurate estimate of soil moisture is critical to better modeling soil C-N cycles in response to environmental changes. Here we attempt to achieve more accurate estimate of daily soil moisture and its effect on soil microbial and C-N dynamics in a long-term grassland experiment (BioCON) in Minnesota, USA. We first adopted the Gradient Boosting Machine (GBM), a machine learning approach, to predict continuous daily soil moisture from antecedent n-day average meteorological data and the measured soil moisture at four different layers (0-20cm; 22-39cm; 42-59cm; 83-100cm) at a weekly time interval. We then use the generated daily soil moisture as input to drive the Microbial-Enzyme Decomposition (MEND) model. We aim to examine the response of soil microbial and C-N processes to soil moisture variation in this grassland system.
Results/ConclusionsThe GBM model results show that the predicted soil moisture agreed well with the observations in both model calibration (R2 = 0.7–0.9) and validation (R2 = 0.55–0.8) across the four soil layers. Apart from predicting the daily soil moisture accurately, the model demonstrated that different layers were related to distinct time-lags (12–56 days) of the meteorological data (precipitation, air temperature, shortwave radiation, and vapor pressure), supporting our hypothesis that time-lags increase with the soil depth. The GBM-based variable importance indicates that precipitation and shortwave radiation are the most important predictors for the upper two soil layers (0–42 cm), and for the lower two soil layers (42–100 cm), precipitation is still the most important predictor, but temperature is more important than shortwave radiation. The MEND modeling further showed strong microbial responses to the changes in soil moisture, e.g. the proportion of active fraction and the rate of organic matter decomposition are higher in a relatively moist soil environment, resulting in lower soil organic carbon. This study presents an efficient machine-learning approach for estimating daily soil moisture from meteorological data. Our results also implied the significance of soil moisture in controlling microbially-mediated soil biogeochemical processes.
Results/ConclusionsThe GBM model results show that the predicted soil moisture agreed well with the observations in both model calibration (R2 = 0.7–0.9) and validation (R2 = 0.55–0.8) across the four soil layers. Apart from predicting the daily soil moisture accurately, the model demonstrated that different layers were related to distinct time-lags (12–56 days) of the meteorological data (precipitation, air temperature, shortwave radiation, and vapor pressure), supporting our hypothesis that time-lags increase with the soil depth. The GBM-based variable importance indicates that precipitation and shortwave radiation are the most important predictors for the upper two soil layers (0–42 cm), and for the lower two soil layers (42–100 cm), precipitation is still the most important predictor, but temperature is more important than shortwave radiation. The MEND modeling further showed strong microbial responses to the changes in soil moisture, e.g. the proportion of active fraction and the rate of organic matter decomposition are higher in a relatively moist soil environment, resulting in lower soil organic carbon. This study presents an efficient machine-learning approach for estimating daily soil moisture from meteorological data. Our results also implied the significance of soil moisture in controlling microbially-mediated soil biogeochemical processes.