Estimating Soil Organic Matter and Total Nitrogen Losses Risk from Arable Lands: A Sight with Comparison of Model Learning Methods
Abstract
Extreme hydrologic events, including drought and flooding, could significantly influence soil organic matter (SOM) and soil total nitrogen (STN) of arable lands, resulting in carbon (C) emissions to the atmosphere and nitrogen (N) loss downstream. Modeling could be a practical approach to estimating the effects of extreme hydrologic on SOM and STN dynamics. However, studies on model development of SOM and STN and its differences between wheat and paddy lands after a long cultivation history (i.e. 20-year) are still limited. Hence, a comparison model of SOM and STN from thirteen soil health monitoring stations in China (0-170 cm with different soil layers) were built by six different learning methods with air temperatures, precipitation, humidity, and atmospheric pressure [i.e., the annual (semiannual/quarterly/ monthly) average (maximum/ minimum) value of a climate factor], and regressions between SOM, STN, climate factors, as well as latitude were analyzed. The models were evaluated by the root mean squared error (RMSE), mean deviation (RMD), mean absolute error (MAE), and model effectiveness (EF). Results showed that the SOM content in wheat lands (17.04±0.17 g kg-1) was significantly lower than that in paddy lands (25.12±0.28 g kg-1), while the STN content in wheat lands (10.8±0.01 g kg-1) was significantly higher than that in paddy lands (1.54±0.02 g kg-1). The XGBoost, LightGBM, Random forest, and Decision tree had better performances than Support Vector Regression and Linear learning methods with the highest r2 of 0.88 (ranged from 0.65 to 0.88, n=5953) for both SOM and STN models. Extreme air temperatures, including monthly, quarterly, and annual maximum values, negatively correlated with SOM and STN. This indicated severe drought of 2022 in southern China, resulting from high air temperatures could lead to unexpected soil C and N losses. Additionally, latitude had a significant negative relation with both SOM and STN, indicating that the losses of soil C and N from arable lands in southern China were higher than in the northern region. It is suggested that a comparison of different machine learning methods will help develop a better model for similar studies.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.U25C2211L