State-space prediction of soil respiration time series in temperate, semi-arid grassland in northern China

The prediction of soil respiration (Rs) has traditionally been studied using classical statistical methods. These methods do not consider temporal/spatial coordinates and assume independence between samples. The aim was to determine the primary factors influencing Rs and to develop a state-space model able to predict soil respiration. This study was conducted during one growing season, from July to October 2010, in temperate, semi-arid grassland. Data were collected for Rs, air temperature, soil surface temperature, soil temperature at a depth of 5 cm, soil moisture, air pressure, and relative humidity. Additionally, a novel autoregressive state-space method was used to simulate and predict Rs based on primary factors, and the quality of prediction was compared with the quality of prediction using classical statistics. Soil surface temperature and soil moisture were identified as primary factors affecting Rs. The state-space model that included soil surface temperature was a simple but effective model, accounting for 95% of the variation in Rs. The classical statistical models, however, represented only 39–69% of the variation in Rs. Furthermore, the quality of prediction of the state-space models was consistently much better than the quality from the classical statistical methods. State-space analysis is an effective tool for studying the temporal relationships between soil respiration and influencing factors. Additionally, the state-space method is recommended for predicting soil respiration using soil surface temperature in semi-arid grassland in northern China.