A mixed-effects regression modeling approach for evaluating paddy soil productivity

Abstract

Soil productivity (SP) is a description of the soil's inherent capacity for crop production and approximates the long-term average crop yield. Knowledge of the key driving factors of SP is essential for short-term soil management and long-term agricultural sustainability. Representative 50-cm intact soil profiles from high-, moderate-, and low-yielding paddy fields with long rice (Oryza sativa L.)-production histories were collected in southern China. Each profile was stratified into 10 layers at 5-cm intervals. Multiple linear (MLM) and mixed-effects (MEM) regression models were developed from the basic soil properties, with the MEM using four different combinations of soil depth of sampling, to evaluate paddy SP. Soil cation exchange capacity (CEC), Ca2+, K+, available potassium (AVK), pH, and clay content were correlated with SP (n = 60, r = 0.25-0.59, p < 0.05), while soil organic C and N contents were poorly related to SP (r = 0.03-0.07, p > 0.05). A MEM with three fixed effects [log (AVK), CEC, and pH] and two random effects [log (Na+) and clay] with two-layer stratification (0-20 and 20-50 cm) best estimated SP (n = 12, R2 = 0.96, p < 0.001). We concluded that the combination of soil stratification and mixed effects could make SP assessment in paddy fields more efficient.