Soil moisture modeling based on stochastic behavior of forces on a no-till chisel opener

• Autoregressive error function combined with neural networks and neuro-fuzzy model were evaluated for soil moisture modeling.
• Inputs were the AREF parameters, speed and angle between chisel force components.
• All models estimated soil moisture better than multiple linear regressions.

Crop-yield variability is frequently associated with soil moisture and its real-time measurement can be an alternative for the automatic control of no-till seeding to improve soil–crop conditions. Soil moisture has a significant influence on soil behavior, markedly on its temporal and spatial variability; however, the measurement of soil moisture is generally time consuming and expensive. Many studies employ electric, electromagnetic, optical, or radiometric sensors for the direct measurement of soil moisture. It is also possible to develop an estimation method employing existing machinery components using mechanical sensors such as load cells. Auto-regressive error function (AREF) combined with computational models is applied in this study for estimating soil moisture using a data set of forces acting on a chisel and speed as inputs to assess the feasibility of achieving more accurate results than previously obtained by Sakai et al. (2005). AREF is a stochastic method that can be applied to the analysis of soil-force patterns acting on a tool. Three computational models are developed, including two artificial neural networks (a Multi-Layer Perceptron (MLP) and a Radial Basis Function (RBF)) and one Neuro-Fuzzy model (ANFIS). These are compared with two multiple linear regression (MLR) models with two and six independent variables. The models’ performances are evaluated using root mean square error (RMSE), determination coefficient (R2), and average percentage error (APE). The computational models demonstrated superior performance compared to MLR, confirming the hypothesis. The neural network models had similar performances with RMSE between 1.27% and 1.30%, R2 around 0.80, and APE between 3.77% and 3.75% for testing data. These results indicate that using AREF parameters combined with computational models may be a suitable technique to estimate soil moisture and has potential to be used in control systems applied to no-till machinery.

Figure optionsDownload as PowerPoint slide