Fusion of multi soil data for the delineation of management zones for variable rate irrigation

Up until now, there have been no multi-sensor approaches used to estimate available water content (AWC) in order to determine variable rate irrigation. This has been a major problem for growers adopting precision farming technologies. The aim of this project is to implement an on-line multi-sensor platform and data fusion approach for the delineation of management zones for site specific irrigation in vegetable crop production systems. This is performed by simultaneous measurement of soil moisture content (MC), organic carbon (OC), clay content (CC), plasticity index (PI) and bulk density (BD) with an on-line visible and near infrared (vis-NIR) spectroscopy sensor and a load cell attached to a subsoiler and frame, which was linked to a three-point linkage of a tractor. The soil apparent Electrical Conductivity (ECa) was separately measured with an Electromagnetic Induction (EMI) device. Measurements were carried out in three fields in Lincolnshire and one in Cambridgeshire. Vis-NIR calibration models of soil properties were developed using partial least squares (PLS) regression. A multiple linear regression analysis (MLR) and an Artificial Neural Network (ANN) was used to derive zones of water holding capacity (WHC), based on correlation between on-line measured OC, CC, PI, BD and ECa with MC. The AWC was calculated with empirical equations, as a function of clay and sand fractions. Result showed that the on-line measurement accuracy for OC and MC were good to excellent (R2=0.71-0.83 and R2=0.75-0.85, RPD=2.00-2.57 and RPD=1.94-2.10 for OC and MC, respectively). For CC and PI, the measurement accuracy (R2=0.64-0.69 and RPD=0.55-0.66 for clay content and PI) was evaluated as moderate. It was observed in the study fields, that the ECa results had a minor response to MC distribution. Furthermore, the fusion of multi-soil data to derive a WHC index with MLR and ANN resulted in successful delineation of homogeneous zones. These were divided into four different normalisation categories of low (0 – 0.25), medium (0.25 – 0.5), high (0.5 – 0.75) and very high (0.75 – 1) of WHC. Spatial similarity between WHC maps with those of CC, IP and MC was documented, and found to be in line with the literature. AWC maps calculated as a function of soil texture classes, showed spatial similarity with WHC maps. Low values of AWC were observed at zones with low WHC index and vice versa. This supports the final conclusion of this work that multi-sensor and data fusion is a useful approach to guide positions of moisture sensor and optimise the amount of water used for irrigation.