Remote and Proximal Sensors Data Fusion: Digital Twins in Irrigation Management Zoning

Area of Science:

• Agricultural Science
• Soil Science
• Geospatial Analysis

Background:

• Precision agriculture requires detailed soil data, but extensive sampling is resource-intensive.
• A trade-off exists between data density, cost, and map accuracy in soil sampling.
• Optimizing input use and reducing environmental impact in agriculture necessitates efficient data acquisition strategies.

Purpose of the Study:

• To evaluate mapping approaches for apparent electrical conductivity (aEC) using sparse soil sampling.
• To compare the performance of kriging with external drift (KED) and geographically weighted regression (GWR) against a reference map derived from exhaustive sampling.
• To assess the effectiveness of these methods in defining management zones (MZs) for irrigation.

Main Methods:

• An exhaustive apparent electrical conductivity (aEC) dataset (3906 points) was collected using an EM38-MK2 sensor.
• A sparse dataset (162 points) was simulated from the exhaustive data.
• Ordinary kriging (OK) was used for the reference map; KED and GWR were applied to the sparse dataset, incorporating remote sensing and terrain covariates.

Main Results:

• The reference aEC map (exhaustive data, OK) achieved high accuracy (R² = 0.97, RMSE = 0.56).
• Kriging with external drift (KED) using sparse data showed good performance (R² = 0.78, MAE = 1.26, RMSE = 1.62), outperforming GWR (R² = 0.57, MAE = 1.78, RMSE = 2.30).
• Management zones derived from KED closely matched the reference map, validating its potential for irrigation management.

Conclusions:

• Sparse soil sampling combined with KED and auxiliary variables offers a viable alternative to exhaustive sampling for aEC mapping.
• KED demonstrates superior performance over GWR in creating accurate soil property maps from limited data.
• The KED method provides reliable guidance for defining management zones, optimizing irrigation strategies in precision agriculture.