Modelling spatial variability of soil nitrogen using geostatistical methods and remote sensing

Abstract

The aim of this study was to explore the use of geostatistical approaches involving remote sensing data to model and map soil nitrogen variability for use in precision agriculture. Precision agriculture is a management strategy aimed at reducing the costs of nitrogen fertilizers by matching site-specific nutrient applications with crop requirements and soil properties as both vary across a field. The study was conducted on a 4-hectare plot at the University of Zimbabwe farm in Zimbabwe. Three prediction models were applied to estimate soil nitrogen variability within the plots and these included; (1) Ordinary kriging with sample nitrogen data only, (2) Ordinary co-kriging with sample nitrogen and remotely sensed data as covariate data, and (3) spatial regression with remotely sensed data only. Landsat 8 was used as a source of remotely sensed data. Results showed that the Coastal/Aerosol, Blue and NIR spectral bands were highly correlated to soil nitrogen, and produced good prediction models in co-kriging with R2 values of 0.8593, 0.8606 and 0.8596 respectively. Ordinary kriging with nitrogen sample data alone also yielded similar results with R2 value of 0.8597. Finally, spatial regression analysis (SAR lag model) using the same spectral bands yielded R2 values of 0.527, 0.517 and 0.545 respectively while the other bands like Red, SWIR-1 and SWIR-2, and indices like NDVI, RVI and SAVI yielded values below 0.5. Comparison of the three prediction models indicated that there was no significant difference in the mean prediction values of the models. Thus, the results suggest that remotely sensed data can successfully be used alone and in combination with field sample data to model and map soil nitrogen variability in soils.