Development of soil spectroscopy models for the Western Highveld region, South Africa: Why do we need local data?

Abstract

The increasing global demand for sustainable agriculture requires accurate and effi-cient soil analysis methods. Conventional laboratory techniques are often time-consuming, costly and environmentally damaging. To address this challenge, wedeveloped and validated locally calibrated mid-infrared (MIR) spectroscopy modelsfor predicting key soil properties pH, phosphorus (P) and exchangeable cations insoil samples from South Africa's Western Highveld region, using a dataset of 979 soilsamples and machine learning algorithms Cubist, partial least squares regression(PLSR) and random forest (RF). A subset of spectra was also submitted to the newlydeveloped Open Soil Spectral Library's (OSSL) prediction models to determinewhether global prediction models could be used for local soil property prediction.Accurate predictions for pH, calcium (Ca) and magnesium (Mg), with coefficient ofdetermination (R2) values exceeding 0.76 were obtained with the local calibrationalgorithms. The predictions for P, potassium (K) and sodium (Na) did not meet therequirements for reliability. Soil spectroscopic prediction models calibrated withlocal soils outperformed the corresponding global prediction models considered. TheOSSL prediction results were inaccurate, with a RPIQ <1, and consistently underpre-dicted all soil properties. Furthermore, the OSSL collection of prediction models doesnot include a pH (KCl) model, the routinely used pH measurement method inSouth Africa. These findings highlight the importance of local calibration for accu-rate soil property prediction and underscore the need for regional representation inglobal spectral libraries. This research serves as the first local calibration of MIR spec-troscopy models for the Western Highveld region of South Africa and provides afoundation for future local soil property inference model development. It also servesas a potential starting point for a comprehensive South African soil spectral librarythat can be contributed to global spectral libraries