Do XRF local models have temporal stability for predicting plant-available nutrients in different years? A long-term study showing the effect of soil fertility management in a tropical field

This study evaluates the temporal stability of X-ray fluorescence (XRF) models for predicting plant-available calcium (av-Ca) and potassium (av-K) in a tropical agricultural field under changing soil management. Understanding this stability is crucial for advancing XRF as a quick and clean tool for soil nutrient monitoring. XRF models were tested across six sampling periods (2015, 2019, 2020, and three in 2022); lime and potash rock powder were applied before 2022 samplings to assess the XRF models response to amendments, which altered the ratio of total to plant-available nutrients (T/A ratio). We evaluated a simple model (M15) calibrated using only samples acquired in 2015 (S15), and two time-specific models (M15+SS and SS models) that incorporate samples collected at each analysis period. All models showed temporal stability when the T/A ratio was consistent, with RMSE values of 3.15─6.95 mmol_c dm⁻³ (1.91 ≤ RPIQ ≤ 4.22) for av-Ca and 1.20─1.64 mmol_c dm⁻³ (1.86 ≤ RPIQ ≤ 2.55) for av-K. However, the application of lime and potash rock powder disrupted the T/A ratio for Ca and K, reducing all models accuracy, with M15’s RMSE increasing to 10.78─40.64 mmol_c dm⁻³ (0.33 ≤ RPIQ ≤ 1.23) for av-Ca and to 1.86─6.37 mmol_c dm⁻³ (0.48 ≤ RPIQ ≤ 1.64) for av-K. Although time-specific models improved accuracy compared to M15, they require frequent recalibration. Overall, XRF models can reliably predict plant-available Ca and K over time if soil management maintains a consistent T/A ratio. This study underscores the need to consider soil amendments when applying XRF models for nutrient monitoring and contributes to the theoretical basis for using XRF in agricultural management.