| Summary: | Knowledge of soil conditions is a critical factor of effective land management to achieve optimum ecosystem services in agriculture. Digital soil modeling that incorporates soil spectral data has become a cost-efficient method compared to traditional wet-chemical soil analysis to produce high quality soil management data at the regional level. This research objective was to develop soil property prediction models to assess the characteristics of paddy field soil conditions in East Java, Indonesia. This study examined the utilization of a hybrid two-step regression (2Step-R) soil modeling approach that incorporates categorical soil-environmental variables and continuous visible-near-infrared (VNIR) spectral data. The integrative 2Step-R method was implemented in two modes—Partial Least Square Regression, PLSRmod, and Sparse Bayesian Infinite Factor, SBIFmod—with ‘mod’ referring to the fusion of important categorical soil-environmental parameters and soil VNIR spectral data. This research successfully developed reliable prediction models for soil organic carbon, nitrogen (N), pH, sum of bases (SB), and cation exchange capacity in the study area, with acceptable model performance (performance to interquartile ratio or RPIQ 1.77 to 2.54; performance to deviation ratio or RPD 1.45 to 1.89; and coefficient of determination or R2 0.53 to 0.72). The fused soil-environmental and VNIR spectral data approach (2Step-R) showed markedly improved model performances when compared to models that solely used VNIR spectra data (PLSR) for all modeled soil properties. This research underpins the utilization of cost-effective soil modeling to reveal existing soil conditions in securing soil ecosystem services for intensive agriculture in data sparse regions such as Indonesia.
|
|---|