Diffuse Reflectance Spectroscopy for Soil Testing Under Different Soil Nutrient Management in Cocoa Production Systems

ABSTRACT The feasibility of using a non‐destructive soil testing approach, like diffuse reflectance spectroscopy (DRS), has rarely been explored in intensively managed smallholder farms. Hence, we examined the applicability of calibrated DRS models for testing soil samples collected before and after different nutrient treatments were allotted at three intensively managed smallholder cocoa ( Theobroma cacao ) farms in Papua New Guinea (PNG). Following baseline soil sampling in 2017, the effects of three soil treatments (trenching, addition of compost and addition of chemical fertilisers containing N, P, K and Mg) on key soil properties were evaluated using soil samples collected following 2 (2019) and 3 years (2020) of continuous application of soil treatments. A total of 1369 soil samples were analysed for 14 soil properties. The DRS approach was evaluated using the Cubist model and spectral reflectance data over the visible near‐infrared (vis‐NIR) region. At most farms, 7 out of 14 soil properties (Soil Organic Carbon: SOC, Cation Exchange Capacity: CEC, available N, Fe, Cu, Mn and Zn) could be estimated with moderate–high accuracy ( R 2 > 0.6) for 2020 samples (after treatment allocation) using a Cubist model trained on 2017 data (before treatment allocation) and spiked with 20% of 2019 samples (after treatment allocation). These improvements were seen both in site‐specific samples and when all samples were pooled to create a single spectral library. Our results, for the first time, demonstrate that an existing spectral library can be augmented by adding soil data undergoing select treatments across multiple years. The combined spectral library may then be used to estimate soil properties in new soil samples. This work has implications for DRS approaches to rapidly and relatively cost‐effectively inform smallholder cocoa farmers of soil management practices, with potential applications in other production systems as well.