Large-scale measurement of soil organic carbon using compact near-infrared spectrophotometers: effect of soil sample preparation and the use of local modelling

Compact near-infrared (NIR) spectrophotometers are low-cost instruments that enable rapid, non-destructive and environmentally friendly measurement of soil organic carbon (SOC). However, several aspects, such as soil sample preparation modes or modelling strategies, related to the use of these instruments in large and heterogeneous data sets are yet to be addressed extensively. This work aimed to evaluate the performance of two compact NIR spectrophotometers (NeoSpectra and NanoNIR) to determine SOC content in a large-scale application. Also, it is important to understand the implications of soil sample preparation (soil grinding and drying) and the use of local partial least squares regression (LOCAL-PLSR) on the accuracy of the models built using these instruments. The soil samples of the calibration (n = 320, selected using the Kennard–Stone algorithm) and validation sets (n = 160) were collected from Minas Gerais state (approximately 589 000 km2), Brazil. Three soil sample preparation modes were considered: air-dried and 2 mm sieved samples, air-dried and finely ground samples, and oven-dried and ground samples. Models to determine SOC were developed using the traditional PLSR (GLOBAL-PLSR) and a new approach based on LOCAL-PLSR, and their performance was evaluated using the root mean square error of prediction (RMSEP). The accuracy of the models built using the compact instruments was compared with the accuracy achieved using a bench Vis-NIR spectrophotometer. The NeoSpectra was the best-performing spectrophotometer, showing values of RMSEP, R2 and bias, respectively, between 5.2 and 6.3 g kg−1, 0.522 and 0.645 and −0.08 and −0.594. Significant enhancements in SOC estimation of up to 13% were found when models were calibrated using LOCAL-PLSR and oven-dried and ground soil samples. Our results showed that compact NIR spectrophotometers are a cost-effective alternative to the Vis-NIR spectrophotometers for large-scale SOC measurement. Models built using these instruments were accurate, mainly when LOCAL-PLSR calibration was used together with oven-dried and ground soil samples.