In precision agriculture, rapid, non-destructive, cost-effective and convenient soil analysis techniques are needed for crop and soil management. However, the spatial variability of soil properties is consider to be high cost and time consuming to characterize using traditional soil analysis method. To achieve cost and time reduction, the potential benefits of in-situ measurement of soil spectra have been recognized.

     This study aims to develop a general online (In-situ) soil sensing approach for various agricultural fields. In this study, we have investigated the effective approach for collecting soil spectra data in a slope filed because there are many limitations on soil sensing in slope fields. We tested the performance of a potable system (ASD FieldSpec4, 350nm~2500nm) for real time prediction of multiple soil properties, and develop soil property maps based on the prediction results. Field experiments have been carried out in a citrus field located in a hill slope in Ehime prefecture in Japan. To develop calibration models, 100 soil samples from the field were collected in 200mm soil depths, 1000mm from the citrus trees. A global Positioning System (GPS) Hiper ?U(TOPCON) is used to determine the local information (Longitude, latitude, elevation) of the sampling points. The soil spectral was measured on the field using the fresh soil samples. To determine the variability of the spectral data, we collected ten soil spectra from ten different points of each soil sample and for each point, we also measured five times. After soil spectra collection, the soil samples were sent to laboratory for chemical analysis and 20 soil properties were analyzed. Partial least squares regression (PLSR) coupled with full cross-validation were used to establish the relationship between soil reflectance spectra and the soil properties.

     The accuracy is based on coefficient of correlation, coefficient of determination (R²), root mean square error and residual prediction deviation (RPD). We have obtained excellent prediction of moisture content (MC), soil organic matter (SOM), pH, electrical conductivity (EC) in our previous study, and in this study, we will determine calibration models for multiple soil properties such as total carbon (C-t), total nitrogen (N-t), nitrate nitrogen, phosphorus absorptive coefficient (PAC) and so on. The potential of the system for making real-time predictions of multiple soil properties will be evaluated by comparing the accuracy with the previous studies including field-based and lab-based results.

     For effective collection of soil samples, we also evaluated the correlations between soil properties using cluster analysis method and the result shows six representative soil properties can be used for other soil properties estimation. Also, we investigated the proper spectra range for making good soil properties prediction, and all the results will contribute to develop spectroscopic models to predict and map at a high spatial resolution soil properties that are useful for site-specific soil management and precision agriculture.