Detection of soil properties with airborne hyperspectral measurements of bare fields

ABSTRACT. Remote sensing with aircraft‐based sensors can provide the fine resolution required for site‐specific farming. The within‐field spatial distribution of some soil properties was found by using multiple linear regression to select the best combinations of wave bands, taken from among a full set of 60 narrow bands in the wavelength range of 429 to 1010 nm. The resulting regression equations made it possible to calculate the value of the soil property at every pixel, with a spatial resolution of 1.2 m. Both surface and subsurface samples of soil were taken from the center of each of 321 equal‐sized grids on 128 ha of recently seeded and nearly bare soil. The soil samples were tested in a laboratory for 15 different properties. The percent sand in surface samples was found to be detectable with a reasonable degree of accuracy with R2 = 0.806 for a four‐parameter model; the best combination of wavelengths was 627, 647, 724, and 840 nm. For silt, clay, chlorides, electrical conductivity, and phosphorous, the results were somewhat less satisfactory with a range of 0.66 < R2 < 0.76. The poorest fit was for carbon with R2 = 0.27. Organic matter and saturation percentage had R2 < 0.49. For the remaining properties, i.e., pH, Ca, Mg, Na, K, and bicarbonates, the correlation was intermediate and statistically significant, but with a great deal of scatter around the regression lines. An example of an image map was produced showing the percent sand at every pixel location in one field. New spectral indices were developed; one index (I = R763- 0.85*R753- 0.24*R657-0.40*R443) was found to work well with five of the soil properties (EC, Ca, Mg, Na, and Cl), indicating some commonality in the manner in which they affected the reflectance from the soil surface, possibly due to a salinity effect. Multiple linea