Seven California soils were studied to determine if near infrared (700 to 2
500 nm) reflectance spectroscopy could be used to determine soil moisture c
ontent. Near infrared absorbance data in the 1400- to 2400-nm region correl
ated well (r(2) = 0.97, SEP = 1.3%, Bias = 0.2%) with soil moisture content
when a partial least squares calibration model was used to estimate the mo
isture content of soil samples of the same soil type and particle size as t
hose in the calibration data set. However, when the model was used to estim
ate the moisture content of a soil sample with a particle size which differ
ed from those included in the calibration set, the performance was degraded
due to large slope and bias errors (Bias = 4.0%, SEP = 2.2%). However, the
high coefficient of determination (r(2) = 0.98) suggested that predictions
for soil samples which differ from those included in the calibration set c
ould be improved if the slope and intercept were corrected for a given site
. An example validation of this type was shown where the SEP and bias were
reduced from SEP = 2.1% and Bias = 6.0% to SEP = 1.0% and Bias = 0.9% after
slope and bias correction.