Heavy metals with a twist

NIR reflectance spectroscopy has shown some promise at detecting heavy metals in agricultural soils, where they can contaminate food crops. Unfortunately, NIR spectroscopy can only do this when the heavy metals are present at fairly high concentrations, above around 1000mg/kg, because heavy metals are not very good at reflecting NIR wavelengths. This is a major limitation, because heavy metals such as lead, zinc, copper and arsenic can cause contamination problems even when present at much lower concentrations.

Now, a team of Chinese researchers led by Wenxiu Gao at Wuhan University has come up with a novel method that seemingly allows NIR spectroscopy to detect low concentrations of heavy metals in agricultural soil. The twist is, however, that it may actually be detecting something else altogether.

Gao and her team took 100 samples of agricultural soil from 30 sites where the composition was likely to differ, including fields growing corn, rice and tea, and woodland and grassland. They then used an atomic spectrometer to measure the concentration of lead, zinc, copper and arsenic in the soil samples, finding that zinc was present at the highest concentrations. Even so, the highest zinc concentration was only 117mg/kg, with an average across all the samples of 54mg/kg, far below the concentrations that can be detected by NIR spectroscopy.

Nevertheless, when Gao and her team analysed the samples using visible and NIR (VNIR) reflectance spectroscopy, they found that the spectral response did seem to vary in line with heavy metal concentrations. Building models from the spectral data for each soil type using a combination of a genetic algorithm and partial least squares regression, with the genetic algorithm selecting the most meaningful wavelengths, they found that these models could fairly accurately predict heavy metal concentrations.

As the researchers report in Geoderma, while most of the models could only produce accurate predictions for the same soil type, three of the models could produce accurate predictions for some other soil types as well. But how could this be happening if the heavy metal concentrations are too low for VNIR reflectance spectroscopy to detect?

A possible answer comes from the fact that the models were least accurate at predicting arsenic concentrations, which is of interest because lead, zinc and copper all bind strongly to organic matter in soil, while arsenic doesn’t. This implies that VNIR reflectance spectroscopy may actually be responding to organic matter levels in the soil, but that these are closely correlated to the heavy metal concentrations.