Evaluating Ecological Condition Using Soil Biogeochemical Parameters and Near Infrared Reflectance Spectra

Rapid, repeatable assessment of ecological condition is critical for quantitative ecosystem monitoring. Soils provide a sensitive, integrative indicator for which sampling and analysis techniques are well defined. We evaluated soil properties as indicators of ecological condition (subjectively classified into minimally/moderately/severely degraded based on vegetative, hydrologic and edaphic cues) at 526 sites within Ft. Benning military installation (Georgia, USA). For each sample, we measured 17 biogeochemical parameters, and collected high-resolution diffuse reflectance spectra using visible/near infrared reflectance spectroscopy (VNIRS). VNIR spectra have been related to numerous soil attributes — we examine them here for diagnosing integrated response (i.e., ecological condition). We used ordinal logistic regression (OLR) and classification trees (CT) to discriminate between condition categories using both sets of predictors (biogeochemistry and spectra). Sixteen biogeochemical parameters were significantly different across condition categories; however, multivariate models greatly improved discrimination ([calibration, validation] accuracy of [69%, 66%] and [96%, 73%] for OLT and CT models, respectively). Important predictors included total C, total P, and Mehlich K/Ca/Mg. VNIR spectra further improved discrimination ([calibration, validation] accuracy of [74%, 70%] and [96%, 75%] for OLR and CT models, respectively). While spectra were comparably effective at discriminating minimally degraded sites, they were significantly more effective at discriminating severely degraded sites. Error rates across confounding factors suggest that watershed of origin and landscape position were the only important confounders, likely due to imbalanced sampling. We conclude that multivariate diagnosis improves accuracy, and that VNIR spectroscopy, which yields substantial cost and logistical improvements over conventional analyses, provides an effective tool for rapid condition diagnosis.