Assessment of Four Near-Surface Soil Freeze/Thaw Detection Algorithms Based on Calibrated Passive Microwave Remote Sensing Data Over China

The near-surface soil freeze/thaw (F/T) cycle affects the surface energy balance, hydrological processes, and soil greenhouse gas release. Passive microwave remote sensing data are the most widely used method for determining the near-surface soil F/T status. While many algorithms have been developed for this purpose, their performance has never been compared using same and large in situ data set. Here, we evaluate and inter-compare the classification results of the four most widely used algorithms using a large ground truth data set covering China. Based on the ground observations, our evaluation shows a wide range of near-surface soil F/T detection performance, with Cohen's kappa coefficient ranging from 0.42 to 0.72 and an overall accuracy between 73.8% and 86.2%. We suggest performing parameter calibration for the decision tree algorithm and the discriminant function algorithm before applying them to areas outside of the training sites. All these four algorithms exhibited remarkable uncertainty in the detection of the onset and offset of near-surface soil freezing with root mean squared errors of more than 27 days. These results suggest that careful cautions should be taken when outputs of these algorithms are used for investigations of long-term changes.