With the rapid spread of time series data, real-time anomaly detection techniques have become increasingly important to ensure reliability. Existing anomaly detection methods are excellent at detecting anomalies, but are limited in their ability to explain the underlying causes of the anomaly. In this study, we propose a new approach that uses dynamic mode decomposition to explain the dynamic characteristics behind the data points identified as anomalies by their singular values. This method computes the singular values that explain the anomaly by comparing the predictions computed by dynamic mode decomposition with the anomalous observed values. While conventional methods explain the anomalies using dimensional subspaces, the proposed method uses singular values to explain the anomalies, allowing for explanations that take into account periodicity and trends. Experiments proved the explanatory power of the proposed method by identifying dimensions with strong influence of singular values as explanations and comparing their error rates with those of dimensions with actual anomalies.