An efficient approach to capture continuous impervious surface dynamics using spatial-temporal rules and dense Landsat time series stacks

Impervious surface dynamics have far-reaching consequences on both the environment and human well-being. The expansion of impervious surface is often spontaneous and conscious, particularly in fast developing regions. Thus, monitoring impervious surface dynamics with high temporal frequency in a both accurate and efficient manner is highly needed. Here, we propose an approach to capture continuous impervious surface dynamics using spatial-temporal rules and dense time series stacks of Landsat data. First, a stable area mask based on image classification in the start and the end years is generated to remove pixels that are persistent or spatially irrelevant. The Continuous Change Detection (CCD) algorithm is then employed to determine the change points when non-impervious cover converts to impervious surface based on the property of temporal irreversibility. Finally, the CCD time series models are calibrated for pixels with no change or multiple changes. We apply and assess the proposed approach in Nanchang (China), which has been experiencing rapid impervious surface expansion during the past decade. According to the validation results, overall accuracies of image classification in the start and the end years are 97.2% and 96.7%, respectively. Our approach generates convincing results for impervious surface change detection, with overall accuracy of 85.5% at the annual scale, which is higher than three commonly used approaches in previous studies. At the continuous scale, the mean biases of the detected time of imperviousness emergence are +0.17 (backward) and −3.42 (forward) Landsat revisit periods (16 days) for pixels with one single change and multiple changes, respectively. The derived impervious surface extent maps exhibit comparable performances with five widely used products. The present approach offers a new perspective for providing timely and accurate impervious surface dynamics with dense temporal frequency and high classification accuracy.