Instability of remote sensing ecological index and its optimisation for time frequency and scale

Abstract

Evaluating regional ecological environmental quality (EEQ) using remote sensing is important for achieving sustainable development. The remote sensing ecological index (RSEI) is a typical EEQ evaluation model used to comprehensively reflect regional ecological quality. However, owing to its complete dependence on remote sensing image information, the RSEI also has inherent issues, including unstable time series and inconsistent resolutions of its four sub-indices. To address these problems, this study used the Google Earth Engine (GEE) to propose an optimisation notion for the harmonic analysis of time series (HANTS) coupled random forest (RF) model and determined the accuracy and image quality of the optimised RSEI (RSEIo) for the Yangtze River Basin (YRB). Based on the findings, HANTS could fill the gaps in the images, effectively reducing the noise and discrete degrees of the four indices and thereby improving the stability of the RSEI. The correlation coefficient (R) between the RSEI and RSEIo was 0.93, with a root mean square error (RMSE) of 0.13, indicating that RSEIo is reliable. The image quality assessment (based on contrast and information entropy) indicates that HANTS combined with the RF model can produce RSEIo images with higher definitions and richer information at a constant spatial resolution. Further, the pixel-by-pixel coefficient of variation evaluation indicates that the RSEIo image was highly stabilised, yielding higher numbers of effective RSEIo images without changing the temporal resolution. Compared with traditional RSEI calculations, the optimisation proposal herein could highlight ecological differences caused by topographic changes in the YRB, which would produce an RSEI closer to actual surface conditions. Further, this proposed method could be used to obtain more detailed ecological information at a constant spatiotemporal resolution, thereby meeting the needs of long-term ecological monitoring in large-scale regions.