Reconstructing And Prediction NDVI Time-series Data Using A Linear Interpolation With Extended Kalman Filter

Vegetation typically elicits dynamics at the seasonal and annual level. Time-series of Normalized Difference Vegetation Index(NDVI) datasets, such as the Pathfinder AVHRR Land(PAL) NDVI, MODIS and SPOT/VEGTATION NDVI dataset, have proven to be appropriate for the detection of long-term vegetation cover changes in regional, continental or global scales. They are also successfully applied to extract the biophysical parameters of vegetation cover. Normally, there are quite frequently fluctuations with marked rises and falls because of variable cloudiness, data transmission errors, incomplete or inconsistent atmospheric correction and bi-directional effects in the NDVI dataset. Scholars have put forward a lot of methods about noise reduction and reconstruction of high quality NDVI time series data, but each method has advantages and disadvantages. Therefore, the reconstruction method to study NDVI time series data and puts forward more optimization method, in order to suppress the abnormal low value and high value anomaly, and improve the reconstruction accuracy of NDVI time series data. In addition, it is capable of avoid changing the normal NDVI values, and construct the high quality time series images. Meanwhile, it is of great practical significance to explore whether the prediction of NDVI time series data can be realized.Taking Guangzhou city for example, the data source is MODIS NDVI and field measured data. Cut the forest image of 100*100 region size which is located in the River Forest National Park as the research object. This paper combined a linear interpolation method with Extended Kalman Filter(EKF) to reconstruct NDVI time series. Then, the extended Kalman filter is used to predict the results of NDVI time series data reconstruction. Finally, the reconstruction and prediction results are verified. And the work is as follow:(1) NDVI time series reconstructionOn the basis of previous studies, this paper combined with the existing NDVI time series model. The linear interpolation method is introduced to improve the precision of the extended Kalman filter to reconstruct the NDVI time series. It combined a linear interpolation method with Extended Kalman Filter(EKF) to reconstruct NDVI time series. This study was conducted in forest areas area of Guangzhou city. The accuracies of the predictions from EKF, Median filter and Image definition were compared based on the field observations collected from sample plots and area. The results of comparing the simulated NDVI values to the observed show that the relative errors varied respectively from-1.91% to 0.93% for the combination of the linear interpolation and EKF, from-3.86% to 5.85% for EKF alone and from-0.28% to 16.30% for Median filter. Comparing the simulated values to the observed, the relative errors from the combination of linear interpolation and EKF varied from 3.89% to 154.09%. The results showed that the combination of linear interpolation and EKF can better perform the approximation of data, fit the peaks of original curve, enhance the overall effect of the curve, and decrease the mean deviation and noise of the original data. Thus, the method can improve the image quality of forested areas.(2) NDVI time series predictionBased on the time series model of NDVI time series reconstruction, the prediction of the NDVI time series data is made by using one step prediction function of the extended Kalman filtering algorithm. The absolute error range of the time series prediction results of the pixel point is between 0.00438~0.094404. Comparing the predicted values with the reference values, the fitting degree was 97.85%. The overall image definition shows a downward trend. The decline range is from-0.1063 to-0.0011, and the reduction range is from-71.04% to-2.90%. It shows that the effect of the extended Kalman filtering on the NDVI time series prediction is not significant, and the accuracy and applicability of the results to be further discussed.Finally, according to the results of the above analysis, the effect of a linear interpolation method with Extended Kalman Filter(EKF) to reconstruct NDVI time series is better. This result is a significant improvement in the effect of the cloud layer and the effect of poor image. At the same time, the image of the original image in the area in the reconstruction of the original image quality can be retained. This study implied that the proposed method increased the reconstructed accuracy of high-quality long time-series NDVI and provided great potential for monitoring forests. However, the effect of the extended Kalman filtering on the reconstruction of NDVI time series data is not significant, which needs further research.