Predicting surface reflectance using time series harmonic model and all available Landsat imagery
2019
A time series harmonic (TSH) model for predicting surface reflectance using all available Landsat 8 images in three years is developed. First, a cloud, cloud shadow, and snow detection algorithm called Fmask is used for eliminating “noisy” observations. Then, a time series model that has components of overall, seasonality and trend are estimated for each pixel of each spectral band. The model is capable of predicting surface reflectance for pixels at any location and any date assuming persistence of land cover. The algorithm was applied to one Landsat 8 scene in North China (WRS Path 123 and Row 32). All available (a total of 63) Landsat 8 images acquired between 2014 and 2016 in Google Earth Engine were used. Three clearest scenes were used for quality assessment of the predicted surface reflectance. The average R2 of 6 bands were 0.75, 0.69 and 0.67 for 04/13/2014, 09/04/2014 and 10/06/2014 scene, respectively. The average Root Mean Square Error (RMSE) of 6 bands were 0.025, 0.026 and 0.028 for 04/13/2014, 09/04/2014 and 10/06/2014 scene, respectively. The results demonstrate that the predicted images are in good agreement with the real images. In addition, the performance of TSH model for cloud-gap fill was assessed and was compared with Best-Available-Pixel (BAP) composite method. The MODIS Nadir BRDF-Adjusted Reflectance product (MCD43A4) was used to assess the error of cloud-gap fill. For the three selected cloud-covered test regions, the average R2/ RMSE were 0.8/0.022, 0.63/0.014 and 0.8/0.018 for TSH model, and were 0.72/0.026, 0.53/0.016 and 0.66/0.032 for BAP composite method. The results conclude that prediction accuracy of TSH model is superior to that of BAP composite method. The TSH model-derived cloud-free images are of great importance for multi-temporal land-cover classification and change detection.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
14
References
0
Citations
NaN
KQI