**6. Conclusions**

High spatio-temporal resolution albedo products are essential for climate simulations, as well as for various agricultural and environmental monitoring applications. In this study, we developed a novel method based on the ensemble Kalman filter algorithm to integrate MODIS high temporal resolution albedo and Landsat high spatial resolution albedo data to estimate high spatial-temporal resolution albedos.

We constructed an albedo background field using MODIS historical surface albedo data, from which the initial predicted albedo was obtained from a dynamic model. TM albedos derived by direct estimation were then used as observation data to update the initial predictions. The error of the dynamic model and observations was generated from field observations. When extending the model to regions where no field observation data were available, we created a "common" error by averaging the error of all test stations. The results were compared with ground measurement data for cropland, deciduous broadleaf forest, evergreen needleleaf forest, grassland, and evergreen broadleaf forest sites. We found that estimation accuracy is high (RMSE < 0.0152) for all land cover types. When applied to large areas, the proposed algorithm also shows high estimation accuracy both for homogeneous and heterogeneous regions. The estimated and TM albedos derived by direct estimation were in accordance with RMSE values of 0.003 to 0.0112, and R<sup>2</sup> values of 0.8584 to 0.9964.

The proposed algorithm has four distinct advantages over other similar methodologies. (1) It can generate reliable estimates of land surface albedo with high spatio-temporal resolution. (2) It can be used for all manner of snow-free vegetation surfaces, even heterogeneous surfaces. (3) It compensates for return-cycle and cloud contamination problems with high spatial resolution satellite image data. (4) The algorithm can be easily extended to other fine-resolution data similar to Landsat data (e.g., Sentinel-2).

The proposed algorithm does still have some drawbacks. For instance, the EnKF starts with the first available observation data to update the dynamic model, so the start time of high accuracy estimation depends on the availability of cloudless Landsat satellite images. The quality and quantity of observation data are crucial to the accuracy of the assimilation results. To further improve estimation accuracy, it is necessary to obtain more, higher-quality albedo data.

**Author Contributions:** Conceptualization, H.Z. and H.X.; Methodology, H.Z. and G.Z.; Software, H.Z. and G.Z.; Validation, G.Z.; Formal Analysis, H.Z. and G.Z.; Investigation, G.Z. and C.W.; Resources, H.Z., H.W. and J.W.; Data Curation, G.D.; Writing-Original Draft Preparation, G.Z.; Writing-Review & Editing, H.Z., J.W. and H.X.; Visualization, G.D. and C.W.; Supervision, H.Z. and H.X.; Project Administration, H.Z.; Funding Acquisition, H.Z.

**Funding:** This research was funded by the National Natural Science Foundation of China gran<sup>t</sup> number 41801242, 41801366, the Key research and development program of China under gran<sup>t</sup> 2016YFB0501502, the Chinese 973 Program under gran<sup>t</sup> 2013CB733403, the Key Scientific and Technological Project of Henan Province under gran<sup>t</sup> 172102110268.

**Conflicts of Interest:** The authors declare no conflict of interest.
