**4. Conclusions**

Bare soil albedo is an important ancillary dataset used in both climate and energy balance models and satellite LAI/fPAR retrieval procedures, but is usually prescribed based on soil type maps at very coarse spatial resolution. This fixed classification is unable to account for variability observed in soil albedo due to soil moisture and other soil properties particularly in the spatial domain. Long-term satellite observations provide a grea<sup>t</sup> opportunity to extract the bare soil albedo information at a much finer resolution. Methods that rely mostly on a single vegetation index (NDVI, LAI, fPAR, etc.) have been widely practiced in producing the existing soil albedo datasets from satellite data, the results of which, however, may suffer from the uncertainties in the upstream vegetation index product. To overcome this problem, a novel method was proposed in this study for the detection and extraction of bare soil albedo from thirteen years of MODIS albedo product and USDA soil type over the CONUS based on the soil line concept, NDVI, and NDWI derived from the MODIS spectral bands. The soil line concept turned out to be very effective in extracting the bare soil information with minimized impact from vegetation.

The validation of the bare soil albedo is quite challenging over a large spatial domain as ground measurements are not widely available. In this study, bare soil pixels were extracted from Landsat data using the classification map from NLCD. Good agreemen<sup>t</sup> has been found between bare soil albedo from Landsat and our estimations from MODIS data. Further work is expected to make comparisons using more Landsat and MODIS data over different regions and soil types.

In most cases, the proposed soil line method in mapping the bare soil albedo is very efficient and effective. However, it is not always effective, especially when the soil is covered by dense evergreen vegetation canopy. Though the NDVI and NDWI thresholds can help exclude the observations with dense vegetation canopy coverage, it is very possible that the regressed soil line detects the observation with least green canopy coverage rather than pure bare soil background. In other words, the derived climatological bare soil albedo in the evergreen vegetation covered area is not as reliable as those over land surface with sparser vegetation. Such an issue also exists in other current bare soil albedo products, which draws attention in applying them over areas covered by evergreen vegetation. Compared with the existing datasets, the maps generated in this study have reduced uncertainty because more background soil can be observed with the finer spatial resolution input data.

A decrease in soil albedo with an increase in soil moisture from AMSR-E data has been demonstrated over a large area. The exponential relationship has been found to be relatively stable for different time and locations. However, more efforts are still needed to improve the quantification of this relationship by using instantaneous (or at least daily) albedo and soil moisture in the future.

Statistics of bare soil broadband albedos were calculated based on soil types and land cover types. The within-class SDEV statistics sugges<sup>t</sup> that: Though both classification schemes could be used as prescribing indicators for soil albedo, land cover type would be a better choice to determine the albedo magnitude for vegetated areas, while soil type is better at characterizing the soil line feature for sparely vegetated areas.

The bare soil broadband albedo could be very useful as one of the key ancillary data for climate models. On the other hand, our proposed method can also generate the bare soil albedo for the spectral bands. The derived bare soil albedo dataset has been demonstrated quite effective in improving the accuracy of satellite LAI and fPAR estimations under low vegetation density conditions [55].

**Author Contributions:** T.H. and F.G. conceived and designed the experiments; T.H. performed the experiments; S.L., and Y.P. contributed to the data analysis and provided comments and suggestions for the manuscript; T.H. wrote the paper.

**Acknowledgments:** This work was supported by the National Natural Science Foundation of China gran<sup>t</sup> (41771379), the Key Laboratory of National Geographic State Monitoring of National Administration of Surveying, Mapping and Geoinformation gran<sup>t</sup> (2017NGCMZD02), the USDA project (No. 12451361002825) from the NASA ROSES gran<sup>t</sup> (NNH09ZDA001N) to the University of Maryland. We appreciate the comments from Crystal Schaaf at University of Massachusetts Boston, Martha Anderson at USDA-ARS and the anonymous reviewers to our manuscript. We thank the MODIS albedo team and AMSR-E soil moisture team. Both satellite products were distributed by the NASA Earth Observing System Data and Information System (EOSDIS) and available at http://reverb.echo.nasa.gov. We also thank the USDA NRCS and the Multi-Resolution Land Characteristics Consortium (MRLC) for maintaining and distributing the soil type map and NLCD2006 data used in this study. USDA is an equal opportunity provider and employer.

**Conflicts of Interest:** The authors declare no conflict of interest. The founding sponsors had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, and in the decision to publish the results.
