**5. Conclusions**

In this paper we explore, different approaches to find the most robust way of applying the VJB correction to AVHRR data. To do this we use AVHRR data from 1982-2015 divided in AVHRR-pre (1982–2000) and AVHRR (2000–2015) and MODIS data (2000–2015) at CMG spatial resolution (0.05◦) in 445 BELMANIP2 sites. In the first place, we compare the effect of using 3-years or 15+ years to derive the BRDF parameters. We found that for coarse spatial resolution, where the vegetation characteristics of the target don't appear to change significantly, it's preferable to use 15+ years of observation. This result was true both for AVHRR and for MODIS data. The differences were higher for the NIR band, where the average noise was reduced by 7% when using the whole time series from AVHRR-pre-data instead of 3-year parameters.

Secondly, we used MODIS data to retrieve relationships between the BRDF parameters. With the information derived from this sensor, we built different models based on the VJB method, which aim to minimize the propagation of the atmospheric errors present in the AVHRR data to the correction of directional effects. We found that for the red and NIR bands, the Stable method provides a robust correction in terms of reducing both the average noise of the pixels considered and the width of the noise distribution. This is the recommended model for surface albedo retrieval for the VJB method using AVHRR data. For the NDVI, however, we found that the lowest average noise is obtained by correcting MODIS data with MODIS derived parameters and AVHRR data with AVHRR derived parameters. These results are true both for AVHRR-pre which has no aerosol or water vapor correction and for AVHRR, which uses MODIS information for the atmospheric correction.

Further studies should focus on the effect of the spatial resolution on the assumptions made by the VJB method, such as the invariability of a certain site during the composite period of the model. This information is especially useful for the derivation of surface albedo. A surface albedo using AVHRR data from 1982–2018 is of grea<sup>t</sup> interest to the scientific community, especially one which relies heavily on observations and on semi-empirical physical models.

**Author Contributions:** Conceptualization, J.L.V.-N., B.F. and E.F.V.; Formal analysis, J.L.V.-N. and E.F.V.; Investigation, J.L.V.-N.; Methodology, J.L.V.-N., B.F. and E.F.V.; Supervision, B.F., E.F.V. and J.-C.R.; Writing—original draft, J.L.V.-N.; Writing—review & editing, J.L.V.-N., B.F., E.F.V. and J.-C.R.

**Funding:** This research received no external funding.

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