**5. Conclusions**

In this paper, we demonstrated that GAMs help us detect (1) the dual phenology of mangrove forests, and (2) seasonal and inter-annual changes in mangrove phenology by using 668 satellite images of di fferent study sites across Australia. The two distinct periods of leaf growth in mangrove forests had not been detected using satellite imager until now. We compared our model to the field and published data to explore which biophysical variables help explain the seasonal changes in EVI. When compared to field data, we found that seasonal and inter-annual variations of EVI correlate well with the leaf production rate, net leaf production of mangrove forests. When compared to published data, we found that there is a time lag between leaf gain and the EVI. Overall, leaf gain and net leaf production are more closely related to higher EVI values than leaf fall. Regarding the phenological metrics, in our Gladstone site, the start of season occurs more frequently between September and October each year and the peak growing season between May and July.

Rather than imposing a parameterized mathematical curve to the data, our study leverages the ability of GAMs to let the data determine the type of relationship between a given spectral index and plant phenology. This data-driven approach helped us detect a bimodal phenology in mangrove forests dominated by *R. stylosa;* bimodal phenology has been reported in the literature but it has never been seen with remote sensing techniques. More importantly, GAMs allowed us to determine that mangrove phenology is site-dependent. Fully parametric methods, when applied to remotely sensed data, have over-simplified the phenology of mangrove ecosystems and other evergreen forests worldwide.

By understanding how phenology changes from site to site, and year to year, this study provides a tool for regional and continental-scale assessments of mangrove phenology. We expect to see an increase in the use of GAMs, especially in conjunction with the Landsat and other long-term, worldwide imagery archives.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2072-4292/12/24/4008/s1, Figure S1: Observed EVI vs Apparent phenology for all sites, Figure S2: Mean Absolute error of the Cross-validation predictions of EVI.

**Author Contributions:** Conceptualization, N.Y., K.E.J., L.L. and S.W.M.; methods, T.D.N., L.L. and N.Y.; field data collection, N.C.D.; data analysis, N.Y., T.D.N., N.C.D.; writing—original draft preparation, N.Y.; review and editing, N.Y., K.E.J., S.W.M., T.D.N., L.L., N.C.D. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received grants the following support Wet Tropics Management Authority Student Research Grant (NY), National Environment Science Program (NESP) Tropical Water Quality (TWQ) Hub Research Grant (NY), and a Centre for Tropical Water & Aquatic Ecosystem Research (TropWater) Student Research Grant (NY).

**Acknowledgments:** This project is supported by NIESGI Cia. Ltda. This research used resources from the National Computational Infrastructure (NCI) and Digital Earth Australia.

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