*4.4. Moving Forward*

Apparent phenology can be detected using a variety of spectral indices, and EVI is only one of several that has been used for coastal ecosystem investigations [14,16,59]. Future studies could use EVI in combination with other spectral indices to improve the detection of phenological events such as accurate measurement of leaf production and different stages of leaf growth. It would also be important to assess whether other spectral indices also display a time lag with relation to net leaf production and whether other phenology models show this temporal shift as well. For example, spectral indices that use the short-wave infra-red region of the spectrum could provide information on water content and indirectly inform the number of leaves in the forest. Establishing this relationship is important, especially in scenarios where mangroves are at risk of massive diebacks such as drought and heatwaves.

Besides temperature, rainfall, and other climate data, other sources of information that can potentially provide additional insights to our model: (1) Fractional Vegetation cover, and (2) radar imagery from Sentinel 1 or Advanced Land Observing Satellite (ALOS) sensors. The use of Radar datasets to monitor mangrove forests has been increasing in the past few years, mainly providing insights on mangrove zonation [60,61], canopy structure and height [62], while Fractional Vegetation Cover informs mangrove dynamics [13]. The spatial resolution of many of these sensors, including Landsat, does not allow the discrimination of species, however, estimating general trends in mangrove phenology could be more important to protect these forests rather than species-specific values.

We have also identified several ways in which the remote sensing and ecology communities can take phenology modelling to the next level. Firstly, we can evaluate accuracy by gathering and/or sharing field data with sufficient temporal resolution to compare it with satellite imagery. These data

could include leaf area index, litterfall, leaf onset, biomass, and other measurements of plant phenology and growth that aid in assessing model accuracy and potential bias. For these e fforts to be successful, data collection has to use identical, or at least comparable, techniques to identify and measure the variables of interest. Agreements on how to define and measure mangrove phenology, coupled with high-resolution imagery could greatly benefit this type of studies.

Secondly, GAMs and high-resolution imagery (i.e., equal or better than 1.5 × 1.5 m) can be used to model phenological changes on individual plants. By modelling phenology and the factors that affect it, users can take preventive or corrective measurements before the plant (or crop) fails or dies. More importantly, high-resolution imagery could potentially be used to create models at the same scale as the data collection plots, could be used to monitor restoration projects [63], and couple phenology to functional traits [64].

Thirdly, incorporating independent datasets to the GAMs will allow us to examine which environmental variables have the most influence on mangrove phenology at a continental scale. These datasets could include parameters like temperature, rainfall, humidity and tidal range. Besides altering spectral reflectance value values in the near and short-wave infrared bands [9], the tidal range at the time of image acquisition may play an important role in mangrove phenology. Just like temperature and rainfall, the tides vary seasonally across Australia [65] and their impact on mangrove phenology is ye<sup>t</sup> to be assessed.

Lastly, we have demonstrated the usefulness of GAMs with a dense time-series of remotely sensed imagery, but the applications of this work could also be used with Moderate-Resolution Imaging Spectroradiometer (MODIS), Sentinel or other satellite sensors. Creating maps of mangroves around the world is important, but we currently have the technology to process large datasets in just hours, so why not model (and forecast) phenology under di fferent climate change scenarios? This means detecting changes in the start of season and peak growing season dates over time and how that may correlate with changing weather and climatic patterns.
