*3.1. Apparent Phenology*

We found that some Australian mangroves display a bimodal seasonality with two periods of high EVI values and two periods of low EVI values, as shown in Figures 3 and 4. In the Gladstone region, the highest EVI values are recorded between May and August ("Peak growing season" in Figure 3), which are immediately followed by the lowest EVI values between September and November ("Start of season/End of season" in Figure 3). During the wet season, EVI values exhibit a second, less pronounced peak between December and January followed by a subtle drop between February and April. This bimodal seasonality refers to two different peaks in leaf production [6] and is also seen through time, with EVI values in mid-year being higher than those at the beginning or end of the year (wet season).

**Figure 4.** Apparent phenology for each study site. Grey dashed circles show examples of year-to-year variations in the apparent phenology. Blue squares represent locations where only published literature was used, while the red square represents the location of the field data site and where published literature was used.

Figure 4 shows the average phenology of all the pixels in each study site and highlights the fact that mangrove phenology varies with location and through time. For example, both Gladstone sites display similar phenology models despite being years apart. When compared to the Hinchinbrook site, however, the models are somewhat different, especially when looking immediately before and after the highest EVI values (i.e., peak growing season). On a greater scale, the phenology models across states differ greatly from one another. The site located in New South Wales has a distinctly smooth phenology curve while the Queensland sites show jagged features and the Northern Territory is in between.

Temporally, GAMs reveal subtle year-to-year differences in the phenology model that cannot be seen with fully parametric models as the latter over-simplify the phenology from satellite images. Grey circles in Figure 4 focus on certain features in the phenology models that change from year to year. Since we created the GAMs on a pixel-by-pixel basis, we can examine each pixel individually and determine the causes of such variations.

#### *3.2. Apparent Phenology and Field Data*

In Figure 5, we show the apparent phenology and the in situ data from [26]. We can see that each field variable has a marked seasonal pattern, where the values of the variable increase and decrease at certain times of the year (see below). Similarly, the apparent phenology shows a seasonal pattern with higher values between May and September and lower values between October and April. Both the monthly mean and the apparent phenology seem to describe some variables better than others as explained below (see also Section 4).

**Figure 5.** Apparent phenology vs. in situ data from [26]. The red line represents the apparent phenology for the Gladstone area (1995–1999). Grey bars and black lines represent the values for each variable and standard error, respectively. On the left panel, the data are grouped by month and on the right panel, the data are presented in chronological order. No in situ data were recorded for April during the experiment. Panels (**A**, **C** and **E**) display the monthly leaves lost, leaves gained, and net leaf production respectively. Panels (**B**, **D** and **F**) display the leaves lost, leaves gained, and net leaf production in chronological order.

#### 3.2.1. Apparent Phenology and Leaves Lost

Visually, the apparent phenology appears to have an inverse relationship with the number of leaves lost. Between November and March, when the number of leaves lost is high (≥3 leaves/m<sup>2</sup>/day), EVI values are often low. In contrast, EVI values are often high between May and October when fewer leaves are lost. This relationship is evident in Figure 5A,B.

#### 3.2.2. Apparent Phenology and Leaves Gained

From Figure 5 C,D we see that the apparent phenology has a closer relationship with the number of leaves gained than with the number of leaves lost. Visually, this relationship is very strong, especially in the second half of the year. Between October and December, the number of leaves gained rises to its maximum value; this number then drops and remains stable until May. Similarly, in October, EVI rises steadily from its lowest value until December where it remains stable until March before rising to its maximum values between May and June before dropping again and restarting the cycle. In Figure 5D the apparent phenology shows peaks that coincide with periods of a high number of leaves produced (e.g., January 1997, December 1997, and May 1998). The same can be said about the troughs in the apparent phenology, which coincide with lower values of leaves produced (e.g., October 1996 and 1997).

#### 3.2.3. Apparent Phenology and Net Leaf Production

Net leaf production presented by [26] shows two distinct peaks (i.e., June and December) and two troughs (i.e., January and November) that coincide with the peaks and troughs of the apparent phenology (Figure 5E). When the data are aggregated by month (Figure 5E), the relationship between EVI and the net leaf production is clear. Similarly, when the data are presented in chronological order, the months where net leaf production is highest (or lowest) coincide with months of high (or low) EVI values (Figure 5F). In some cases, high and low EVI values precede the highest and lowest values of net leaf production by about a month, however, this is not consistent over time.

#### 3.2.4. Validation: Apparent Phenology vs. In Situ Variables

We validated the apparent phenology against in situ data by running a linear regression between the apparent phenology and every in situ variable from [26]. When the data are grouped by date (i.e., chronological order), the highest correlations with the apparent phenology come from the leaf production rate (R<sup>2</sup> = 0.20), total leaf area (R<sup>2</sup> = 0.16) and net leaf production (R<sup>2</sup> = 0.11). When the data are aggregated by month, however, the correlation of the variables with EVI increases in most cases (e.g., leaf production rate (R<sup>2</sup> = 0.27), standing stock (R<sup>2</sup> = 0.14)).

We also validated our model using non-linear regressions between the apparent phenology and each variable. In Table 3 we show the results from the support vector regression using RBF, linear, and polynomial kernels. For brevity, we only show the results for the regression between net leaf production and apparent phenology using the polynomial kernel, because those results show the best results. Again, monthly net leaf production has a slightly higher correlation and explained variance than chronological net leaf production.

**Table 3.** Explained variance, R2, and Mean Absolute Error resulting from the support vector regression between Net leaf production and apparent phenology.


#### *3.3. Apparent Phenology and Published Literature*

Similar to our field data site, the apparent phenology shows a bimodal phenology curve across all sites described in the selected literature (Table 1, Figure 6). In general, the phenology models have either (1) an inverse relationship or (2) a time lag with respect to the intensity of leaf fall reported in the literature. Most studies report higher leaf fall rates between October and March and lower leaf fall rates between April and September (Figure 6), which denotes an inverse relationship with EVI. By shifting the models by three months, the visual relationship between leaf fall and EVI becomes stronger, especially for the data presented by [31,32] and [8].

**Figure 6.** Panels (**A**–**F**) display the a qualitative measure of Leaf fall, Leaf gain and Net Leaf Production for each study site on the left, right and center respectively. Each panel represents a different study site. The red line represents the monthly value of the apparent phenology from the GAMs and the blue dotted line represents the apparent phenology shifted by three months.

Models of EVI seem to be better predictors of leaf gain intensity when compared to leaf fall intensity. On most sites, leaf gain intensity is highest between November and April and lowest between May and October. Higher values of leaf gain intensity relate well with high values of EVI. However, their timing does not match exactly. Visually, the shifted apparent phenology shows a much closer relationship with leaf gain intensity across all sites than the models with no time shift (Figure 6). Regarding net leaf production, sites in Gladstone (QLD) and Darwin (NT) show that the peaks and troughs of the apparent phenology coincide with the highest and lowest values of leaf production (Figure 6C,D). In contrast, the shifted models shown in Figure 6A,B,E have a better visual relationship with net leaf production.

#### 3.3.1. Validation: Apparent Phenology vs. Published Data

With the exception of [29], all sites have higher correlation values with leaf gain or net leaf production when the apparent phenology is shifted by two or three months. For example, the apparent phenology correlates better with leaf gain values shifted by two months in the case of [31,33] and tree months in the case of [30,32]. The high R<sup>2</sup> values in Table 4 demonstrate that, in mangrove forests, the EVI response to leaf gain intensity and net leaf production is not immediate but delayed by two to three months.

**Table 4.** Correlation coe fficients of the apparent phenology versus net leaf production, leaf fall and leaf gain for each site. Highest R<sup>2</sup> values per site are shown in bold.


In summary: (1) the apparent phenology resulting from the GAMs is a good predictor of leaf gain and net leaf production across our study sites; and (2) apparent phenology does not respond immediately to leaf gain and in most cases has a two- to three-month delay after mangrove forests show signs of leaf gain or increased net leaf production.
