**4. Discussion**

#### *4.1. Change Dynamics in Estuarine Ecosystems*

Old world tropical mangroves found in the Indo-Pacific, including tropical Australia, possess notable attributes of species diversity, richness, abundance, and succession, and they are therefore considered to be the most dominant and important mangroves globally [63,64]. We examined changes in their vegetation structure and connectivity within a spatially extensive estuarine region of Central Queensland, Australia by using four methods of change detection. This work has relevance to the maintenance of biodiversity and ecological processes because it explores: (1) the distribution of critical wetland habitats in relation to their proximity to threats from human development; (2) temporal change in the distribution and abundance of wetland habitats correlated (spatially) with temporal change in human activities of varying types (fishing, coastal development, agriculture, erosion, and hydrology modification); and (3) interactions that occur at scales larger than a protected area's boundary that affect the maintenance of biodiversity values.

An analysis of classified maps revealed that gradual ecosystem change occurred across large areas and various habitats. Other studies such as that by Kanniah et al. [65] in the Southern Peninsula, Malaysia, and that of Tran and Fischer [66] in Ca Mao Province, Vietnam, confirm that protected area status does not guarantee the encroachment control of long-term anthropogenic

influence, and the downsizing of mangrove communities continues. Competing demands for available resources, especially in coastal provinces, drives change in hydrology and land use outside a protected areas' administrative boundary, thus a ffecting ecological processes within such as movement of organisms, water availability, and connectivity functions [31]. Mangroves are well recognised as fragile ecosystems that play an important role in linking marine and terrestrial systems [67]. Nevertheless, it is apparent from our results and most previously published research that mangrove ecosystems are in decline [68–70]. The mangroves in our study region are located in the mid and lower intertidal zone and are constrained not by land surface temperature, as in semi-arid regions [71,72], but by air/seawater temperature, freshwater levels, and other geospatial properties. Whilst air/seawater temperature was not measured in this study, other studies have shown a linkage between mangrove deforestation and anthropogenic climate change [73]. The frequency and intensity of cyclones and storms has increased as a consequence of greater sea-surface temperature, with further escalation predicted [74,75]. The Landsat 2017 image was captured in April immediately following the impact of a severe tropical cyclone [16]. We sugges<sup>t</sup> that the mangrove decline in our study (1147 hectares) was a result of change in sediment profiles, defoliation, and inundation from a coastal cyclone [74,76]. Furthermore, because effects outside the boundary of a protected area manifest themselves within that boundary, alterations in hydrology for pasture and direct trampling may be linked to the decline in mangroves, saltmarsh grass, and estuarine wetlands. Similarly, Al-Hamdan et al. [77], by using Landsat satellite data in Tanzania during the period of 2000–2010, found a net deforestation of mangroves with a net agriculture expansion. Our result for mangrove transition (1015 hectares) from the thematic change analysis is comparable to Chen et al. [78], who observed the transition of mangroves to other land uses during the period of 1985–2013 in the Honduras.

River and stream flow regulation is pervasive in Australia. River regulation a ffects riverine vegetation by fundamentally altering the flow regime, thus changing the hydrology and flow across a range of di fferent spatial and temporal scales [79]. Though there is recognition by governmen<sup>t</sup> agencies of the alienation of flow-dependent ecosystems that are attributed to anthropogenic barriers [22], regulating structures such as culverts, pipes, road crossings, weirs, and ponded pasture still potentially cause connectivity disruption in our study region [80]. Further, the inclusion of regulating structures interrupts stream flow regimes, floodplain–wetland connections, biotic responses, channel formation, and sediment transfer [81]. The fragmentation of riverine vegetation with corresponding environmental degradation from flow regulation has been observed with Landsat imagery in other studies, such as those by Das and Pal [82] in India and Antwi et al. [83] in Ghana. We sugges<sup>t</sup> that the reduction in estuarine wetland in our study (1496 hectares) including endangered ecosystems, in combination with a significant declining trend in vegetation extent (thinning) (Table 1, Figure 2), was primarily due to the alteration of natural flow regimes through stream regulation, which a ffected the processes that sustain riparian vegetation communities.

Saltmarsh grass is recognized as providing climate benefits through carbon sequestration as well as other ecosystem benefits, e.g., storm surge erosion protection and ontogenetic habitat for fisheries species [9,84]. Tropical saltmarsh grass is poorly represented in protected areas and crudely acknowledged for its ecosystem services when confronted with alternative land uses. In a recent review, Wegscheidl et al. [85] identified a lack of quantitative information needed to substantiate the value of Australia's saltmarshes, both locally and regionally. Likewise, we found a low occurrence of saltmarsh grass in Cape Palmerston National park, and there was an apparent decreasing trend in the vegetation extent (thinning) of saltmarsh grass throughout the study site. Saltpan and bare mudflat areas exhibited inconsistency across the study site. There were high levels of bare mudflat accretion in the main channels and areas of coastline; however, stream inlets and drainage networks showed inter-annual variability (long-term trend increases and decreases) due to tidal fluctuations and a decrease through time. Similar to those reported by previous studies in tropical regions [86], our results sugges<sup>t</sup> that the saltmarsh–mudflat system in the landward region of the Cape Palmerston National Park shows instability and is degrading over time, possibly due to climatic factors such as recent cyclonic activity, sea level variability, and prolonged inundation.

The need for scientifically-based regional-scale land use planning around protected areas is integral in human-dominated landscapes to balance conservation goals with livelihood needs for crops, pasture, and other ecosystem services [31]. The decline in wetland ecosystems in our study could be attributed to both direct and indirect e ffects. Direct e ffects could include altered vegetation composition and structure from trampling by grazing animals and the modification of ground morphology. An indirect effect could include the draining and hydrological disturbances that convert wetlands to agricultural and grazing land, resulting in tidal disruption and vegetation fragmentation. Across the sub-catchment, the cumulative area of open forest, estuarine wetland, and saltmarsh grass (1628 hectares) was converted to pasture. Riverine landscapes are highly valued in Australia for grazing and are often preferred by livestock because of their vegetation, shade, and water [87]. Though the Sarina Inlet–Ince Bay Aggregation is a designated important wetland under Australian federal biodiversity conservation policy [46], implementation is lacking [88]. The land classes that are open forest and estuarine wetland transition to cropping/grazing is a similar result to that obtained by Haque and Basak [37], who found that forested land transitioned to either shallow water or settlement in Bangladesh during 1980–2010. The result by Toure et al. [89] in Senegal with Landsat imagery and ML classification highlighted the unexpected transition of agriculture to saltpan, as was the case for areas of cropping/grazing in our study (192 hectares).

The significant declining trend observed for open forest, fringing mangroves, estuarine wetlands, and vegetation levels in scattered grazing sites was inconsistent across the study area. This inconsistency illustrates how multiple forms of change can co-occur within relatively close proximity. We sugges<sup>t</sup> that the decline in shoreline vegetation cover was the direct result of a severe tropical cyclone that impacted the coast in March 2017, and we also sugges<sup>t</sup> that grazing-induced, ubiquitous vegetation degradation contributed to and will continue to exacerbate the loss of resilience in these systems.

#### *4.2. Comparison with SLATS*

The SLATS program was initiated by the Queensland Government to provide factual information on land cover and trends in land clearing, tree growth, and regrowth on public and private lands [90]. The SLATS data are based on the supervised classification of multiple Landsat satellite images and digital terrain models at a resolution of approximately 30 m, with maps on woody vegetation clearing (and replacement land cover) that are the result of the anthropogenic removal of vegetation [91]. SLATS has clear di fferences with our study in that SLATS does not include any vegetation loss caused by natural tree death or natural disasters (e.g., cyclones) when calculating woody vegetation clearing rates. Further, SLATS applies radiometric standardisation to the Landsat images. Finally, topographic corrections are used to increase accuracy in areas of high slope [92]. However, as our study area is generally of low, flat elevation, we deemed the correction unnecessary. An inspection of SLATS maps from 2004 to 2017 in ArcGIS displayed similarities with our results with many sites cleared of woody vegetation and converted to pasture, particularly along the boundary of the national park, in the north-east, north-west, south, and central areas. According to SLATS, the total converted vegetation in the Plane Creek catchment is 3536 hectares, and, by digitizing the Rocky Dam Creek sub-catchment pasture polygons in ArcGIS, we found a total of 1100 hectares. Though the total SLATS pasture profile is smaller than our results for the thematic change (1628 hectares) and time series (1896 hectares), our results nevertheless reflect a variable but significant impact on the coastal region that was likely caused by an intense climatic event.

#### *4.3. Limitations of the Study*

Remote sensing data and tools are fundamental methods for measuring LULC, but there are critical drawbacks in the change detection of wetlands. The first drawback is that classification errors from the individual-date images can a ffect the final change detection accuracy, and, although ground-truth data engender the development of accurate LULC classification and accuracy assessments, errors can still occur [93]. Foody [94] argued that accuracy values cannot be appropriately interpreted by readers or users unless a detailed account of the approach to accuracy assessment is provided. The lack of robust validation could have serious implications for some users and may lessen their confidence in remote sensing as a source of land cover data. Therefore, the validation methods that detail the user's and producer's accuracies of change with Kappa coe fficient and which include the confusion matrix for the 2004 and 2017 images (Table 4) have been given to allow for replication. The second drawback is that during high tides, there can be a sharp decline in the spectral reflectance of mangroves, especially in the NIR and SWIR regions [95,96]. Our study used a combined binary change detection and time series analysis approach, illustrating that it is beneficial to use multiple images in change detection research since apparent changes between any two images could be due to irrelevant causes such as tide, sea surface state, and water constituents. The third drawback is that, ideally, change detection requires precise image alignment, which is di fficult to achieve, and the fourth drawback is that post classification comparison-based binary approaches that are used for hard classifications, i.e., comparatively broad scale classifications, may not detect subtle transformations in land cover modification in which the land cover type may have been altered but not changed (e.g., a thinned forest or saltmarsh degradation), ensuing an inappropriate representation [97].

#### *4.4. Implications for the Conservation of Estuarine Ecosystems*

Quantifying the level of coastal wetland fragmentation and landscape connectivity is an essential component of contemporary strategies that are aimed at biodiversity conservation and fishery sustainability [98]. The results presented here are noteworthy from two viewpoints. The first is nationally—in Australia, there is no nationally consistent approach to quantify the area of mangrove or saltmarshes, and historical benchmarks are scarce [99]. Our results inform the Australian inventory of spatio–temporal distribution, as they show important changes in the representation of coastal vegetation classes, particularly mangroves and saltmarsh grasses, in the tropical catchments of the Eastern seaboard. The second viewpoint is regionally—natural resource managemen<sup>t</sup> is hampered by complex managemen<sup>t</sup> arrangements that provide challenges to achieving environmental sustainability and are additional to increasing pressures from natural and anthropogenic forces [100]. Our findings raise concerns that lands surrounding the Cape Palmerston National Park are under threat, and, because interactions occur at scales larger than a protected areas' boundary, repercussions arise for the environmental stability of the entire region. Watson et al. [101] argue that the occurrence of threatened species is widespread outside protected areas, and plants are one of the most poorly represented taxonomic groups. Furthermore, protected areas are not exempt from anthropogenic impacts; for example, Jones et al. [102] identified an increase of human pressure of 1.5% on IUCN listed protected areas categories I and II between 1993 and 2009. Particularly evident in our study was the decline of estuarine wetlands, which include endangered ecosystems: the broad leaf tea-tree *Malaleuca viridiflora* and semi-evergreen microphyll vine thicket-to-vine forest [103] (Figure 2).

There is a need for a more comprehensive understanding of the ecosystem value assigned to Australia's coastal landscapes. This information is a high priority and needed to support evidence-based decision-making and conservation actions that attribute socio–economic value, warranting ecosystem protection and repair [85]. For example, the Australian Government listed subtropical and temperate coastal saltmarsh as a vulnerable ecological community under the Environment Protection and Biodiversity Conservation Act 1999 (EPBC) in 2013 [104]. Carbon sequestration pathways designate saltmarshes (and other coastal wetlands) as disproportionality valuable in sequestering carbon dioxide compared to terrestrial ecosystems [105]. Therefore, it is an opportune time to apply protection to these communities. We propose that the vulnerable listing be extended to tropical saltmarsh regions. The 1998–2003 historical occurrence of mangrove dieback in local estuaries, which a ffected >30 km<sup>2</sup> of remnant mangrove cover [106,107], failed to conclusively identify the causative agen<sup>t</sup> (agricultural herbicides and flooding were implicated in the event). However, recent northern Australian mangrove

dieback has been linked to climate change as the most likely cause [108]. The declining trend in fringing mangroves found in our study is concomitant to a loss of ecosystem services that are provided by the coastal habitat–fishery linkage, as the service value of mangroves has been observed to be higher at the seaward edge [109]. Notwithstanding the Australian government's e fforts to provide protection to Great Barrier Reef catchments [16], ecosystem degradation is ongoing.

Two key factors determine the extent to which coastal habitats can recover and the associated fauna rejuvenate from a major acute (pulse-like) disturbance such as a cyclone: (1) the time window until the next major acute disturbance [110] and (2) the extent and intensity of chronic (press-like) disturbances, such as disruption in sediment/water profiles [74] and elevated mean seawater temperatures suppressing recovery rates [64] during that window. Predictions that tropical cyclones will increase in frequency and intensity in Australia in the coming decades [111] have been accompanied by projections of an escalation in storm surges and extreme sea-levels under future climate change [112]. Improving the resilience of Great Barrier Reef coastal ecosystems requires active landscape protection and restoration approaches to maintain as many biodiversity and ecosystem functions as possible [113].
