Three Decades of Inundation Dynamics in an Australian Dryland Wetland: An Eco-Hydrological Perspective

Abstract

Wetland ecosystems are experiencing rapid degradation due to human activities, particularly the diversion of natural flows for various purposes, leading to significant alterations in wetland hydrology and their ecological functions. However, understanding and quantifying these eco-hydrological changes, especially concerning inundation dynamics, presents a formidable challenge due to the lack of long-term, observation-based spatiotemporal inundation information. In this study, we classified wetland areas into ten equal-interval classes based on inundation probability derived from a dense, 30-year time series of Landsat-based inundation maps over an Australian dryland riparian wetland, Macquarie Marshes. These maps were then compared with three simplified vegetation patches in the area: river red gum forest, river red gum woodland, and shrubland. Our findings reveal a higher inundation probability over a small area covered by river red gum forest, exhibiting persistent inundation over time. In contrast, river red gum woodland and shrubland areas show fluctuating inundation patterns. When comparing percentage inundation with the Normalized Difference Vegetation Index (NDVI), we observed a notable agreement in peaks, with a lag time in NDVI response. A strong correlation between NDVI and the percentage of inundated area was found in the river red gum woodland patch. During dry, wet, and intermediate years, the shrubland patch consistently demonstrated similar inundation probabilities, while river red gum patches exhibited variable probabilities. During drying events, the shrubland patch dried faster, likely due to higher evaporation rates driven by exposure to solar radiation. However, long-term inundation probability exhibited agreement with the SAGA wetness index, highlighting the influence of topography on inundation probability. These findings provide crucial insights into the complex interactions between hydrological processes and vegetation dynamics in wetland ecosystems, underscoring the need for comprehensive monitoring and management strategies to mitigate degradation and preserve these vital ecosystems.

1. Introduction

The increasing demand for freshwater for agricultural, industrial, and municipal use has caused rapid degradation of riparian wetland ecosystems due to the diversion of natural flows [1,2]. This has resulted in adverse effects on several eco-hydrological functions in wetland systems, including detrimental impacts on biodiversity, vegetation health, hydrology, and the breeding events of fish, birds, and amphibian species. Periodic inundation events driven by natural flows and localized rainfall are the main drivers of maintaining vegetation health and germination [3,4,5] while recharging groundwater [6,7] to maintain vegetation health during prolonged droughts [8,9]. Wetland inundation provides necessary habitats for the amphibian, turtle, and fish species [10,11], as well as for the waterbirds who need a certain depth and duration of inundation beneath nests for successful breeding and fledging [12,13,14]. Therefore, evaluating the eco-hydrological responses with respect to the inundation dynamics is important to understand wetland dynamics for conservation and decision making in wetland management.

Mapping long-term wetland inundation dynamics over both spatial and temporal domains is important for quantifying the relationships between stream flows and flooding regimes in riparian wetland floodplains [15,16,17] and evaluating their characteristic metrics and ecological response [18]. Such knowledge and insights are important for formulating sustainable water policy and for designing an effective environmental water allocation framework, such as the amount and time to make environmental water releases [19,20]. However, mapping wetland inundation by fieldwork is impractical across large wetland areas over long time scales. With the advancements of satellite remote sensing technologies, mapping inundation and surface water was carried out at different scales, i.e., from global [21], subcontinental, and national [22,23] catchment [15,24,25,26] to site scales [25,26]. However, mapping wetland inundation is complicated and challenging compared with mapping surface waterbodies due to the tree canopies blocking the satellite view capturing the below canopy inundation and mixed spectral responses from dissolved matter and algae. Thus, researchers have developed different approaches to map wetland inundation (e.g., [25,26,27,28,29,30]). The work presented here is based on the inundation maps produced by Senanayake et al. [31] over the Macquarie Marshes in southeastern Australia using the random forest-based multi-index classification (RaFMIC) algorithm. The RaFMIC framework [31,32,33] was built on the Google Earth Engine (GEE) [34] by developing a time series of image composites consisting of Landsat-based spectral indices and bands associated with water and vegetation spectral responses. Subsequently, training samples were collected from selected composites representing a wide range of inundation and vegetation scenarios, including mixed spectral responses. Then, the random forest algorithm was applied to the image composites to create a time series of binary inundation maps. This work provided a temporally dense, long-term time series of inundation maps ranging from 1987 to 2020 by jointly using Landsat 5, 7, and 8 datasets over the Macquarie Marshes in southeastern Australia. These inundation maps showed good accuracy when compared against high spatial resolution aerial imagery and existing inundation maps, indicating their potential to be used for wetland eco-hydrological analysis.

Wetland ecology has a strong relationship with the frequency of inundation. Over a large wetland, the frequency of inundation can vary both spatially and temporally. Therefore, analyzing the eco-hydrological response to inundation dynamics of this region can provide valuable insights for ecological management, guiding environmental water releases and enhancing the understanding of wetland ecosystems. However, the lack of long-term inundation information poses a challenge to this endeavor, highlighting the need for reliable and continuous wetland inundation maps to improve our understanding of wetland ecosystems. Given this, the Landsat-based long-term series of inundation maps developed by Senanayake et al. [31] provides a robust dataset for assessing the historical trends and patterns of inundation dynamics in the region, serving as a valuable resource to analyze the eco-hydrological response. Demonstrating the applicability of long-term time series of inundation maps over large wetland areas is important for assessing the wetland environment. This is particularly important in understanding eco-hydrological responses and inundation probabilities during extreme climatic events. In light of this, this follow-up article to Senanayake et al. [31] aims to carry out the following:

• Assess the RaFMIC method [31] as a robust approach for creating a long-term, dense time series of inundation maps using earth observation data and validate the effectiveness of the Landsat-based wetland inundation maps generated by the RaFMIC method in assessing eco-hydrological responses across large wetland areas.
• Analyze long-term inundation dynamics across the catchment and compare these with climatic events, vegetation changes, and inflows.
• Classify the wetland into zones based on the long-term probability of inundation and investigate their response to climatic variability and inflows. These zones will be useful for wetland management and assessment. Examining how the flood characteristics at each inundation class have influenced the development and distribution of wetland vegetation will shed light on the relationship between inundation dynamics and vegetation response. Vegetation patches defined by Quijano-Baron et al. [35] and the Normalized Difference Vegetation Index (NDVI) were used to analyze the vegetation response to inundation dynamics.

Further, the influence of the wetness index on vegetation dynamics is also examined in this study to examine the topographic influence on inundation dynamics.

Overall, the insights from this study deepen the understanding of wetland ecosystems to enhance ecological and hydrological management strategies, contributing to more informed practices. The wetland mapping approach and eco-hydrological analysis will contribute to global wetland monitoring and assessment systems under the Ramsar Convention, such as the GlobalWetland project [36,37], as well as to continental, national, regional, and local-scale wetland monitoring efforts (e.g., [38,39,40,41,42,43,44,45,46,47,48,49,50]) by using Earth observation technologies. The Northern Marshes in the Macquarie Marshes wetland are highly prone to inundation events and are recognized as an area with highly dynamic ecology [16,51,52]; therefore, they are used as the study area in this work.

2. Materials and Methods

2.1. Study Area

The Macquarie Marshes (~2000 km²) (herein called Marshes) are a large Ramsar-listed, riparian dryland wetland located in the Macquarie-Bogan River catchment of the Murray–Darling River Basin (MDB) of southeastern Australia [53,54,55], extending from latitudes 30.19°S to 31.33°S and longitudes 147.24°E to 147.80°E (Figure 1a–c). This area receives an average annual rainfall of 400–500 mm with high evapotranspiration rates (pan evaporation rates of 1800–2400 mm). The average temperature of the area varies from 2–3 °C to 30–36 °C in the winter and summer, respectively [55]. The Marshes have a gentle northward slope with elevation varying from 115 to 194 m (Figure 1d).

The vegetation of the Marshes consists mainly of river red gum (Eucalyptus camaldulensis) woodlands, common reed (Phragmites australis), mixed marsh/water couch beds (Typha sp., sedges, Phragmites australis, Paspalum distichum), and terrestrial vegetation (Chenopod-Chenopodiaceae) [56]. The Marshes provide habitats for at least 76 waterbird species, fish, and frogs [57,58]. An overview of the Marshes and their biodiversity can be found in Thomas and Ocock [55]. The average NDVI values of the area as captured by the Landsat 8 mission from 2013 to 2020 are shown in Figure 1e. Here, it is evident that most of the region encompasses low NDVI values, except for the Northern Marshes, where high NDVI values are observed. This implies the Northern Marshes are an area with a high vegetation density, thereby supporting more ecological functions. Therefore, a number of researchers have used the Northern Marshes as a key study area in their research (e.g., [52,59,60,61]).

The natural flows to the Marshes and their hydrology were significantly disrupted after the functioning of the Burrendong and Windamere Dams upstream to the Marshes in 1967 and 1984, which diverted the flows for agricultural and municipal uses [62,63]. This has resulted in the degradation of wetland ecosystems and biodiversity [13], affecting various aspects such as vegetation health, waterbird breeding events [51,57,64], and fish populations [65]. Particularly, during the prolonged droughts like the Millenimum Drought from 2001 to 2009 [66], the situation worsened [13]. As a solution, environmental water management plans, including the Murray–Darling Basin (MDB) Plan and its provisions to purchase back the water licenses, were instituted to restore the flows into the Marshes [67,68,69].

2.2. Datasets

2.2.1. Inundation Maps

Inundation maps produced by Senanayake et al. [31] at 30 m spatial resolution using RaFMIC algorithm were used as the baseline dataset in this study to capture the inundated extent. Inundated and noninundated pixels are denoted, respectively, by ones and zeros in these binary inundation maps. The inundated maps developed by using Landsat 5 image collections extends from 1987 to 2011, Landsat 7 from 1999 to 2020, and Landsat 8 from 2013 to 2020. Ideally, maps derived from one satellite product should have the same temporal resolution (i.e., 18 days with Landsat 5 and 16 days with Landsat 7 and 8). However, the temporal frequency of each Landsat dataset used in developing this dataset is lower than its actual revisit time due to cloud cover and missing data.

2.2.2. Landsat Imagery

The NDVI [70] layers were calculated from the Landsat Surface Reflectance Image collections available in the GEE (https://developers.google.com/earth-engine/datasets/catalog, accessed on 15 February 2021). NDVI value of a pixel i can be calculated as

$$NDV{I}_i={\displaystyle \frac{NI{R}_i-Re{d}_i}{NI{R}_i+Re{d}_i}}$$

(1)

where NIR_i and Red_i are the spectral reflectances of Near-Infrared and Red wavelength ranges at pixel i. NDVI is the most commonly used satellite-data-based vegetation index to identify vegetation and assess vegetation health. NDVI was developed based on the disparity between NIR and red reflectance from healthy, green vegetation. Generally, NDVI values vary from −1 to +1, with negative values depicting water, positive values near zero depicting bare land or urban areas, mid NDVI values depicting sparse vegetation, and high NDVI values depicting dense, healthy vegetation. Landsat red and NIR bands with 30 m spatial resolution were used to calculate NDVI layers.

2.2.3. Stream Discharge Data

Daily stream discharge data at Marebone Weir (#421090) and Marebone Break (#421088) (see Figure 1c for locations) for the study period were obtained from Water NSW (https://realtimedata.waternsw.com.au/, accessed on 12 February 2021).

2.2.4. Precipitation Data

Precipitation data from both the rain gauges and PERSIANN-CDR [71] products were used in this study. Gauged rainfall data at stations #051042 and #051057 (see Figure 1c for locations) for the study period were obtained from the Bureau of Meteorology, Australia. These two rainfall stations were chosen due to the continuous data record over the period between 1988 and 2020. PERSIANN provides a satellite-based precipitation product by using adaptive neural network algorithm to combine information from various satellites [72,73]. Here, PERSIANN-CDR product was obtained from the GEE Data Catalog. This daily dataset was developed for hydrological and climatic studies, which need consistent long-term datasets available from 1983 at 0.25° spatial resolution.

2.2.5. LiDAR-Derived Digital Elevation Model (DEM)

The 1 m LiDAR-derived DEM over the study area was obtained from the ELVIS—Elevation and Depth Foundation Spatial Data platform (https://elevation.fsdf.org.au, accessed on 20 February 2021). This 1 m DEM was developed by compiling and resampling the national 5 m bare Earth DEM from LiDAR surveys carried out over the area.

2.3. Methodology

2.3.1. Comparison of Landsat Inundation Maps

Ideally, the inundation maps derived from Landsat 5, 7, and 8 on a particular day should be identical. However, in reality, inundation maps created from these three missions have disparities driven by the differences in mission characteristics such as bandwidths, sensor characteristics, etc. Therefore, inundation maps were prepared by training the classification models separately for each Landsat mission. Landsat-7 Enhanced Thematic Mapper Plus (ETM+), on the other hand, suffers from scan line corrector (SLC) failure, causing stripes of missing data on the images since May, 2003 [74], i.e., SLC-off data. Further, the inundation maps derived from individual Landsat missions may exhibit potential disparities, attributed to different image acquisition dates within their 16-day temporal resolution. As a result, certain inundation events missed by one mission have been captured by the other.

2.3.2. Calculating the Probability of Inundation

The probability of inundation of each 30 m × 30 m pixel over the Marshes was calculated for each year using the timeseries of inundation maps produced by Senanayake et al. [31] to obtain an overview to the inundation dynamics over the period of 1988 to 2020 over the Marshes, especially with respect to the annual rainfall and streamflow data. The probability of inundation of pixel i, i = {1, …, n} (n = number of pixels in a map) is calculated by

$$P_i={\displaystyle \frac{\sum _{t=t_1}^{t_2}D{N}_i}{n_t}}$$

(2)

where P_i is the probability of inundation of the pixel i over the time period of t₁ to t₂, DN_i is the pixel value of i, DN_i = {1 or 0}, and n_t is the number of maps from t₁ to t₂ period.

Subsequently, the probability of inundation was computed separately using the inundation maps derived from Landsat 5, 7, and 8 (based on the available data period for each collection) and classified into 10 equal interval probability classes. Then, the percentage inundation over each of these classes was calculated for each inundation map, i.e., for a particular day, percentage inundation for probability class x is (PI_{class_x}),

$$P{I}_{class\_x}={\displaystyle \frac{I{A}_{di}}{A_{classx}}}$$

(3)

where IA_di is the inundated area of the day d_i in class x (IA_di = $A_p \sum _{i=1}^{n}D{N}_i$), A_p is the size of a pixel, DN_i is the value of pixel i, DN_i = {0 or 1}, i = 1:n is the number of pixels in class x, and A_classx is the area of class x.

2.3.3. Influence of Wetland Inundation Probability on Vegetation Dynamics

The inundation probability classes were then compared against the vegetation maps developed by Bowen et al. [75] over the study area to compare the vegetation types against the inundation dynamics. To further evaluate vegetation response to the inundation dynamics, three simplified vegetation patches in the Northern Marshes, i.e., (i) river red gum forest, (ii) river red gum woodland, and (iii) shrubland vegetation (encompassing common reed, water couch, mixed marshes, and terrestrial vegetation) were examined. These patches were delineated based on the 49-patch model defined by Quijano-Baron et al. [35] based on the observed inundation response and the criteria defined by Sandi et al. [56] on the basis of the vegetation survey of Bowen et al. [75]. Here, percentage inundation over each vegetation patch was calculated from the inundation maps in a similar manner as of Equation (3). Then, the spatial average NDVI value of each class patch was calculated by using the Landsat image collections. Afterwards, regressions were built between percentage inundated area and average NDVI values from the timeseries of images to examine their correlation.

The three vegetation patches were also compared against the probability of inundation of a wet (2016), dry (2019), and intermediate (2015) year as captured by Landsat 8-based inundation maps to compare the inundation extent of these conditions with vegetation dynamics. The years representing three climatic scenarios were chosen based on the annual rainfall data, discharge data (i.e., by comparing the annual rainfall and discharge against the overall mean), and information given in the Bureau of Meteorology, Australia website (http://www.bom.gov.au/climate/current/statement_archives.shtml, accessed on 12 February 2021).

Two drying down scenarios were selected to examine the probability of inundation during a drying event, which offers insights into the hydro-period. The first, from 8 September 1990 to 30 January 1991, encompasses six inundation maps, and the second, from 29 August 1998 to 5 February 1999, comprises five inundation maps. These maps were compared against the three vegetation patches.

2.3.4. Wetness Index on Vegetation Dynamics

The three vegetation patches and inundation dynamics were compared against the SAGA (System for Automated Geoscientific Analyses) Wetness Index (SWI) [76,77,78] to evaluate the influence of the topography. SWI uses the catchment contributing area and the slope to estimate a value representative of the degree of soil wetness. This is similar to the Topographic Wetness Index (TWI) but with a modified catchment area calculation which does not treat the flow as a thin film as is the case in the calculation of catchment areas in conventional algorithms, resulting in an improved performance for the lateral redistribution of water in flat landscapes [79,80]. Given the flat terrain across the Marshes, SWI was chosen to identify the degree of soil wetness based on the catchment contributing area and slope.

3. Results

3.1. Comparison of Inundated Areas

The results of the comparison of inundated area values calculated from Landsat 5-, 7-, and 8-based maps are shown in Figure 2. It is evident that the inundated areas calculated using the maps derived from the three Landsat data collections closely follow a similar trend. However, Figure 2 reveals some mismatches, primarily attributable to disparities in the imaging days of different Landsat missions, resulting in the inability to capture certain inundation events.

Additionally, the comparison of Landsat 5- and Landsat 7-based inundated areas (date matched by linear temporal interpolation) showed a coefficient of determination (R²) value of 0.83. The R² value between the inundated areas calculated from Landsat 7 and Landsat 8-based maps was 0.88, indicating a strong agreement between the inundated areas derived from different Landsat datasets.

3.2. Analysing the Probability of Inundation

Figure 3 shows the annual probability of inundation as captured by the inundation maps derived from Landsat 5, 7, and 8 image collections over the entire Marshes. Due to the SLC-off data of Landsat-7, the Landsat 5- and 8-based maps are used in this figure over Landsat 7 for the years with mission overlaps. The inundation probability maps indicate Northern Marshes (see Figure 1c) as the region with a high frequency of inundation and a high inundation probability, even in the dry years. This is in agreement with the high NDVI values shown in Figure 1e, which shows the vegetation response to these frequent inundations in the Northern Marshes, i.e., river red gum forests. The areas with high inundation probability are expanding from the Northern Marshes mainly toward the north and south in wet years. The variable inundation of the Southern Marshes located in the southwestern region of the Marshes is also evident in the maps shown in Figure 3.

The annual probability of inundation maps demonstrates a good agreement with the annual stream discharge to the Marshes and rainfall data shown in Figure 4. The maps show an overall low probability of inundation during the Millennium Drought (from 2001 to 2009) and during the drought in 2019 [81], as well as a high probability of inundation during the strong La Niña conditions and high rainfall events, e.g., 2010–2012 [82], 2016, and 2020 (http://www.bom.gov.au/climate/current/statement_archives.shtml?region=nsw&period=annual, http://www.bom.gov.au/climate/history/rainfall/, accessed on 10 February 2021).

The ten equal-interval probability of inundation classes over the focus area, the Northern Marshes, derived from Landsat 5-, 7-, and 8-based inundation maps are shown in Figure 5a–c, respectively, whereas Figure 5d shows the inundation probability classes as captured collectively by all the inundation maps. The inundation classes show similar patterns in Figure 5a–c but with higher inundation probability over the southwestern area in the Landsat 8-based map. The higher number of wet years (e.g., 2016, 2017, 2020) captured by the Landsat 8-based inundation maps can be the potential reason for this high inundation probability of the Landsat 8-derived map (Figure 5c).

Figure 5e shows the main vegetation groups over the Northern Marshes as per the vegetation survey in 2013 [75]. It can be seen that the river red gum forests are located over the region having high inundation probability (>70%). Also, the variable inundation probability over the semipermanent wetlands is also evident in Landsat 5-, 7-, and 8-based inundation probability maps. Flood-dependent woodland and shrublands are located in regions with moderate inundation probability. River red gum woodlands are mainly distributed over the areas with low inundation probability.

Figure 6 illustrates the time record of the inundated area for each inundation probability class derived from combined Landsat inundation maps. It reveals a noticeable trend, wherein the total inundated area generally decreases from the lowest inundation probability class to the highest. Notably, a substantial expanse falls within the lowest inundation probability class (i.e., 0–10%), while the highest inundation probability class (i.e., 90–100%) encompasses only a comparatively small area.

Furthermore, Figure 6 highlights the persistent inundation of the inundation probability class associated with the river red gum forest (i.e., >70% inundation probability, as shown in Figure 5) throughout the recorded timeframe. This consistent inundation suggests a vital ecological characteristic of the area, indicating its importance as a habitat or crucial ecosystem component.

In contrast, regions characterized by moderate inundation probability (i.e., 30–70%) exhibit notable fluctuations over time. These fluctuations create a dynamic environment supportive of flood-dependent species, such as waterbirds, and extend over a reasonably large area compared with regions with the highest inundation probability. Such variability in inundation levels provides diverse ecological niches, fostering biodiversity and supporting various species dependent on floodplain habitats.

3.3. Vegetation Dynamics over the Northern Marshes and SWI

Inundation over three simplified vegetation patches was examined to further analyze the association of vegetation dynamics with inundation probability. Figure 7 shows the comparison between the percentage areal inundation over those three vegetation patches (i. river red gum forest, ii. river red gum woodland, and iii. Shrubland) in the Northern Marshes as captured collectively by the Landsat-based inundation maps. In general, temporal inundation trends over all three vegetation patches show an agreement with each other, obviously due to their correlation with in-flows to the Marshes. However, the temporal fluctuations of percentage areal inundation over each patch shows a clear difference. The river red gum forest patch shows a higher percentage of inundation compared with the other two vegetation patches. The river red gum woodland and shrubland patches showed higher temporal fluctuations of inundated area compared with the river red gum forest patch.

Figure 8a–c shows the temporal dynamics of percentage areal inundation and spatial average NDVI values over these three vegetation patches in the Northern Marshes as captured by the Landsat 7-based inundation maps. An agreement between the NDVI values and the percentage of inundation over the patches can be observed, implying the relationship between the environmental flows to the marshes and vegetation health. Here, high, persistent NDVI values can be observed over the river red gum forest patch compared with the river red gum woodland patch and the shrubland patch, obviously due to persistent inundation. The river red gum woodland patch shows higher variability of NDVI over time. River red gums are highly susceptible to changes in flow regimes; therefore, the variable inundation dynamics in this area can cause fluctuations in NDVI values [83].

Generally, a slight delay can be observed between the peaks of NDVI and the peaks of percentage inundation, suggesting a lag in vegetation response to the flows. High NDVI values are noticeable even during the dry periods in the shrubland patch (refer to Figure 8c). The encroachment of invasive vegetation species into the wetlands during these dry periods [75] might sometimes account for such anomalies, potentially leading to increased NDVI values without any major inundation events. Regressions between the percentage areal inundation and the NDVI values over these three vegetation patches are shown in Figure 8d–f. Here, the river red gum woodland patch shows a strong correlation with an average coefficient of determination (R²) value of 0.61 compared with river red gum forest and shrubland patches, which showed low correlations (R² = 0.18 and 0.17, respectively) with the NDVI. High susceptibility to inundation dynamics and less attenuation of the spectral bands used to calculate NDVI from persistent inundation can be the main reasons for this high correlation between percentage inundation and NDVI over the river red gum woodland patch.

Figure 9a–c shows Landsat 8-based probability of inundation maps over these vegetation patches during dry, intermediate, and wet years, i.e., 2019, 2015, and 2016, respectively (see Figure 10d for rainfall and stream discharge values in these years). Figure 9d shows the rainfall and discharge time series, depicting the scenarios during these three years. This shows that most of the shrubland areas were inundated in all three scenarios, including the dry year. Although the river red gum forest patch showed a persistently high inundation probability over time, it showed a low inundation probability in the dry year (2019). The river red gum forest patch exhibited a high inundation probability during the intermediate (2015) and wet (2016) years. This suggests that groundwater recharge during wet years likely plays a significant role in sustaining the health of vegetation in this patch. Examining the hydroperiod during a drying event would aid in understanding the duration of standing water after a flood event, providing an indication of the groundwater recharge potential. The river red gum woodland patch mainly gets inundated during the wet year, but it showed low inundation probability during the dry and intermediate years.

The probability of inundation during two drying events (see Figure 10a) as captured by the Landsat 5-based inundation products over the Northern Marshes are shown in Figure 10b,c. Here, the inundation probability was calculated by using the Landsat 5-based inundation maps within the drying period. The river red gum woodland and forest patches show a high inundation probability during drying compared with the shrubland patch, which implies a longer period of standing water after a flood event in river red gum patches. Higher evaporation rates due to the higher exposure to solar radiation can be a potential reason for lower inundation probability at shrubland patches during a drying event. The groundwater recharge driven by the long hydro period can be identified as a main reason for maintaining the health of river red gum patches in this region. On the other hand, the SWI (Figure 10d) indicates higher values over the river red gum forest patch and shrubland patch compared with the river red gum woodland patch. This suggests that the SWI values, influenced by topography, align more closely with the long-term inundation probability, as shown in Figure 5a–d, while inundation during drying events is primarily driven by factors such as evaporation.

4. Discussion

The results of this study provide valuable insights into the eco-hydrological dynamics of the Marshes wetland, particularly in the context of wetland inundation mapping and its implications for ecological management. Diversion of river flows due to increasing demand for freshwater for various uses and flood control has led to the degradation of riparian wetland ecosystems [1,51,52,84], underscoring the importance of understanding and monitoring wetland inundation dynamics. Since the construction of the Burrendong Dam in 1967 and subsequent water diversions for irrigation, the Marshes have experienced a long-term reduction in natural flows by approximately 40–50% [51,62]. These wetland ecosystems are highly sensitive to changes in the flow regime, impacting the functioning of these ecosystems [85]. Many previous studies investigating hydrology and vegetation dynamics in wetland systems have employed model-based approaches (e.g., [16,86,87,88]), since reliable, long-term inundation information is difficult to obtain. This study presents a data-driven analysis spanning the last three decades by using a Landsat-based inundation dataset, which fills a critical gap left by the limitations of model-based approaches. Among the entire Marshes, the Northern Marshes exhibit the highest inundation probability and NDVI. Consequently, most of the ecological functions occur in this specific region of the Marshes. Therefore, this study focused on evaluating the inundation dynamics and ecological response over the Northern Macquarie Marshes. The findings align with numerous studies focused on eco-hydrological conditions specifically within the Northern Marshes [52,59,60,61,88].

The comparison of inundated areas calculated from Landsat 5-, 7-, and 8-based maps demonstrated a close agreement in trends, indicating the potential application of Landsat-based maps for wetland monitoring. However, disparities observed in certain areas highlight the need for further refinement and consistency in mapping methodologies. The subtle dynamics in inundation and vegetation might not have been captured adequately by Landsat-based products due to their spatial resolution of 30 m. For a more focused examination of eco-hydrological responses, it is recommended to use Sentinel-2-based inundation and NDVI products for smaller study areas. Furthermore, incorporating the newly developed Harmonized Landsat and Sentinel-2 (HLS) products for mapping inundation dynamics holds great potential for creating a much denser temporal series of inundation maps, which can minimize the effects of temporal data gaps due to clouds and other issues. Such datasets can enable a more comprehensive analysis of eco-hydrological responses to inundation dynamics.

The annual probability of inundation maps provided insights into the spatial and temporal variations in wetland inundation probability and also showed the Northern Marshes as the region with high inundation frequency in the Marshes. Ralph et al. [59] demonstrate alterations in flooding patterns across southern marshes over time, indicating that the frequency of inundation affects the spatial distribution of vegetation communities. The time series of inundation maps used in this study and the inundation probability maps can be used to analyze the inundation dynamics and ecological response over other regions of the Marshes, e.g., the Southern Marshes.

The agreement between inundation probability maps and environmental factors such as stream discharge and rainfall underscores the importance of these maps for understanding wetland dynamics and informing management practices. The inundation time series over specific areas not only provides valuable insights into hydroperiod but also serves as a baseline dataset for analyzing the lag time of ecological responses to inundation events. This enables a more comprehensive understanding of the complex interactions between hydrological processes and ecosystem dynamics within wetland environments.

The analysis of vegetation dynamics over the Northern Marshes revealed correlations between environmental flows, vegetation health, and inundation frequency. The results show that the river red gum woodland patch retains water for a longer duration after a flood event, indicating a high potential for groundwater recharge. Further investigation involving soil properties and groundwater data is necessary to validate these findings. Inundation probability shown in Figure 5e shows a good agreement with the probability of river red gum persistence throughout the Marshes, derived from a 10-year (1997–2006) historical analysis of inundation frequency by Catelotti et al. [83]. Generally, vegetation communities in the area exhibit a strong correlation with inundation frequency, particularly noticeable within the river red gum woodland patch. Catelotti et al. [83] explain that the condition of river red gum communities is degrading due to the decrease in natural flows to the Marshes. River red gums are long-lived species that rely on regular flooding patterns that are crucial for both their own survival and the overall health of wetland ecosystems. They are highly sensitive to alterations in flow regimes, requiring flooding approximately every three years to sustain their survival and reproductive cycles [89]. Therefore, the status of river red gum forests offers a valuable indicator of ecosystem vitality, especially when expanded to encompass insights into reproduction, germination, and recruitment patterns concerning various environmental flow scenarios. The lack of historic vegetation mapping data is one of the major limitations in investigating the distribution and changes in vegetation communities over time in response to inundation dynamics. This emphasizes the importance of frequent vegetation mapping endeavors to assist in wetland management. Methods such as measuring the chlorophyll fluorescence emitted by plants [90] can provide a more precise understanding of photosynthetic activity. This approach could improve the representation of vegetation conditions in wetland areas and be valuable for future frameworks assessing vegetation conditions.

It is important to note that NDVI sensitivity can decrease at higher biomass levels due to saturation issues. However, this study is primarily concerned with changes in NDVI under drought conditions, rather than addressing sensitivity variations at high biomass levels. The temporal trends in inundation and NDVI values showed a delay in vegetation response to inundation, suggesting a lag in vegetation growth following inundation events. The lag time between vegetation and hydrological or seasonal influences has been studied by a number of researchers in different vegetation communities. For example, Wei and Wan [91] found a 3-month lag time between NDVI and terrestrial water storage at the Weihe River Basin in China. Feng et al. [92] found a one-month lag effect between NDVI and precipitation in the Mara River Basin, East Africa. Some of the regions in the Marshes which were frequently inundated have become drier areas over time, probably due to the diversion of flows for agricultural uses. For example, the drying over Hunt’s Woodland area is evident between the Landsat 5-derived inundation probability map and the other two maps in Figure 5a–c. This is in agreement with the findings of Brandis et al. [85], who explained that the natural flows to this region were altered by the construction of a bypass channel, which changed its vegetation and the reliability of water during small floods. This region was a breeding colony for colonial waterbirds but has not functioned as a breeding colony since 1993. Similar declines in waterbird populations due to wetland deterioration have been recorded in many parts of the world, e.g., the Lowbidgee floodplain in Australia [93], western North America [94], and Rio Cruces wetland, Chile [95]. Many of these are migratory birds whose breeding events depend on the periodic inundation of semipermanent wetlands. The inundation probability maps provide valuable insights into inundation dynamics over these areas for ecological conservation and management.

Overall, the results of this study deepen our understanding of wetland ecosystems and provide valuable insights for ecological and hydrological management strategies. The development of a long-term series of inundation maps using Landsat image collections provided a robust dataset for assessing historical trends and patterns of inundation dynamics, serving as a valuable resource for further research and management decision making. Therefore, testing the RaFMIC method for mapping long-term wetland inundation dynamics by incorporating localized training datasets and Landsat-based indices and bands, and subsequently assessing eco-hydrological responses in various regions globally, will help evaluate its broader applicability.

5. Conclusions

This study examines the inundation dynamics and vegetation response in the Northern Macquarie Marshes over the last three decades by using a Landsat-based time series of wetland inundation maps and NDVI data to analyze the relationship between inundation frequency and vegetation dynamics.

The Northern Marshes exhibit the highest inundation probability and NDVI among the entire Macquarie Marshes, highlighting the ecological importance of this region. However, some areas within the Marshes have become drier over time, possibly due to water diversions for agricultural purposes. This study also reveals that the river red gum woodland patch retains water for a longer duration after flood events, indicating a high potential for groundwater recharge. Vegetation communities in the area show a strong correlation with inundation frequency, particularly noticeable within the river red gum woodland patch.

Overall, this study provides valuable insights into the eco-hydrological responses of the Macquarie Marshes, emphasizing the importance of maintaining natural flows for the health and functioning of wetland ecosystems. The findings can inform conservation efforts and water management practices in the region. The approach and results presented here are valuable for assessing the eco-hydrological status of wetland systems and for designing a decision support framework for effective environmental water releases.