Status of the undisturbed mangroves at Brunei Bay, East Malaysia: a preliminary assessment based on remote sensing and ground-truth observations

Study area

Brunei Bay is extended over 2,500 km² where much of its aquatic and terrestrial land belongs to East Malaysia (Fig. 1). The bay area is also known as a significant habitat for marine biological diversity in the South China Sea (Vo, Pernetta & Paterson, 2013; Joseph et al., 2016). The southern limit of the bay is flanked by lush green mangroves, especially on the lower reaches of Limbang, Sundar, Weston and Menumbok rivers (BirdLife International, 2015). The Limbang estuary—emerging from the three rivers, namely Sungai Manunggul, Sungai Limbang and Sungai Pandaruan—is orientated in a north-south direction whereas the Sundar estuary—emerging from Sungai Trusan—is stretched between Kuala Trusan and Tanjung Perepat in a west-east direction. Both Weston and Menumbok estuaries are formed by Sungai Padas and Sungai Klias, respectively, and run in a north-east direction. The mangrove vegetation, together with seagrass beds and coral reefs in Brunei Bay, are providing several eco-socio-economic benefits to the local communities (Ahmad-Kamil et al., 2013). This study was conducted with the permission of the State Forestry Department of Sarawak (# NCCD.907.4.4 (Jld. 10)-294).

The climate of Brunei Bay is influenced by tropical weather with two monsoonal regimes: northeast (mid-December to mid-March) and southwest (mid-May to the end of October) (Malik, 2011). The historical weather data (2005–2015) showed an average highest temperature of 33.5 °C for April–May and the highest precipitation of 72.9 mm for June (WU, 2015). Intense seawater current, accompanied by strong winds, can be observed every year during the northeast monsoon (Nelson et al., 2015). Monsoonal impact in the areas facing the South China Sea is evident by coastal erosion and in less and patchy distribution of the mangrove cover (Hamdan et al., 2012).

Remote sensing data and analysis

For mangrove species-level mapping at Brunei Bay, the Advanced Visible and Near-Infrared Radiometer type 2 (AVNIR-2) data acquired from the ALOS (spatial resolution: 10 m) (dated 1st September 2010) were used. The ALOS data—subjected to both atmospheric and geometric corrections—were provided by the Japan Science and Technology Agency (JST). However, to ensure a good match of the land-use/cover features in the imagery with the ground-truth observations, the data have been georeferenced again with WGS_1984 coordinate system using a toposheet (1:50,000) obtained from the Department of Survey and Mapping Malaysia (RMS error: 0.682) (ArcMap v.10). In order to have a better image processing and analysis, the mangrove areas adjacent to the bay and river channels were digitised on screen. For the medium resolution remote sensing data like ALOS, on screen digitisation through visual interpretation is indeed beneficial to separate mangrove and non-mangrove areas (Kuenzer et al., 2011). In this context, the false colour composite (FCC) (with 4-3-2 band combination) of the ALOS data was used to recognise the mangroves visible in a darker shade than of nearby terrestrial vegetation due to less spectral reflectance (Spalding, Kainuma & Collins, 2010; Zhang et al., 2017). The polygon features, as a new shapefile (with WGS_1984 coordinate system), were created only for mangroves under the jurisdiction of East Malaysia, and extracted the raster cells using spatial analyst tools (with ‘extract by polygon’ function in ArcMap v.10; ArcGIS, Redlands, CA, USA).

Species-level classification using all four (blue, green, red and near-infrared) bands was carried out through the maximum-likelihood algorithm that known to facilitate a robust classification for mangroves (Wang, Sousa & Gong, 2004; Shafri, Suhaili & Mansor, 2007; Satyanarayana et al., 2011; Kuenzer et al., 2011; Nguyen et al., 2013; Khatami, Mountrakis & Stehman, 2016; Mafi-Gholami, Mahmoudi & Zenner, 2017). In this context, the training samples for the most dominant mangrove taxa such as Nypa, Rhizophora, Sonneratia and Xylocarpus spp., were assigned based on ground knowledge acquired through the PCQM (68 sample points: Limbang—9, Sundar—19, Weston—20 and Menumbok—20) (details are given in ‘Ground inventory’). In addition, tonality and textural characteristics (in FCC) of the dominant species were considered (cf. Dahdouh-Guebas et al., 2000; NRC, 2006). For instance, Nypa vegetation was represented by a dark red colour with coarse texture, whereas Rhizophora by bright red with fine texture, Sonneratia by light red with open spaces (especially at river mouths), and Xylocarpus by a bright red colour with coarse texture (especially at back mangrove area). Based on our ground knowledge and visual interpretation of the satellite data, we assigned 20–25 training samples (with a pixel count of 125,135–572,692) for the widespread mangrove species (Nypa, Rhizophora spp.), and 10–15 (with a pixel count of 98,610–102,029) for the locally distributed species (Sonneratia, Xylocarpus spp.). The classified image was then subjected to an accuracy assessment through a confusion matrix (Congalton, 1991), for which 144 additional ground control points (GCPs) (Limbang—60, Sundar—45, Weston—26 and Menumbok—13) collected from both mangrove and non-mangrove areas were used. The GCPs were collected randomly as per the forest condition and accessibility. The location of the GCPs was marked on the ALOS imagery (hardcopy) and recorded the type of land-use/cover (i.e., mangrove or non-mangrove), together with species composition in the case of mangrove, for the accuracy assessment. Most locations in mangrove from where the GCPs were collected have shown the distribution of dominant species in more than 10 m × 10 m land-cover area.

For the accuracy assessment, we also report quantity (%)—the amount of pixels that differed between reference data and classification per class, exchange (%)—the allocated error by number of pixels that interchanged between two classes, and shift (%)—the other allocation differences that were not included in the exchange difference (cf. Pontius Jr & Millones, 2011; Pontius Jr & Santacruz, 2014). Any disagreements among quantity, exchange and shift variables are useful to learn the sources of error in the classification in a more interpretable manner (Pontius Jr & Millones, 2011). Estimates of quantity, exchange and shift were made from the PontiusMatrix41 (Microsoft Excel file) developed by RG Pontius (https://www.clarku.edu/).

Area statistics showing the total mangrove cover (with a species-level demarcation for each estuary) at Brunei Bay were derived from the supervised classification. For this purpose, the mangrove extent was treated under three sectors: Limbang, Sundar (up to Sipitang village) and Weston + Menumbok together (due to no clear-cut mangrove boundary between these two estuaries). However, there were a few limitations identified with the ALOS data. Firstly, the image was four years old at the time of fieldwork. Second, a (minor) patch of mangrove was missing on the northern most corner of Menumbok (Fig. 1). The former concern was evaluated through our ground inventory by observing the mangrove cover changes (if any), while the latter was identified from Google Earth Pro (image dated 29th November 2014) by digitising the missing (mangrove) area.

Ground inventory

For the ground data collection (5th–19th August 2014), two 100 m length transects—one close to the river mouth and another in mid-forest—were chosen from each estuary i.e., Limbang, Sundar, Weston and Menumbok (eight transects in total). Sampling at the river mouth was chosen to identify the pioneer group of vegetation (i.e., mangrove succession by select species), whereas in the mid-forest to identify other available species in the vicinity. We took a 100 m transect in order to cover at least ten sample points with four quadrants of the PCQM (cf. Engeman et al., 1994; Satyanarayana et al., 2002; Satyanarayana et al., 2011). Different adult tree (i.e., ≥1.3 m height with a diameter (D₁₃₀) of ≥2.5 cm or girth (G₁₃₀) ≥8 cm) structural parameters such as density (trees 0.1 ha⁻¹), basal area (m² 0.1 ha⁻¹), relative density (%), relative dominance (%), relative frequency (%), species’ importance value (IV) (relative density + relative dominance + relative frequency), and the complexity index (CI) were estimated using the P-DATA PRO v. 5.01 interface developed by Dahdouh-Guebas & Koedam (2006). In the case of Nypa—as its stem remains underground—the diameter of all leaf shoots was considered to determine the (average) basal area. For mangrove species identification, the nomenclature suggested by Tomlinson (1986) and Duke (2006) was followed. Each tree height was measured with the help of a clinometer (Suunto PM-5, Finland). A hand-held Global Positioning System (Garmin 45; Garmin, Olathe, KS, USA) was used for navigation and to obtain the latitude and longitude positions of the sampling points.

Statistical analysis

Variation between the tree structural parameters, like density and basal area, at Limbang, Sundar, Weston and Menumbok estuaries was tested through one-way ANOVA (OriginPro v. 9.1).

Remote sensing based observations

From the supervised classification of the ALOS data (Fig. 2), it was possible to recognise the predominance of Sonneratia caseolaris (L.) Engler, Rhizophora apiculata Bl. and Nypa fruticans (Thunb.) Wurmb., along with S. alba J Smith and Xylocarpus granatum König species at Brunei Bay (accuracy: 80% and Kappa index: 0.714) (Table 2). The difference in the amount of pixels between the reference data and the classification per class (= quantity) was found to be 9%, whereas the error due to interchanged pixels between two classes (= exchange) was 7% and the allocation difference other than to the exchange difference (= shift) was 1%. Among the five mangrove species, N. fruticans showed the highest quantity (9%) and exchange (7%) differences (Fig. 3). In fact, there was a considerable overlap in the (visible range) spectral reflectance values of the dominant mangrove species (Fig. 4). The Malaysian mangrove cover at Brunei Bay was found to be ca. 35,183.74 ha of which Limbang occupied 5,011.42 ha, Sundar (up to Sipitang village) 9,606.46 ha, and Weston + Menumbok 20,565.86 ha (Table 3). If the spatial extent of each mangrove species is considered, N. fruticans shows a widespread distribution (14,879.27 ha), followed by R. apiculata (12,801.89 ha), S. caseolaris (5,533.10 ha), X. granatum (993.10 ha) and S. alba (976.38 ha) (Table 3). Area wise, Limbang, Weston and Menumbok were dominated by N. fruticans (as an important species), and Sundar by R. apiculata (Table 3).

Ground inventory observations

Ground-truth observations found several non-dominant mangrove species like Avicennia alba Blume, Bruguiera gymnorrhiza (L.) Lamk., B. cylindrica (L.) Blume, Ceriops sp., Heritiera littoralis Dryand and Kandelia candel (L.) Druce, along with the associates such as Acanthus ilicifolius L., Acrostichum aureum L., Derris trifoliata Lour. and Hibiscus tilliaceus L., in the vicinity. Out of 11 dominant and non-dominant mangrove species, only nine were encountered in the vegetation (PCQM) sampling points (Table 4). Among the four estuaries, Limbang had the highest mangrove basal area (120.17 m² 0.1 ha⁻¹ with a density of 163 trees 0.1 ha⁻¹), followed by Menumbok (81.17 m² 0.1 ha⁻¹ with a density of 132 trees 0.1 ha⁻¹), Weston (35.10 m² 0.1 ha⁻¹ with a density of 130 trees 0.1 ha⁻¹), and Sundar (33.24 m² 0.1 ha⁻¹ with a density of 134 trees 0.1 ha⁻¹). Basal area at Limbang and Menumbok was significantly different from Sundar and Weston (one-way ANOVA, P = 0.08). Sonneratia caseolaris holds the highest importance value for Limbang, while R. apiculata for both Sundar and Menumbok, and N. fruticans for Weston. The mangrove stand structural complexity (derived from the complexity index) was high in the order of Limbang >  Menumbok >  Sundar >  Weston (Table 4). In terms of the diameter class distribution, more than 50% of trees at Limbang were represented by 31–90 cm (range: 9.5–198.9 cm), Sundar 2.5–40 cm (range: 3.2–186.1 cm), Weston 2.5–60 cm (range: 3.0–190.1 cm), and Menumbok 2.5–120 cm (range: 2.5–250 cm) (Fig. 5).

The ground inventory also revealed no detectable changes in relation to the four-year old satellite (ALOS) data. The core mangroves were found intact in almost all locations between 2010 and 2014. The only sign of change was observed at Limbang, Sundar and Weston river mouths where shrubby vegetation (as understood from the visual interpretation of the ALOS data before the fieldwork) had become grown-up Sonneratia trees (height: 9–13 m). The missing mangrove patch (observed from the Google Earth) on the northern most corner of Menumbok was ca. 3,741 ha and found to be co-dominated by R. apiculata and N. fruticans (similar to other mangrove patches at Menumbok).