Estimation and mapping of above-ground biomass of mangrove forests and their replacement land uses in the Philippines using Sentinel imagery

https://doi.org/10.1016/j.isprsjprs.2017.10.016Get rights and content

Abstract

The recent launch of the Sentinel-1 (SAR) and Sentinel-2 (multispectral) missions offers a new opportunity for land-based biomass mapping and monitoring especially in the tropics where deforestation is highest. Yet, unlike in agriculture and inland land uses, the use of Sentinel imagery has not been evaluated for biomass retrieval in mangrove forest and the non-forest land uses that replaced mangroves. In this study, we evaluated the ability of Sentinel imagery for the retrieval and predictive mapping of above-ground biomass of mangroves and their replacement land uses. We used Sentinel SAR and multispectral imagery to develop biomass prediction models through the conventional linear regression and novel Machine Learning algorithms. We developed models each from SAR raw polarisation backscatter data, multispectral bands, vegetation indices, and canopy biophysical variables. The results show that the model based on biophysical variable Leaf Area Index (LAI) derived from Sentinel-2 was more accurate in predicting the overall above-ground biomass. In contrast, the model which utilised optical bands had the lowest accuracy. However, the SAR-based model was more accurate in predicting the biomass in the usually deficient to low vegetation cover non-forest replacement land uses such as abandoned aquaculture pond, cleared mangrove and abandoned salt pond. These models had 0.82–0.83 correlation/agreement of observed and predicted value, and root mean square error of 27.8–28.5 Mg ha−1. Among the Sentinel-2 multispectral bands, the red and red edge bands (bands 4, 5 and 7), combined with elevation data, were the best variable set combination for biomass prediction. The red edge-based Inverted Red-Edge Chlorophyll Index had the highest prediction accuracy among the vegetation indices. Overall, Sentinel-1 SAR and Sentinel-2 multispectral imagery can provide satisfactory results in the retrieval and predictive mapping of the above-ground biomass of mangroves and the replacement non-forest land uses, especially with the inclusion of elevation data. The study demonstrates encouraging results in biomass mapping of mangroves and other coastal land uses in the tropics using the freely accessible and relatively high-resolution Sentinel imagery.

Introduction

Mangroves are an important coastal resource in the tropics. They provide many ecosystem goods and services including the provision of wood for construction and fuel, habitat of coastal fauna and nursery of juvenile marine organisms, carbon (C) storage in biomass and soil, protection from strong winds during typhoons and coastal erosion mitigation (Alongi, 2009, Donato et al., 2011). However, there has been a large reduction in the global mangrove forest cover due to conversion to non-forest land uses such as aquaculture, perennial agriculture and clearing for human settlement (FAO, 2007). This is especially true in tropical Southeast Asia where more than 100,000 ha of mangroves were deforested and converted to other land uses during the last 15 years, notably for aquaculture and agriculture (Richards and Friess, 2016). It is crucial, therefore, to monitor the mangroves against land use change and forest degradation.

Empirical studies that quantify the carbon stocks of mangroves and the land uses that replaced them are needed in order to provide emission estimates based on actual measures of carbon stocks and reduce the uncertainty of the estimate. In addition, climate mitigation programs such as Reducing Emissions from Deforestation and Forest Degradation Plus (REDD+) are being proposed to prevent large emissions from deforestation and forest degradation in the tropics. These programs will require accurate assessment and mapping to establish the baseline biomass and C stocks against which to monitor future changes (Maraseni et al., 2005). Integrated coastal management would also require relevant maps such as biomass maps for better planning and decision-making.

Above-ground biomass is one of the important carbon pools in mangrove ecosystem (Kauffman and Donato, 2012, Howard et al., 2014). There has been a growing body of literature on mangrove biomass and their carbon stocks (e.g. Donato et al., 2011, Abino et al., 2014, Tue et al., 2014, Phang et al., 2015, Stringer et al., 2015, Vien et al., 2016). However, only a few studies have quantified biomass and carbon stock of mangrove forest side by side their replacement land uses such as aquaculture pond (Kauffman et al., 2013, Bhomia et al., 2016, Duncan et al., 2016) and cattle pastures (Kauffman et al., 2016). Such studies could help in quantifying the differences in carbon stock and hence the emission from conversion (Maraseni et al., 2008). These studies, however, have utilised field plots to estimate biomass and infer the stock for the whole study site. This approach is sufficient only for a few hectares, but costly and slow if implemented over large areas. It is also difficult to implement in remote and treacherous portions in a larger landscape. The use of satellite remote sensing techniques offers cost and time advantages in implementing large-scale biomass assessment. For this approach, remote sensing-based biomass assessment utilises the relationships between field-measured biomass data, imagery and other thematic maps to develop models that predict biomass in different locations of the study site. The outcome of remote sensing-based biomass estimation is a spatially-explicit pattern of the total above-ground biomass and its variations for the entire area (Samalca, 2007).

Satellite image-based biomass prediction models can be derived from radar backscatter polarisations, multispectral bands, vegetation index [e.g. Normalised Difference Vegetation Index (NDVI)], and vegetation cover biophysical variables [e.g. Leaf Area Index (LAI)]. These models can be developed with or without ancillary thematic map data (e.g. elevation; Lu et al., 2004, Simard et al., 2006, Fatoyinbo et al., 2008, Kumar et al., 2012, Sarker et al., 2012, Jachowski et al., 2013, Dube and Mutanga, 2015, Dusseux et al., 2015, Dube and Mutanga, 2016). In the tropics, however, during the rainy season where clouds are persistent, the use of multispectral image is challenging. In contrast, data from space-borne synthetic aperture radar (SAR) sensors are independent of daytime and weather conditions, and can provide valuable data for the monitoring of land cover.

Previous satellite remote sensing-based biomass retrieval and mapping studies in coastal areas have dealt mostly on mangrove forest alone (Simard et al., 2006, Proisy et al., 2007, Fatoyinbo et al., 2008, Jachowski et al., 2013, Kovacs et al., 2013, Aslan et al., 2016, Pham and Brabyn, 2017), and did not cover the land uses that replaced mangroves. This gap could be an important basis for productivity quantification and comparison with original land use. Simard et al. (2006) utilised SRTM elevation data to map the height of mangroves in the Everglades using linear regression with field data and used that mangrove height map and a local mangrove tree height-biomass equation to eventually map the biomass of mangrove therein. Proisy et al. (2007) used high-resolution IKONOS imagery and field data, and employed Fourier-based textural ordination from canopy grain analysis to model and map the mangrove biomass in French Guiana. Fatoyinbo et al. (2008) also used STRM elevation data and field data to map the mangrove height in Mozambique using linear regression and applied a general mangrove height-biomass equation to map the mangrove biomass in the area. In contrast, Jachowski et al. (2013) made use of high-resolution GeoEye-1 imagery and field data to estimate and map the biomass mangroves in Thailand using a suite of machine learning algorithm. Aslan et al. (2016) also used SRTM elevation and field data to map the mangrove height in Indonesia using linear regression but utilised non-linear quartile regression to generate biomass map of mangroves in the area using the mangrove height map and field biomass. More recently, Pham and Brabyn (2017) used SPOT images and object-based approach in combination with machine learning algorithms to estimate the biomass change of a mangrove forest in Vietnam.

The recent launch of the new-generation Sentinel-1 (SAR) and Sentinel-2 (multispectral) satellite missions of the Copernicus program of the European Space Station is expected to provide new capabilities for monitoring and mapping of biomass in the coastal zone of the tropics. Sentinel-1 provides radar imagery with HH+HV or VV+VH polarisations in C-band (Sentinel-1_Team, 2013) while Sentinel-2 offers 13 multispectral bands, including three vegetation red edge bands and two infrared bands, in addition to visible and near infrared bands (Sentinel-2_Team, 2015). However, to our knowledge, the retrieval and mapping of the biomass of mangrove forest and land uses that replaced them from data acquired by these newly launched multispectral and SAR instruments onboard the Sentinel-1 and Sentinel-2 satellite missions have not been reported yet in the scientific literature. Therefore, pioneering studies are needed to assess these new-generation satellite imagery.

In this study, we evaluated the ability of data from Sentinel-1 and Sentinel-2 imagery for the retrieval and predictive mapping of above-ground biomass of mangroves and the associated replacement land uses in a coastal area in the tropics. The specific objectives of the study included the following: (1) to determine and model the relationship between field-measured above-ground biomass and Sentinel-1 SAR backscatter coefficients and Sentinel-2 multispectral reflectance from mangrove forest and replacement non-forest land uses, (2) to evaluate the accuracy of the biomass prediction models, and (3) to evaluate the accuracy of the output predictive biomass maps. We developed and evaluated above-ground biomass models and predictive maps derived from Sentinel-1 SAR imagery, Sentinel-2 multispectral bands, Sentinel-2-derived vegetation indices (e.g. NDVI) and Sentinel-2-derived vegetation biophysical variables (e.g. LAI). The novelty of this paper is the use of Sentinel-1 and Sentinel-2 imagery in the estimation and mapping of the biomass of mangrove forests and non-forest land uses that replaced mangroves. This study attempted to contribute in developing remote sensing-based biomass predictive mapping techniques for mangrove area. It is a pioneering study that utilised Sentinel-1 SAR and Sentinel-2 optical data for biomass modelling and mapping of mangrove forests and non-forest land uses that replaced mangroves in tropical areas.

Section snippets

Study site

The study site is situated on the southern coast of Honda Bay within the administrative jurisdiction of Puerto Princesa City in the island province of Palawan, Philippines. It is geographically located between latitude 9.8028° to 9.9612°N and longitude 118.725° to 118.805°E (Fig. 1). The city is located in the central part of the province and is approximately 567 km south-west of Manila, the country’s capital. The climate in the study site is tropical, with two seasons (dry and wet) and under

Sentinel-1 (SAR) polarisations

There was an increase in the backscatter values as the above-ground biomass increases, i.e. from nil in aquaculture pond, salt pond and cleared mangrove to low biomass in the coconut plantation, and low to high biomass in mangroves (Fig. 4). Backscatter (dB) values of vegetated mangrove and coconuts were comparable, −13.41 and −13.19 in VH polarisation and −7.51 and −7.86 in VV polarisation, respectively. Non-vegetated areas under aquaculture pond, salt pond and cleared mangrove had lower

Relationship of field biomass with sentinel SAR polarisations and multispectral bands

Single-date SAR VH channel had a higher correlation with biomass than VV polarisation. However, the combination of multi-temporal VH and VV polarisations correlated better with biomass than single-date SAR imagery. Proisy et al. (2003) also reported similar observation for the C-band of the airborne-AIRSAR where cross-polarised HV channel had better correlation with biomass than co-polarised VV channel for both mangrove forests in Northern Australia and French Guiana. Also, Kumar et al. (2012)

Conclusion

Sentinel-1 (SAR) and Sentinel-2 (multispectral) image data can be used for biomass retrieval and mapping of the coastal land uses, mangrove and non-mangrove alike, of Honda Bay and adjacent coastal areas in Puerto Princesa City, Philippines. The prediction accuracy is comparable to imagery from current commercial sensors. High correlation values (r = 0.84) between biomass and Sentinel imagery data were obtained from the combination of dual-date SAR VH and VV channels, red and red edge bands,

Acknowledgement

We would like to thank Australia Awards Scholarship for the full-time scholarship of the first author (ST000FN15) and for funding his round-trip airfares during the fieldwork in the Philippines, and the University of Southern Queensland, Australia for contributing some funds for the study. We are also grateful to Ecosystems Research and Development Bureau (ERDB) management and staff in the Philippines for providing logistics and fieldwork support. The help and company of Elmer Caliwagan,

References (57)

  • M. Sibanda et al.

    Examining the potential of Sentinel-2 MSI spectral resolution in quantifying above ground biomass across different fertilizer treatments

    ISPRS J. Photogrammetry Remote Sens.

    (2015)
  • C.E. Stringer et al.

    Carbon stocks of mangroves within the Zambezi River Delta, Mozambique

    Forest Ecol. Manage.

    (2015)
  • R.B. Thapa et al.

    Potential of high-resolution ALOS–PALSAR mosaic texture for aboveground forest carbon tracking in tropical region

    Remote Sens. Environ.

    (2015)
  • N.T. Tue et al.

    Carbon storage of a tropical mangrove forest in Mui Ca Mau National Park, Vietnam

    CATENA

    (2014)
  • A.C. Abino et al.

    Species diversity, biomass, and carbon stock assessments of a natural mangrove forest in palawan, philippines

    Pak. J. Bot.

    (2014)
  • D. Alongi

    The Energetics of Mangrove Forests

    (2009)
  • R.K. Bhomia et al.

    Impacts of land use on Indian mangrove forest carbon stocks: implications for conservation and management

    Ecol. Appl.

    (2016)
  • Brown, S., 1997, Estimating Biomass and Biomass Change of Tropical Forests: A Primer, vol. 134, Food & Agriculture...
  • J. Chave et al.

    Tree allometry and improved estimation of carbon stocks and balance in tropical forests

    Oecologia

    (2005)
  • Cintron, G., Novelli, Y.S., 1984, 'Methods for Studying Mangrove Structure', in Mangrove Ecosystem: Research Methods,...
  • D.C. Donato et al.

    Mangroves among the most carbon-rich forests in the tropics

    Nat. Geosci.

    (2011)
  • Dube, T., Mutanga, O., 2015. 'Investigating the robustness of the new Landsat-8 Operational Land Imager derived texture...
  • Dube, T., Mutanga, O., 2016, 'The impact of integrating WorldView-2 sensor and environmental variables in estimating...
  • Dube, T., Gara, T.W., Mutanga, O., Sibanda, M., Shoko, C., Murwira, A., Masocha, M., Ndaimani, H., Hatendi, C.M., 2016,...
  • ESA, 2016, Sentinels Scientific Data Hub, European Space Agency, viewed August 1,...
  • FAO

    The World’s Mangroves 1980–2005

    (2007)
  • T.E. Fatoyinbo et al.

    Landscape-scale extent, height, biomass, and carbon estimation of Mozambique's mangrove forests with Landsat ETM+ and Shuttle Radar Topography Mission elevation data

    J. Geophys. Res.: Biogeosci.

    (2008)
  • M. Hall et al.

    The WEKA data mining software: an update

    ACM SIGKDD Explorations Newsl

    (2009)
  • Cited by (209)

    View all citing articles on Scopus
    View full text