Sanderman 2018 Environ. Res. Lett. 13 055002

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Environmental Research Letters

LETTER • OPEN ACCESS Related content


- A global predictive model of carbon in
A global map of mangrove forest soil carbon at 30 mangrove soils
Sunny L Jardine and Juha V Siikamäki
m spatial resolution - Estimating mangrove aboveground
biomass from airborne LiDAR data: a case
study from the Zambezi River delta
To cite this article: Jonathan Sanderman et al 2018 Environ. Res. Lett. 13 055002 Temilola Fatoyinbo, Emanuelle A
Feliciano, David Lagomasino et al.

- Measurement and monitoring needs,


capabilities and potential for addressing
reduced emissions from deforestation and
View the article online for updates and enhancements. forest degradation under REDD+
Scott J Goetz, Matthew Hansen, Richard A
Houghton et al.

This content was downloaded from IP address 217.138.30.50 on 15/05/2018 at 12:37


Environ. Res. Lett. 13 (2018) 055002 https://doi.org/10.1088/1748-9326/aabe1c

LETTER

A global map of mangrove forest soil carbon at 30 m


OPEN ACCESS
spatial resolution
RECEIVED
29 June 2017 Jonathan Sanderman1,21 , Tomislav Hengl2, Greg Fiske1, Kylen Solvik1, Maria Fernanda Adame3, Lisa
REVISED Benson5,6, Jacob J Bukoski7, Paul Carnell8, Miguel Cifuentes-Jara9, Daniel Donato19, Clare Duncan4,8,
10 April 2018
Ebrahem M Eid10,20, Philine zu Ermgassen17,18, Carolyn J Ewers Lewis8, Peter I Macreadie8 , Leah Glass5,
Selena Gress11, Sunny L Jardine12, Trevor G Jones5,13, Eugéne Ndemem Nsombo14, Md Mizanur Rahman15,
ACCEPTED FOR PUBLICATION
13 April 2018
PUBLISHED
Christian J Sanders16, Mark Spalding17 and Emily Landis17
30 April 2018 1 Woods Hole Research Center, 149 Woods Hole Road, Falmouth, MA 02540, United States of America
2 ISRIC — World Soil Information, Wageningen, The Netherlands
3 Australian Rivers Institute, Griffith University, Nathan, QLD, Australia
Original content from 4 Institute of Zoology, Zoological Society of London, Outer Circle, Regent’s Park, London NW1 4RY, United Kingdom
this work may be used 5 Blue Ventures Conservation, London, United Kingdom
under the terms of the 6
Creative Commons Centre for Environment, Fisheries and Aquaculture Science, Lowestoft, United Kingdom
Attribution 3.0 licence. 7 Department of Environmental Science, Policy and Management, University of California, Berkeley, CA, United States of America
8 Centre for Integrative Ecology, School of Life and Environmental Sciences, Deakin University, Burwood, VIC, Australia
Any further distribution
of this work must 9 Forests, Biodiversity and Climate Change Program, CATIE 7170, Turrialba, Costa Rica
maintain attribution to 10 Botany Department, Faculty of Science, Kafr El-Sheikh University, Kafr El-Sheikh, Egypt
the author(s) and the 11 School of Applied Sciences, Edinburgh Napier University, Edinburgh, Scotland
title of the work, journal 12
citation and DOI. School of Marine and Environmental Affairs, University of Washington, Seattle, WA, United States of America
13 Department of Forest Resources Management, University of British Columbia, Vancouver, BC, Canada
14 Institute of Fisheries and Aquatic Sciences, University of Douala, Doula, Cameroon
15 Graduate School of Agriculture, Kyoto University, Kyoto, Japan
16 National Marine Science Centre, Southern Cross University, Coffs Harbour, NSW, Australia
17 The Nature Conservancy, Arlington, VA, United States of America
18 School of Geosciences, University of Edinburgh, Edinburgh, United Kingdom
19 Washington State Department of Natural Resources, Olympia, WA United States of America
20 Biology Department, College of Science, King Khalid University, Abha 61321, Saudi Arabia
21 Author to whom any correspondence should be addressed.

E-mail: [email protected]

Keywords: blue carbon, carbon sequestration, land use change, machine learning

Supplementary material for this article is available online

Abstract
With the growing recognition that effective action on climate change will require a combination of
emissions reductions and carbon sequestration, protecting, enhancing and restoring natural carbon
sinks have become political priorities. Mangrove forests are considered some of the most
carbon-dense ecosystems in the world with most of the carbon stored in the soil. In order for
mangrove forests to be included in climate mitigation efforts, knowledge of the spatial distribution of
mangrove soil carbon stocks are critical. Current global estimates do not capture enough of the finer
scale variability that would be required to inform local decisions on siting protection and restoration
projects. To close this knowledge gap, we have compiled a large georeferenced database of mangrove
soil carbon measurements and developed a novel machine-learning based statistical model of the
distribution of carbon density using spatially comprehensive data at a 30 m resolution. This model,
which included a prior estimate of soil carbon from the global SoilGrids 250 m model, was able to
capture 63% of the vertical and horizontal variability in soil organic carbon density (RMSE of
10.9 kg m−3 ). Of the local variables, total suspended sediment load and Landsat imagery were the
most important variable explaining soil carbon density. Projecting this model across the global
mangrove forest distribution for the year 2000 yielded an estimate of 6.4 Pg C for the top meter of soil
with an 86–729 Mg C ha−1 range across all pixels. By utilizing remotely-sensed mangrove forest cover
change data, loss of soil carbon due to mangrove habitat loss between 2000 and 2015 was 30–122 Tg C
with >75% of this loss attributable to Indonesia, Malaysia and Myanmar. The resulting map products

© 2018 The Author(s). Published by IOP Publishing Ltd


Environ. Res. Lett. 13 (2018) 055002

from this work are intended to serve nations seeking to include mangrove habitats in payment-for-
ecosystem services projects and in designing effective mangrove conservation strategies.

1. Introduction carbon storage in tall mangroves but 96%–99% of


total ecosystem carbon in medium and low stature
Mangrove forests, occupying less than 14 million ha mangrove stands. Importantly, Kauffman et al (2014)
(Giri et al 2011), just 2.5% of the size of the Ama- found that conversion of these mangrove forests to
zon rainforest, provide a broad array of ecosystem shrimp ponds resulted in the loss of 90% of this car-
services (Barbier et al 2011). Mangroves are critical bon from the top 3 m of soil (612–1036 Mg C ha−1 ).
nursery habitats for fish, birds and marine mammals In addition to avoided emissions, many mangrove for-
(Mumby et al 2004, Nagelkerken et al 2008), act as est soils are accreting as sea level rises (Krauss et al
effective nutrient filters (Robertson and Phillips 1995), 2014), providing continual carbon sequestration on
buffer coastal communities from storm surges (Gedan the order of 1.3–2.0 Mg C ha−1 yr−1 (Breithaupt et al
et al 2011) and support numerous rural economies 2012, Chmura et al 2003). Clearly, there can be a
(Spalding et al 2014, Temmerman et al 2013). These major climate benefit to halting or even slowing the rate
ecosystem service benefits have been valued at an aver- of mangrove conversion, with a rough potential esti-
age of 4200 US$ ha−1 yr−1 in Southeast Asia (Brander mated to be 25–122 Tg C yr−1 (Pendleton et al 2012,
et al 2012). Despite these ecosystem service bene- Siikamäki et al 2012). For nations with large mangrove
fits, mangroves are highly threatened by both urban holdings, protection and restoration can make major
expansion and other ‘higher value’ land uses because contributions to meeting climate mitigation targets
of their close proximity to major human settlements. (Herr and Landis 2016).
There are no reliable estimates of original mangrove While many mangrove forests do accumulate large
cover, but some authors have suggested that 35% or quantities of soil carbon, others do not. There can be
more of original cover may have been lost and wider significant variability in soil carbon stocks across dif-
areas have been degraded (Valiela et al 2001, Spald- ferent mangrove forests (Jardine and Siikamäki 2014)
ing et al 2010). Loss rates have slowed dramatically but also within the same mangrove forest (Adame
in the past 10–20 years in most areas, however they et al 2015, Kauffman et al 2011). Understanding the
remain considerable, with rates up to 3.1% annu- distribution of soil carbon in mangrove forests will be
ally in some countries (Hamilton and Casey 2016). very important in prioritizing protection and restora-
The major drivers of loss are conversion for aquacul- tion efforts for climate mitigation. The controls on soil
ture, especially shrimp farming, agriculture and urban carbon stocks are diverse and are likely scale dependent;
development (Alongi 2002, Valiela et al 2001, Spald- however, some generalizations can be made. Man-
ing et al 2010, Richards and Friess 2016) but loss due grove forests, no matter how productive, will struggle
to extreme climatic events are also becoming more to have high soil carbon stocks in the upper meter
common (Duke et al 2017). of soil if they receive large annual sediment loads.
With the growing recognition that effective action Mangrove forests in river deltas, such as the Sundar-
on climate change will require a combination of bans (Banerjee et al 2012) and the Zambezi river delta
emissions reductions and removals (Rockström et al in Mozambique (Stringer et al 2016), typically only
2017), protecting, enhancing and restoring natural car- contain a few percent organic carbon throughout the
bon sinks have become political priorities (Boucher soil profile. These locations may still have very high
et al 2016, Grassi et al 2017). Mangrove forests can carbons stocks, but the density of carbon is low due
play an important role in carbon removals; in addition to the high allocthonous input of mineral sediments.
to being some of the most carbon-dense ecosystems Conversely, forests with moderately low productivity
in the world (Donato et al 2011), if kept undisturbed, can accumulate large amounts of soil carbon if they
mangrove forest soils act as long-term carbon sinks are in an isolated hydrogeomorphic setting (Ezcurra
(Breithaupt et al 2012). As such, there is strong interest et al 2016). Within the same mangrove forest there
in developing policy tools to protect and restore man- are typically steep hydrogeomorphic gradients from
groves through payment for ecosystem services (Friess the seaward to landward extent of the forest which
et al 2016, Howard et al 2017). results in zonation of both vegetation (Snedaker 1982)
Mangroves can store significant amounts of car- and soil carbon storage (Kauffman et al 2011, Ouyang
bon in their biomass (Hutchison et al 2014); however, et al 2017, Ewers Lewis et al 2018) but not necessarily
the vast majority of the ecosystem carbon storage is for the same reasons. Within a similar hydrogeomor-
typically found in the soil (Donato et al 2011, Mur- phic position, forest productivity and soil edaphic
diyarso et al 2015, Sanders et al 2016). For example, conditions (e.g. redox potential, pH, salinity) driving
Kauffman et al (2014) found that within the same estu- decomposition rates are often the dominant controls
ary, soil carbon contributed 78% of total ecosystem on soil carbon density. Consideration of this nested

2
Environ. Res. Lett. 13 (2018) 055002

hierarchy of controls will be necessary to successfully of OCD in three dimensions (soil depth used as a
capture the variability in soil carbon at both local and predictor) using all soil horizons layers at different
global scales. depths, which means that a single statistical model
Accurate estimates and an understanding of the can be used to predict OCD at any arbitrary depth.
spatial distribution of mangrove soil carbon stocks This 3D approach to modeling OCD reduces the need
are a critical first step in understanding climatic and for making complex assumptions about the downcore
anthropogenic impacts on mangrove carbon stor- trends in OCD, and maximizes the use of collected
age and in realizing the climate mitigation potential data.
of these ecosystems through various policy mecha- The derived spatial prediction model is then used
nisms (Howard et al 2017). Previous global estimates to predict OCD at standard depths 0, 30, 100, and
(Atwood et al 2017, Jardine and Siikamäki 2014), 200 cm, so that the organic carbon stock (OCS) can
do not capture enough of the finer scale spatial be derived as a cumulative sum of the layers down
variability that would be required to inform local deci- to the prediction depth for every 30 m pixel identi-
sions on siting protection and restoration projects. fied as having mangrove forest in the year 2000 (Giri
To close this information gap, we have: (1) com- et al 2011). Importantly, we found that there is a spatial
piled and published a harmonized global database of mismatch between the global mangrove forest dis-
the profile distribution of soil carbon under man- tribution (GMFD) of Giri et al (2011) and satellite
groves, (2) used this database to develop a novel imagery (figure S2). To best resolve this spatial mis-
machine-learning based data-driven statistical model match, we have adjusted the GMFD by growing all
of the distribution of carbon density using spatially vectors by one pixel (∼30 m) and then filtering out any
comprehensive data at an ∼30 m resolution, (3) pro- pixel that falls over water by using Landsat NIR band
jected the model results across global mangrove habitat (see SI for more details).
for the year 2000 (Giri et al 2011), and (4) over- Environmental covariates have been compiled to
laid estimates of mangrove forest change between represent the postulated major controls on OCS in
2000 and 2012 (Hamilton and Casey 2016) to esti- soils generally (McBratney et al 2003) and specifi-
mate potential soil carbon emissions from recent forest cally for mangrove ecosystems (Balke and Friess 2016).
conversion. Covariates included:

2. Methods 1. Vegetation characteristics including percent


forest cover (Hansen et al 2013) and Landsat
2.1. Mangrove soil carbon database bands 3 (red), 4 (near infrared), 5 (shortwave
A harmonized globally representative database (avail- infrared) and 7 (shortwave infrared) for the year
able at: 10.7910/DVN/OCYUIT) was compiled from 2000 (Hanson et al 2013) retrieved from http://
peer-reviewed literature, grey literature and from con- earthenginepartners.appspot.com/science-2013-gl
tributions of unpublished data from a number of obal-forest/download_v1.3.html;
researchers and organizations. Details of database 2. Digital elevation data, which at or near sea-level
development and a statistical summary of the data approximately follows forest canopy height (Simard
are given in the supplemental information available et al 2006), from the shuttle radar topography mis-
at stacks.iop.org/ERL/13/055002/mmedia. sion (SRTM GL1; NASA, 2013) was retrieved from
https://lpdaac.usgs.gov, maintained by the NASA
2.2. Spatial modelling of soil organic carbon EOSDIS Land Processes Distributed Active Archive
In order to maximize the utilization of available soil Center (LP DAAC) at the USGS/Earth Resources
carbon data, we developed a machine learning-based Observation and Science (EROS) Center, Sioux
model of organic carbon density (OCD) which models Falls, South Dakota;
OCD as a function of depth (d), an initial estimate of 3. Long-term averaged (1990–2010) monthly sea sur-
the 0–200 cm organic carbon stock (OCS) from the face temperature (SST) averaged into four seasons
global SoilGrids 250 m model (Hengl et al 2017), and were generated in Google Earth Engine from NOAA
a suite of spatially explicit covariate layers (X𝑝 ): AVHRR Pathfinder Version 5.2 Level 3 Collated
data (Casey et al 2010) and downscaled to 30 m
OCD(𝑥𝑦𝑑) = 𝑑 + OCSSG + 𝑋1 (𝑥𝑦) resolution using bicubic resampling;
+𝑋2 (𝑥𝑦) + ...𝑋𝑝 (𝑥𝑦)
4. The M2 tidal elevation amplitude product
(FES2012) from a global hydrodynamic tidal model
where OCSSG is the aggregated organic carbon stock which assimilates altimetry data from multiple plat-
estimated for 0–200 cm depth using global SoilGrids forms was used to represent tidal range at each
250 m approach down-sampled from 250 m–30 m res- location. The FES2012 product was produced by
olution, and xyd are the 3D coordinates northing Noveltis, Legos and CLS Space Oceanography Divi-
easting and soil depth (measured to center of a hori- sion and distributed by Aviso, with support from
zon). Note here that we model spatial distribution Cnes (www.aviso.altimetry.fr/);

3
Environ. Res. Lett. 13 (2018) 055002

5. Averaged (2003–2011) monthly total suspended 2.3. Data analysis


matter (TSM) averaged into four seasons estimated Soil carbon stocks were calculated for the global extent
from MERIS imagery collected by the European of mangroves for the year 2000 by summing the OCS in
Space Agency’s Envisat satellite. Processed and vali- each pixel for 1 and 2 m depths. Country level carbon
dated TSM data was retrieved from the GlobColour stocks were also calculated for the same depths. Given
project (http://hermes.acri.fr). the fringing nature of mangroves, a global spatial vector
6. A mangrove typology map delineating mangroves data layer was built that allocated the offshore area for
into estuaries and then either organogenic or each country where mangrove forests can be found. It
mineralogenic based on an analysis of TSM and was derived from the Exclusive Economic Zone for each
tidal amplitude data (Zu Ermgassen, unpublished country. This layer was then dissolved with the onshore
data). areas for each associated country and subsequently used
to quantify mangrove OCS tonnage and areal extent.
Potential loss of OCS due to mangrove habitat
Sea surface temperature, tidal amplitude and TSM conversion was calculated between 2000 and 2015 by
are 4 km resolution ocean products and needed to summing the OCS in mangrove forest pixels which
be extrapolated to each pixel containing mangrove were identified to be deforested. While this analysis
forest. Missing values in the sea surface tempera- cannot distinguish between natural and anthropogenic
ture, tidal amplitude and TSM were first filled-in disturbance, human-driven land use change is believed
using spline interpolation in SAGA GIS, then down- to be by far the dominant driver of deforestation
scaled to 30 m resolution using bicubic resampling in in mangrove ecosystems (Alongi 2002, Murdiyarso
GDAL. Including SoilGrids and depth, there were a et al 2015). We define deforested using the Global
total of 20 covariates used in building the mangrove Forest Change dataset (Hansen et al 2013) available
OCD model. online from: http://earthenginepartners.appspot.com/
The ability of the training points to represent science-2013-global-forest. We chose to use this
the entire covariate space of the global mangrove approach for estimating deforestation instead of using
domain was assessed by conducting a principal com- the derived mangrove tree cover loss data produced
ponents analysis (PCA) on 15 000 randomly selected by Hamilton and Casey (2016) as used by Atwood
points and the 1613 points used in the spatial et al (2017) because the Hamilton and Casey (2016)
model. Spatial variables were detrended and centered analysis only considered forested area as area actually
by subtracting the mean and dividing by the stan- covered by trees (i.e. if a 100 ha forest has 80% tree
dard deviation (s.d.) before entering into the PCA cover then it is counted as 80 ha of forest). In our opin-
analysis. ion, this definition mischaracterizes forest area extent.
Soil carbon typically varies in highly non-linear Next, an estimate of the soil carbon emissions associ-
ways with depth and across the landscape and as ated with land use conversion is needed. The amount
such the ability of standard parametric models to cap- of OCS lost can be highly variable and depends on
ture this variation is limited (Jardine and Siikamäki the new land use (Kauffman et al 2014, 2016b, Jones
2014, Hengl et al 2017). Here we model the spatial et al 2015) and probably on soil properties. Pendle-
(xyd) distribution of OCD using a machine learn- ton et al (2012) used a 25%–100% loss range. Donato
ing random forest model implemented in the ranger et al (2011) used a low estimate of 25% of the OCS
package (Wright and Ziegler 2015) in the R environ- in top 30 cm and 75% in top 30 cm + 35% from
ment for statistical computing (R Core Team 2000). deeper layers as a high estimate. Expanding on ear-
Given the clustered nature of the point data, we have lier work, Kauffman et al (2017) found that on average
implemented a spatially balanced random forest model 54% of belowground carbon (soil + roots) to 3 m was
design. Model performance was assessed with a 5 lost after conversion to shrimp ponds and pastures.
fold (Leave-Location-Out) cross-validation procedure Given the limited number of studies comparing soil
where 20% of complete locations were withheld in each OCS change with land use change, in this work we
model refitting (Gasch et al 2015). The relative impor- adopt the same 25%–100% range as used by Pendleton
tance of using SoilGrids as a covariate was assessed by et al (2012) applied to the first meter of soil. Finally,
implementing the cross-validation procedure with and country level statistics for OCS loss were calculated as
without this variable. described above. All global and country level analy-
Finally, prediction error of OCS for 0–1 m depth ses were performed on the 30 m resolution dataset in
was derived for ±1 s.d. based on the quantile regression Google Earth Engine (Gorelick et al 2017).
approach of Meinshausen (2006) and implemented
in R via the ranger package. This procedure is rela-
tively computationally demanding so a random subset 3. Results
of approximately 15 000 points were selected to cal-
culate prediction errors. All modeling was run on 3.1. Model results
ISRIC High Performance Computing servers with The random forest model was successful in captur-
48 cores of 256 GB RAM. ing the major variation in OCD across the mangrove

4
Environ. Res. Lett. 13 (2018) 055002

Figure 1. Model fitting results: the corresponding 3D Random Forest model for soil organic carbon density (a) with cross-validation
results in (b), and relative variable importance plot (c). TSM = total suspended matter, SST = sea surface temperature (numbers
following TSM and SST refer to quarter of the year), NIR = Near Infrared, SW1 = Short wave mid infrared.

database (figure 1(a)) with an R2 of 0.84 and a root seen (figure 2). Mangrove forests dominated by sed-
mean square error (RMSE) of 6.9 kg m−3 compared to iment laden fluvial inputs typically have consistently
the mean OCD value of 29.6 kg m−3 . Cross-validation low OCS as seen in the Sundarbans and Madagascar
results yield an R2 of 0.63 and an RMSE of 10.9 kg m−3 (figures 2(a) and (e)). In non-deltaic mangroves, the
(figure 1(b)), which is the de-facto mapping accu- model appears to have captured the large zonal vari-
racy to be expected on the field. Low OCD values ation in OCS that is often observed in field studies
were slightly over-predicted and high OCD values (figures 2(b) and (c).
were under-predicted (figure 1(b)). The initial OCS
prediction from SoilGrids 250 m was the most impor- 3.3. Soil carbon loss due to habitat loss
tant variable explaining mangrove OCD. Running Utilizing the Hanson et al (2013) global deforesta-
the 5 fold cross-validation without and with Soil- tion analysis (figure S8), we found that 278049 ha
Grids indicated that this single variable explained (1.67% of total) of area identified as mangrove habitat
improved model performance by nearly 50% (R2 in the year 2000 was deforested resulting in the com-
increased from 0.42–0.63). Seasonal total suspended mitted emission of 30.4–122 Tg C (111–447 Tg CO2 )
matter (TSM), depth of sample, mangrove tree cover, from mangrove forest soils due to land use change
Landsat Red band, sea surface temperature (SST), and between 2000 and 2015 (figure 3). The relative rank of
tidal range were the next ten most important vari- nations in terms of loss of mangrove forest area and
ables, respectively (figure 1(c)). Quantile regression OCS were often not the same (table S1). Indonesia
analysis indicated that the full uncertainty (±1 s.d.) alone was responsible for 52% of this global loss with
about a mean prediction of carbon stocks to 1 m depth Malaysia and Myanmar representing another 25% of
averaged 40.4% of the mean OCS with lowest rel- the global total loss (figure 3(c)). When visualized as
ative uncertainty in the most carbon-rich mangrove a percent loss from year 2000 stocks, a slightly differ-
forests (figure S7). ent pattern emerged (figure 3(d)). Guatemala had the
highest percent loss of mangrove OCS (0.9%–6.8%)
followed by several southeast Asian nations, but high
3.2. Mangrove soil carbon storage
percent losses were also found in several Caribbean
Projection of the mangrove OCD model to global
island nations as well as the United States and
mangrove forests revealed the distribution of soil car-
several west African countries.
bon storage in these ecosystems (figure 2). The mean
(±1 s.d.) OCS to 1 m depth was 361 ± 136 Mg C ha−1
with a range of 86–729 Mg C ha−1 . At the national 4. Discussion
level (table S1), Bangladesh had the lowest per ha
stocks, averaging just 127 Mg C ha−1 followed by China 4.1. Amount and distribution of Mangrove SOC
and the nations bordering the Persian Gulf and Red Our new estimate of global mangrove OCS of
Sea with an average OCS of 214 and 233 Mg C ha−1 , 6.4 Pg C in the upper meter and 12.6 Pg C to 2 m is
respectively. The highest per ha stocks were found largely consistent with past efforts to calculate this
in many of the pacific island nations, averaging value (Donato et al 2011, Jardine and Siikamäki
505 Mg C ha−1 with much of Southeast Asia ranking 2014, Sanders et al 2016). However, our estimate is
well above the global mean. double that of Atwood et al (2017) primarily due to
While the national level comparisons are reveal- their use of the Hamilton and Casey (2016) estimate of
ing, by modeling at a 30 m resolution much richer mangrove extent instead of Giri et al (2011). Impor-
details of potential within forest variation in OCS are tantly, by using an environmental covariate model,

5
Environ. Res. Lett. 13 (2018) 055002

Figure 2. Global distribution of mangrove soil carbon stocks for the top meter of soil (hex bin area ∼19 000 km2 ) and detailed maps
(30 m resolution) for selected mangrove regions of the world: (1) Sundarbans along the India/Bangladesh border, (2) Bahı́a de los
Muertos, Pacific coast of Panama, (3) southwest coast of Papua, Indonesia, (4) Hinchinbrook Island, Queensland, Australia, (5)
Ambaro Bay, Madagascar, and (6) Guinea-Bissau and Guinea along the West African coast. In top panel, data presented as mean stock
(Mg C ha−1 ) for mangrove forest area only within each hex bin.

we have been able to make plausible estimates for Jardine and Siikamäki (2014). This difference in
regions where no sampling has taken place instead of range was likely due to the inclusion of more data
relying on global mean values (i.e. Atwood et al 2017). from sub-tropical and temperate mangroves (figure
The total amount of soil carbon was similar in S3).
our analysis and the most comparable analysis, that The depth trend analysis (figure S6) and ran-
of Jardine and Siikamäki (2014), but the spatial distri- dom forest variable importance (figure 1(b)) both
bution of carbon-rich versus carbon-poor mangroves indicated that depth should be considered in calcu-
varied substantially. For example, we found much lation of OCS. For locations that were either stable
higher OCS levels in West Africa than in East African peat domes or located in estuaries receiving large
nations (figure 2 and table S1) but the reverse was annual sediment loads, a stable OCD profile distri-
found by Jardine and Siikamäki (2014). Large dis- bution would be expected and this was found for many
crepancies were also found for Colombia, Sri Lanka sites (figures S6(a) and (e)). However, where man-
and many of the countries bordering the Red Sea. groves are growing in a mineral matrix that is receiving
These differences were most likely driven by lack of only low sediment loads, a decline in OCD may be
data in those regions at the time of the analysis by expected as the carbon inputs from the productive
Jardine and Siikamäki (2014) given that nearly half mangrove forest would be concentrated in the sur-
the data in our database was collected after their face horizons (figure S6(b)). Still in other cases (figure
study was published. Additionally, our analysis sug- S6(c)), changes in hydrologic/sediment regimes can
gested a much larger range in OCS (86–729 Mg C ha−1 ) lead to irregular depth patterns or even an increase in
compared to 272–703 Mg C ha−1 in the analysis of OCD with depth.

6
Environ. Res. Lett. 13 (2018) 055002

Figure 3. Top 20 nation rankings for (a) total mangrove area lost between the years 2000 and 2012, (b) area loss as a percent of year
2000 mangrove area, (c) total soil organic carbon stocks, (d) carbon loss as a percent loss of year 2000 soil carbon stock. Range in
values for (c) and (d) come from 25%–100% loss of carbon in upper meter of soil in pixels identified as being deforested between the
years 2000 and 2015.

While total area of mangroves was a key deter- analysis has demonstrated mangrove soil carbon is
minant of total soil carbon storage, amongst the top highly variable and many mangroves actually store
25 mangrove OCS holding nations, there was a nearly fairly modest amounts of carbon in the upper one
even split between nations with smaller area of high or two meters of soil. While not the focus of this
soil carbon density forests and those nations with lots analysis, it is important to point out that while some
of low soil carbon density forests (figure 4). Indone- mangrove forests store modest levels of OCS in the
sia was the clear exception to this trend with the upper meter of soil, they can have high sequestration
largest mangrove holdings which also contain rich car- rates and conversely carbon-dense mangroves can have
bon stocks resulting in Indonesia alone holding nearly low annual sequestration rates (Lovelock et al 2010,
25% of the world’s mangrove OCS. MacKenzie et al 2016).
Compared to terrestrial carbon pools, mangrove
forests rank low due to their limited spatial extent. For 4.2. Drivers of soil carbon storage
example in the upper meter of soil, permafrost affected Our spatial modelling framework, in which global pre-
soils are estimated to store 472 ± 27 Pg C (Hugelius dictions were combined with local high resolution
et al 2014), tropical forests contain ∼188 Pg C, and images, was successful as the general patterns of car-
soils under permanent cropping contain ∼150 Pg C bon variation from SoilGrids 250 m were maintained,
(table 1). However, on an equal area basis, man- while the spatial detail was significantly improved by
grove forests on average store more soil carbon than moving from 250 m–30 m spatial resolution. The ini-
most other ecosystems (table 1). Importantly, our tial SoilGrids 250 m OCS prediction (Hengl et al 2017)

7
Environ. Res. Lett. 13 (2018) 055002

Table 1. Soil organic carbon stocks (mean with 5th-95th percentile in parentheses) and total storage for different terrestrial ecosystems
compared to mangrove forests.

1 m soil organic carbon stock


Land cover category (IGBP code)a Area (106 ha) Pg C Mg C ha−1
Mangrove forestb 16.6 6.4 361 (94–628)
Gelisols (permafrost soils)c 1878 472.0 389 (178–691)
Evergreen Needleleaf forest (1)d 286 60.0 210 (121–346)
Evergreen Broadleaf forest (2) 1248 188.4 151 (85–271)
Deciduous Needleleaf forest (3) 116 29.3 253 (163–412)
Deciduous Broadleaf forest (4) 165 22.1 134 (83–223)
Mixed forest (5) 771 152.8 198 (93–343)
Closed shrublands (6) 56 6.2 110 (39–223)
Open shrublands (7) 1933 325.8 169 (49–329)
Woody savannas (8) 1179 185.8 158 (82–274)
Savannas (9) 1010 112.9 112 (52–201)
Grasslands (10) 1810 280.1 155 (56–289)
Permanent wetlands (11)e 104 25.1 241 (114–474)
Croplands (12) 1177 149.6 127 (60–200)
Cropland/Natural veg. mosaic (14) 868 117.7 136 (58–238)
a data for MODIS-based IGBP land cover classes (Friedl et al 2010) extracted from 1 m OCS map for the year 2010 produced by Sanderman
et al (2017).
b mangrove area and OCS data from this study.
c permafrost area from Tarnocai et al (Tarnocai et al 2009), OCS data from Hugelius et al (2014).
d some overlap between class 1 (evergreen Needleleaf forest) and gelisols.
e class 11 (permanent wetlands) likely has overlap with mangrove area.

stocks. The mangrove typology ended up being unin-


formative likely because most of the data that went into
this classification was already captured in the model.
Covariates related to mangrove biomass (SRTM
elevation and Landsat bands) were also important in
explaining the local distribution of OCD (figure 1(b)).
However, as pointed out by Bukoski et al (2017), it is
unclear whether the importance of these vegetation-
related data are causal drivers of differences in OCD
or they just happen to co-vary in the same way as
OCD. To further explore the relationship between
forest biomass carbon and soil carbon storage, we
extracted aboveground biomass data from Hutchison
et al (2014) and compared it to our OCS results (fig-
ure 5). While a clear positive trend was found between
Figure 4. Rank of nations by mangrove area plotted against biomass and soil carbon storage (R2 = 0.26), there is
rank by soil carbon density for all nations containing
>30 Tg C. Bubble size is proportional to total carbon stock clearly a lot of variance especially at lower biomass levels
(Tg C) within each nation. where nearly the full range in OCS can be found.
Depth below the soil surface was an important
covariate in modeling OCD distribution (figure 1(b)).
This finding was supported by the database depth
was based upon machine learning algorithms using
trend analysis (figure S6) which indicated that a flat
237 covariates that covered the main state factors of
depth distribution of OCD would be an incorrect
soil formation (Jenny 1994)—climate, relief, living
assumption 37%–64% of the time. These findings sug-
organisms (vegetation) and parent material—and was
gest that the simple scaling performed in figure S5
four times more important in predicting mangrove
and in several previous assessments (Bukoski et al
OCD than any of the local covariates. However,
2017, Jardine and Siikamäki 2014, Atwood et al 2017)
the local covariates allowed for a much more refined
may not be an accurate estimation of total OCS
picture of the spatial variation in OCS within man-
especially when extrapolating from a surface horizon
grove forests (figure 2) which was not captured in the
sample alone.
250 m resolution SoilGrids 250 m prediction. Covari-
ates related to hydrogeomorphology (TSM and tidal
range), as hypothesized, were important predictors of 4.3 Soil carbon loss due to land conversion (2000–
local variation in OCD. Both TSM and tidal range 2015)
were strongly negatively correlated with OCD suggest- Our analysis suggests that mangrove soils have lost
ing that locations with either high sediment loads or or are at least committed to losing 30.4–122 Tg C due
strong tidal flushing do not accumulate large carbon to the land use conversion that occurred between the

8
Environ. Res. Lett. 13 (2018) 055002

Figure 5. Comparison of aboveground biomass carbon (Hutchison et al 2014) with organic carbon stocks (OCS) for the top meter
of soil. Points were generated by casting 10 000 random points into mangrove areas and extracting values from both maps. Linear
regression R2 = 0.26.

years 2000 and 2015 (figure 3). Given that at the global disturbance (Cahoon et al 2003) likely leads to more
level the rate of mangrove forest lost was consistent moderate emissions as decomposition and erosion
over this time period (Hamilton and Casey 2016), exceed new plant carbon inputs.
we estimated an annual soil carbon emission of 2.0– It is important to note that nearly all available
8.1 Tg C yr−1 . This value is significantly lower than data on OCS loss due to conversion to other land
previous estimates (Donato et al 2011, Pendleton uses come from organogenic mangrove forests. In
et al 2012) for two reasons. First, we use remote a mineral-dominated mangrove systems with only a
sensing-based measurements of actual mangrove loss few percent sediment OCC, we would not expect the
instead of applying a large range of annual conver- same level of carbon loss as when peat deposits are
sion rates, which are notoriously variable according to drained or removed. In fact, reclamation of deltaic sed-
their source (Friess and Webb 2011). Second, we have iments for paddy rice cultivation can lead to increases
summed the actual OCS values for each of the pix- in OCS (Kalbitz et al 2013), although methane emis-
els where land conversion has taken place (i.e. figure sions would be expected to increase. Additionally, if
S8) instead of applying a conversion rate to a mean mangrove habitat is lost due to deforestation with-
OCS value. out a change in hydrologic regime, mineral-dominated
The three nations of Indonesia, Malaysia and mangroves can continue to accrete carbon, but at
Myanmar contributed 77% of global mangrove OCS a lower rate than in a system that has additional
loss for this time period (figure 3). Despite similar area organic matter inputs from the mangroves themselves
loss (figure 3(a)), Malaysia lost approximately twice as (Pérez et al 2017).
much soil carbon as Myanmar due to the large dif-
ferences in carbon density between these two nations
4.4. Limitations and uncertainties
(mean OCS = 485 ± 57 versus 245 ± 63 Mg C ha−1 ,
While we endeavored to ensure that the model input
respectively). This comparison highlights the impor-
data was of the highest quality possible, there undoubt-
tance of using local OCS values for estimating carbon
edly remain unknown errors in the database which
emissions attributed to mangrove conversion.
are contributing to model error. Machine learning
Not all land use conversions result in equal loss
models are particularly sensitive to outlier values
of OCS. Conversion of mangrove forest to shrimp
and extrapolation (Murphy 2012). Various research
ponds results in a rapid and near complete loss of car-
groups use different methods for determining the
bon in the upper meter of soil (Kauffman et al 2014),
organic carbon concentration (OCC) of a sample
as well as losses deeper in the soil profile (Kauffman
with not all publications reporting whether or not
et al 2017). Conversion to other agricultural uses such
results were corrected for occurrence of inorganic car-
as pasture for beef production (Kauffman et al 2016b)
bon or whether or not roots were excluded before
and cereal crops (Andreetta et al 2016) also appear
further processing. Bulk density (BD) is a difficult
to result in large soil carbon emissions. However,
parameter to measure accurately in many soils, and
mangrove degradation and loss due to over harvest-
based on our analysis of BD versus OCC (figure S1)
ing for fuelwood (Jones et al 2015) or due to natural
some reported data are unlikely to be accurate. We

9
Environ. Res. Lett. 13 (2018) 055002

developed a procedure to correct potential BD errors, Acknowledgments


but a pedotransfer function gives only an approxi-
mation of the true value. Given the importance of We would like to thank A Andreetta, M Osland, J
depth in our models (figure 1(b)), it was unfortu- M Smoak, A DelVecchia, M E Gonneea, R K Bho-
nate that so many investigations only report OCS for mia and J Kelleway for providing additional data
large depth increments. We suggest that future studies from their publications. S-T Kang for translating
using the common practice of collecting subsamples and extracting data from Chinese language papers.
within larger horizon increments (e.g. Kauffman and R K Bhomia acknowledges CIFOR SWAMP project,
Donato 2012, Kauffman et al 2016a) report the spe- USFS International Program and USAID for fund-
cific depth increment of the sample rather than that of ing. M Rahman acknowledges support from USAID
the entire core. This additional level of transparency in for funding. PM and CJS acknowledge support from
the data would allow mass-preserving splines (Bishop the Australian Research Council (DE130101084 and
et al 1999) to be fit through the distinct measurement LP160100242) and (DE160100443 and DP150103286),
intervals, resulting in unbiased estimates of OCS. respectively. CD acknowledges support from the Ruf-
The largest uncertainty in the input data likely ford Foundation and Darwin Initiative for funding.
resulted from imperfect information on plot location. JS, TH, GF and KS were supported by an anonymous
Whether, accidental or purposeful (i.e. not wanting gift to The Nature Conservancy. MFA was sup-
to identify exact locations), spatially misplaced data in ported by funding from the Queensland Government
publications are of limited utility in geospatial appli- through the Advance Queensland Fellowship. ISRIC
cations. All of the covariate data for each mangrove is a non-profit organization primarily funded by the
point were selected from spatial layers resulting in Dutch government.
the potential for a mismatch between the recorded
OCS and the spatial predictors. In this study, we visu- Data availability
ally inspected all coordinates against Google Earth
imagery and our adjusted mangrove domain, and then The mangrove soil carbon database and model
contacted many authors to seek further information outputs can be downloaded from Harvard dataverse at
and manually adjusted coordinates when we were https://dataverse.harvard.edu/dataset.xhtml?persistent
confident that the adjustments lead to better spatial Id=doi:10.7910/DVN/OCYUIT.
location. In the final analysis, 199 soil profiles had to
be excluded from analysis because we could not confi-
dently locate these plots within the adjusted mangrove ORCID iDs
spatial domain.
Jonathan Sanderman https://orcid.org/0000-0002-
3215-1706
5. Conclusions Peter I Macreadie https://orcid.org/0000-0001-7362-
0882
This work has produced three resources which we
hope to be of significant value to the blue carbon References
research and management communities: 1) a large har-
monized database of soil carbon data from mangrove Adame M F, Santini N S, Tovilla C, Vázquez-Lule A, Castro L and
Guevara M 2015 Carbon stocks and soil sequestration rates of
ecosystems; 2) high-resolution (30 m) predictions tropical riverine wetlands Biogeosciences 12 3805–18
with error of soil carbon stocks across all mangrove Alongi D M 2002 Present state and future of the world’s mangrove
forests globally; 3) estimates of potential soil carbon forests Environ. Conserv. 29 331–49
losses due to mangrove habitat loss between 2000 Andreetta A, Huertas A D, Lotti M and Cerise S 2016 Land use
changes affecting soil organic carbon storage along a
and 2015. By using a statistical data-driven model, mangrove swamp rice chronosequence in the Cacheu and Oio
we have been able to produce credible estimates regions (northern Guinea-Bissau) Agric. Ecosyst. Environ. 216
OCS for the numerous mangrove regions where no 314–21
field data exist. We found that mangrove OCS is Atwood T B et al 2017 Global patterns in mangrove soil carbon
stocks and losses Nat. Clim. Change 7 523–8
highly variable (86–729 Mg C ha−1 in the top meter)
Balke T and Friess D A 2016 Geomorphic knowledge for mangrove
but that much of the variability could be captured restoration: a pan-tropical categorization Earth Surf. Process.
using spatially-comprehensive predictors in a machine- Landforms 41 231–9
learning framework. Of the 6400 Tg C in the upper Banerjee K, Chowdhury M R, Sengupta K, Sett S and Mitra A 2012
Influence of anthropogenic and natural factors on the
meter of soil, 30–122 Tg have likely been lost due to mangrove soil of Indian Sundarbans wetland Arch. Environ.
deforestation since the year 2000 with 77% of this Sci. 6 80–91
loss attributed to Indonesia, Malaysia and Myanmar. Barbier E B, Hacker S D, Kennedy C, Koch E W, Stier A C and
These spatially-explicit estimates of mangrove soil car- Silliman B R 2011 The value of estuarine and coastal
ecosystem services Ecol. Monogr. 81 169–93
bon storage and loss will provide a practical first step Bishop T F A, McBratney A B and Laslett G M 1999 Modelling soil
for enabling nations to prioritize mangrove protection attribute depth functions with equal-area quadratic smoothing
as part of their climate mitigation and adaptation plans. splines Geoderma 91 27–45

10
Environ. Res. Lett. 13 (2018) 055002

Boucher O, Bellassen V, Benveniste H, Ciais P, Criqui P, Guivarch Grassi G, House J, Dentener F, Federici S, den Elzen M and Penman
C, Le Treut H, Mathy S and Séférian R 2016 Opinion: in the J 2017 The key role of forests in meeting climate targets
wake of Paris Agreement, scientists must embrace new requires science for credible mitigation Nat. Clim. Change 7
directions for climate change research Proc. Natl Acad. Sci. 220–6
USA 113 7287–90 Hamilton S E and Casey D 2016 Creation of a high spatio-temporal
Brander L, Wagtendonk A, Hussain S, McVittie A, Verburg P H, de resolution global database of continuous mangrove forest
Groot R S and van der Poeg S 2012 Ecosystem service values cover for the 21st century (CGMFC-21) Glob. Ecol. Biogeogr.
for mangroves in Southeast Asia: a meta-analysis and value 25 729–38
transfer application Ecosyst. Serv. 1 62–9 Hansen M C et al 2013 High-resolution global maps of 21st century
Breithaupt J L, Smoak J M, Smith T J, Sanders C J and Hoare A forest cover change Science 342 850–3
2012 Organic carbon burial rates in mangrove sediments: Hengl T et al 2017 SoilGrids 250 m: Global gridded soil information
strengthening the global budget Glob. Biogeochem. Cycles 26 based on machine learning PLoS ONE 12 e0169748
GB3011 Herr D and Landis E 2016 Coastal blue carbon ecosystems.
Bukoski J J, Broadhead J S, Donato D C, Murdiyarso D and Opportunities for Nationally Determined Contributions. Policy
Gregoire T G 2017 The use of mixed effects models for Brief (Gland: IUCN and Washington, DC: TNC)
obtaining low cost ecosystem carbon stock estimates in Howard J, Sutton-Grier A, Herr D, Kleypas J, Landis E, Mcleod E,
mangroves of the Asia Pacific PLoS ONE 12 e0169096 Pidgeon E and Simpson S 2017 Clarifying the role of coastal
Cahoon D R, Hensel P, Rybczyk J, McKee K L, Proffitt C E and and marine systems in climate mitigation Front. Ecol. Environ.
Perez B C 2003 Mass tree mortality leads to mangrove peat 15 42–50
collapse at Bay Islands, Honduras after Hurricane Mitch J. Hugelius G et al 2014 Estimated stocks of circumpolar permafrost
Ecol. 91 1093–105 carbon with quantified uncertainty ranges and identified data
Casey K S, Brandon T B, Cornillon P and Evans R 2010 The past, gaps Biogeosciences 11 6573–93
present, and future of the AVHRR pathfinder SST program Hutchison J, Manica A, Swetnam R, Balmford A and Spalding M
Oceanography from Space (Dordrecht: Springer Netherlands) 2014 Predicting global patterns in mangrove forest biomass
pp 273–87 Conserv. Lett. 7 233–40
Chmura G L, Anisfeld S C, Cahoon D R and Lynch J C 2003 Global Jardine S L and Siikamäki J V 2014 A global predictive model of
carbon sequestration in tidal, saline wetland soils Glob. carbon in mangrove soils Environ. Res. Lett. 9 104013
Biogeochem. Cycles 17 1111 Jenny H 1994 Factors of soil formation: a system of quantitative
Donato D C, Kauffman J B, Murdiyarso D, Kurnianto S, Stidham M pedology (Courier Corporation)
and Kanninen M 2011 Mangroves among the most Jones T et al 2015 The dynamics, ecological variability and
carbon-rich forests in the tropics Nat. Geosci. 4 293–7 estimated carbon stocks of mangroves in Mahajamba Bay,
Duke N C, Kovacs J M, Griffiths A D, Preece L, Hill D J E, Van Madagascar J. Mar. Sci. Eng. 3 793–820
Oosterzee P, Mackenzie J, Morning H S and Burrows D 2017 Kalbitz K et al 2013 The carbon count of 2000 years of rice
Large-scale dieback of mangroves in Australia’s Gulf of cultivation Glob. Change Biol. 19 1107–13
Carpentaria: a severe ecosystem response, coincidental with an Kauffman J B, Arifanti V B, Basuki I, Kurnianto S, Novita N,
unusually extreme weather event Mar. Freshw. Res. 68 Murdiyarso D, Donato D C and Warren M W 2016a
1816–29 Protocols for the measurement, monitoring, and reporting of
Ewers Lewis C J, Sanderman P E, Baldock J and Macreadie P I 2018 structure, biomass, carbon stocks and greenhouse gas
Variability and vulnerability of coastal blue carbon stocks: a emissions in tropical peat swamp forests Working paper 221
case study from Southeast Australia Ecosystems 21 263–79 (Bogor: CIFOR)
Ezcurra P, Ezcurra E, Garcillán P P, Costa M T and Kauffman J B, Hernandez Trejo H, del Carmen Jesus Garcia M,
Aburto-Oropeza O 2016 Coastal landforms and Heider C and Contreras W M 2016b Carbon stocks of
accumulation of mangrove peat increase carbon sequestration mangroves and losses arising from their conversion to cattle
and storage Proc. Natl Acad. Sci. 113 4404–9 pastures in the Pantanos de Centla, Mexico Wetl. Ecol.
Friedl M A, Sulla-Menashe D, Tan B, Schneider A, Ramankutty N, Manage. 24 203–16
Sibley A and Huang X 2010 MODIS collection 5 global land Kauffman J B, Arifanti V, Trejo H, del Carmen Jesus ́ Garcı́a M,
cover: algorithm refinements and characterization of new Norfolk J, Hadriyanto D, Cifuentes-Jara M, Murdiyarso D and
datasets Remote Sens. Environ. 114 168–82 Cifuentes-Jara M 2017 The jumbo carbon footprint of a
Friess D A, Thompson B S, Brown B, Amir A A, Cameron C, shrimp: carbon losses from mangrove deforestation Front.
Koldewey H J, Sasmito S D and Sidik F 2016 Policy challenges Ecol. Environ. 15 183–8
and approaches for the conservation of mangrove forests in Kauffman J B and Donato D 2012 Protocols for the measurement,
Southeast Asia Conserv. Biol. 30 933–49 monitoring and reporting of structure, biomass and carbon
Friess D A and Webb E L 2011 Bad data equals bad policy: how to stocks in mangrove forests Working paper 86 (Bogor: CIFOR)
trust estimates of ecosystem loss when there is so much Kauffman J B, Heider C, Cole T G, Dwire K A and Donato D C
uncertainty? Environ. Conserv. 38 1–5 2011 Ecosystem carbon stocks of micronesian mangrove
Gasch C K, Hengl T, Gräler B, Meyer H, Magney T S and Brown D J forests Wetlands 31 343–52
2015 Spatio-temporal interpolation of soil water, temperature, Kauffman J B, Heider C, Norfolk J and Payton F 2014 Carbon
and electrical conductivity in 3D+ T: the cook agronomy farm stocks of intact mangroves and carbon emissions arising from
data set Spat. Stat. 14 70–90 their conversion in the Dominican Republic Ecol. Appl. 24
Gedan K B, Kirwan M L, Wolanski E, Barbier E B and Silliman B R 518–27
2011 The present and future role of coastal wetland vegetation Krauss K W, McKee K L, Lovelock C E, Cahoon D R, Saintilan N,
in protecting shorelines: answering recent challenges to the Reef R and Chen L 2014 How mangrove forests adjust to
paradigm Clim. Change 106 7–29 rising sea level New Phytol. 202 19–34
Giri C, Ochieng E, Tieszen L L, Zhu Z, Singh A, Loveland T, Masek Lovelock C E, Sorrell B K, Hancock N, Hua Q and Swales A 2010
J and Duke N 2011 Status and distribution of mangrove Mangrove forest and soil development on a rapidly accreting
forests of the world using earth observation satellite data Glob. shore in New Zealand Ecosystems 13 437–51
Ecol. Biogeogr. 20 154–9 MacKenzie R A, Foulk P B, Klump J V, Weckerly K, Purbospito J,
Gorelick N, Hancher M, Dixon M, Ilyushchenko S, Thau D and Murdiyarso D, Donato D C and Nam V N 2016
Moore R 2017 Google Earth engine: planetary-scale Sedimentation and belowground carbon accumulation rates
geospatial analysis for everyone Remote Sens. Environ. 202 in mangrove forests that differ in diversity and land use: a tale
18–27 of two mangroves Wetl. Ecol. Manage. 24 245–61

11
Environ. Res. Lett. 13 (2018) 055002

McBratney A, Mendonça Santos M and Minasny B 2003 On digital Sanders C J, Maher D T, Tait D R, Williams D, Holloway C, Sippo J
soil mapping Geoderma 117 3–52 Z and Santos I R 2016 Are global mangrove carbon stocks
Meinshausen N 2006 Quantile regression forests J. Mach. Learn. driven by rainfall? J. Geophys. Res. Biogeosci. 121 2600–9
Res. 7 983–99 Siikamäki J, Sanchirico J N and Jardine S L 2012 Global economic
Mumby P J et al 2004 Mangroves enhance the biomass of coral reef potential for reducing carbon dioxide emissions from
fish communities in the Caribbean Nature 427 533–6 mangrove loss Proc. Natl Acad. Sci. 109 14369–74
Murdiyarso D, Purbopuspito J, Kauffman J B, Warren M W, Simard M, Zhang K, Rivera-Monroy V H, Ross M S, Ruiz P L,
Sasmito S D, Donato D C, Manuri S, Krisnawati H, Taberima Castañeda-Moya E, Twilley R R and Rodriguez E 2006
S and Kurnianto S 2015 The potential of Indonesian Mapping height and biomass of mangrove forests in
mangrove forests for global climate change mitigation Nat. Everglades National Park with SRTM elevation data
Clim. Change 5 8–11 Photogramm Eng. Remote Sens. 72 299–311
Murphy K P 2012 Machine Learning: A Probabilistic Perspective Snedaker S C 1982 Mangrove Species Zonation: Why?
(Cambridge, MA: MIT) (Netherlands: Springer) pp 111–25
Nagelkerken I et al 2008 The habitat function of mangroves for Spalding M, Kainuma M and Collins L 2010 World Atlas of
terrestrial and marine fauna: a review Aquat. Bot. 89 155–85 Mangroves (London: Earthscan)
Ouyang X, Lee S Y and Connolly R M 2017 Structural equation Spalding M, McIvor A, Tonneijck F, Tol S and van Eijk P 2014
modelling reveals factors regulating surface sediment organic Mangroves for Coastal Defence. Guidelines for Coastal
carbon content and CO2 efflux in a subtropical mangrove Sci. Managers and Policy Makers (Wetlands International and The
Total Environ. 578 513–22 Nature Conservancy)
Pendleton L et al 2012 Estimating global ‘blue carbon’ emissions Stringer C E, Trettin C C and Zarnoch S J 2016 Soil properties of
from conversion and degradation of vegetated coastal mangroves in contrasting geomorphic settings within the
ecosystems PLoS ONE 7 e43542 Zambezi River Delta, Mozambique Wetl. Ecol. Manage. 24
Pérez A, Machado W, Gutierrez D, Stokes D, Sanders L, Smoak J M, 139–52
Santos I and Sanders C J 2017 Changes in organic carbon Tarnocai C, Canadell J G, Schuur E A G, Kuhry P, Mazhitova G and
accumulation driven by mangrove expansion and Zimov S 2009 Soil organic carbon pools in the northern
deforestation in a New Zealand estuary Estuar. Coast. Shelf Sci. circumpolar permafrost region Glob. Biogeochem. Cycles 23
192 108–16 GB2023
Richards D R and Friess D A 2016 Rates and drivers of mangrove R Core Team 2000 R Language Definition (Vienna: R Found. Stat.
deforestation in Southeast Asia, 2000–2012 Proc. Natl Acad. Comput.)
Sci. 113 344–9 Temmerman S, Meire P, Bouma T J, Herman P M J, Ysebaert T and
Robertson A I and Phillips M J 1995 Mangroves as filters of shrimp De Vriend H J 2013 Ecosystem-based coastal defence in the
pond effluent: predictions and biogeochemical research needs face of global change Nature 504 79–83
Hydrobiologia 295 311–21 Valiela I, Bowen J L and York J K 2001 Mangrove forests: one of the
Rockström J, Gaffney O, Rogelj J, Meinshausen M, Nakicenovic N world’s threatened major tropical environments BioScience 51
and Schellnhuber H J 2017 A roadmap for rapid 807–15
decarbonization Science 355 1269–71 Wright M N and Ziegler A 2015 ranger: a fast implementation of
Sanderman J, Hengl T and Fiske G J 2017 Soil carbon debt of 12 000 random forests for high dimensional data in C++ and R
years of human land use Proc. Natl Acad. Sci. 114 9575–80 (arXiv:1508.04409)

12

You might also like