Improvement of Mangrove Soil Carbon Stocks Estimation in North Vietnam Using Sentinel-2 Data and Machine Learning Approach
Total Page:16
File Type:pdf, Size:1020Kb
See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/346107063 Improvement of Mangrove Soil Carbon Stocks Estimation in North Vietnam Using Sentinel-2 Data and Machine Learning Approach Article in GIScience & Remote Sensing · November 2020 DOI: 10.1080/15481603.2020.1857623 CITATIONS READS 0 84 8 authors, including: Tien Dat Pham Naoto Yokoya Florida International University The University of Tokyo 44 PUBLICATIONS 614 CITATIONS 145 PUBLICATIONS 2,915 CITATIONS SEE PROFILE SEE PROFILE Trang Thu Nga Le VNU University of Science Institute of Mechanics 2 PUBLICATIONS 1 CITATION 8 PUBLICATIONS 83 CITATIONS SEE PROFILE SEE PROFILE Some of the authors of this publication are also working on these related projects: Mangrove Mapping View project Assessing impact of Urban growth on Urban Air Quality in Indian Cities View project All content following this page was uploaded by Pham Tien Duc on 24 December 2020. The user has requested enhancement of the downloaded file. GISCIENCE & REMOTE SENSING https://doi.org/10.1080/15481603.2020.1857623 Improvement of Mangrove Soil Carbon Stocks Estimation in North Vietnam Using Sentinel-2 Data and Machine Learning Approach Tien Dat Pham a, Naoto Yokoya a, Thi Thu Trang Nguyen b, Nga Nhu Le c, Nam Thang Ha d, Junshi Xia a, Wataru Takeuchi e and Tien Duc Pham b aGeoinformatics Unit, The RIKEN Center for Advanced Intelligence Project (AIP), Chuo-ku, Tokyo, Japan; bFaculty of Chemistry, VNU University of Science, Vietnam National University, Hanoi, Vietnam; cDepartment of Marine Mechanics and Environment, Institute of Mechanics, Vietnam Academy of Science and Technology (VAST), Ba Dinh, Hanoi, Vietnam; dFaculty of Fisheries, University of Agriculture and Forestry, Hue University, Hue, Vietnam; eInstitute of Industrial Science, The University of Tokyo, Meguro-ku, Tokyo, Japan ABSTRACT ARTICLE HISTORY Quantifying total carbon (TC) stocks in soil across various mangrove ecosystems is key to under Received 9 July 2020 standing the global carbon cycle to reduce greenhouse gas emissions. Estimating mangrove TC at a Accepted 23 November 2020 large scale remains challenging due to the difficulty and high cost of soil carbon measurements when KEYWORDS the number of samples is high. In the present study, we investigated the capability of Sentinel-2 Soil carbon stocks; CatBoost; multispectral data together with a state-of-the-art machine learning (ML) technique, which is a sentinel-2 MSI; machine combination of CatBoost regression (CBR) and a genetic algorithm (GA) for feature selection and learning; mangrove optimization (the CBR-GA model) to estimate the mangrove soil C stocks across the mangrove ecosystem; Vietnam ecosystems in North Vietnam. We used the field survey data collected from 177 soil cores. We compared the performance of the proposed model with those of the four ML algorithms, i.e., the extreme gradient boosting regression (XGBR), the light gradient boosting machine regression (LGBMR), the support vector regression (SVR), and the random forest regression (RFR) models. Our proposed model estimated the TC level in the soil as 35.06–166.83 Mg ha−1 (average = 92.27 Mg ha−1) with satisfactory accuracy (R2 = 0.665, RMSE = 18.41 Mg ha−1) and yielded the best prediction performance among all the ML techniques. We conclude that the Sentinel-2 data combined with the CBR-GA model can improve estimates of the mangrove TC at 10 m spatial resolution in tropical areas. The effectiveness of the proposed approach should be further evaluated for different mangrove soils of the other mangrove ecosystems in tropical and semi-tropical regions. 1. Introduction overexpansion (Duke et al. 2007; Giri and Muhlhausen 2008; Sasmito et al. 2019; Friess et al. 2019). In Southeast Mangrove forests in the intertidal zones of the tropical Asia, over 100,000 ha of mangrove forests have been and semi-tropical areas are highly productive ecosys lost between 2000 and 2012 (Richards and Friess 2016; tems; they provide a wide range of functions and vital Hamilton and Friess 2018). The mangrove forest area in services to coastal populations such as reducing the Vietnam in the early 20th century has decreased drasti effects of tsunamis (Danielsen et al. 2005) and mitigat cally by 400,000 ha (Le Xuan et al. 2003), and in North ing the damage of tropical cyclones (Mazda et al. 1997). Vietnam, from 1964 to 1997, it decreased by 17,094 ha These forests can store a significant amount of carbon in because of the conversion to aquaculture (Pham and soil sediments (Donato et al. 2011) and thus are consid Yoshino 2016). ered a key component of “Blue Carbon,” which plays an Traditionally, the soil total carbon (TC) content of important role in mitigating the impacts of global warm mangroves is estimated through soil sample collection ing and climate change (Boone et al. 2013; Alongi 2012). and laboratory analysis. The straightforward measure Understanding mangrove soil carbon stocks plays ments can be accurate; however, they cannot be used important role in sustainably conserving mangroves for large-scale and rapid monitoring of TC during man and protecting these forests from deforestation and grove conversion processes or mangrove changes forest degradation as mangroves have been destroyed because they are costly and time-consuming, particu in the past five decades as a result of human activities, larly in mixed and dense mangrove forests. As a result, rapid urbanization, weak governance, and aquaculture the spatial distribution and reliable statistical data of CONTACT Tien Dat Pham [email protected]; Tien Duc Pham [email protected] © 2020 Informa UK Limited, trading as Taylor & Francis Group 2 T. D. PHAM ET AL. mangrove TC stocks have rarely been reported in the multispectral data in the quantitative analysis of current literature. Many studies have performed soil soil carbon remains challenging due to some limita carbon estimation using the optical proximal emersion tions associated with the use of satellite sensors to and remote sensing techniques (Gholizadeh and capture soil carbon such as atmospheric, radiometric Kopačková 2019). The optical proximal emersion tech corrections, soil moisture, and forest cover niques consider the applications of different sensors to (Angelopoulou et al. 2019), particularly in mangrove obtain signals from the soil using the sensor’s receiver in ecosystems. Thus, in this study, we developed a pre contact with the soil (Gholizadeh et al. 2018; Gholizadeh diction model and a novel framework based on and Kopačková 2019) and use spectral-based measure CatBoost regression (CBR) and genetic algorithms ments such as earth observation (EO) data. These (GA), namely CBR-GA, to quantify the total soil car approaches provide cost-effective methods for monitor bon (TC) content in mangrove ecosystems across the ing dynamic changes in mangrove ecosystems. The northern coast of Vietnam using Sentinel-2 (S-2) remote sensing techniques use electromagnetic radia data. tion to obtain information regarding an object such as We selected the CBR model because it is an soil type without any physical contact (Jin et al. 2017). In advanced gradient boosting decision trees (GBDT) other studies, many spectral bands were employed for algorithm recently proposed by Prokhorenkova et soil carbon estimation using visible-near infrared (VI- al. (2018) that can handle many features and achieve NIR) and short-wave infrared (SWIR) imaging systems promising results in numerous classification and mounted on space-borne, airborne sensors, and regression tasks in a variety of machine learning unmanned aerial systems (UAS) (Pinheiro et al. 2017; techniques (Dorogush, Ershov, and Gulin 2018). The Liu, Min, and Buchroithner 2017; Gholizadeh et al. CBR model has recently been applied as an effective 2018; Angelopoulou et al. 2019). Nonetheless, these method for nonlinear supervised learning problems methods produce relatively low performance and can in different domains with noisy labels and complex not be used for estimating TC on a large scale, resulting dependencies (Prokhorenkova et al. 2018) such as in the lack of spatial distribution of TC in mangrove athletes’ gender prediction (Walsh, Heazlewood, forests. Thus, accurate, cost-effective, rapid, and nondes and Climstein 2019), evapotranspiration (ET) estima tructive prediction models that use EO data to estimate tion (Huang et al. 2019), and mangrove aboveground the mangrove TC across different ecosystems in the biomass estimation (Pham et al. 2020b). Despite its tropics are needed for sustainable conservation pro strong predictive performance and robustness, the grams. Such models will support efforts to mitigate CBR model has never been used for retrieving man the impacts of climate change and develop strategies grove TC stocks. We hypothesize that the CBR algo for pilot Blue Carbon projects and the Reducing rithm may be useful for estimating mangrove TC due Emissions from Deforestation and Forest Degradation to the unique characteristics of mangrove soil prop (REDD+) programs (Pendleton et al. 2012; Ahmed and erties. Furthermore, the ability of existing GBDT algo Glaser 2016). rithms for estimating mangrove TC stocks has not Multispectral EO data have been widely employed been quantitatively evaluated in the current litera in monitoring soil carbon because of the advantages ture. More importantly, a quantitative comparison of over proximal and airborne hyperspectral remotely the GBDT algorithms and traditional