Earth observation for ecosystem accounting: spatially explicit national seagrass extent and carbon stock in Kenya, Tanzania, Mozambique and Madagascar

Abstract

Seagrass ecosystems are globally significant hot spots of blue carbon storage, coastal biodiversity and coastal protection, rendering them a so-called natural climate solution. Their potential as a natural climate solution has been largely overlooked in national and international climate strategies and financing. This stems mainly from the lack of standardized, spatially explicit mapping and region-specific carbon inventories. Here, we introduce a novel seagrass ecosystem accounting framework that harnesses machine learning, big satellite data analytics and open region-specific reference data within the Google Earth Engine cloud computing platform. Leveraging a biennial percentile composite, assembled from 16 453 Sentinel-2 surface reflectance image tiles at 10-m spatial resolution, and 20 820 reference data points, we applied the cloud-native framework to produce the first national inventories of seagrass extent and total seagrass carbon stocks in Kenya, Tanzania, Mozambique and Madagascar. We estimated 4316 km 2 of regional seagrass extent (mean F1-score of 59.3% and overall accuracy of 84.3%) up to 23 m of depth. Pairing country-specific in situ carbon data and our spatially explicit seagrass extents, we calculated total regional seagrass blue carbon stocks between 11.2-40.2 million MgC, with the largest national carbon pool in Kenya (8-29.2 million MgC). We envisage that improvements in the remote sensing components of the framework guided by a necessary influx of region-specific data on seagrass stocks and fluxes could reduce uncertainties in our current spatially explicit ecosystem extent and carbon accounts, enhancing the incorporation of seagrasses into Multilateral Environmental Agreements for future resilient ecosystems, societies and economies.

Bahamian seagrass extent and blue carbon accounting using Earth Observation

“…Utilizing region-specific seagrass carbon data, this spatially-explicit extent translates to a Tier 2 blue carbon assessment up to 771.4 million Mg C and an annual blue carbon sequestration rate of 123 Mt CO 2 (Table 1). The second most important finding is that our seagrass ecosystem accounting efforts enrich our previously published research in 27 different tropical and temperate countries across the Mediterranean, East Africa and the Western Indian Ocean (Lee et al, 2022;Traganos et al, 2022a;Traganos et al, 2022b). The common denominator of all 28 national seagrass ecosystem accounting endeavors is the utilization of a standardized technological Earth Observation framework which supports an effective qualitative and quantitative comparison of its yielded physical accounts by minimizing both technological and environmental differences.…”

“…This provides an alternative to traditionally expensive and time-consuming surveying methods (Hossain et al, 2015;Veettil et al, 2020). The incorporation of country-scale seagrass carbon data with high-quality mapping products supports the assessment of the seagrass ecosystems' importance and its implementation into national climate agendas, and benefits relevant seascape management and conservation efforts (Roelfsema et al, 2014;McKenzie et al, 2020;Traganos et al, 2022b).…”

“…The filtered archive was then statistically reduced to a single multi-temporal composite using the 20 th percentile of each pixel's spectral value. This approach further reduces high-reflectance obstacles like remaining clouds and sun glint effects, waves, turbidity, and haze (Thomas et al, 2021;Traganos et al, 2022b). To further minimize misclassification, optically deep water, land and land water pixels were masked.…”

“…All thresholding methods were based on suitable image subsets which allowed for a strong bi-modal histogram distribution and therefore better masking results. Input parameters like the Canny edge sensitivity threshold and the number of buckets for the Otsu-histograms were chosen based on an iterative approach (Traganos et al, 2022b). Land water bodies were masked using an edited version of the Global Administrative Unit Layers 2015 (Dataset FAO UN, 2014).…”

“…The lack of recognition and protection arises from insufficient knowledge of spatially explicit seagrass distribution and related carbon accounts, especially in remote and/or data-poor areas (McKenzie et al, 2020;Macreadie et al, 2021). Recent advances in Earth Observation utilize high temporal and spatial resolution satellite data archives, cloud computing, and machine learning to minimize mapping uncertainties (Lyons et al, 2020;Traganos et al, 2022a;Traganos et al, 2022b). This provides an alternative to traditionally expensive and time-consuming surveying methods (Hossain et al, 2015;Veettil et al, 2020).…”