Modelagem da biomassa de mangues utilizando sensores remotos e inteligência artificial: estudo de caso na Reserva de Desenvolvimento Sustentável Municipal Piraquê-Açu-Mirim no estado do Espírito Santo, Brasil
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidade Federal de Viçosa
Abstract
Os manguezais são ecossistemas costeiros de elevada relevância ecológica, destacando-se como sumidouros de carbono azul com capacidade de sequestro superior à maioria das florestas tropicais. No entanto, sua estimativa de biomassa acima do solo (AGB) por métodos tradicionais é onerosa e limitada espacialmente. Esta pesquisa desenvolveu e validou um modelo preditivo de AGB para o manguezal da Reserva de Desenvolvimento Sustentável Municipal Piraquê-Açu-Mirim (ES), utilizando exclusivamente dados gratuitos de sensores remotos (Sentinel-1, Sentinel- 2 e GEDI) e algoritmos de aprendizado de máquina (Random Forest) na plataforma Google Earth Engine. A metodologia integrou índices espectrais de vegetação, bandas de reflectância e retroespalhamento SAR, tendo como variável dependente as amostras de densidade de biomassa do GEDI. O modelo foi aplicado em dois cenários (pré e pós-evento climático extremo ocorrido em junho de 2016), permitindo quantificar a perda de biomassa. Os resultados indicaram redução de aproximadamente 7.581 toneladas de biomassa total em seis meses, com redução da densidade média de 31,9 t/ha para 27,8 t/ha. As métricas de validação apresentaram R² de 49,98% (pré-evento) e 45,99% (pós-evento), com RMSE de 19,17 e 21,55 Mg/ha, respectivamente, evidenciando subestimação em altos valores de AGB e superestimação em baixos valores, consistentes com o efeito de compressão à média e limitações do GEDI em dosséis densos. Conclui-se que a abordagem multisensor e de baixo custo é viável para o monitoramento espaço- temporal da biomassa em manguezais, fornecendo subsídios para a gestão do carbono azul e a conservação costeira no Brasil. Palavras-chave: carbono azul; Random Forest; GEDI; sensoriamento remoto; manguezais
Mangroves are coastal ecosystems of high ecological relevance, standing out as blue carbon sinks with a sequestration capacity that surpasses that of most tropical forests. However, estimating their above-ground biomass (AGB) through traditional methods is costly and spatially limited. This research developed and validated a predictive AGB model for the mangrove forest of the Piraquê-Açu-Mirim Municipal Sustainable Development Reserve (ES), using exclusively freely available remote sensing data (Sentinel-1, Sentinel-2, and GEDI) and machine learning algorithms (Random Forest) on the Google Earth Engine platform. The methodology integrated spectral vegetation indices, reflectance bands, and SAR backscatter, with GEDI biomass density samples as the dependent variable. The model was applied in two scenarios (pre- and post-extreme climatic event occurring in June 2016), allowing for the quantification of biomass loss. The results indicated a reduction of approximately 7,581 tons of total biomass over six months, with a decrease in mean density from 31.9 t/ha to 27.8 t/ha. Validation metrics yielded an R² of 49.98% (pre-event) and 45.99% (post-event), with RMSE of 19.17 and 21.55 Mg/ha, respectively, evidencing underestimation at high AGB values and overestimation at low values, consistent with the compression-to-the-mean effect and GEDI limitations in dense canopies. It is concluded that the low-cost, multi-sensor approach is feasible for the spatiotemporal monitoring of biomass in mangroves, providing support for blue carbon management and coastal conservation in Brazil. Keywords: blue carbon; Random Forest; GEDI; remote sensing; mangroves
Mangroves are coastal ecosystems of high ecological relevance, standing out as blue carbon sinks with a sequestration capacity that surpasses that of most tropical forests. However, estimating their above-ground biomass (AGB) through traditional methods is costly and spatially limited. This research developed and validated a predictive AGB model for the mangrove forest of the Piraquê-Açu-Mirim Municipal Sustainable Development Reserve (ES), using exclusively freely available remote sensing data (Sentinel-1, Sentinel-2, and GEDI) and machine learning algorithms (Random Forest) on the Google Earth Engine platform. The methodology integrated spectral vegetation indices, reflectance bands, and SAR backscatter, with GEDI biomass density samples as the dependent variable. The model was applied in two scenarios (pre- and post-extreme climatic event occurring in June 2016), allowing for the quantification of biomass loss. The results indicated a reduction of approximately 7,581 tons of total biomass over six months, with a decrease in mean density from 31.9 t/ha to 27.8 t/ha. Validation metrics yielded an R² of 49.98% (pre-event) and 45.99% (post-event), with RMSE of 19.17 and 21.55 Mg/ha, respectively, evidencing underestimation at high AGB values and overestimation at low values, consistent with the compression-to-the-mean effect and GEDI limitations in dense canopies. It is concluded that the low-cost, multi-sensor approach is feasible for the spatiotemporal monitoring of biomass in mangroves, providing support for blue carbon management and coastal conservation in Brazil. Keywords: blue carbon; Random Forest; GEDI; remote sensing; mangroves
Description
Citation
SILVA, Matheus Guimarães Cancela. Modelagem da biomassa de mangues utilizando sensores remotos e inteligência artificial: estudo de caso na Reserva de Desenvolvimento Sustentável Municipal Piraquê-Açu-Mirim no estado do Espírito Santo, Brasil. 2026. 57 f. Dissertação (Mestrado em Geografia) - Universidade Federal de Viçosa, Viçosa. 2026.
