Fechar

@MastersThesis{Ruiz:2022:UsMéFe,
               author = "Ruiz, Isadora Haddad",
                title = "Uso de m{\'e}tricas fenol{\'o}gicas calculadas de diferentes 
                         {\'{\i}}ndices de vegeta{\c{c}}{\~a}o da 
                         constela{\c{c}}{\~a}o de sat{\'e}lites PlanetScope para 
                         classifica{\c{c}}{\~a}o de fitofisionomias do Cerrado",
               school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
                 year = "2022",
              address = "S{\~a}o Jos{\'e} dos Campos",
                month = "2022-02-11",
             keywords = "fenologia da vegeta{\c{c}}{\~a}o, planetScope, Cerrado, 
                         constela{\c{c}}{\~a}o de sat{\'e}lites, 
                         classifica{\c{c}}{\~a}o, ensemble metrics, land surface 
                         phenology, random Forest, satellite constellation, Savannas.",
             abstract = "O mapeamento da vegeta{\c{c}}{\~a}o nativa do Cerrado no Brasil 
                         {\'e} desafiador, n{\~a}o havendo um consenso sobre a melhor 
                         estrat{\'e}gia de sensoriamento remoto para lidar com a 
                         variabilidade espacial de algumas fitofisionomias e a similaridade 
                         espectral de outras. Neste estudo, avaliou-se o desempenho de 12 
                         m{\'e}tricas fenol{\'o}gicas (Land Surface Phenology - LSP) 
                         calculadas a partir de tr{\^e}s diferentes {\'{\i}}ndices de 
                         vegeta{\c{c}}{\~a}o (IV) da constela{\c{c}}{\~a}o de 
                         sat{\'e}lites PlanetScope (PS). As m{\'e}tricas foram usadas 
                         como vari{\'a}veis de entrada para o algoritmo de aprendizado de 
                         m{\'a}quina Random Forest (RF), visando classificar oito 
                         fitofisionomias do Cerrado. A {\'a}rea de estudo foi a 
                         Esta{\c{c}}{\~a}o Ecol{\'o}gica de {\'A}guas Emendadas (ESAE), 
                         localizada na regi{\~a}o central do Brasil. Testou-se a 
                         classifica{\c{c}}{\~a}o LSP na esta{\c{c}}{\~a}o de 
                         crescimento 2017-2018 contra a classifica{\c{c}}{\~a}o IV na 
                         esta{\c{c}}{\~a}o seca de 2017, usando um mapa 
                         dispon{\'{\i}}vel de vegeta{\c{c}}{\~a}o como refer{\^e}ncia 
                         para avalia{\c{c}}{\~a}o da precis{\~a}o dos resultados. 
                         Al{\'e}m disso, analisou-se o desempenho do uso combinado (todos 
                         os IVs ou m{\'e}tricas LSP em conjunto) e individual (cada IV ou 
                         m{\'e}tricas LSP usadas separadamente) das vari{\'a}veis na 
                         classifica{\c{c}}{\~a}o RF das fitofisionomias. Os resultados 
                         mostraram que a acur{\'a}cia total (OA) da 
                         classifica{\c{c}}{\~a}o RF usando 12 imagens PS adquiridas na 
                         esta{\c{c}}{\~a}o seca de 2017, variou de 0,56 para o Green-Red 
                         Normalized Difference (GRND) a 0,57 e 0,61 para o Enhanced 
                         Vegetation Index (EVI) e Normalized Difference Vegetation Index 
                         (NDVI), respectivamente. As m{\'e}tricas LSP, determinadas 
                         durante a esta{\c{c}}{\~a}o de crescimento de 2017-2018, 
                         produziram ganhos de 19,3% (EVI), 13,1% (NDVI) e 5,4% (GRND), 
                         quando comparadas com o uso isolado de IVs da esta{\c{c}}{\~a}o 
                         seca. Mantendo o EVI da esta{\c{c}}{\~a}o seca como 
                         refer{\^e}ncia para compara{\c{c}}{\~a}o, o uso combinado dos 
                         IVs (OA = 0,70) ou m{\'e}tricas LSP (OA = 0,73) produziu ganhos 
                         na OA de 22,8% e 28,1%, respectivamente. As vari{\'a}veis mais 
                         significativas para o modelo RF empregando conjuntamente as 
                         m{\'e}tricas LSP foram obtidas principalmente do NDVI e EVI, 
                         sendo elas: o valor m{\'{\i}}nimo (TRG) e m{\'a}ximo (PEAK) de 
                         IV; o valor m{\'e}dio na primavera (MSP); o valor m{\'e}dio na 
                         esta{\c{c}}{\~a}o de crescimento (MGS); e a taxa de verdejamento 
                         na primavera (RSP). Os resultados mostraram a import{\^a}ncia de 
                         se utilizar dados de alta resolu{\c{c}}{\~a}o espacial e 
                         temporal da constela{\c{c}}{\~a}o de sat{\'e}lites PlanetScope 
                         para classificar fitofisionomias de Cerrado, usando 
                         informa{\c{c}}{\~o}es de fenologia da vegeta{\c{c}}{\~a}o. 
                         Al{\'e}m disso, quando s{\'e}ries temporais densas n{\~a}o 
                         estiverem dispon{\'{\i}}veis para calcular adequadamente as 
                         m{\'e}tricas LSP, uma alternativa {\'e} o uso combinado de IVs 
                         com sensibilidades diferentes aos par{\^a}metros 
                         biof{\'{\i}}sicos da vegeta{\c{c}}{\~a}o. Isto {\'e} 
                         v{\'a}lido especialmente para dados de sat{\'e}lite adquiridos 
                         durante a esta{\c{c}}{\~a}o seca local, quando a frequ{\^e}ncia 
                         de cobertura de nuvens {\'e} reduzida. ABSTRACT: Mapping of 
                         savannas in Brazil is challenging since there is no consensus on 
                         the best remote sensing strategy to deal with the spatial 
                         variability of some physiognomies and the spectral similarity 
                         among others. In this study, we evaluated the performance of 12 
                         land surface phenology (LSP) metrics calculated from three 
                         PlanetScope (PS) vegetation indices (VIs) for Random Forest (RF) 
                         classification of eight savanna physiognomies. At the protected 
                         Ecological Station of {\'A}guas Emendadas (ESAE) located in 
                         central Brazil, we tested the LSP classification in the 2017-2018 
                         growing season against the dry-season VI classification in 2017 
                         using an available vegetation map for accuracy assessment. 
                         Furthermore, we analyzed the performance of individual (each set 
                         of VIs or LSP metrics used separately) and combined (all VIs or 
                         LSP metrics used together) metrics for RF classification of the 
                         savanna physiognomies. The results showed that the overall 
                         accuracy (OA) of RF classification using 12 PS images acquired in 
                         the 2017 dry season ranged from 0.56 for the Green-Red Normalized 
                         Difference (GRND) to 0.57 and 0.61 for the Enhanced Vegetation 
                         Index (EVI) and Normalized Difference Vegetation Index (NDVI), 
                         respectively. The LSP metrics retrieved during the 2017-2018 
                         growing season produced gains in OA of 19.3% (EVI), 13.1% (NDVI) 
                         and 5.4% (GRND) when compared to the individual use of VIs in the 
                         dry season. Keeping the dry-season EVI as a reference of 
                         comparison, the combined use of VIs (OA = 0.70) or LSP metrics (OA 
                         = 0.73) also generated gains in OA of 22.8% and 28.1%, 
                         respectively. The most important ranked LSP metrics from the 
                         combination of this type of variable were mainly calculated from 
                         the NDVI and EVI: the minimum (TRG) and maximum (PEAK) VI values; 
                         the mean Spring (MSP); the mean growing season (MGS); and the rate 
                         of spring green up (RSP). The results show the importance of the 
                         combined use of high spatial and temporal resolution data of the 
                         Planets satellite constellation for the classification of 
                         Brazilian savannas using the vegetation phenology information. 
                         Besides that, when dense time series are not available for 
                         retrieving the LSP metrics, an alternative is the combined use of 
                         different VIs for satellite data acquired during the dry season 
                         when the frequency of cloud cover is reduced.",
            committee = "K{\"o}rting, Thales Sehn (presidente) and Galv{\~a}o, L{\^e}nio 
                         Soares (orientador) and Breunig, F{\'a}bio Marcelo (orientador) 
                         and Bourscheidt, Vandoir",
         englishtitle = "Use of phenological metrics calculated from different vegetation 
                         indices of the planetscope satellite constellation for classifying 
                         Savanna physiognomies",
             language = "pt",
                pages = "71",
                  ibi = "8JMKD3MGP3W34T/46DD25B",
                  url = "http://urlib.net/ibi/8JMKD3MGP3W34T/46DD25B",
           targetfile = "publicacao.pdf",
        urlaccessdate = "28 mar. 2024"
}


Fechar