Banca de QUALIFICAÇÃO: LUCAS SANTOS HONDA

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
STUDENT : LUCAS SANTOS HONDA
DATE: 26/06/2024
TIME: 13:30
LOCAL: Departamento de Agrotecnologias e Sustentabilidade
TITLE:

EVALUATION OF THE USE OF REFERENCE AREA IN MAPPING
OF C STOCK IN CERRADO SOILS

KEY WORDS:

organic carbon; pedotransfer function, digital soil mapping

PAGES: 17
BIG AREA: Ciências Agrárias
AREA: Agronomia
SUBÁREA: Ciência do Solo
SPECIALTY: Manejo e Conservação do Solo
SUMMARY:

Mapping soil carbon stocks (ECS) represents an important
baseline for the development of public policies within the scope of mitigation of
greenhouse gases (GHG), since there is more carbon stored in the soil than in
atmosphere and plant biomass together. Another role in which inventory mapping
of carbon provides is an understanding of the amount of carbon(C) in the soil and how the
is affected by soil use and coverage, representing an important ally in
soil security and, consequently, food security. One of the difficulties
faced in monitoring soil C is the high cost of data collection
in situ, due to numerous factors such as territorial extension and accessibility, generating
ECSa maps from relatively sparse data or equations of
pedotransference.Digital soil mapping (MDS) arises from the demand for
information about the spatial distribution of soil properties and classes.
In the last two decades, MDS has developed a powerful ally in the
machine learning (ML), which through the use of sensor-derived covariates
remote, are capable of performing tasks of predicting classes and attributes of
soils. One of the approaches employed in the MDS involves extending the soil-landscape relationships of a previously mapped area, identified as a reference area (AR),
to other regions where these relationships still maintain their validity. The use of techniques
of AM, combined with the use of data collected extensively in a reference area (AR) and
specialized pedological knowledge represents a possible solution to generating
maps for ECS with greater detail, especially for large
extensions such as the Cerrado biome, which has been suffering from continuous transformations
of its native vegetation, often replaced by unproductive pastures, altering the
dynamics of soil organic matter and contributing to GHG emissions.
Considering these aspects, we raise the hypothesis that it is possible to produce
ECS maps for the Cerrado with greater precision and accuracy based on pre-existing estimates, using an extensive database created from a survey of
conventional soil associated with the use of machine learning algorithms. Thus, the
the objective of the present work was to evaluate: 1 – the use of the reference area (AR) for the
prediction of carbon stocks in the Cerrado; 2 – the transferability and performance of 2
machine learning algorithms: random forest (RF) and regression tree (RT).

COMMITTEE MEMBERS:
Presidente - 1740899 - ERIKA FLAVIA MACHADO PINHEIRO
Externo ao Programa - 1315209 - MAURO ANTONIO HOMEM ANTUNES - UFRRJExterno à Instituição - GUSTAVO DE MATTOS VASQUES - EMBRAPA