UNIVERSIDADE ESTADUAL PAULISTA - UNESP 

CÂMPUS DE JABOTICABAL 

 

 

 

 

 

 

CARBON DIOXIDE DYNAMICS THROUGH ORBITAL AND 

TERRESTRIAL MEASUREMENTS ON AMAZONIAN CROP 

AND FOREST LANDS 

 

 

 

Fernando Saragosa Rossi 

Computer science 

 

 

 

 

 

 

 

 

 

 

2023



 

UNIVERSIDADE ESTADUAL PAULISTA - UNESP 

CÂMPUS DE JABOTICABAL 

 

 

 

 

 

 

CARBON DIOXIDE DYNAMICS THROUGH ORBITAL AND 

TERRESTRIAL MEASUREMENTS ON AMAZONIAN CROP 

AND FOREST LANDS 

 

 

 

Fernando Saragosa Rossi 
 

Advisor: Prof. Dr. Newton La Scala Júnior 
 

Co-Advisor: Prof. Dr. Carlos Antonio da Silva Junior  
 
 

The thesis presented to the College of 
Agricultural and Veterinarian Sciences – 
UNESP, Jaboticabal Campus, as partial 
fulfillment of the Doctor degree in Agronomy 
(Soil Science). 

 

 

 

 

 

2023 



 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 



 



 

AUTHOR'S CURRICULUM DATA 
 

FERNANDO SARAGOSA ROSSI – Son of Osvaldo Saragosa Rossi and Ana 

Aparecida Bandini Rossi, he was born in Ivaté - PR, on March 11, 1986. He attended 

minor elementary school at the Escola Estadual Ouro Verde and the middle school at 

the Colégio Ágora, in that same school he finished high school in 2003. From 2005 to 

2006, he attended the technician/professional course in Technician in Informatics. 

Escola Tecnica de Viçosa, ETEV, Brazil and also took a technical/professional course 

in Technician in Electrotechnics. Charles Babbage Teaching Institution, UNIORKA, 

Brazil. In 2007, he joined the Computer Science course at Universidade Paulista, 

UNIP, Brazil with the course completion work entitled "Prototype in Interactivity 

Software for Digital TV" under the guidance of professor Mário Henrique Souza Pardo. 

He worked from 2011 to 2015 at the company Odebrecht Energia e Infraestrutura. In 

2018, he started the Master's degree in Mestrado em Biodiversidade e 

Agroecossistemas Amazônicos, at the Universidade do Estado de Mato Grosso, 

UNEMAT, Campus of Alta Floresta - MT. On February 12, 2017, he submitted to the 

Master's defense, approved as a Master in Biodiversidade e Agroecossistemas 

Amazônicos. In March 2020, he started his Ph.D. course in Agronomy (Soil Science) 

also from FCAV-UNESP. In January 2023, he submitted the doctoral thesis to an 

examination panel, and received his Ph.D. degree in Agronomy (Soil Science) at 

UNESP/FCAV.



 

“You can look at a mistake as bullshit to be forgotten or as a result that points 

you in a new direction.” 

Steve Jobs 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 



 

I DEDICATE 

To my father Osvaldo Saragosa Rossi and my mother Ana Aparecida 

Bandini Rossi for their unconditional support in all the difficult moments of my life. 

This work is dedicated to them. Eternal gratitude.  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

  



 

ACKNOWLEDGMENTS 

 

Firstly, to my parents, Ana and Osvaldo, for always believing in me, for still supporting 

me, for their love, understanding, trust and especially for the teachings they gave me 

for life, you are the most influential teachers I have in my life. My sister Fernanda too, 

for her love and friendship. 

 

I thank my companion Auana Vicente Tiago for all her patience and dedication to me. 

 

To my advisor and friend, Dr. Newton La Scala Júnior, for their patience, guidance, 

trust, support, friendship. Thank you very much for contributing to my professional, 

personal and intellectual development. I will be eternally grateful to you for my training 

as a scientist. 

 

To my co-advisor, Dr. Carlos Antonio da Silva Junior, for having received me so 

fraternally in his life, in the laboratory at UNEMAT -SINOP / Alta Floresta and always 

ready to clarify all my doubts, his contribution was indispensable. Mainly for his 

friendship besides being a teacher. Thank you very much. 

 

Special thanks to a friend who welcomed me with open arms from the beginning of my 

doctoral journey, Gustavo André de Araújo Santos. 

 

To the members who participated in my qualification exam: Dr. Glauco De Souza Rolim 

and Dr. Alan Rodrigo Panosso.    

 

To the examination panel members of the thesis defense: Dr. Glauco De Souza Rolim, 

Dr. Alan Rodrigo Panosso, Dr Paulo Eduardo Teodoro and Dr José Francisco de 

Oliveira Júnior. 

 

To my colleagues at the GAAF (UNEMAT-SINOP). 

 



To the professors of the Departament of Engineering and Mathematical Sciences, 

thanks for the pleasant coexistence during my journey. 

Thanks also at Universidade Estadual Paulista, Câmpus de Jaboticabal and Programa 

de pós-graduação em Agronomia (Ciência do Solo) for the opportunity to be able to 

study his doctorate in one of the best programs in the country.  

This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal 

de Nível Superior - Brasil (CAPES) - Finance Code 001.

The databases SCOPUS, ScienceDirect, WEB OF SCIENCE, SCIELO, and the 

Periódico Capes facilitated the search for my literature review and understanding 

of my results. 

And to all who contributed directly and indirectly to the conduct of my study. 



xi 

SUMMARY 

 
                                                                                                                               Page 

LIST OF ABBREVIATIONS AND ACRONYMS........................................................ xv 

LIST OF FIGURES ................................................................................................. xviii 

LIST OF TABLES .................................................................................................... xxi 

CHAPTER 1 – GENERAL CONSIDERATIONS ......................................................... 1 

1 INTRODUCTION .................................................................................................. 1 

1.3 LITERATURE REVIEW ..................................................................................... 2 

1.3.1 Greenhouse Gases (GHGs) ............................................................................ 2 

1.3.2 Carbon in soil .................................................................................................. 3 

1.3.3 Land use and land cover change .................................................................... 5 

1.3.4 Remote Sensing ............................................................................................. 6 

CHAPTER 2 – VARIATIONS OF CO2 EMISSIONS ON DIFFERENT VEGETATIONS 

COVERS IN THE BRAZILIAN AMAZON ................................................................... 9 

ABSTRACT .............................................................................................................. 9 

2.1 INTRODUCTION ............................................................................................. 10 

2.2 MATERIAL AND METHODS ............................................................................ 12 

2.2.1 Studied area .................................................................................................. 12 

2.2.2 Vegetation Indices ........................................................................................ 13 

2.2.3 Planet© Images ............................................................................................ 15 

2.2.4 Gross primary production (GPP) ................................................................... 15 

2.2.5 CO2 Model Data ........................................................................................... 16 

2.2.6 Assessing soil CO2 emission, temperature, and moisture ............................ 17 

2.2.7 Statistical analysis ......................................................................................... 20 

2.3 RESULTS ........................................................................................................ 20 

2.4 DISCUSSION ................................................................................................... 31 

2.5 CONCLUSIONS ............................................................................................... 33 

CHAPTER 3 – RELATION BETWEEN SOIL RESPIRATION AND MULTIPLE 

VARIABLES ON EUCALYPTUS SPECIES CROP IN SOUTHWESTERN BRAZILIAN 

CERRADO ................................................................................................................ 35 

ABSTRACT ............................................................................................................ 35 

3.1 INTRODUCTION ............................................................................................. 36 

3.2 MATERIAL AND METHODS ............................................................................ 37 

3.2.1 Study area ..................................................................................................... 37 

3.2.2 In loco performance of experiment ................................................................ 38 

3.2.3 Multispectral airborne images acquisition and vegetation indices ................. 41 

3.2.4 Statistical analysis ......................................................................................... 45 

3.3 RESULTS ........................................................................................................ 45 

3.4 DISCUSSION ................................................................................................... 52 

3.5 CONCLUSIONS ............................................................................................... 54 



xii 

CHAPTER 4 – CARBON DIOXIDE SPATIAL VARIABILITY AND DYNAMICS FOR 

CONTRASTING LAND USES IN CENTRAL BRAZIL AGRICULTURAL FRONTIER 

FROM REMOTE SENSING DATA ........................................................................... 55 

ABSTRACT ............................................................................................................ 55 

4.1 INTRODUCTION ............................................................................................. 56 

4.2 MATERIAL AND METHODS ............................................................................ 58 

4.2.1 Study area ..................................................................................................... 58 

4.2.2 Orbiting Carbon Observatory-2 ..................................................................... 59 

4.2.3 Enhanced Vegetation Index time-series ....................................................... 59 

4.2.4 Detection of Land Use via Normalized Difference Fraction Index ................. 60 

4.2.5 Rainfall data and Standardized Precipitation index ....................................... 62 

4.2.6 CO2 Model Data ........................................................................................... 64 

4.2.7 Statistical analysis ......................................................................................... 65 

4.3 RESULTS ........................................................................................................ 65 

4.3.1 Temporal Dynamics ...................................................................................... 65 

4.3.2. Spatial dynamics .......................................................................................... 70 

4.4 DISCUSSION ................................................................................................... 77 

4.4.1 Temporal Dynamics ...................................................................................... 77 

4.4.2 Spatial Dynamics .......................................................................................... 79 

4.5 CONCLUSIONS ............................................................................................... 80 

CHAPTER 5 – GENERAL CONCLUSIONS ............................................................. 82 

References............................................................................................................. 84 

 



xiii 

DINÂMICA DO DIÓXIDO DE CARBONO ATRAVÉS DE MEDIÇÕES ORBITAIS E 
TERRESTRES EM TERRAS DE CULTIVO E FLORESTAS AMAZÔNICAS 

 
 

ABSTRACT - A emissão de dióxido de carbono (CO2) do solo é reconhecida 
como o segundo maior fluxo de carbono (C) entre os ecossistemas terrestres e a 
atmosfera. Nas áreas agrícolas, é um processo resultante da interação de diferentes 
fatores, tais como clima, condições do solo e investimento tecnológico. Assim, estudos 
capazes de caracterizar a dinâmica do CO2 são importantes na busca da 
compreensão das relações entre os atributos do solo em diferentes sistemas de 
produção. O trabalho foi realizado em quatro capítulos, desenvolvidos no estado de 
Mato Grosso e Mato Grosso do Sul. No capítulo 1, foram apresentadas as 
considerações gerais do estudo. No capítulo 2, avaliamos a variabilidade temporal da 
emissão de CO2 do solo e sua relação com variáveis relacionadas como fluxo de CO2 
(CO2Flux) derivado de processos fotossintéticos, Índice de Vegetação Melhorada 
(EVI), Produtividade Primária Bruta (GPP), Índice de Área Foliar (LAI), Temperatura 
do Solo (Ts) e Umidade do Solo (Us), no ano safra 2020/2021 na Fazenda Aurora, 
que está localizada no município de Cláudia no Estado do Mato Grosso com os 
diferentes usos e ocupação do solo: soja de baixo potencial de rendimento (SB); 
floresta nativa (FN); soja de alto potencial de rendimento (SA); bom pasto (PB) e pasto 
degradado (PD). Influência significativa (p < 0,01) foi encontrada neste capítulo para 
todas as variáveis analisadas e entre os diferentes usos do solo e cobertura da terra. 
Os valores FCO2 para SA e SB com os valores mais baixos e para FN com os valores 
mais altos sendo o oposto para CO2Flux. No capítulo 3, foi avaliada a relação entre o 
fluxo de dióxido de carbono no solo e os índices multiespectrais da vegetação em um 
sistema de monocultura de espécies de eucalipto, no campo experimental da 
Universidade Federal de Mato Grosso do Sul, Campus de Chapadão do Sul onde a 
área é composta por cinco espécies de eucalipto: E. camaldulensis, E. uroplylla, E. 
saligna, E. grandis e E. urograndis e Corymbria citriodora. No capítulo 3, a umidade 
do solo apresenta uma correlação negativa entre FCO2 e significativa (p < 0,05), 
provou ser um dos fatores importantes no controle da respiração do solo. Os índices 
de vegetação NDRE (Normalized Difference Red-Edge Index), NDVI (Normalized 
Difference Vegetation Index), SAVI (Soil-Adjusted Vegetation Index) e MSAVI 
(Modified Soil-Adjusted Vegetation Index) mostraram maior correlação negativa com 
FCO2, e uma influência significativa de p < 0,001 para as espécies E. camaldulensis, 
E. saligna e E. uroplylla. No capítulo 4, a detecção remota foi usada para avaliar a 
dinâmica espaço-temporal do CO2 de 2015 a 2018 no estado de Mato Grosso, que é 
a principal fronteira agrícola do Brasil. Os resultados do capítulo 4 mostraram que a 
variabilidade temporal do fluxo de CO2 está positivamente correlacionada com a 
precipitação, enquanto que o XCO2 está negativamente correlacionado com a 
precipitação. Portanto, não somente o XCO2, mas também o fluxo de CO2 estão 
diretamente relacionados ao uso da terra e à mudança de cobertura do solo (LULCC) 
em sistemas complexos que são afetados por variáveis e processos climáticos, como 
a fotossíntese e a respiração do solo.   
 
 
Keywords: Fluxo de CO2, Uso da terra e cobertura da terra, Agricultura, Gases de 

efeito estufa  



xiv 

CARBON DIOXIDE DYNAMICS THROUGH ORBITAL AND TERRESTRIAL 
MEASUREMENTS ON AMAZONIAN CROP AND FOREST LANDS 

 
 

ABSTRACT - The soil carbon dioxide (CO2) emission, is recognized as the 
second largest carbon (C) flux between terrestrial ecosystems and the atmosphere. In 
agricultural areas it is a process resulting from the interaction of different factors, such 
as climate, soil conditions and technological investment. Thus, studies capable of 
characterizing the dynamics of CO2 are important in the search for understanding the 
relationships between soil attributes in different production systems. The work was 
conducted in four chapters, developed in the state of Mato Grosso and Mato Grosso 
do Sul. In chapter 1, the general considerations of the study were presented. In chapter 
2, we evaluated the temporal variability of CO2 emission from soil and its relation with 
related variables as CO2 flux (CO2Flux) derived from photosynthetic processes, 
Enhanced vegetation index (EVI), Gross Primary Productivity (GPP), Leaf Area Index 
(LAI), Soil Temperature (Ts) and Soil Moisture (Ms), in the crop year 2020/2021 in 
Aurora Farm, which is located in the municipality of Cláudia in the State of Mato Grosso 
with the different land uses and occupation: low yield potential soybean (SB); native 
forest (FN); high yield potential soybean (SA); good pasture (PB) and degraded 
pasture (PD). Significant influence (p < 0.01) was found in this chapter for all variables 
analyzed and between the different land uses and land cover. The FCO2 values for SA 
and SB with the lowest values and for FN with the highest values being the opposite 
for CO2Flux. In chapter 3, were evaluated the relation between the flux of carbon 
dioxide from soil and multispectral indices of vegetation in a monoculture system of 
eucalypt species, in the experimental field of the Federal University of Mato Grosso do 
Sul, Campus de Chapadão do Sul where the area is composed of five species of 
eucalyptus: E. camaldulensis, E. uroplylla, E. saligna, E. grandis and E. urograndis 
and Corymbria citriodora. In chapter 3 soil moisture has a negative correlation between 
FCO2 and significant (p < 0.05), proved to be one of the important factors in controlling 
soil respiration. The vegetation indices NDRE (Normalized Difference Red-Edge 
Index), NDVI (Normalized Difference Vegetation Index), SAVI (Soil-Adjusted 
Vegetation Index) and MSAVI (Modified Soil Adjusted Vegetation Index) showed 
higher negative correlation with FCO2, and a significant influence of p < 0.001 for E. 
camaldulensis, E. saligna and E. uroplylla species. In chapter 4, remote sensing was 
used to evaluate the spatio-temporal dynamics of CO2 from 2015 to 2018 in the state 
of Mato Grosso, which is the main agricultural frontier in Brazil. The results of chapter 
4 showed that the temporal variability of CO2 flux is positively correlated with 
precipitation, while XCO2 is negatively correlated with precipitation. Therefore, not only 
XCO2, but also CO2flux are directly related to land use and land cover change (LULCC) 
in complex systems that are affected by climatic variables and processes, such as 
photosynthesis and soil respiration.   
 
 
Keywords: CO2 flux, Land use and land cover, Agriculture, Greenhouse Gas 

 

 



xv 

LIST OF ABBREVIATIONS AND ACRONYMS 

 
Al - Aluminum 

ANOVA - Analysis of Variance 

APAR - Absorbed Photosynthetically Active Radiation 

ARVI2 - Atmospherically Resistant Vegetation Index 2 

ATSAVI - Ajusted Transformed Soil-Ajusted VI 

BWDRVI - Blue-Wide Dynamic Range Vegetation Index 

CCCI - Canopy Chlorophyll Content Index 

CHIRPS - Climate Hazards Group Infrared Precipitation with Station 

CIgreen - Chlorophyll Index Green 

CIrededge - Chlorophyll Index RedEdge 

CVI - Chlorophyll Vegetation Index 

DFOV - Diagonal Field of View  

DVI - Difference Vegetation Index 

ESA - European Space Agency 

EVEI2 - Enhanced Vegetation Index 2 

EVI - Enhanced vegetation index 

EVI - Improved Vegetation Index 

fAPAR - Fraction of photosynthetically active radiation absorbed by vegetation 

FCO2 - Soil CO2 emission 

FN - Native Forest 

GCP - Ground Control Points  

GDVI - Difference NIR/Green Difference Vegetation Index 

GEMI - Global Environment Monitoring Index 

GHG - Greenhouse Gas 

GNDVI - Green Normalized Difference Vegetation Index 

GPP - Gross Primary Productivity 

GRNDVI - Green-Red NDVI 

GRVI - Green-Red Vegetation Index) 

GSAVI - Green Soil Adjusted Vegetation Index 

GTVI - Green Triangle Vegetation Index 

HFOV - Horizontal Field of View  



xvi 

HV - Healthy Vegetation 

IBGE - Brazilian Institute of Geography and Statistics 

INMET - National Institute of Meteorology 

IPCC - Intergovernmental Panel on Climate Change 

IPVI - Infrared Percentage Vegetation Index 

LAI - Leaf Area Index 

LAPIG - Image Processing and Geoprocessing Laboratory 

LogR - Log Ratio 

LSMM - Linear Spectral Mixture Model  

LULC - Land Use and Land Cover 

MODIS - Moderate Resolution Imaging Spectroradiometer 

MSAVI - Modified Soil Adjusted Vegetation Index 

MSRNir_Red - Modified Simple Ratio NIR/RED 

NASA/POWER - National Aeronautics and Space Administration / Prediction of 

Worldwide Energy Resources 

NDFI - Normalized Difference Fraction Index 

NDRE - Normalized Difference Red-Edge Index 

NDVI - Normalized difference vegetation index 

NGRDI - Normalized Green-Red Difference Index 

NIR - Near-infrared 

NOAA - National Oceanic and Atmospheric Administration 

NormR1 - Normalized G 

NormR2 - Normalized NIR 

NormR3 - Normalized R 

OCO-2 - Orbiting Carbon Observatory-2 

PAR - Photosynthetically Active Radiation 

PB - Productive Pasture  

PCA - Principal Component Analysis 

PD - Degraded Pasture  

PDF - Probability Density Function 

PRI - Photosynthetic Vegetation Index 

PVC - Polyvinyl Chloride 



xvii 

RGR - Red Green Ratio Index 

RI - Redness Index  

RTK - Real Time Kinematic 

SA - High-Yield Potential Soybean  

SAVI - Soil Adjusted Vegetation Index  

SB - Low-Yield Potential Soybean  

SM - Soil moisture 

SMAP - Soil Moisture Active Passive 

SMT - State of Mato Grosso 

SPI - Standardized Precipitation Index 

sPRI - Scaled Value Of PRI 

ST - Soil temperature 

TDR - Time Domain Reflectometry 

TES - Tropospheric Emission Spectrometer  

VFOV - Vertical Field of View  

VIs - Vegetation Indices  

VPDscalar - Water Vapor Pressure Deficit 

WDRVI - Wide Dynamic Range Vegetation Index 

ε - Light-Use Efficiency 

  



xviii 

LIST OF FIGURES 

 
Page 

CHAPTER 2 – IMPLICATIONS OF CO2 EMISSIONS ON THE MAIN LAND AND 
FOREST USES IN THE BRAZILIAN AMAZON1 

Figure 1. Study area located in the State of Mato Grosso with the different land uses 
and occupation in the Amazon ecoregion: a) low-yield potential soybean (SB); b) native 
forest (FN); c) high-yield potential soybean (SA); d) productive pasture (PB) and e) 
degraded pasture (PD) in the municipality of Cláudia. .............................................. 13 

Figure 2. LI-COR model LI-8100 portable system (a) connected to the soil chamber 
(b), soil moisture sensor (c) and soil temperature sensor (d). ................................... 18 

Figure 3. Locations of the land uses and land cover of the experiment. a) native forest; 
b) productive pasture; c) degraded pasture; d) high-yield potential soybean; e) low-
yield potential soybean. ............................................................................................. 19 

Figure 4. Grouping of means and the standard deviation of the variables a) Soil 
Moisture (Ms) and b) Soil Temperature (Ts) evaluated on different dates in the land 
uses and land cover (LULC) of native forest (FN), productive pasture (PB), degraded 
pasture (PD), high-yield soybean (SA) and low-yield soybean (SB). Means followed by 
uppercase letters do not differ in land use and land cover within each day and 
lowercase letters do not differ for the dates by Scott-Knott test at 5% probability. .... 22 

Figure 5. Grouping of means and the standard deviation of the variable Soil 
Temperature (Ts) and Soil Moisture (Ms) evaluated on different dates in the land uses 
and land cover (LULC) of native forest (FN), productive pasture (PB), degraded 
pasture (PD), high-yield soybean (SA) and low-yield soybean (SB). Means followed by 
uppercase letters do not differ in land use and land cover within each day and 
lowercase letters do not differ for the dates by Scott-Knott test at 5% probability. .... 23 

Figure 6. Grouping of means and the standard deviation of the variables a) FCO2, b) 
EVI, c) GPP, d) CO2Flux and e) LAI evaluated on different dates in the land uses and 
land cover (LULC) of native forest (FN), productive pasture (PB), degraded pasture 
(PD), high-yield soybean (SA) and low-yield soybean (SB). Means followed by 
uppercase letters do not differ in land use and land cover within each day and 
lowercase letters do not differ for the dates by Scott-Knott test at 5% probability. .... 26 

Figure 7. Principal component analysis for the variables FCO2, Ms, Ts, CO2Flux, EVI, 
LAI and GPP evaluated at different dates in the land uses and land cover (LULC) of 
native forest (FN), productive pasture (PB), degraded pasture (PD), high-yield soybean 
(SA) and low-yield soybean (SB). .............................................................................. 28 

Figure 8. Pearson correlations for the variables FCO2, Ms, Ts, CO2Flux, EVI, LAI and 
GPP evaluated at different dates in the land uses and land cover (LULC) of native 
forest (FN), productive pasture (PB), degraded pasture (PD), high-yield soybean (SA) 
and low-yield soybean (SB). ...................................................................................... 30 

 
1 This chapter corresponds to the scientific article submitted to the Environmental Research, Volume 
220, 01 March 2023, 8.431 Impact Factor. 



xix 

CHAPTER 3 – RELATION BETWEEN SOIL RESPIRATION AND MULTIPLE 
VARIABLES ON EUCALYPTUS SPECIES CROP IN SOUTHWESTERN BRAZILIAN 
CERRADO 

Figure 1. Study area at the Universidade Federal de Mato Grosso do Sul, Chapadão 
do Sul campus. An area composed by five eucalyptus species: E. camaldulensis, E. 
uroplylla, E. saligna, E. grandis, E. urograndis and Corymbria citriodora. ................. 38 

Figure 2. Experimental local of the six eucalyptus species. ...................................... 39 

Figure 3. LI-COR model LI-8100 portable system (a) connected to the soil chamber 
(b), soil moisture sensor (c) and soil temperature sensor (d). ................................... 40 

Figure 4. Remotely Piloted Fixed-Wing Aircraft Sensefly eBee RTK loaded with 
Sensefly Sequoia sensor. .......................................................................................... 42 

Figure 5. Pearson correlation network among the analyzed variables. Each point 
represents an index and the lines between the points are correlations between the 
indices, with thicker lines indicating a correlation greater than 0.6 and the red and green 
color indicating negative and positive correlation respectively. ................................. 46 

Figure 6. Canonical variable analysis considering eucalyptus species and FCO2, soil 
moisture (Ms) and temperature (Ts), spectral bands (green, red, NIR, and red-edge), 
and vegetation indices (NDRE, NDVI, SAVI, and MSAVI). ....................................... 48 

Figure 7. The spatialization of the vegetation indices that best correlated significantly 
with FCO2 among the eucalyptus species, NDRE, NDVI, SAVI, and MSAVI. ........... 49 

Figure 8. Scatterplot of Pearson correlations and variables (FCO2, soil temperature 
and moisture, spectral bands, and vegetation indices) and the distribution of variables 
according to the canonical variable analysis modeling considering the eucalyptus 
species. ..................................................................................................................... 51 

CHAPTER 4 – CARBON DIOXIDE SPATIAL VARIABILITY AND DYNAMICS FOR 
CONTRASTING LAND USES IN CENTRAL BRAZIL AGRICULTURAL FRONTIER 
FROM REMOTE SENSING DATA2 

Figure 1. Location of study area in central-western Brazil, Mato Grosso State based 
on the average Normalized Difference Vegetation Index (NDVI) for March 2018. .... 58 

Figure 2. Temporal series of XCO2 (ppm), including (A) daily observations, (B) annual 
averages, and monthly averages for (C) 2015, (D) 2016, (E) 2017, and (F) 2018 for the 
State of Mato Grosso, Brazil...................................................................................... 66 

Figure 3. Seasonal variability of atmospheric XCO2 with (a,b,c,d) the standardized 
precipitation index (SPI), (e,f,g,h) rainfall (i,j,k,l) CO2 flux, and (m,n,o,p) the enhanced 
vegetation index (EVI) for the State of Mato Grosso, Brazil. ..................................... 67 

Figure 4. Temporal series of land use according to the enhanced vegetation index 
(EVI) for the State of Mato Grosso, Brazil. ................................................................ 69 

 
2 This chapter corresponds to the scientific article published in the Journal of South American Earth 
Sciences https://doi.org/10.1016/j.jsames.2022.103809. 

https://doi.org/10.1016/j.jsames.2022.103809


xx 

Figure 5. Heat-map of Pearson’s correlation matrix for XCO2 with CO2 flux (FCO2), 
rainfall, and standardized precipitation index (SPI-12) for the State of Mato Grosso, 
Brazil. ........................................................................................................................ 70 

Figure 6. Land use in the State of Mato Grosso, Brazil during 2015, 2016, 2017, and 
2018. ......................................................................................................................... 71 

Figure 7. Average annual rainfall for the State of Mato Grosso, Brazil via Climate 
Hazards Group InfraRed Precipitation with Station (CHIRPS) data for 2015, 2016, 
2017, and 2018. ........................................................................................................ 72 

Figure 8. Standardized precipitation index (SPI-12) values for the State of Mato 
Grosso, Brazil via Climate Hazards Group InfraRed Precipitation with Station 
(CHIRPS) data for 2015, 2016, 2017, and 2018. ....................................................... 74 

Figure 9. CO2 flux for the State of Mato Grosso, Brazil for 2015, 2016, 2017, and 2018.
 .................................................................................................................................. 75 

Figure 10. Spatial patterns of atmospheric XCO2 concentration for the State of Mato 
Grosso, Brazil for 2015, 2016, 2017, and 2018. ........................................................ 77 

 

 



xxi 

LIST OF TABLES 

 
                                                                                                                               Page 

CHAPTER 2 – IMPLICATIONS OF CO2 EMISSIONS ON THE MAIN LAND AND 
FOREST USES IN THE BRAZILIAN AMAZON3 

Table 1. Collection dates for each land use and land cover and soybean soil 
preparation, sowing, and harvesting periods. ............................................................ 19 

Table 2. Analysis of variance for the seven variables among the five land uses and 
land cover (LULC) for the dates. ............................................................................... 21 

CHAPTER 3 – RELATION BETWEEN SOIL RESPIRATION AND MULTIPLE 
VARIABLES ON EUCALYPTUS SPECIES CROP IN SOUTHWESTERN BRAZILIAN 
CERRADO 

Table 1. Vegetation indices over Parrot Sequoia imagery. ....................................... 43 

Table 2. Multivariate analysis of variance considering eucalyptus species and the 
variables FCO2, soil temperature and moisture, spectral bands (green, red, infrared, 
and red edge), and vegetation indices (NDRE, NDVI, SAVI, and MSAVI). ............... 47 

 
 

 
3 This chapter corresponds to the scientific article submitted to the Environmental Research, Volume 
220, 01 March 2023, 8.431 Impact Factor. 



1 

CHAPTER 1 – GENERAL CONSIDERATIONS 

 

1 INTRODUCTION  

The conversion of land use and land cover is currently one of the main impacts 

to the environment, and as consequence of the loss of local biodiversity and carbon 

emissions. This data survey is essential to report on the fulfillment of the goals signed 

with the civil society and international organizations, besides being important in the 

elaboration of strategies for the improvement of public policies. While the preservation 

and conservation of vegetation is necessary for CO2 recycling, there is a need for 

large-scale food production, relative to the world population increase. 

The expansion and modernization of agriculture implies on socio-environmental 

impacts such as the different areas occupied by soybean cultivation and thus, forest 

areas were replaced by agriculture, promoting intensification of the forest 

fragmentation process (ITAQUI, 2002; SILVA; LIMA, 2018b). However, the factors 

favorable to soybean farming and the support of the Federal Government were 

fundamental for its rapid expansion in the country (BATISTA FARIAS, 2018; SOARES; 

SPOLADOR; OTHERS, 2016). 

The soil is an important carbon pool, and its quantity variation has direct 

interference in CO2 concentrations, and can signal low or high concentrations of this 

important greenhouse gas (GHG) in the system. Therefore, soil CO2 emission rates 

(FCO2) is directly related to loss or gain of soil organic carbon, arising from the 

complex dynamics among soil attributes and the climatic conditions (LAL; NEGASSA; 

LORENZ, 2015; SILVA-OLAYA et al., 2013b) related to emissions variations, due to 

its great variability in time and space (DE BORTOLI TEIXEIRA et al., 2011; EPRON 

et al., 2006).  

For better estimates of soil respiration, it is first necessary to describe its 

temporal variability and the relation among soil respiration and environmental 

variables, such as soil temperature and soil moisture (SILVA et al., 2019; XAVIER et 

al., 2020), both with orbital and in situ sensors. In this context, several studies seek 

the comprehension of agricultural practices and their respective productivities effects 

on increasing atmospheric concentrations of GHGs, particularly CO2 (LA SCALA; 

BOLONHEZI; PEREIRA, 2006; MOITINHO et al., 2015; SILVA et al., 2019). 

Thus, few works have sought to compare environmental variables assessed via 

remote sensing techniques against those obtained in situ with the flux chamber, and 



2 

especially to compare CO2 emission data in areas of different levels of soybean crop 

productivity. Understanding the spatial variability of CO2 emission at different levels of 

data acquisition in agricultural areas becomes important for the conduction of a 

controlled and sustainable management of the culture for the preservation of carbon 

in the soil, thus contributing to the reduction of the greenhouse effect and the agility in 

measurement with orbital data. 

Sensors onboard land monitoring satellites obtain data that assist in 

understanding the global distribution of vegetation types, agricultural crops, as well as 

their biophysical and structural properties in spatial-temporal variations (DIDAN et al., 

2015). Vegetation indices allow monitoring seasonal, interannual and long-term 

variations in vegetation structure, phenological and biophysical parameters (HUETE 

et al., 2002), as well as changes in climatic variables. 

The hypothesis of this study is that changes in land use affects either the soil 

CO2 emission (FCO2) or the CO2 flux model (CO2Flux) and those could be related to 

vegetative aspects like Enhanced Vegetation Index (EVI) and Leaf Area Index (LAI), 

resulting from modifications of the land use and land cover (LULC) system and 

environmental factors, which exhibit spatial and temporal variability. 

Thus, the objective of the study is to evaluate the spatial and temporal 

characteristics of FCO2 and CO2Flux as function of climate and multispectral 

vegetation indexes in contrasting land uses in crop and forest lands of the state of 

Mato Grosso and Mato Grosso do Sul, Brazil.  

 

1.3 LITERATURE REVIEW 

1.3.1 Greenhouse Gases (GHGs)  

In recent decades, climate change results from increasing concentrations of the 

carbon dioxide (CO2), methane (CH4), nitrous oxide (N2O) and the 

Chlorofluorocarbons (CFC's) among others in the atmosphere (IPCC, 2020). The rates 

of CO2, CH4 and N2O concentrations in 2019 reached levels of 409.9 (±0.4) parts per 

million (ppm), 1866.3 (±3.3) parts per billion (ppb) and 332.1 (±0.4) ppb respectively 

with an increase from 1750 to 2019 of 131, 6 ± 2.9 ppm (47.3%), 1137 ± 10 ppb (156%) 

and 62 ± 6 ppb (23.0%) and an effective radiative forcing (ERF) of halogenated 

components in 2019 was 0.4 W m-2, an increase of 3.5 % since 2011 (BYERS et al., 

2022). The increase of these gases concentration in the atmosphere has been 

significant (IPCC; OTHERS, 2013; ZHENG et al., 2019), where Earth's global 



3 

temperature has increased by 0.5 to 1 °C in the last 100 years (CHEN; XIN, 2017; 

MIKHAYLOV et al., 2020). 

Anthropic actions such as fossil fuel burning, agriculture and land use change 

are the main sources of GHGs with 66%, 20% and 14% respectively of CO2 emissions 

to the atmosphere (BRUHWILER et al., 2021; SHUKLA et al., 2019). In Brazil, the fires 

from deforestation account for 180 to 200 Mt C year-1 (megatons of carbon per year) 

being higher than the burning of fossil fuel in the country of 70 to 90 Mt C year-1 , 

making Brazil the 5th largest emitter of the world (SANQUETTA et al., 2022; SANTILLI 

et al., 2005). The contributions of total CO2 emissions in Brazil by agriculture is 

represented by 20 to 25% for global warming (HOUGHTON et al., 2001). 

Soil tillage by plowing and harrowing leads to a decrease in organic material 

and an increase in the oxidation of organic carbon into CO2 by breaking up soil 

aggregates. The dynamics of N in the soil is intrinsically related to the emission of N2O, 

which is influenced by the presence of nitrogen-fixing plants and additions of mineral 

nitrogen fertilizer. The degradation process by agriculture soil management causes 

the microbiota under anaerobic conditions to oxidize CO2 to CH4 (GUENET et al., 

2021; TIWARI; SINGH; SINGH, 2020). 

Terrestrial ecosystems composed of vegetation and soil are currently 

considered important carbon sinks, particularly soils. There are several ways in which 

proper management of the earth's biosphere, particularly soils, can significantly 

reduce the increase in GHGs.  

 

1.3.2 Carbon in soil  

Soil is an important carbon reservoir, in the first 100 cm depth, globally between 

1,300 - 2,000 Pg (petagram) of carbon are stored, corresponding to twice the 

concentration of atmospheric carbon (LAL, 2003; MAYER et al., 2019). Changes in 

land use and land cover can have significant impacts on the concentration of carbon 

dioxide (ROSSI et al., 2022) and other GHGs in the atmosphere (SCHUMAN; 

JANZEN; HERRICK, 2002; SCHWANDNER et al., 2017). 

Recently, research has pointed out that soil management is a decisive factor in 

the soil organic carbon sequestration, such as the no-till farming system, which allows 

crop rotation, straw maintenance on the soil surface with the use of cover crops and 

sowing without soil disturbance, providing a significant increase in soil C stock 

(TONELLO; AZEREDO, 2021). 



4 

Tropical soils store 506 Pg of carbon (MARÍN-SPIOTTA; SHARMA, 2013) of 

which 66 Pg of carbon are in soils of the Amazon (BATJES, 1999), and 47 Pg of carbon 

in the Brazilian legal Amazon (BATJES, 1999). Considering the first 30 cm of depth, 

the soils of the Brazilian territory, store around 36,4 Pg of carbon (BERNOUX et al., 

2002). However, assessing changes in soil C stocks still has its barriers, due to the 

difficulties in detecting small modifications (OLIVEIRA, 2018). 

 Soil carbon is present in living organic matter, which accounts for less than 4 

percent of total soil organic carbon, and in dead organic matter, which returns most of 

the total soil organic carbon (about 98 percent) (REGO et al., 2013; SIQUEIRA et al., 

2021). 

 The soil organic C stock is determined by the annual balance of the addition 

of photosynthesized C and the loss of organic C due to its oxidation to CO2 by 

heterotrophic microorganisms. Decomposition processes and transformation rates are 

strongly influenced by climate, type and quality of organic matter, chemical and 

physicochemical associations of organic matter with soil mineral components, and the 

location of organic matter in the soil (ALMEIDA; SANCHES, 2014; MÜTZENBERG, 

2020). 

 The three main processes responsible for carbon sequestration in soils are 

humification, aggregation and sedimentation. At the same time, the processes 

responsible for carbon losses in the soil are erosion, decomposition, volatilization and 

leaching (LAL, 1997; RUFINO et al., 2022). The physical protection of soil organic 

matter by the mineral fraction of the soil during the process of aggregate formation has 

been a fundamental process for the increase of soil C stocks (DE M. SÁ et al., 2001). 

The conversion of a natural ecosystem into agriculture and/or pasture may have 

an important influence on the fate of the C stored in the soil (FERNANDES; 

FERNANDES, 2008; NEILL et al., 1997; RODRIGUES, 2021). In soils under natural 

vegetation, the preservation of organic matter tends to be maximum, because soil 

disturbance is minimal, and the contribution of carbon in forests is higher than in 

cultivated areas. In cultivated areas, the levels of organic matter, as a rule, decrease, 

since the organic fractions are more exposed to attack by microorganisms, due to 

greater soil disturbance and de-structuring (CHRISTENSEN, 2020). 

However, several factors such as soil and climatic conditions can lead to an 

increase or decrease in soil C contents. Thus, studies in broad scenarios are essential, 



5 

in order to help reduce the variability of results and promote the homogeneity of 

information (CIDIN, 2016). 

 
1.3.3 Land use and land cover change 

Forest ecosystems, collaborate through their functions in reducing climate 

change, mainly by absorbing a significant part of carbon dioxide emissions and storing 

large amounts of carbon in biomass and soils (COLLALTI et al., 2019; LE QUÉRÉ et 

al., 2018), in addition to the importance for watershed protection, soil, climate 

regulation and biodiversity conservation (BONAN, 2008; ROY; RAVAN, 1996). 

Forests store up to about 45% of terrestrial carbon (YAO et al., 2018). 

Approximately 2.4 ± 0.4 Pg of Carbon per year is absorbed by the global forest 

ecosystem, reaching the equivalent of 60% of the cumulative carbon emissions from 

global fossil consumption (FU et al., 2015). The Amazon presents itself as one of the 

great potentialities to capture carbon, since it is estimated as the largest tropical forest 

in the world, and responsible for sequestering 14% of the planet's carbon (BRIENEN 

et al., 2015). 

During the development process, trees absorb CO2 and release O2 in 

photosynthesis, on the contrary, in respiration, they release carbon dioxide, in reduced 

amounts in relation to that absorbed (BARBOSA, 2015; PIRES, 2019). Thus, one can 

consider the practices for climate change mitigation, reforestation, restoration of 

degraded forests and forest protection (BERENGUER et al., 2014; HOUGHTON; 

UNRUH; LEFEBVRE, 1993; ZHANG; XU, 2003). 

Forestry and agribusiness have driven the production of soybean and 

eucalyptus monocultures in Brazil (SANTOS, 2019). The increase in soybean and 

eucalyptus planting is as a result of the demand for the spread of this activity largely 

influenced by the Brazilian state (LAMEIRA et al., 2017). 

The expansion of soybean cultivation in Brazil in the last two decades has 

reached large proportions, bringing important changes to the model of territorial 

occupation and the development of the national economy. Soybean expansion differs 

from other crops in relation to its impact on land use change, especially in the Amazon 

and Cerrado biomes (VALE; CARVALHO; ABDALA, 2021). Second largest world 

soybean producer and largest exporter since 2003, Brazil has increased its production 

following the global trend of growth in demand and supply of soybeans (LIMA, 2006). 



6 

Soybean (Glycine max (L.) Merr.) is one of the commodities that presents the 

greatest prominence in the national and international market, being one of the most 

consumed and produced grains in the world (FAOSTAT, [s.d.]). In Brazil it has a 

significant socioeconomic importance, thus becoming one of the main vehicles of 

development and the main product of agriculture in the country (HIRAKURI; 

LAZZAROTTO, 2014). 

In addition to soybeans, Brazil also stands out as one of the largest producers 

of eucalyptus (Eucalyptus spp.), with high levels of productivity due to favorable soil 

conditions, climate, and land availability (ABREU; FEIO, 2021; IBÁ, 2020). The 

increase in forest plantations in Brazilian territory reaches approximately 8.7 million 

hectares, with eucalyptus plantations predominantly occupying 7.47 million hectares, 

corresponding to 33% of the total area of eucalyptus plantations in the world 

(ALMEIDA; VIEIRA, 2022; MAPBIOMAS, 2021; ZHANG; WANG, 2021). 

Strengthening the production of paper, pulp, furniture and biofuels (LIMA et al., 2020). 

However, one of the main concerns about GHG's has been the conversion of 

natural vegetation for agriculture and cattle ranching (MCTI, 2014; SILVA; LIMA, 

2018a; SILVA et al., 2018), which result in significant amounts of CO2 emissions to 

the atmosphere, either by burning the aerial biomass or subsequent loss in soil carbon 

stocks (IPCC, 2014). This influence still seems to be indirect, since it is being 

established over degraded pasture areas from failed cattle ranching projects, in most 

cases. 

Carbon loss is faster in the conversion from native vegetation to an agricultural 

system located in the tropics, due to the combination of high humidity and temperature 

throughout the season (SCHOLES; VAN BREEMEN, 1997). The main losses of soil 

organic matter are stimulated by soil disturbance, greater variations in soil temperature 

and moisture, breakdown of aggregates, and the decrease in soil cover. The reduction 

of the most active fraction of soil organic matter affects several functions in the soil, 

being composed of humic substances of low molecular weight, as well as plant and 

animal residues and their primary decomposition products. 

 

1.3.4 Remote Sensing  

Remote sensing is a technique that is becoming prominent in agricultural 

activities, in plant evaluations through biophysical parameters, and in several other 

fields of science through ground, airborne or orbital sensors that act actively or 



7 

passively, recording the interaction of electromagnetic energy with the environment 

(DA SILVA JUNIOR et al., 2018; XU et al., 2020). The phenomena of absorption, 

reflection and transmission relies on the interaction of electromagnetic energy with the 

targets of interest (LIU, 2015). The ability of an object to absorb, reflect, and transmit 

electromagnetic radiation is referred to as absorptance, reflectance, and 

transmittance, respectively (XUE; SU, 2017). Transmission is done through the 

components that make up the leaf, when the radiation is reflected by the leaves (TSAI 

et al., 2019). Among the three phenomena, the energy reflected by objects is the most 

used in studies by most researchers using orbital and sub-orbital sensors (DENG et 

al., 2019). 

In photosynthesis, plants have higher absorbance in the wavelength 

corresponding to the blue and red bands and higher reflectance in the near infrared 

range (NIR) (GITELSON et al., 2021). Much of this absorbed radiation and made by 

the pigments of chlorophylls "a" (430 - 660 nm) and "b" (450 - 650 nm) existing in 

chloroplasts, with high absorption in the blue and red bands (GITELSON et al., 2006). 

One of the major factors of the low absorption and reflectance of incident energy is the 

pigmentation contained in the leaves. With the loss of chlorophyll from a plant indicates 

that it is in the senescence stage, thus having a reflectance in the red range 

(MERZLYAK et al., 1999; MURAOKA et al., 2010). 

In remote sensing, the vegetation is a complex target, since it presents leaf 

manifestations non-uniformity of structure inter and intra vegetation in the same plant, 

besides the soil spectral interference. The variability of the red and near infrared 

energy reflected by the vegetation that reaches the sensor will vary with solar 

irradiance, atmospheric conditions, substrate type, canopy structure and composition. 

The radiance is received by the optical components and then converted into electrical 

signals, then transformed into a digital numerical value associated with the pixels of 

the satellite images at the time of imaging. 

By using satellite images, we can perform analysis by means of vegetation 

indices (IR) to highlight the spectral behavior of biomass in relation to the soil and 

other targets of the earth's surface, in order to assess natural resources and monitor 

vegetation cover. Commonly, vegetations indices rely on the arithmetic operations 

combination by using two or more spectral bands, in order to provide stable information 

about the surface compared to reflectance measurements (DA SILVA et al., 2017). 

Studies use this combination of spectral bands to analyze biomass and carbon stock 



8 

over coffee plantation areas by using the Normalized Difference Vegetation Index 

(NDVI), the Photosynthetic Vegetation Index (PRI) and the CO2flux model, which is 

based on the NDVI and PRI by means of Landsat images (COLTRI et al., 2009). 

 The discrimination between deficient and adequate boron levels in 

Eucalyptus spp. juveniles can be based on the wavelengths of the MSI/Sentinel-2 

sensor system with the data from this sensor support the method of discrimination of 

adequate B nutrition in areas with planting of the species (DAMASCENO et al., 2023). 

In other studies used structural equation modeling (SEM) and factor analysis on corn 

varieties allowed to identify which agronomic variables and which NDVI, NDRE and 

SAVI indices are most associated with crop grain yield (SANTANA et al., 2022). 

Nowadays, there are several vegetation indices developed for different 

applications and research assessment abroad the science. This proves that the use 

of remote sensing techniques is capable of supplying research on spatial and temporal 

scales of vegetation, both locally and globally.  



9 

CHAPTER 2 – VARIATIONS OF CO2 EMISSIONS ON DIFFERENT VEGETATIONS 
COVERS IN THE BRAZILIAN AMAZON4 
 
 

ABSTRACT: Soil carbon dioxide (CO2) emission in agricultural areas is a 
process resulting from the interaction of several factors, such as climate, soil and 
management conditions. Agricultural practices directly affect the carbon dynamics 
between soil and atmosphere. Here, we evaluated the temporal variability (2020/2021 
crop season) of soil CO2 emission and its relationship with related variables as CO2 
flux model, Enhanced vegetation index (EVI), Gross Primary Productivity (GPP), Leaf 
Area Index (LAI) from orbital data and soil temperature, soil moisture and soil CO2 
emission in situ collection, over native forest, productive pasture, degraded pasture, 
high-yield potential soybean and low-yield potential soybean areas. Significant 
influence (p < 0.01) was observed for all variables and between the different land uses 
and land occupation. September and October presented lower soil CO2 emission, 
recording low means of soil moisture and soil temperature, and no differences were 
observed among treatments. On the other hand, there was a significant effect of CO2 
flux model in the areas of productive pasture, high-yield potential soybean and low-
yield potential soybean. The months with the highest CO2 flux model values, 
regardless the land use and land cover, were October and November, the beginning 
of the rainy season. There were positive correlations between soil CO2 emission and 
GPP (0.208), LAI (0.354), EVI (0.363) and soil moisture (0.280), and negative 
correlations between soil CO2 emission and soil temperature (-0.240) and CO2 flux 
model (-0.314). The land uses and land covers showed negative correlations between 
these variables, except for the CO2 flux model variable. Soil CO2 emission values were 
lower for high-yield potential soybean (averages from 0.834 to 6.835 μmol m-2 s-1) and 
low-yield potential soybean (from 0.943 to 5.686 μmol m-2 s-1) and higher for native 
forest (from 2.279 to 8.131 μmol m-2 s-1), whereas for CO2 flux model the opposite 
behavior was observed. 
 
Keywords: Land use and land cover; Sustainability; climate change; greenhouse 
gases   

 
4 This chapter corresponds to the scientific article submitted to the Environmental Research, Volume 
220, 01 March 2023, 8.431 Impact Factor. 



10 

2.1 INTRODUCTION 

Climate change is one of the main challenges for society and is probably the 

most widespread threat to humanity due to the earth's growing population and the 

resulting increased concentration of carbon dioxide (CO2) released into the 

atmosphere (NASROLLAHI et al., 2020; RIBEIRO; RYBSKI; KROPP, 2019). As the 

world's population increases, the need for large-scale food production arises, 

agricultural expansion and increased food production have been among the major 

impacts of humanity on the environment that include climate change, greenhouse gas 

(GHG) emissions (AWUCHI et al., 2020; REHMAN et al., 2021a, 2021b, 2022). Land 

use and land cover change for commodity production is one of Brazil's main sources 

of GHGs, making the country responsible for 2.8% of global emissions (DE AREA 

LEÃO PEREIRA et al., 2020). 

Soybean (Glycine max (L.) Merr.) is one of the most important commodities in 

the national and international market, being one of the most consumed and produced 

grains in the world (FAOSTAT, [s.d.]). In Brazil it has an expressive socioeconomic 

importance, thus becoming one of the main vehicles of development and the main 

agricultural product of the country (HIRAKURI; LAZZAROTTO, 2014). However, one 

of the main concerns about GHG's has been the conversion of natural vegetation for 

agriculture and cattle ranching (MCTI, 2014; SILVA; LIMA, 2018a; SILVA et al., 2018), 

that result in significant amounts of CO2 emissions to the atmosphere, either by 

burning the above-ground biomass or subsequent loss in soil carbon stocks (IPCC, 

2014). 

Soil CO2 emission in agricultural areas has proven to be a complex 

phenomenon, as it has great spatial and temporal variability due to its relationship with 

various soil properties (LA SCALA; BOLONHEZI; PEREIRA, 2006; SILVA et al., 2019, 

2020). For better estimates of soil respiration, it is first necessary to describe its 

temporal variability and the relationships with environmental variables that can be 

continuously monitored, such as temperature and humidity (SILVA et al., 2019; 

XAVIER et al., 2020), both with orbital sensors and in loco, which makes it possible to 

monitor other related parameters.  

In this context, several studies have elucidated the effects of agricultural 

practices and their respective productivity on the increasing atmospheric 

concentrations of GHGs, particularly CO2 (LA SCALA; BOLONHEZI; PEREIRA, 2006; 



11 

MOITINHO et al., 2015; SILVA et al., 2019). In the soil, CO2 is produced through 

biochemical reactions directly related to the biological activity of microorganisms and 

root respiration, which are mainly influenced by soil temperature and humidity (BOND-

LAMBERTY; THOMSON, 2010; OERTEL et al., 2016; PRIES et al., 2017). 

According to Intergovernmental Panel on Climate Change (2014), Inadequate 

agricultural land management practices are responsible for approximately a quarter of 

global GHG emissions, including deforestation, agricultural methane emissions, and 

nitrous oxide emissions in fertilized soils. In this context, soil carbon accumulation in 

agricultural systems is a promising strategy to mitigate the atmospheric increment of 

CO2 (DOSSOU-YOVO et al., 2016), thus reducing the greenhouse gas balance and 

the carbon footprint associated with agricultural production, as examples in meat and 

sugarcane production respectively (DE FIGUEIREDO et al., 2017). 

Nevertheless, environmental variables that tangent to this prospect demand 

comparison, where in loco data obtained with the soil respiration chamber can be 

related to carbon sequestration efficiency estimates related to vegetation in 

agricultural areas. Understanding the spatial and temporal variability of CO2 at 

different levels of data acquisition in farming lands becomes important for conducting 

controlled and sustainable crop management for soil carbon preservation, thus 

contributing to the mitigation of the extended greenhouse effect. 

For the generation of objective information in near-real-time, remote sensing is 

a tool that provides and contributes to the distinction and characterization of forests, 

agriculture, climate, among others (ALMEIDA et al., 2019; LIU et al., 2020b). On-board 

sensors from ground monitoring satellites obtain data that assist in understanding the 

global distribution of vegetation types, agricultural crops, as well as their biophysical 

and structural properties in spatio-temporal variations (DIDAN, 2015). For example, 

one of the products generated by Moderate-Resolution Imaging Spectroradiometer 

(MODIS) sensor are the vegetation indices (VIs), which perform spectral 

transformations of two or more bands and enable the visualization of vegetation 

properties and temporal inter-comparisons of terrestrial photosynthetic activity and 

canopy structural variations, enabling for example the monitoring of areas with 

soybean (DA SILVA et al., 2017). The VIs allow monitoring of seasonal, inter-annual 

and long-term variations in vegetation structure, phenological and biophysical 

parameters (HUETE et al., 2002), besides changes in climatic variables, such as 

precipitation (GOUVEIA et al., 2008). Furthermore, it is also possible to acquire and 



12 

estimate other variables through spectral bands, such as Gross Primary Productivity 

(GPP) (ROSSI; SANTOS, 2020) and carbon dioxide flux (CO2Flux) through 

photosynthesis and respiration (DELLA-SILVA et al., 2022; ROSSI et al., 2022; SILVA 

JUNIOR et al., 2019). 

In view of the biogeochemical carbon cycle and its great relation to climate, the 

observation of soil CO2 emission, moisture and soil temperature dynamics, as well as 

the relation of variables obtained by means of remote orbital sensing techniques 

related to the emission and/or absorption of atmospheric carbon in the biotic carbon 

pool are potentially associated to the type of land use and occupation. 

Thus, our objective was to verify the relation of soil carbon emission with 

remotely sensed variables of CO2 flux model (CO2Flux), GPP, vegetation indices and 

leaf area index, as well as in situ metrics of soil moisture and temperature, conditioned 

to different land uses and land covers of an agricultural are in northern Mato Grosso. 

 

2.2 MATERIAL AND METHODS 

2.2.1 Studied area 

The study was carried out in the 2020/2021 crop season at Aurora Farm, 

located in the municipality of Cláudia, Midwest region of Brazil (Figure 1) southern 

Amazon ecoregion, with a total area of 3.843,561 km² (IBGE, 2021) According to 

Köppen's classification, the region's climate is type Aw (tropical climate, with dry 

winter) (ALVARES et al., 2013), with a rainy period from October to April and a dry 

period from May to September. Annual rainfall is approximately 1970 mm, and 

average monthly temperatures range between 24 and 27 °C (SOUZA et al., 2013a). 

The soil of the area was classified as yellow latosol, according to the Brazilian Soil 

Classification System (SIBCS) (EMBRAPA_SOLOS, 2020; EMBRAPA, 2018). 

 



13 

 

Figure 1. Study area located in the State of Mato Grosso with the different land uses 
and occupation in the Amazon ecoregion: a) low-yield potential soybean (SB); b) 
native forest (FN); c) high-yield potential soybean (SA); d) productive pasture (PB) and 
e) degraded pasture (PD) in the municipality of Cláudia. 

 

Five areas with distinct uses were chosen according to availability and ease of 

access. The native forest is in an initial regeneration stage due to environmental 

intervention, such as clear-cutting, agricultural cultivation, or fire/burning that has 

occurred in the last 10 years. The pasture areas are in the process of changing to rice 

in the first crop grown on the site, and the later to soybean and corn crops. Previously, 

there was pasture in the area, and farming on the site began in 2015. 

 

2.2.2 Vegetation Indices 

Vegetation indices (VIs) have been developed to highlight the combinations of 

surface reflectance at two or more wavelengths designed to highlight a particular 

property of vegetation in order to exclude soil and other terrestrial targets. The 

numerical values of these indices can vary between -1 and 1, in which vegetation is 

commonly indicated by positive values, as there is greater reflectance in the near-

infrared wavelength than in the red (HUETE et al., 2002; SOLANO et al., 2010). Some 

targets, such as water and snow, exhibit higher reflectance at the red (600 – 750 nm) 



14 

wavelength region, as compared with the near-infrared (750 – 1100 nm) region. With 

the emergence of the Moderate-Resolution Imaging Spectroradiometer (MODIS) 

sensor, Huete et al. (1997) proposed an Enhanced Vegetation Index (EVI) (Equation 

1) that minimizes the effects of soil and atmosphere by considering the blue spectral 

band. 

 

𝐸𝑉𝐼 = 𝑔
𝜌𝑁𝐼𝑅 − 𝜌𝑅𝐸𝐷

𝜌𝑁𝐼𝑅 + (𝑐1 ∙ 𝜌𝑅𝐸𝐷) − (𝑐2 ∙ 𝜌𝐵𝐿𝑈𝐸) + 𝐿
 (1) 

 

where 𝜌𝑁𝐼𝑅, 𝜌𝑅𝐸𝐷, and 𝜌𝐵𝐿𝑈𝐸 are the reflectances in the near-infrared, red, and blue 

spectral ranges, respectively; g is the gain factor (2.5); c1 and c2 are the correction 

coefficients for red (6) and blue (7.5) atmospheric effects, respectively; and L is the 

correction factor for soil interference (1). The coefficients adopted in the EVI equation 

according to Huete et al. (1994, 1997). 

 

The Soil Adjusted Vegetation Index (SAVI) (Equation 2), takes into account the 

effects of exposed soil in the analyzed images (that is, when the surface is not 

completely covered by vegetation) to adjust Normalized Difference Vegetation Index 

(NDVI). 

 

𝑆𝐴𝑉𝐼 =
(1 + 𝐿𝑠)(ρ𝑁𝐼𝑅 − ρ𝑅𝐸𝐷)

(𝐿𝑠+ρ𝑁𝐼𝑅 + ρ𝑅𝐸𝐷)
 

(2) 

 

where Ls is a constant called the adjustment factor of SAVI and can assume values 

of 0.25–1 depending on the soil cover. According to Huete (1988), a value for Ls of 

0.25 is indicated for dense vegetation, of 0.5 for intermediate density, and of 1 for low 

density vegetation. When Ls is equal to 0, the assigned SAVI value is equal to the 

NDVI. Huete (1988) suggested an optimal value of Ls = 0.5 to account for first-order 

soil background variations. 

 

Leaf Area Index (LAI) (Equation 3) is a biophysical index defined by the ratio 

between the leaf area of a vegetation per unit area used by this vegetation, being an 

indicator of the biomass of each image pixel, as computed by the following empirical 

equation (ALLEN et al., 2002). Surface Energy Balance Algorithm for Land (SEBAL) 



15 

was designed to calculate the components that make up the energy balance locally 

and regionally, from limited surface data, by using empirical relationships and physical 

parametrization (BASTIAANSSEN et al., 1998a, 1998b; MARTINS; GALVANI, 2020). 

 

𝐿𝐴𝐼 =
ln (

0.69 − 𝑆𝐴𝑉𝐼
0.59

)

0.91
 

(3) 

 

The constants used in equation 3 are assigned according to (ALLEN et al., 

2002). 

 

2.2.3 Planet© Images  

Planet© is a California-based company that has manufactured and launched 

several nano-satellites such as "Doves" or PlanetScope that constitute the largest 

constellation of orbital imaging satellites on the earth. PlanetScope can image any part 

of the earth and provides data in four spectral bands between 454 and 860 nm with a 

3-meter spatial resolution. Atmospheric corrections of the PlanetScope images were 

carried out at the stage of preliminary processing and correction factors for each 

spectral band are present in the metadata of each scene. 

 Planet platform (https://www.planet.com/products/platform) is a fully 

automated cloud-based platform for downloading, processing, and managing images 

online. The images acquired by the platform were needed to run the CO2Flux, EVI, 

and LAI models, where NDVI is based on red and NIR, and PRI is based on blue and 

green bands.  

 

2.2.4 Gross primary production (GPP) 

The MOD and MYD17A2H products related to gross primary production is a 

cumulative composite of GPP (kg C m-2 d-8) values based on the concept of the solar 

radiation-use efficiency by vegetation (ε). By this logic, primary production is linearly 

related to absorbed photosynthetically active radiation (APAR) (RUNNING et al., 1999; 

RUNNING; ZHAO, 2019). APAR can be calculated as the product of the incident 

photosynthetically active radiation (PAR) in the visible spectral range 0.4 μm - 0.7 μm, 

assumed as 45% of the total incident solar radiation, and the fraction of 



16 

photosynthetically active radiation absorbed by the vegetation cover (FPAR) 

(HEINSCH, 2003). 

One of the major challenges in using such models is to obtain the light-use 

efficiency ε over a large area, due to its dependence on environmental factors and the 

vegetation itself. One solution is to relate ε according to its maximum value (εmax) plus 

the environmental contributions synthesized by the minimum air temperature 

(Tminscalar) and the water status of the vegetation (VPDscalar - water vapor pressure 

deficit) (FIELD; RANDERSON; MALMSTRÖM, 1995). 

In this study, MODIS GPP (Gross Primary Productivity), version 6 was used 

with the 8-day average. Pixel values referring to digital numbers from MODIS images 

were converted into biophysical values from kg (kilogram) to g (gram) by multiplying 

0.0001 and transformed to daily by dividing by 8 (8-day), at the end we have the 

transformed GPP measurement unit from kg C m-2 d-8 to g C m-2 day-1 (Equation 4). 

 

𝐺𝑃𝑃 =
𝐺𝑃𝑃 ∙ 0.0001

8
 

(4) 

 

2.2.5 CO2 Model Data 

The CO2Flux model was used (RAHMAN et al., 2001) to estimate the efficiency 

of the vegetation-related carbon sequestration process, which is the photosynthesis 

rate derived from photosynthetic processes. For this, the photochemical reflectance 

index (PRI) (Equation 5) was calculated (GAMON; SERRANO; SURFUS, 1997) using 

the green and blue spectral bands. The PRI is related to the amount of carotenoid 

pigments in leaves, which indicates the CO2 storage rate (BARNES et al., 2017; 

DELLA-SILVA et al., 2022; RAHMAN et al., 2001). 

 

𝑃𝑅𝐼 =
ρ𝐵𝐿𝑈𝐸 − ρ𝐺𝑅𝐸𝐸𝑁

ρ𝐵𝐿𝑈𝐸 + ρ𝐺𝑅𝐸𝐸𝑁
 (5) 

 

where ρ𝐵𝐿𝑈𝐸 is the blue spectral band reflectance and ρ𝐺𝑅𝐸𝐸𝑁 is the green spectral 

band reflectance. 

 



17 

Since our goal was to use PRI as an efficiency factor, we derived a scaled value 

of PRI (sPRI) to represent an adjusted range of 0 to 1. Therefore, it is necessary to 

generate the sPRI (Equation 6) (MARTINS; BAPTISTA, 2013). 

 

𝑠𝑃𝑅𝐼 =
𝑃𝑅𝐼 + 1

2
 

(6) 

 

Hence, the CO2Flux (µmol m-2 s-1) is the multiplied result of the NDVI and sPRI. 

There is a relationship where it may be possible to capture the carbon sequestration 

absorptions using the PRI index, which indicates the efficient use of light during 

photosynthesis, and the NDVI, which demonstrates the vigor of the photosynthetically 

active vegetation (RAHMAN et al., 2001). For more coherent estimates of CO2Flux in 

Brazil, the fitting of flux data associated with the CO2Flux model with direct 

measurement of CO2 fluxes by Turbulent Vortex Covariance (Eddy Covariance) 

(GOULDEN et al., 1996) was applied. Santos, (2017) determined a coefficient of R2 = 

0, 8185 and the regression model y = -66.207x + 13.63 between the direct 

measurement of CO2 fluxes and the CO2Flux model, the best correlation is given in 

Equation (7). 

 

𝐶𝑂2𝐹𝑙𝑢𝑥 = 13.63 − (66.207 ∙ (𝑁𝐷𝑉𝐼 ∙ 𝑠𝑃𝑅𝐼)) (7) 

  

In this model, the linear and angular coefficients are presented as NDVI and 

PRI vegetation models and should be fitted according to environmental characteristics. 

The CO2Flux model expressed in Equation (7) was fitted to Amazonian standards, 

where negative values (-) indicate net absorption of carbon (photosynthesis) by the 

surface and positive values (+) indicate loss of carbon to the atmosphere (respiration). 

 

2.2.6 Assessing soil CO2 emission, temperature, and moisture 

For the assessments of soil CO2 emission (FCO2), we used the LI-COR portable 

system, model LI-8100 (LI-COR Bioscience, Nebraska, USA) (Figure 2a), which 

monitors the CO2 concentration variations inside the soil respiration chamber (Figure 

2b) by means of optical absorption spectroscopy in the infrared spectral region. The 

soil respiration chamber is a closed system with an internal volume of 854.2 cm3 and 



18 

a circular contact area of 83.7 cm2 and is coupled over PVC collars that were 

previously inserted into the soil 24 hours before collection. 

 

 

Figure 2. LI-COR model LI-8100 portable system (a) connected to the soil chamber 
(b), soil moisture sensor (c) and soil temperature sensor (d). 
 

The on-site measurements were performed in a 100-meter transect with 20 

sample points for each land use and land cover (LULC) (Figure 3), with seven 

collections comprising before sowing, during the phenological stage, and after 

soybean harvest (Table 1), carried out in the morning from 8 am to 10 am in each 

LULC. The spatial distribution of the in situ collection points were distributed every five 

meters, so that each point could be represented in a different pixel. 

 



19 

 

Figure 3. Locations of the land uses and land cover of the experiment. a) native forest; 
b) productive pasture; c) degraded pasture; d) high-yield potential soybean; e) low-
yield potential soybean. 

 

Table 1. Collection dates for each land use and land cover and soybean soil 
preparation, sowing, and harvesting periods. 

Collection date Julian Day Period 

09/26/2020 270 Before soybean sowing 

10/08/2020 282 During the soybean phenological stage 

11/05/2020 310 During the soybean phenological stage 

11/26/2020 331 During the soybean phenological stage 

01/21/2021 21 During the soybean phenological stage 

03/06/2021 65 During the soybean phenological stage 

03/24/2021 83 After soybean harvesting 

 

The FCO2 was evaluated at each point by a fit of the CO2 air concentration 

inside the chamber as a function of a polynomial regression over time after its closure. 

The measurement mode for determining the soil CO2 emission took 90 seconds at 

each sampling point, and the CO2 concentration inside the chamber was determined 

every 2.5 seconds, approximately (SÁ et al., 2019).  

The Soil temperature (Ts) was monitored using a Digital Spit-Type 

Thermometer (DELLT DT-625) (Figure 2d). It consists of a 20 cm rod inserted into the 

soil at 5 cm from where the PVC collars were previously installed for the evaluation of 

soil CO2 emission. Similarly, the Soil Moisture (Ms) was recorded using a TDR – Time 

Domain Reflectometry (Hydrosense TM, Campbell Scientific, Australia) (Figure 2c), 

consisting of a probe with two 12-cm rods inserted inside the soil, perpendicular to the 

surface at 5 cm from the PVC collars. The soil moisture value is derived from the time 

it takes for an electric current to travel the distance of 32 mm from one rod to the other. 

The soil temperature and moisture evaluations were performed concurrently with the 

soil CO2 emission evaluations. 



20 

 

2.2.7 Statistical analysis 

Data were submitted to analysis of variance (ANOVA) for comparison of land 

use and land cover (LULC) at each evaluation date. Means were grouped by the Scott-

Knott test at 5% probability. Subsequently, to verify the interrelationship between the 

variables and LULC of each experiment, data were analyzed by principal component 

analysis (PCA), and a biplot was generated with the first two components due to the 

straightforward interpretation of these results. In this biplot, confidence ellipses were 

constructed for each LULC evaluated. Additionally, Pearson's correlation was applied 

to verify the relationship between the studied variables, and then a plot was generated 

to analyze the results. All analyses were performed on the R 3.6.3 (R CORE TEAM, 

2022) software using the packages "FactoMineR" (LÊ; JOSSE; HUSSON, 2008), 

"factoextra" (KASSAMBARA; MUNDT, 2020), "ExpDes.pt" (FERREIRA; 

CAVALCANTI; NOGUEIRA, 2021) and "GGally" (SCHLOERKE et al., 2021). 

 

2.3 RESULTS  

Analysis of variance (ANOVA) was the statistical test used to determine the 

difference between the means of land use and land cover and observed days. The 

analysis of variance revealed significance level (p < 0.01) for the variables atmospheric 

(FCO2 and CO2Flux), vegetative (LAI, EVI and GPP), and physical (Ms) variables. The 

physical variable Ts differed from the others by presenting significance at 5% 

probability level (p < 0.05). On the other hand, all variables analyzed showed the same 

significance level (p < 0.01) among the different LULC (FN, PB, PD, SA and SB) and 

for the evaluations from sowing to soybean harvesting, besides a significant interaction 

between these effects (LULC and Date) for the variables analyzed (Table 2). Factor 

analysis was used to describe the variability among the independent and possibly 

correlated variables. The analysis showed that land use and land cover (LULC) and 

time series dates showed a maximum significant value for all variables atmospheric, 

vegetative, and physical as described in the Table 2. Therefore, the pattern of temporal 

variation of the variables differs depending on the soil cover on the days evaluated 

according to the significant F-test, rejecting the hypothesis of H0, and applying the test 

for comparison of means. 

 



21 

Table 2. Analysis of variance for the seven variables among the five land uses and 
land cover (LULC) for the dates. 

Source of 

variationϮ 
DF 

Mean Square 

Ms Ts FCO2 CO2Flux LAI EVI GPP 

Date 6 1396.358** 529.818* 132.437** 645.805** 60.200** 1.584** 0.040686** 

LULC 4 59.345** 514.757** 44.044** 961.383** 71.224** 1.254** 0.005449** 

LULC x Date 24 50.474** 185.559** 25.581** 142.735** 12.318** 0.386** 0.002095** 

Residual 665 8.905 0.671 8.309 0.426 0.024 0.001 0.000002 

Mean  8.519 27.443 3.884 1.623 2.267 0.380 0.035 

 Ϯ Statistical significances of p ≤ 0.05 and p < 0.01 are represented by * and **, 
respectively. 

 

The mean soil moisture (Ms), regardless of the LULC and Julian days 

evaluated, was 8.52% (Table 2). The Ms presented significant differences throughout 

the days studied especially after November 5th, and the highest difference between 

the LULC is noted for March 6th. This differentiation was not observed for Julian days 

270 (September 26th), 282 (October 08th) and 83 (March 24th) (Figure 4a). For the 

different LULC among the Julian days (lowercase letters) with SA (high-yield potential 

soybean), the highest Ms values was observed in January and March (Julian days 21 

and 83) (Figure 4a), unlike SB (low-yield potential soybean) and PD (degraded 

pasture), which maintained higher Ms between November, January and March (Julian 

days 310, 21, 65, and 83). The highest Ms for FN (native forest) was recorded in March 

(Julian days 65 and 83), as well as for PB (productive pasture). 

 



22 

 

Figure 4. Grouping of means and the standard deviation of the variables a) Soil 
Moisture (Ms) and b) Soil Temperature (Ts) evaluated on different dates in the land 
uses and land cover (LULC) of native forest (FN), productive pasture (PB), degraded 
pasture (PD), high-yield soybean (SA) and low-yield soybean (SB). Means followed 
by uppercase letters do not differ in land use and land cover within each day and 
lowercase letters do not differ for the dates by Scott-Knott test at 5% probability. 

 

For the periods evaluated (near sowing, during profiling, and post-harvest), the 

mean Soil Temperature (Ts) obtained in the LULC was 27.44 °C. In Ts there were 

differences between the days observed for all types of LULC (Figure 4b). In 

September, October and March (Julian day 270, 282 and 83), the Ts were higher for 

the PD soil and lower for FN. SA obtained the highest records on Julian day 310 

(05/November) relative to FN, PB, PD and SB, and SB on 331 and 65 (26/November 

and 06/March) relative to FN, PB, PD and SA. For Julian day 21 (21/January), the 

lowest Ts record was observed for FN, while there was no difference among the other 

LULC. This means that Ts were high for the other environments.  

When evaluating the Ts of each LULC within the observed days, FN recorded 

higher Ts on Julian days 310 and 331 (November 05th and November 26th), unlike PB, 

which was on Julian days 270 and 282 (September 26th and October 08th). The PD, 

SA, and SB recorded only a single day of high Ts (Julian days 270, 310, and 331, 

respectively). 



23 

Figure 5 shows the grouping of Ms and Ts for the different days and LULC 

studied. The mean soil moisture was significantly higher for Julian days 65 and 83 

(Figure 5), with no difference between the LULC regardless of the days observed, but 

in the overall mean, the highest moisture was obtained for PB (Figure 5).  

The lowest mean Ts were recorded from Julian day 21 onwards, corresponding 

with the highest mean humidity recorded. On the other hand, the highest mean Ts 

(30.3 °C and 30.2 °C) were on Julian days 270 and 282, which is probably related to 

the end of the dry season and beginning of the rainy season, as well as the lowest 

mean Ms.  

The highest Ts was for PD (29.4 °C), followed by PB and SB (28.3 °C and 27.9 

°C, respectively). The FN soil had the lowest temperature (24.3 °C), allowing it to 

maintain good Ms. 

 

Figure 5. Grouping of means and the standard deviation of the variable Soil 
Temperature (Ts) and Soil Moisture (Ms) evaluated on different dates in the land uses 
and land cover (LULC) of native forest (FN), productive pasture (PB), degraded 
pasture (PD), high-yield soybean (SA) and low-yield soybean (SB). Means followed 
by uppercase letters do not differ in land use and land cover within each day and 
lowercase letters do not differ for the dates by Scott-Knott test at 5% probability. 

 

Regarding the variable FCO2, the mean value recorded during the time series 

considering all treatments was 3.88 µmol m-2 s-1 (Table 2). On Julian days 270 

(September 26th) and 310 (November 05th), FCO2 was highest for FN, on Julian days 

282 (October 08th) and 331 (November 26th) for FN, PB and PD had the highest values 

of FCO2 and lowest for SA and SB. PD, SA and SB treatments recorded the highest 

FCO2 on Julian day 21 (January 21st) and the lowest for FN and PB. As for Julian days 

65 (March 06th) and 83 (March 24th), FN and SA and PB and SA, respectively, 

corroborated for higher FCO2 emissions (Figure 4a).  



24 

When we analyze the LULC, the FN had the highest emissions recorded on 

Julian days 310 and 65, while for PB the highest emission days were recorded on four 

Julian days 310, 331, 65, and 83. In PD and SB, the highest emission was observed 

on Julian day 21, different from what was observed for SA, with the highest emissions 

on Julian days 21, 65 and 83 (Figure 4a).  

In general, total soil CO2 emission (FCO2) over time was higher in the FN area 

and in the SA and SB land uses, a lower amount than the other environments when 

compared to the overall means. However, when compared statistically, there is no 

significant difference among LULC for FCO2 emission under the conditions 

established in this study. 



25 

 



26 

Figure 6. Grouping of means and the standard deviation of the variables a) FCO2, b) 
EVI, c) GPP, d) CO2Flux and e) LAI evaluated on different dates in the land uses and 
land cover (LULC) of native forest (FN), productive pasture (PB), degraded pasture 
(PD), high-yield soybean (SA) and low-yield soybean (SB). Means followed by 
uppercase letters do not differ in land use and land cover within each day and 
lowercase letters do not differ for the dates by Scott-Knott test at 5% probability. 

 

The EVI data show a mean of 0.38, with the highest EVI recorded for FN (0.52). 

Subsequently, PB, PD, SA and SB showed an EVI of 0.40, 0.40, 0.31 and 0.28, 

respectively.  

Analyzing the LULCs for the time series, the EVI index for FN was higher from 

Julian days 270 to 331 compared to the other LULCs (Figure 6b). There was variation 

among the LULCs from Julian day 21, in which SA and SB obtained higher EVI in this 

period, while for PD, the EVI was higher on Julian days 65 and 83. When observed 

the days within the LULCs, FN, SA and SB had highest EVI on Julian day 21, contrary 

to PD (Julian days 65 and 83) and PB (Julian day 83). 

The mean GPP was 0.036, with the LULC averaging from 0.029 to 0.046 (SB 

and PD). In FN, PB and SB, the GPP was 0.033, 0.037 and 0.032, respectively. GPP 

values were higher for PD on five Julian days of the seven evaluated compared to FN, 

PB, SA and SB (Figure 6c). The other LULCs had higher averages on Julian day 282 

(PB) and 21 (FN, SA and SB). When relating each LULC to the observed days, it can 

be seen that Julian day 21 had the highest mean GPP for FN, PB, SA, and SB, while 

PD had the highest mean GPP on Julian day 331. 

The variability of CO2Flux for each LULC is presented in Figure 6d. The mean 

CO2Flux for all LULCs in the time series was 1.62 µmol m-2 s-1. SB had the highest 

CO2Flux storage on six of the seven Julian days, with the highest mean recorded on 

Julian day 310, corresponding to November. However, on Julian day 282, the SB did 

not differ from the SA, i.e., both showed equal means for CO2Flux. The PB has the 

highest CO2Flux storage only on Julian day 21. 

Observing the different LULC for the seven days, the FN and PB recorded 

CO2Flux in larger amounts on Julian day 331, different from PD, which showed higher 

means on Julian days 270 and 282. For SA and SB, the highest values were on Julian 

days 282 and 310, respectively. Overall, FN and PB were the LULCs with the lowest 

CO2Flux values, with the lowest concentration occurring in January.  

The results found through the LAI in the different LULCs show a variation from 

0.90 to 4.70, with an mean of 2.27. The FN stands out with the highest mean in the 



27 

different days observed (3.37) (Figure 6e). Only on Julian days 21 and 65, the means 

were higher for SA, SB and PD. When we analyze the LAI for each LULC, there is a 

significant difference between the days observed, with the highest values verified on 

days 21 (FN, SA and SB), 65 (PB and PD) and 83 (PB). 

Through PCA, we observed that the ordering of variables in each principal 

component (PC) of the axes was influenced by the degree of vegetation cover and 

moisture of land uses and land cover (Figure 7). Thus, PC1 contributed more to the 

variability of variables related to photosynthesis and LULC. Therefore, in PC2, the 

climatic factors affecting LULC influenced the variables humidity and temperature. 

The first two PCs accumulated at least 67.3% of the total data variance across 

all dates (Figure 7). The first principal component (PC1) accounted for at least 50.4% 

of the total variance, while the second principal component (PC2) accounted for at 

least 16.1% of the total variance across collection days. 

For Julian day 270 (September 26th), FN and PB were associated with the 

variables FCO2, LAI and EVI. The other LULC were close to the variables GPP, 

CO2Flux, Ms and ST with emphasis on soybean with high productive potential, which 

showed great variability among the points collected. The variable CO2Flux had the 

highest positive contribution, and the variables EVI and LAI had the highest negative 

contribution for PC1. On Julian day 282 (October 8th), FN and PB were associated 

with the variables EVI, GPP, FCO2 and Ms. The other LULC were close to the 

variables ST and CO2Flux with emphasis on native forest (FN) and high yield potential 

soybean (SA), which showed great variability among the points collected. CO2Flux 

had the highest positive contribution, and the EVI had the highest negative contribution 

for PC1. 

For Julian day 310 (November 5th), the FN, PB and PD were associated with 

the variables FCO2, LAI, EVI and GPP. The other LULC were close to the variables 

CO2Flux, Ms and ST with emphasis on the low-yield potential soybean (SB), which 

showed great variability between the points collected. The CO2Flux had the highest 

positive contribution, and the variables EVI and LAI the highest negative contribution 

for PC1. On Julian day 331 (November 26th), FN, PB and PD were associated with 

the variables FCO2, GPP, Ms, EVI and LAI. The other LULC were close to the 

variables CO2Flux and Ts with emphasis on SA, which showed high variability among 

the points collected. The EVI and LAI had the highest positive contribution, and the 

CO2Flux and Ts variables had the highest negative contribution for PC1. 



28 

For Julian day 21 (January 21st), the PB and PD were associated with the 

variables CO2Flux and Ms. The other LULC were close to the variables GPP, LAI, EVI, 

FCO2 and Ts with emphasis on the degraded pasture (PD), which showed high 

variability among the points collected. The CO2Flux had the highest positive 

contribution and the variables GPP, EVI and LAI the highest negative contribution for 

PC1. On Julian day 65 (March 6th), SA and SB were associated with the variables 

CO2Flux and Ts. The other LULC were close to the variables FCO2, GPP, Ms, EVI 

and LAI with emphasis on the native forest, which showed great variability among the 

points collected. The CO2Flux and Ts had the highest positive contribution and the 

variables EVI and LAI the highest negative contribution for PC1. On Julian day 83 

(March 24th), FN, PB and PD were associated with the variables LAI, EVI, GPP, Ts, 

and Ms. The other LULC were close to the variables CO2Flux and FCO2 with emphasis 

on SA, which showed high variability among the points collected. The variable EVI and 

LAI had the highest positive contribution and the CO2Flux the highest negative 

contribution for PC1. 

 

 

Figure 7. Principal component analysis for the variables FCO2, Ms, Ts, CO2Flux, EVI, 
LAI and GPP evaluated at different dates in the land uses and land cover (LULC) of 
native forest (FN), productive pasture (PB), degraded pasture (PD), high-yield 
soybean (SA) and low-yield soybean (SB). 

 

For Ms, there was low-magnitude correlations between the variables GPP 

(0.477), LAI (0.414), EVI (0.381), FCO2 (0.280), Ts (-0.444), and CO2Flux (-0.243) 

(Figure 8). The LULCs showed positive correlations between these variables, except 



29 

for Ts and CO2Flux variables. The boxplot shows that Ms values are clustered in the 

first and the second quartiles for all LULC, as shown in the histogram and density plot 

of the continuous variables. The correlation between Ts and GPP (-0.491), LAI (-

0.561), EVI (-0.533), Ms (-0.444) and FCO2 (-0.240) were negative, while for CO2Flux 

(0.459) were positive, but of low magnitude. LULC showed negative correlations 

between these variables, except for moisture and CO2Flux variables. The boxplot 

shows low Ts values for FN, while PD and PB have the highest values, which clustered 

in the third quartile, as shown in the histogram and density plot of the continuous 

variables. 

There were low magnitude correlations between FCO2 and GPP (0.208), LAI 

(0.354), EVI (0.363) Ms (0.280), Ts (-0.240), and CO2Flux (-0.314). The LULC showed 

negative correlations between these variables, except for the CO2Flux variable. From 

the boxplot, it can be seen that the FCO2 values were lowest for SA and SB and 

highest for FN. There were low magnitude correlations between CO2Flux and GPP (-

0.582), LAI (-0.857), EVI (-0.890), FCO2 (-0.314), Ms (-0.243), and Ts (0.459). From 

the boxplot, it can be seen that the CO2Flux values were lowest for FN and highest for 

SA and SB. 

There were positive correlations between EVI and GPP (0.715), LAI (0.979), 

FCO2 (0.363) and Ms (0.381) and negative correlations between EVI and CO2Flux (-

0.890) and Ts (-0.533), with high magnitude for GPP, LAI and CO2Flux. LULC showed 

positive correlations between these variables, except for Ts and CO2Flux. From the 

boxplot, it can be seen that EVI values were lower for SA and SB and higher for FN, 

as shown in the histogram and density plot of the continuous variables. 

The overall correlation between LAI and GPP was positive and significant 

(0.70). All LULC showed positive correlations between these variables. The boxplot 

shows that the LAI values were lower for SA and SB and higher for FN, as evidenced 

in the histogram and density plot of the continuous variables. 



30 

 

Figure 8. Pearson correlations for the variables FCO2, Ms, Ts, CO2Flux, EVI, LAI and GPP evaluated at different dates in the land 
uses and land cover (LULC) of native forest (FN), productive pasture (PB), degraded pasture (PD), high-yield soybean (SA) and low-
yield soybean (SB). 
 



31 

2.4 DISCUSSION 

Climate change worldwide is caused by anthropogenic actions such as land 

occupation (HE; ZHANG, 2022; HOUGHTON et al., 2012), ecosystem degradation 

(TCHOBSALA et al., 2021), urbanization and exploitation of natural resources 

(DANISH; ULUCAK; KHAN, 2020), which are enhanced by the growing world 

population (FRONTISTIS, 2021). This entire scenario goes through the carbon cycle 

dynamics, where solutions for reducing the concentrations of this element in the 

atmosphere are required. 

The mean values of Soil Moisture (Ms) and Soil Temperature (Ts) were higher 

and lower, respectively, in the native forest than in the other land use and land cover 

(Figure 8). This fact was already expected since the native forest provides a higher 

volume of vegetation cover. No difference was found for the Ms variable between the 

first Julian days 270 and 282 (September and October), as well as the last Julian day 

83 (March), which may be related to the previous dry period and the end of the rainy 

period. However, it can be seen a greater influence of the dry-rainy transition period 

on the degraded pasture area since the mean temperatures were higher than the other 

environments (Figure 8).  

 In our study, September and October (Julian days 270 and 282) had the lowest 

FCO2 emission, recording low average humidity and high soil temperatures. 

Therefore, we found that the variable soil CO2 emission (FCO2) showed no difference 

for the LULCs, but did show a difference between the days observed. Throughout the 

days, there was a tendency to increase the FCO2. This can be explained after the 

rainfall events during the observed days, increasing the FCO2 flux. Soil temperature 

and soil moisture are influencing factors in the CO2 (carbon dioxide) emission 

variability in soils and can be altered after a rainfall (PANOSSO et al., 2007). This 

occurs because the water drains into the soil and causes the FCO2 found in the pores 

to escape, making a barrier that prevents the FCO2 emission (ZANCHI et al., 2003).  

Taking into account that the higher the values obtained for the Leaf Area Index 

(LAI), the better the ability of the vegetation to develop by absorbing sunlight and, 

therefore, the higher the CO2 storage capacity, we highlight that the native forest (FN) 

was the environment with the highest LAI and, in a general way, the environment with 

the highest CO2 concentration, regardless of the days observed. Castro Pereira et al. 

(2020) found that areas with high LAI values are sites where much of the vegetation 



32 

is developing, absorbing a higher flux of energy radiation and a higher amount of 

carbon in the biomass.  

Thus, the CO2flux index (CO2 flux) allows the estimation of carbon absorbed by 

the vegetation (DELLA-SILVA et al., 2022). Vegetated areas have high CO2flux 

because the plants retain more carbon dioxide (TEOBALDO; BAPTISTA, 2016), 

allowing better development conditions for the vegetation. Thus, we verified in this 

study a difference in the CO2 flux values between PB (productive pasture), SA (high-

yield potential soybean) and SB (low-yield potential soybean). The months with the 

highest CO2flux values, regardless the LULC, were recorded in October and 

November (Julian days 282 and 310), a transition period from dry to rainy season.  

The interaction between land cover and atmosphere is weighted by the type of 

land cover, that is, its litter stock and organic matter. Principal component analysis 

allowed relating all LULCs and variables simultaneously (Figure 7). Overall, all LULCs 

during the time series were allocated to different quadrants, except native forest, which 

remained almost always in the same quadrant due to its high correlation with LAI and 

EVI. Similar results were reported by Jardim et al. (2022), who observed similarity in 

the occurrence of increased vegetative vigor in the forest canopy. According to these 

authors, such findings can be explained, in general, by climatic conditions, where the 

rainfall rate directly influences the vegetation cover in the area. 

The dynamics of the variables shown in Figure 7 are strongly influenced by the 

seasonality of rainfall and, hence by soil moisture conditions, which is a determining 

factor for the seasonality of leaf cover for each land use and land cover. Over the tim- 

series, we observed a marked positive trend in CO2Flux for the areas where the soil 

was exposed or in vegetation senescence. This positive trend is due to the 

vegetation's efficiency of the carbon sequestration process, as already evidenced in 

previous studies (DELLA-SILVA et al., 2022; ROSSI et al., 2022). Rainfall and soil 

moisture dynamics in the study area affect phenology, stomatal conductance 

seasonality, and photosynthesis (MARQUES et al., 2020; MENDES et al., 2017). It is 

essential to point out that photosynthesis decreases with the closing of the stomas due 

to the reduction of CO2 absorption in the dry season (BAI et al., 2017; PINHO-

PESSOA et al., 2018; ZHANG et al., 2014), which leads to leaf loss (senescence) 

under drought stress conditions. Besides the senescence of leaf tissue, there is a 

severe reduction of remaining leaves due to soil moisture deficits or soybean 



33 

defoliation, resulting in lower EVI and LAI values. These findings indicate different 

plant-environment interactions affecting the variables in the study areas. 

The variables Ms and Ts are important for the CO2 emission process during the 

time-series, directly influencing the decomposition of organic matter and the microbial 

and root activities (SILVA-OLAYA et al., 2013a). Ms and Ts are always on opposite 

axes of the PCA (Figure 7) and are, therefore, negatively correlated (Figure 8). On the 

other hand, no high-magnitude correlation was found between Ts and Ms with FCO2, 

CO2Flux, and the LULCs, possibly due to the low variation over the time-series period 

(MOITINHO et al., 2021). The effect of Ms may partially hide the effects of Ts on FCO2, 

as these are interdependent variables and commonly change simultaneously (DING 

et al., 2010). 

The contribution of soil FCO2 obtained during the experimental period will 

increase, due to the increased activity of the soil microbiota (CHEN et al., 2014; 

STRUECKER; DYCKMANS; JOERGENSEN, 2017) and priming effect (SHAHBAZ et 

al., 2018). The concentration of FCO2 values for the LULC are different. For the native 

forest, the high values come from the higher amount of roots (NICHOLS et al., 2016) 

and the intense rotation of soil organic matter (CHEN et al., 2016; CRAIG et al., 2022). 

The increased soil FCO2 occurs as a function of carbon inputs into the soil, either from 

roots of cover crops, autotrophic respiration (FERREIRA et al., 2018; HU et al., 2018; 

WENCHEN et al., 2017) or by the decomposition of residues deposited on the soil 

surface. Aerobic conditions favor the decomposition of soil organic carbon (JAIN et al., 

2016). 

 

2.5 CONCLUSIONS 

The link among land use and land cover to soil CO2 emissions were high in our 

assessment, and closely related to biotic pool of carbon. In descending order, in the 

native forest, degraded pasture, productive pasture, high-yield potential soybean, and 

low-yield potential soybean areas are listed in order of greatest FCO2 value, which 

means that these values may have been affected by the increased stock of soil organic 

matter and roots within the chamber. 

The positive values of CO2 flux model suggest carbon loss towards atmosphere 

due to plant respiration in the soybean areas and lower in the native forest area, even 

in a landscape with abrupt changes of phytophysiognomy, i.e., land use and land 



34 

cover. Thus, areas with large forest areas had lower concentrations of carbon dioxide, 

mitigating climate change. Furthermore, future works comparing the variables 

considered in our study for other landscapes may contribute to the carbon dynamics 

comprehension.  



35 

CHAPTER 3 – RELATION BETWEEN SOIL RESPIRATION AND MULTIPLE 
VARIABLES ON EUCALYPTUS SPECIES CROP IN SOUTHWESTERN 
BRAZILIAN CERRADO 
 

 
ABSTRACT – Vegetation indices (VIs) are spectral models that allow the 

measurement of chlorophyll content, crop health, predict dry biomass, allowing a priori 
decision making. An important factor in planted areas is the soil CO2 emission, also 
known as soil respiration, which is the result of the decay of labile organic matter and 
root respiration and is related to species, temperature, and soil moisture. This study 
aims to verify if soil carbon dioxide (CO2) flux (FCO2) and VIs obtained via airborne 
remote sensor over a eucalyptus species area (E. camaldulensis, E. uroplylla, E. 
saligna, E. grandis, E. urograndis and Corymbria citriodora). A multispectral imagery 
based the VIs estimation (green, red-edge and near infrared) from the Parrot Sequoia 
sensor, and on the same day FCO2, soil moisture and temperature were collected. 
The six species of eucalyptus were randomly arranged in blocks with four repetitions 
and each block repetition divided into four zones, totaling 96 samples that were 
compared by statistical metrics to the data estimated by the. Soil moisture has a 
negative and significant correlation compared to FCO2 (p < 0.05), proved to be one of 
the main factors that controls soil respiration. The vegetation indices NDRE 
(Normalized Difference Red-Edge Index), NDVI (Normalized Difference Vegetation 
Index), SAVI (Soil-Adjusted Vegetation Index) and MSAVI (Modified Soil Adjusted 
Vegetation Index) showed higher negative correlation to FCO2, for E. camaldulensis, 
E. saligna and E. uroplylla species. Multivariate Analysis of Variance showed 
significance (p < 0.01) for the species factor, which indicates there are differences 
when considering all variables simultaneously. Results achieved in this study allow us 
to differentiate FCO2 among eucalyptus speci