Merge pull request #187 from ronnyhdez/T186

RF suggestions

_quarto.yml

@@ -2,7 +2,7 @@ project:
   type: book
 
 book:
-  title: "Assesing uncertanties related to satellite remote sensing indices to estimate Gross Primary Production"
+  title: "Assesing uncertanties related to the use of satellite remote sensing indices to estimate Gross Primary Production"
   subtitle: "By"
   author: 
     - name: "Ronny A. Hernández Mora"

abstract.qmd

@@ -2,16 +2,17 @@
 
 Methods to quantify Gross Primary Production (GPP) are classified into two
 categories: Eddy Covariance techniques (EC) and satellite data-driven. EC
-techniques can measure carbon fluxes directly, albeit of spatial constraints.
+techniques can measure carbon fluxes directly, albeit with spatial constraints.
 Satellite data-driven methods are promising because they overcome spatial
-constraints associated with EC techniques. However, there are challenges
-associated with an increase in uncertainty when estimating GPP from
-satellite-driven products such as mixed pixels, cloud cover, and the ability of
-the sensor to retrieve vegetation under saturation conditions in high biomass
-environments. Therefore an effort to analyze and quantify the uncertainty of GPP
-products derived from satellite platforms is needed. Here we present how
-commonly used satellite vegetation indices (NDVI, EVI, CCI, and NIRv) can impact
-the uncertainty in the GPP estimation compared with direct methods such as EC
+constraints associated with EC techniques. However, satellite-driven products
+have potentially greater uncertainty than EC methods for GPP estimation such as
+mixed pixels, cloud cover, and the ability of the sensor to retrieve vegetation
+under saturation conditions in high biomass environments. Therefore, an effort
+to analyze and quantify the uncertainty of GPP products derived from satellite
+platforms is needed. This study quatifies the uncertainty of commonly used
+satellite vegetation indices such as Normalized Difference Vegetation Index (NDVI), Enhance
+Vegetation Index (EVI), Chlorophyll/Carotenoid Index (CCI), and Near-Infrared Reflectance Index (NIRv)
+for GPP estimation compared with direct methods such as EC
 measurements. We conduct this study on three different sites: University of
 Michigan Biological Station (USA), the Borden Forest Research Station flux-site
 (Canada) and Bartlett Experimental Forest (USA) using traditional regression

chapter_1_intro.qmd

@@ -3,48 +3,49 @@
 The study and assessment of ecosystem dynamics have greatly benefited from the
 development and advances of the remote sensing field [@running_continuous_2004; @baldocchi_how_2020-1]
 coupled with the growing freely available remote sensing 
-resources [@montero_standardized_2023]. Elements in the Earth surface exhibit
+data [@xiao_remote_2019]. Elements on the Earth surface exhibit
 distinctive spectral signatures arising from their physical properties and
-interactions with electromagnetic radiation. Environmental factors can affect
-these interactions modifying the spectral signatures which can provide insights
-about surface process when measured by remote
-sensing instruments [@montero_standardized_2023].
+interactions with electromagnetic radiation [@montero_standardized_2023].
+Environmental factors (fires, floods, drought, etc) can affect these
+interactions modifying the spectral signatures which can provide insights about 
+surface process when measured by remote sensing instruments [@montero_standardized_2023].
 
 The launch in the 1970's of the  Earth Resources Technology Satellite (ERTS),
-later known as Landsat-1 [@houborg_advances_2015], initiated a new era in the 
+later known as Landsat-1 [@donato_landsat_nodate], initiated a new era in the 
 remote sensing field of the terrestrial biosphere, with a special focus on 
 vegetation monitoring [@montero_standardized_2023]. With information available
 about spectral bands, studies about their relations to biophysical
 characteristics became an object of study to estimate biomass and photosynthetic
 activity [@myneni_estimation_1997]
 
 In fact, one application of remote sensing have been the development of
-vegetation indices to study the characteristics of vegetation canopies and
-functioning [@houborg_advances_2015] Vegetation indices can give us information 
-about the status of the vegetation given the information derived from mathematical
-equations from the values of two or more spectral bands, typically derived from
-satellite images. [@zeng_optical_2022]. Their interpretation is based on the
-theoretical background about the interaction of the light with vegetation
-[@zeng_optical_2022]
+vegetation indices (VI) to study the characteristics of vegetation canopies and
+functioning [@houborg_advances_2015]. VIs are a function of two or more 
+spectral bands, typically derived from satellite images [@zeng_optical_2022]. 
+that can contain information about the status of the vegetation, biochemical
+characteristics, structural characteristics such as Leaf Area Index 
+(LAI), among others [@huete_overview_2002].
 
 Early studies about spectral bands relations, lead to the development of the
-NDVI which showed how the NIR and Red ratio had advantages over other bands
-ratios for monitoring vegetation [@tucker1979red]. It rationale lays in the
-interplay of  the chlorophyll absorbing red spectral region with the
-non-absorbing leaf reflectance signal in the NIR region, that can give optical
-measurements of the vegetation greenness and other biophysical processes.
-[@ramachandran_modis_2010]
-
-One of the advantages of NDVI is that the ratio normalize values with large
-variances [@ramachandran_modis_2010]. Nonetheless, this
-VI have disadvantages such as sensitivy to soil background and saturation at
-high biomass [@huete_overview_2002]. To overcome
-this problem, the EVI was designed to improve vegetation characterizations such
+NDVI, which showed how the NIR and Red ratio had advantages over other bands
+ratios for monitoring vegetation biomass and structure given surface reflectance
+measurements in the shortwave spectrum [@tucker1979red; @richardson1977distinguishing].
+The basis of NDVI rests in the fact that chlorophyll absorption leads to a
+decrease in red reflectance with increasing green vegetation relative to the
+NIR reflectance, that is minimally impacted by chlrophyll absorption [@ramachandran_modis_2010].
+
+The use of a relative indicator serves to reduce the effect of factors such as 
+canopy spatial structure and non-photosynthetic vegetation on relationships 
+between NDVI and biophysical variables such as biomass or absorbed 
+photosynthetically active radiation variances [@sellers_canopy_1987; @ramachandran_modis_2010].
+Nonetheless, NDVI is sensitive to soil background and saturation at
+high biomass [@huete_overview_2002]. To overcome this problem, the Enhance
+Vegetation Index (EVI) was designed to improve vegetation characterizations such
 as canopy greenness without soil background sensitivity and aerosol variations
 by combining the blue, red, and NIR bands [@huete1988soil]. EVI is also less 
 prone to saturate at high biomass instances [@gao_opticalbiophysical_2000]
 
-The ability to combine spectral regions, coupled with advancements and  the
+The ability to combine reflectance from multiple spectral bands, coupled with advancements and  the
 development of new sensor technologies introducing additional bands or different
 spectral characteristics, has created opportunities to develop multiple VIs  and
 the total number of indices have grown steadily [@zeng_optical_2022]. These 
@@ -54,48 +55,55 @@ enhance existing estimations across different ecosystems [@ramachandran_modis_20
 According to the Awesome Spectral Indices open community catalogue (version
 4.0.0), there are 127 spectral indices designed to monitor vegetation. The aim
 behind the ongoing development of new VIs is to achieve a more accurate mapping
-of ecosystem processes **such as GPP**. [@montero_standardized_2023]. These 
+of ecosystem variables such as GPP. [@montero_standardized_2023]. These 
 indices can be classified into 3 main categories: biophysical, biochemical and 
 physiological properties [@zeng_optical_2022] **Describe the categories**
 
-One important metric to understand ecosystems processes is GPP. At the time, VIs
+One important variable to understand ecosystems processes is GPP. At present, VIs
 derived from satellites cannot directly estimate GPP [@gensheimer_convolutional_2022] 
 and instead, instruments like eddy covariance have been the traditional method 
 to estimate GPP more directly [@baldocchi_how_2020-1].
 
-GPP represents the overall quantity of leaf photosynthesis, playing a crucial
-role in the carbon cycle [@xiao_remote_2019]. GPP is responsive to a range
-of factors, including abiotic elements such as radiation, temperature, and
-precipitation [@beer_terrestrial_2010], as well as biotic factors like
+GPP represents the total amount of carbon compunds produced by plant photosynthesis
+in a given period of time [@ashton_managing_2012]. GPP plays a crucial
+role in the carbon cycle since it quantifies the ability of vegetation to fix 
+carbon from the atmosphere using only solar energy and nutrients [@xiao_remote_2019]. 
+GPP is responsive to a range of factors, including abiotic elements such as 
+radiation, temperature, and precipitation [@beer_terrestrial_2010], as well as biotic factors like
 vegetation type, leaf chemical traits, and species composition [@musavi_stand_2017].
 
 Some of the carbon taken in by GPP is release back to the atmosphere through
 plant respiration (autotrophic respiration), and the difference between GPP and
 plant respiration  is called Net Primary Productivity (NPP) [@lieth1975modeling; @xiao_remote_2019].
 GPP, along with Ecosystem Respiration (ER), which includes both autotrophic and
 heterotrophic respiration, together determines Net Ecosystem Exchange (NEE)
-[@xiao_remote_2019].
-
-EC is a method used to measure the exchange of energy and materials between
-ecosystems and the atmosphere within an area that can range from a few hundred
-meters to a few kilometers [@pabon-moreno_potential_2022]. Specifically, EC
-systems measure NEE, which has to be break down into its parts: GPP and
-respiration **for quantifycation** [@reichstein_separation_2005]. Despite being
-a more directly method to estimate GPP, EC systems have a spatial constrain that
-depends on the tower flux footprint [@baldocchi_how_2020-1].
+[@xiao_remote_2019]. NEE is fundamental to the planetary carbon cycle as it 
+represents the net accumulation or loss of carbon by the ecosystem. Measurement 
+of GPP and NEE are critical to understanding the role of the biosphere in the 
+carbon cycle as well as the status and trends of vegetation productivity [@schimel_terrestrial_1995]
+
+Eddy Covariance (EC) is a method used to measure the exchange of energy and 
+materials between ecosystems and the atmosphere within an area that can range 
+from a few hundred meters to a few kilometers [@pabon-moreno_potential_2022].
+Specifically, EC methods can estimate GPP in-situ at temporal resolution of 
+seconds, by subtracting estimates of respiration from chamber measurements or models
+from direct measurements of NEE [@reichstein_separation_2005]. However, EC 
+estimates are both, costly and limited in spatial coverage [@baldocchi_how_2020-1].
+Also, uncertainties from this method lies in the typical increase in spatial and temporal 
+variability in GPP over heteregeneous landscapes such as forest boundaries [@reinmann_edge_2017].
+The prevalent practice of taking measurements primarily from intact forests 
+introduces the risk of potentially misleading estimations for the broader forest 
+being quantified, given the potential differences in dynamics between edge and
+interior environments [@smith_piecing_2018].
 
 Since direct GPP measurements from satellites are not presently available, one
-of the primary objectives of remote sensing is to enhance GPP estimation
-accuracy without the spatial constriction [@xiao_remote_2019]. In this context,
-EC values prove invaluable as a comparative metric when contrasting VIs values
+of the primary objectives of remote sensing is to enhance the spatial coverage
+of EC GPP estimates [@xiao_remote_2019]. In this context,
+EC values prove invaluable as information for calibrating VIs values
 with GPP inferred from flux towers [@ramachandran_modis_2010]. The utilization
 of EC measurements bridges the gap in GPP assessment, providing a practical
 means to evaluate and refine the reliability of remotely sensed data against
-ground-based observations [@xiao_remote_2019]. This methodological approach
-aligns with the overarching goal of leveraging remote sensing techniques to
-advance our understanding  of the magnitude, spatial patterns, inter annual
-variability and long-term trends of ecosystem Carbon dynamics at landscape,
-regional and global scales [@xiao_remote_2019].
+ground-based observations [@xiao_remote_2019]. 
 
 Various methods exist for estimating GPP through satellite measurements.
 Satellite-derived VIs serve as commonly employed proxies for GPP, whereas
@@ -123,42 +131,25 @@ physiological processes governing vegetation productivity.
 However, both Light Use Efficiency (LUE) models and process-based models demand
 a multitude of inputs, some of which may not be readily available for all sites.
 In such cases, a data-driven approach becomes particularly advantageous, and VIs
-serve as effective proxies for GPP estimation. Notably, emerging VIs like NIRv
-demonstrate a capability to account for 68% of the monthly variability in GPP
-across fluxnet sites [@badgley_terrestrial_2019]. Additionally, indices like
-CCI, a pigment-based index designed to capture carotenoid/chlorophyll ratios
+serve as effective proxies for GPP estimation. Notably, emerging VIs like 
+the Near-Infrared Reflectance Index (NIRv) demonstrate a capability to account 
+for 68% of the monthly variability in GPP across 105 fluxnet sites [@badgley_terrestrial_2019]. 
+Additionally, indices like CCI, a pigment-based index designed to capture carotenoid/chlorophyll ratios
 during seasonal photosynthetic activity in evergreen leaves [@gamon_remotely_2016],
 have demonstrated proficiency as estimators of GPP in certain forest types. These
 indices offer valuable insights into the dynamics of vegetation productivity.
 
 Despite the notable progress in VIs development and their user-friendly
-applications, their effective use demands a experience and a strong theoretical
-background to navigate potential pitfalls and ensure accurate and meaningful
-interpretations of the data [@zeng_optical_2022]. There is a consensus that no
-single VI can be universally applied to address every problem or situation.
-Thus, a meticulous understanding of the specific context and characteristics of
-the site is essential to derive more accurate and insightful information from
-the data [@zeng_optical_2022].
-
-Uncertainty poses additional challenges to remote sensing applications, with
+applications there is a consensus that no single VI can be universally applied 
+to address every GPP estimation across 
+species and time periods, even for a single biome [@zeng_optical_2022].
+VI measurement error poses additional challenges to remote sensing applications, with
 diverse sources affecting satellite-derived data, including atmospheric effects,
 retrieval errors, cloud contamination, and sensor degradation
-[@van_leeuwen_multi-sensor_2006; @fang_theoretical_2012]. Another contributing
-factor to uncertainties lies in the variations of carbon uptake and storage,
-which can be higher in forest edges. [@reinmann_edge_2017] The prevalent
-practice of taking measurements primarily from intact forests introduces the
-risk of potentially misleading estimations for the broader forest being
-quantified, given the potential differences in dynamics between edge and
-interior environments [@smith_piecing_2018].
-
-Moreover, while emerging indices show promise as better proxies for
-physiological processes, ongoing technological developments have limitations in
-terms of spatial/temporal resolution which can still be not the best to conduct
-more studies that can be applied easily to more ecosystems [@zeng_optical_2022].
-One notable example is SIF, which holds the advantage of being directly related
-to photosynthetic activity [@magney_mechanistic_2019]. In contrast, Vegetation
-Indices (VIs) are primarily associated with photosynthetic capacity
-[@sellers_canopy_1985]
+[@van_leeuwen_multi-sensor_2006; @fang_theoretical_2012]. Thus, a quantitative 
+and rigorous understanding of the specific context and 
+characteristics of multiple sites is essential to understand the potential and 
+limits of VIs for GPP estimation over large extents of space and time. [@zeng_optical_2022].
 
 Together with the advancement of new VIs, there has been a upsurge in data
 analysis methods, specifically Machine Learning (ML) approaches, has garnered
@@ -171,10 +162,14 @@ sites play a pivotal role in influencing the accuracy and variability of
 regional flux estimates derived from machine learning methods
 [@papale_effect_2015].
 
-In this study, I assess the advantages and contraints associated with the
-utilization of 4 widely employed VIs for estimating GPP. The study specifically
-focuses on VIs derived from the MODIS, which have long standing temporal records
-and acceptable spatial resolution to study extended forest areas. The analysis
+The goal of this study is to assess the advantages and contraints associated with the
+utilization of 4 widely employed VIs for estimating GPP. The study is limited 
+to temperate broadleaf forests of Eastern United States of America and Canada 
+as it required consistent multi-annual EC GPP measurements over multiple sites 
+to quantify uncertainty related to spatial location of calibration data. The 
+VIs are derived from the Moderate Resolution Imaging Spectroradiometer (MODIS),
+which have long standing temporal records and acceptable spatial resolution to 
+study extended forest areas. The analysis
 employs both traditional regression methods and ML techniques to evaluate the
 performance of these indices in estimating GPP. This methodological approach
 seeks to provide a nuanced understanding of the efficacy and limitations of the
@@ -186,4 +181,4 @@ particularly in the context of temperate broadleaf forests.
 ## References
 
 ::: {#refs}
-:::
+:::