Phenology and Seasonal Ecosystem Productivity in an Amazonian Floodplain Forest
, Ricardo Dalagnol 2, Yadvinder Malhi 3, Sami W. Rifai 3, Gabriel B. Costa 4, Thiago S. F. Silva 5, Humberto R. Da Rocha 6, Iane B. Tavares 1 and Laura S. Borma 1Round 1
Reviewer 1 Report
The manuscript with the title “Phenology and seasonal ecosystem productivity in an Amazonian floodplain forest” presents a study that evaluates the seasonality of ecosystem productivity from eddy covariance measurements, environmental drivers and phenological patterns obtained from the field and satellite measurements in an Amazonian floodplain forest. The paper takes in an interesting and current topic, within the context of climate change, is well organized, and the structure is strong. I believe that the manuscript would make a nice contribution to the Remote Sensing Journal.
I just have minor corrections/notes
Line 154-155 Equation (2)
Should be PAR instead of RFA? According to Cabral et al. 2011
Personal comment: Did you consider using Landsat data (30m resolution) to derive EVI and GPP remote sensing products instead of MODIS?
Author Response
Response: First of all, thank you very much for your review. Regarding your question, we have used MODIS data due to the high revisit of the sensor (from 1 to 2 days), which makes it possible to create composition images every 16 to 30 days, especially over tropical forests that are often cloud covered. Therefore, it allowed us to verify seasonal and inter-annual phenology variations, while it is difficult to obtain feasible images from TM, ETM+ and OLI sensors (Landsat), due to the sensor revisit time of 16 days, providing few scenes without cloud cover. Moreover, our data were retrieved considering only cloud-free and low atmospheric turbidity, following the MAIAC quality flags. We added the following description about the number of samples for our MODIS data in the 2.4 methods section (Lines 183-185): “For the MODIS-MAIAC pixel containing the flux tower, the mean number of samples per 16-day composite was 4.6 samples for the flooding period (Feb-Mai) and 12 samples for the non-flooding period.”
Author Response File: 
Author Response.docx
Reviewer 2 Report
General comments:
This manuscript assessed the association of the seasonality of GPP, environmental drivers and phenological patterns in an Amazonian floodplain forest. The authors found that GPP is limited by soil moisture in two different ways, and EVI was positively associated with litterfall and with GPP. They further verified that EVI was able to describe the inter-annual variations on forest responses to environmental drivers. These results are original and very interesting. However, the results are not clearly presented, especially in Section 3.1.
Specific comments:
Line 226, Figure 2d? It could be Figure 2e.
Line 226-230, I can not observe the results as you described.
Figure 2 and 7, I suggest to use April as the start and March as the end for the abscissa.
Author Response
This manuscript assessed the association of the seasonality of GPP, environmental drivers and phenological patterns in an Amazonian floodplain forest. The authors found that GPP is limited by soil moisture in two different ways, and EVI was positively associated with litterfall and with GPP. They further verified that EVI was able to describe the inter-annual variations on forest responses to environmental drivers. These results are original and very interesting. However, the results are not clearly presented, especially in Section 3.1.
Response: We thank the reviewer for the comments. We realized that our results were not clearly presented, and we have improved our methods description section (showing data use/availability), the consistency of our analysis and results. We further inserted more information about the MODIS MAIAC data quality (Lines 176-185) and the percentage of forest cover in the MODIS pixels, delineating permanent water channels using high-resolution images (Lines 200-203).
We included field measurements of a forest inventory performed in 2015 reported by Tavares et al. (2019) (Table S2), to support our assumptions that the forest has a similar canopy structure to that of the LiDAR transect (Lines 359-361). Therefore, we could argue that the percentage of gaps found in the transect is also representative of the tower footprint.
We also improved our figures regarding changes in our dataset, such as the CWD recalculated with ET measurements from the tower (Figure 2, 4 and 7) and the water channels insertion in figure 3 (Line 310-311). We have changed the x and y variables in figure 4, allocating the independent variable in the abscissa, which did not change our correlation values. We also change the x and y variables in the phenology scatterplot (Figure 5, line 347) to adjust the independent variable. In figure 5, we also changed the name of the variable to represent better our assumptions (litterfall = leaf litter mass), which also did not compromise our correlation, because our values were representing the leaf litter mass.
Tavares, I. B.; Borma, L. S.; Fonseca, L. D. M.; Collicchio, E.; Domingues, T. F.; Rocha, H. R. The growth pattern of the forest located in a southeast Amazonian floodplain during the 2015/2016 ENSO year. Ecohydrology. (under review). 2019.
Specific comments:
Line 226, Figure 2d? It could be Figure 2e.
Response: Corrected
Line 226-230, I can not observe the results as you described.
Response: Indeed, there was no bimodal pattern on air temperature seasonality. We excluded this comment from our results and improved our analyses, as described in section 3.1. We rewrite the entire section to make it clear and better describe the results presented in figure 2. We added two new paragraphs to introduce our analyses through the following sentences (Lines 266-278): “The seasonal variation of GPP and its interaction with environmental drivers (rainfall, evapotranspiration, cumulative water deficit, vapour pressure deficit, soil moisture, net radiation and temperature) and phenology (described through EVI), showed different patterns among periods of the year. We observed an extended dry season from May to September, when rainfall was below 100 mm.month-1, while ET remained high through all seasons (mean of 110 mm.month-1), especially during the flooding (Figure 2a). This high ET values indicated that soil moisture was sufficient to attend the atmospheric demands, which were described by the increased VPD (Figure 2c) and decreased CWD, from April to August (Figure 2b).”
“Although ET at this site exhibited a flat pattern, slight maximum values coincided with maximum GPP (Figure 2f), but the first and more pronounced peak occurred in March (123 mm), mid-flooding, in response to an increase of Rn at this month (Figure 2e). The second peak occurred in June / July (116 mm), mid-dry season, and a third in December, mid rainy season (122 mm), before the period of flooding.”
Figure 2 and 7, I suggest to use April as the start and March as the end for the abscissa.
Response: We thank the reviewer for this suggestion, however, after organizing our results, we considered that the way they were presented starting in January and ending in December could benefit the visualization of the bimodal GPP and EVI pattern. It also provided better visualization of the flooded period, because in average, the flood height starts to increase in January (<= 0.5 m) reaching the peak in March (1.3 m) (Figure 2d). Moreover, it draws attention to the extended dry season presented in the middle of the figure (May to September). Although it is a flooded forest, we found that the dry season also limits productivity and drives the canopy phenology, highlighting its importance.
Author Response File: Author Response.docx
Reviewer 3 Report
General comments
The topic of the manuscript is very interesting, it analyzes peculiar flooded forests of Amazon basin by integrating field and satellite data.
The manuscript is well written and addresses the topic of forest productivity (GPP) by analyzing different influence parameters from different points of view.
Unfortunately, by in depth analyzing the study, there are many points that are confused (e.g. data availability) and not sufficiently reliable to support the conclusions. The relationship analysis of satellite vegetation index (EVI MODIS MAIAC) with in situ phenological patterns is based only on seven MODIS pixels in which the water signal presence is not completely excluded.
Moreover, for the peculiar atmospheric conditions of the Amazon basin, it is required an accurate check of MODIS data quality flags, in particular during the wet season, to avoid conclusions based on correlations with filtered/modelled data.
Therefore, I suggest a careful revision of the manuscript to make it publishable in Remote Sensing.
Specific comments
- Figure 1: Please highlight with a box the Cantão State park into the image B (biomes sub-figure).
- Lines 125-204 (data sections): since there are many parameters measured from different sources (flux tower, satellite, and in-field) with different characteristics and acquisition periods, and used directly, or for deriving other parameters, it would be useful to add a table summarizing info per parameter: input data/acquisition, start-end years, usage in the study (direct or for computing other parameters). As an example:
Parameter | Acquisitions | Start year | End year | Usage |
GPP | Tower | 2011 | 2013 | Correlation variable |
CWD | Tower/Satellite | 2004 | 2016 | Correlation variable |
PAR | Tower | NEE computation |
- Lines 133-134: Specific humidity and pressure are input parameter for VPD, but their acquisition is not specified in the data.
- Line 149: The filling procedure for CO2 fluxes is based on the average of valid values within 5-day window. The extension of such a window up to 31 days in case of absence of valid values seems to be too large to obtain reliable values for daily and monthly correlations.
Moreover, are you able to provide the number of filled CO2 data?
- Formula 2: Please verify RFA acronym (declared as PAR).
- Lines 166-176: As the authors presented satellite vegetation indices, It is not clear if EVI data are computed by the authors from MODIS MAIAC reflectances, or they used the ready EVI dataset for South America (MODIS EVI-MAIAC) by the cited Dalagon et al. (2019), and thus, the comments in these lines simply report the Dalagon et al procedure. Please specify the actual input data.
In both the cases, the authors have to check the data quality flags to verify the number of reliable acquisitions within the 16-days composite. This is particularly relevant for the rainy/wet period where the EVI values could be temporally filtered from contiguous (time and space) cloud and haze free acquisitions, and therefore these values are not suitable for 1:1 correlation with tower measurements.
Moreover, why did not you elaborate MAIAC data at 500m spatial resolution?
- Lines 177-181:
The GFC mask was used by the author to identify areas with permanent water. The reliability of a map implemented for global scale requirements is not enough for an application at very local scale as the current study. As visible in Sentinel 2 images containing SWIR data (as in Figure 1), there are large water areas around the tower, particularly in the MODIS pixels North of the tower. To obtain a more detailed map for the investigated MODIS pixels, I suggest the use SWIR Sentinel data combined with high resolution satellite image, such as Bing or Google satellite, (swir data are needed since thy are acquired during the dry period) to digitalize water surfaces. Free data services for GIS environment are:
Bing VirtualEarth http://ecn.t3.tiles.virtualearth.net/tiles/a{q}.jpeg?g=1
Google Satellitehttps://mt1.google.com/vt/lyrs=s&x=%7Bx%7D&y=%7By%7D&z=%7Bz%7D
Alternatively, the authors can use the very high resolution (~3m resolution) Planet data set (https://www.planet.com/products/planet-imagery/) free for researchers.
See in this image sequence flooded areas in February 2016 and March 2017.
Apart the permanent waters, as the authors stated, the characteristic of the study area is just the seasonality of flood events. Therefore, on the basis of rainfall entities and the morphological structure of the site, the flooded areas change among the seasons and from one year to another.
It is not clear how the authors use the measured flood heights. Are the flooded areas evaluated monthly using SRTM data?
- Lines 189-193: For computing the Climatological Water Deficit (CWD), why do you use a fixed evapotranspiration ET value? As shown in Figure 2a, you have ET monthly data available from the tower. Such an approximation only show the rain variability, whose anomalies can be directly computed form TRMM data. Being ET influenced also by temperature dynamics, to avoid over/under estimation periods in CWD, I suggest to recalculate it with the correct monthly ET.
- Lines 213-214: What do you mean with “ The annual cycle is referred to as “inter-annual””? This means that correlations (e.g., in section 3.2) are based on the whole time series and not on the mean annual cycle (e.g., correlation EVI vs GPP for the period 2004-2015 is based on 120 monthly data). If it is right, as the correlation periods strongly differ among the variables, please add the number of correlation points jointly with rho/r and p into the results section.
Moreover, clarify if the correlation of spaceborne EVI is based on the mean of the seven MODIS pixels around the tower or only on the monthly mean of the pixel in which the tower is located.
- Lines 226-230 : Comments on the bimodal temporal pattern of temperature seem to be too hasty. In Figure 2d, the total annual amplitude Tmax-Tmin seems to be about 2°C (annual range ~26-28°C), and the differences between May, June, and July are very subtle and surely in the measurement error of Ta. Thus, Ta in the period May-July is quite constant and the sentence “We observed a bimodal pattern of temperature .., with a pronounced decline in February and a secondary minimum occurring in July” have to be rewrite. Consequently, comments on Rn relationship have to be rearranged.
- Figure 2 Caption: As the mean values of the reported annual cycles represent different periods depending on the variables, please add for each variable the relative years in brackets.
- Line 274: (caption Figure 3) Please specify that the level 181 m refers to the height at the base of the tower and are meters a.s.l.
- Lines 312-320: Since the study is based on only 7 MODIS pixels, it makes little sense to evaluate the canopy structure in a completely different area. Moreover, the evaluated LIDAR strip (500m width) represents half of a MODIS pixel.
To evaluate the influence of water signal in analyzed MODIS pixels, the authors can use satellite SWIR data at higher spatial resolution (e.g. from Landsat series 30m, Sentinel2 20m) for a clear day in the flooding period.
Moreover, another issue of the adopted approach, which correlate MODIS pixels on the LIDAR strip with the pixel of the tower, is represented by the values of correlations. Along the LIDAR transect, correlation seems to vary between 0.6-0.75. Even if the equations are not shown, such values suggest that EVI differences between these pixels and the tower pixel are higher than the annual amplitude ~ 0.06 (Figure 3). The authors can estimate the correlation by separating the dry and wet period. (By considering the water surface structure in the Planet and Bing imagery), it is expected that the correlation is higher during the dry period.
Supplementary material
Table S1: specify in the caption mean annual data to which years it refers to.
Table S2: Data for 2013 are missing.
Author Response
We thank the reviewer for the comments. They contributed a lot to improve the manuscript, clarifying the data use/availability and the consistency of our analysis and results. We addressed the majority of the reviewer concerns during our revision. The content we put here is the same as the attached document, but please see the attachment to visualize the figures we plot.
General comments
The topic of the manuscript is very interesting, it analyzes peculiar flooded forests of Amazon basin by integrating field and satellite data. The manuscript is well written and addresses the topic of forest productivity (GPP) by analyzing different influence parameters from different points of view.
Unfortunately, by in depth analyzing the study, there are many points that are confused (e.g. data availability) and not sufficiently reliable to support the conclusions. The relationship analysis of satellite vegetation index (EVI MODIS MAIAC) with in situ phenological patterns is based only on seven MODIS pixels in which the water signal presence is not completely excluded.
Moreover, for the peculiar atmospheric conditions of the Amazon basin, it is required an accurate check of MODIS data quality flags, in particular during the wet season, to avoid conclusions based on correlations with filtered/modelled data.
Therefore, I suggest a careful revision of the manuscript to make it publishable in Remote Sensing.
Response: We thank the reviewer for the comments. They contributed a lot to improve the manuscript, clarifying the data use/availability and the consistency of our analysis and results. We addressed the majority of the reviewer concerns during our revision.
We have improved the methods text to better clarify the periods for each dataset measurements, how the data were obtained, filtered and analyzed. To address the concern regarding MODIS data availability and quality flags, the composites were retrieved considering only cloud-free and low atmospheric turbidity, following the MAIAC quality flags. Now, we are showing statistics of the number of samples within the 16-day MODIS-MAIAC composites for the flooding and non-flooding period. The pixels used in the experiment to test the relationship between tower-GPP and satellite-EVI were defined considering the limits of the Flux Tower footprint and filtering of forest cover areas (>90%). This ensures our analysis is consistent, and note that we have not used any interpolated value.
To further address the concern regarding the influence of water signal in the analyzed MODIS pixels, we obtained the Landsat-8 surface reflectance time series data for the same area of the MODIS pixels within the tower footprint. We masked the water channels and compared the Landsat-8 EVI signal with and without the mask. We observed a similar signal (R² = 0.97) between them (we better describe this in a further comment). This means that exposed water channels did not influence on MODIS EVI signal.
We also improved our figures regarding changes in our dataset, such as the CWD recalculated with ET measurements from the tower (Figure 2, 4 and 7) and the water channels insertion in figure 3 (Line 310-311). We have changed the x and y variables in figure 4, allocating the independent variable in the abscissa, which did not change our correlation values. We also change the x and y variables in the phenology scatterplot (Figure 5, line 347) to adjust the independent variable. In figure 5, we also changed the name of the variable to better represent our assumptions (litterfall = leaf litter mass), which also did not compromise our correlation, because our values were actually representing the leaf litter mass.
Specific comments
Figure 1: Please highlight with a box the Cantão State park into the image B (biomes sub-figure).
Response: Corrected
Lines 125-204 (data sections): since there are many parameters measured from different sources (flux tower, satellite, and in-field) with different characteristics and acquisition periods, and used directly, or for deriving other parameters, it would be useful to add a table summarizing info per parameter: input data/acquisition, start-end years, usage in the study (direct or for computing other parameters). As an example:
Parameter | Acquisitions | Start year | End year | Usage |
GPP | Tower | 2011 | 2013 | Correlation variable |
CWD | Tower/Satellite | 2004 | 2016 | Correlation variable |
PAR | Tower | NEE computation |
Response: Thank you for this great suggestion. We added this table below in the main manuscript at the beginning of the statistical analysis, section 2.5 (Lines 247-248):
Table 1. Availability and usage description of tower, satellite and field data.
Parameter | Acquisitions | Start year | End year | Usage |
LE | Tower | 2004 | 2014 | ET computation |
PAR | Tower | 2011 | 2013 | NEE computation |
Press | Tower | 2004 | 2014 | RH computation |
q | Tower | 2004 | 2016 | RH and VPD computation |
Rn | Tower | 2004 | 2014 | Correlation variable |
GPP | Tower | 2011 | 2013 | Productivity estimate / Correlation variable |
ET | Tower | 2004 | 2014 | Correlation variable |
VPD | Tower/Satellite | 2004 | 2016 | Correlation variable |
Tar | Tower/Satellite | 2004 | 2016 | VPD and ET computation / Correlation variable |
Rainfall/TRMM | Tower/Satellite | 2004 | 2014 | Correlation variable |
CWD | Tower/Satellite | 2004 | 2016 | Correlation variable |
EVI | Satellite | 2004 | 2016 | Phenology and productivity proxy / Correlation variable |
SM | Field | 2014 | 2016 | Correlation variable |
Litterfall | Field | 2004 | 2005 | Phenology proxy / Correlation variable |
Flood height | Tower | 2004 | 2016 | Define seasonal flooding |
Lines 133-134: Specific humidity and pressure are input parameter for VPD, but their acquisition is not specified in the data.
Response: Specific humidity and pressure were obtained through the eddy covariance system composed of a three-axis sonic anemometer (CSAT3, CSI) and an open-path infrared gas analyzer (Li-Cor 7500, Li-Cor, Lincoln, Nebraska, USA) connected to a second datalogger (CR-5000, CSI). It also measured the wind speed velocity components, virtual air temperature and concentrations of water and carbon dioxide at a rate of10 Hz (Borma et al., 2009). We added these parameters in the 2.2 methods section (Lines 133-134).
Borma, L.S.; Da Rocha, H.R.; Cabral, O.M.; Von Randow, C.; Collicchio, E.; Kurzatkowski, D.; Brugger, P.J.; Freitas, H.; Tannus, R.; Oliveira, L.; et al. Atmosphere and hydrological controls of the evapotranspiration over a floodplain forest in the Bananal Island region, Amazonia. J. Geophys. Res. Biogeosciences 2009, 114, 1–12.
Line 149: The filling procedure for CO2 fluxes is based on the average of valid values within 5-day window. The extension of such a window up to 31 days in case of absence of valid values seems to be too large to obtain reliable values for daily and monthly correlations.
Moreover, are you able to provide the number of filled CO2 data?
Response: The problem of low fluxes or missing data on carbon balance uncertainty is widely discussed in researches that involve eddy-covariance technique (Aubinet, 2008; Hutyra et al., 2007, Sakai et al., 2004. , Goulden et al., 2004). The 31 days window is the most used night gap filling technique, using valid data for windows that do not exceed 31 days of analysis around the day to be filled (15 days before, the day in question and 15 days later), as reported in Restrepo-coupe et al. (2013) and Hutyra et al. (2008). The correction performed in this paper involved 35% of the total night data for a friction velocity (u*) threshold of 0.19 m s-1 for the rainy season and 0.17 m s-1 for the dry season. Our data can be considered below the thresholds and percentage of gaps found in the literature for other sites in the Amazon region (HAYEK et al., 2018).
To clarify this in the text, we have added this sentence in the 2.2.1 methods section (Lines 156-159): “The correction performed in this paper involved 35% of the total night data for a friction velocity (u*) threshold of 0.19 m s-1 for the rainy season and 0.17 m s-1 for the dry season. Our data can be considered below the thresholds and percentage of gaps found in the literature for other sites in the Amazon region (HAYEK et al., 2018).”
Hayek, N. H.; Longo, M.; Wu, J ; Smith, M. N. ; Restrepo-coupe, N.; Tapajós, R.; Da Silva, R.; Fitzjarrald, D. R.; Camargo, P. B.; Hutyra, L. R.; Alves, L. F.; Daube, B.; Munger, J. W.; Wiedemann, K. T.; Saleska, S. R.; Wofsy, S. C. Carbon exchange in an Amazon forest: from hours to years. Biogeosciences, v. 15, p. 4833-4848, 2018.
Formula 2: Please verify RFA acronym (declared as PAR).
Response: Corrected
Lines 166-176: As the authors presented satellite vegetation indices, It is not clear if EVI data are computed by the authors from MODIS MAIAC reflectances, or they used the ready EVI dataset for South America (MODIS EVI-MAIAC) by the cited Dalagon et al. (2019), and thus, the comments in these lines simply report the Dalagon et al procedure. Please specify the actual input data.
In both the cases, the authors have to check the data quality flags to verify the number of reliable acquisitions within the 16-days composite. This is particularly relevant for the rainy/wet period where the EVI values could be temporally filtered from contiguous (time and space) cloud and haze free acquisitions, and therefore these values are not suitable for 1:1 correlation with tower measurements.
Moreover, why did not you elaborate MAIAC data at 500m spatial resolution?
Response: We have used the product processed and described in Dalagnol et al. 2018 and available in the open-access repository of Dalagnol et al. 2019. The composites were retrieved considering only cloud-free and low atmospheric turbidity according to MAIAC quality flags. For the MODIS-MAIAC pixel containing the flux tower, the mean number of samples per 16-day composite was 4.6 samples for the flooding period (Feb-Mai) and 12 samples for the non-flooding period.
To address this concern, we have adjusted this dataset description to the following text (Lines 176 - 185): “We acquired the MODIS-MAIAC product Enhanced Vegetation Index (EVI) for the study area from the period of 2004 to 2017 [50]. In this product, prior to EVI calculation, the surface reflectance data were normalized to nadir target and 45-degree solar zenith angle through the Bidirectional Reflectance Distribution function at a spatial resolution of 1 km and aggregated to biweekly (16-day) composites using the median values. The EVI was calculated using equation 3 [13]. Further information on image processing and correction are described in Dalagnol et al. [49,50]. The composites were retrieved considering only cloud-free and low atmospheric turbidity according to MAIAC quality flags. For the MODIS-MAIAC pixel containing the flux tower, the mean number of samples per 16-day composite was 4.6 samples for the flooding period (Feb-Mai) and 12 samples for the non-flooding period.”
Regarding the MODIS-MAIAC data with 500 m spatial resolution, the MAIAC parameters for BRDF correction are made available from NASA with 5 km spatial resolution. During the processing described in Dalagnol et al. 2018 they were interpolated to 1 km to match the MODIS-MAIAC surface reflectance data. MODIS-MAIAC composites with 1 km spatial resolution were applied in several papers to study the seasonality of Amazon forest (e.g., Hilker et al., 2014; Moura et al., 2015). To refine the spatial resolution to 500 m, we would have to force the assumption that the BRDF parameters, originally obtained at 5 km resolution, are still representative for this scale.
Dalagnol, R.; Wagner, F.H.; Galvão, L.S.; Nelson, B.W.; Oliveira, E. Life cycle of bamboo in southwestern Amazon and its relation to fire events. Biogeosciences. 2018, 1–28.
Dalagnol, Ricardo; Wagner, Fabien Hubert; Galvão, Lênio Soares; Aragão, Luiz Eduardo Oliveira e Cruz. (2019). "The MANVI product: MODIS (MAIAC) nadir-solar adjusted vegetation indices (EVI and NDVI) for South America". (Version v1) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.3159488
Hilker, T.; Lyapustin, A.I.; Tucker, C.J.; Hall, F.G.; Myneni, R.B.; Wang, Y.; Bi, J.; Mendes de Moura, Y.; Sellers, P.J. Vegetation dynamics and rainfall sensitivity of the Amazon. Proc. Natl. Acad. Sci. 2014, 111, 16041–16046.
de Moura, Y.M.; Hilker, T.; Lyapustin, A.I.; Galvão, L.S.; dos Santos, J.R.; Anderson, L.O.; de Sousa, C.H.R.; Arai, E. Seasonality and drought effects of Amazonian forests observed from multi-angle satellite data. Remote Sens. Environ. 2015, 171, 278–290.
Lines 177-181: The GFC mask was used by the author to identify areas with permanent water. The reliability of a map implemented for global scale requirements is not enough for an application at very local scale as the current study. As visible in Sentinel 2 images containing SWIR data (as in Figure 1), there are large water areas around the tower, particularly in the MODIS pixels North of the tower. To obtain a more detailed map for the investigated MODIS pixels, I suggest the use SWIR Sentinel data combined with high resolution satellite image, such as Bing or Google satellite, (swir data are needed since thy are acquired during the dry period) to digitalize water surfaces. Free data services for GIS environment are:
Bing VirtualEarth http://ecn.t3.tiles.virtualearth.net/tiles/a{q}.jpeg?g=1
Google Satellitehttps://mt1.google.com/vt/lyrs=s&x=%7Bx%7D&y=%7By%7D&z=%7Bz%7D
Alternatively, the authors can use the very high resolution (~3m resolution) Planet data set (https://www.planet.com/products/planet-imagery/) free for researchers.
See in this image sequence flooded areas in February 2016 and March 2017.
Response: We agree with the reviewer. To clarify the percentage of water channels in the analyzed MODIS pixels, we have conducted two additional analysis as suggested by the reviewer and added the following text to the 2.4 methods section (Line 200-203): “To further check the presence of water channels in these pixels, we used the high resolution Bing Virtual Earth image on QGIS 2.18 to visual interpret and manually delineate permanent water bodies, and we found more than 94% of forest cover”.
We also added the following table and the map to the supplementary material (Figure S2 and Table S1).
Table S1. Percentage of permanent water channels and forest cover in MODIS pixels as delineated with high-resolution Bing Virtual Earth image on QGIS.
Pixel ID | Water channels (km²) | Modis pixel area (km²) | Water channels (%) | Forest cover (%) |
1 | 0.07 | 1.01 | 6.42 | 94.88 |
2 | 0.02 | 1.01 | 1.65 | 98.95 |
3 | 0.01 | 1.00 | 1.03 | 99.27 |
4 | 0.01 | 1.01 | 0.96 | 99.54 |
5 | 0.02 | 1.00 | 1.61 | 98.39 |
6 | 0.02 | 1.01 | 1.54 | 99.06 |
7 | 0.01 | 1.00 | 0.92 | 99.48 |
Figure S2. Bing Virtual Earth image. Modis pixels (red squares) around the LBA tower (Yellow dot) and mapped water channels.
Apart the permanent waters, as the authors stated, the characteristic of the study area is just the seasonality of flood events. Therefore, on the basis of rainfall entities and the morphological structure of the site, the flooded areas change among the seasons and from one year to another.
It is not clear how the authors use the measured flood heights. Are the flooded areas evaluated monthly using SRTM data?
Response: The flood height was manually recorded by observations of hydrometric rulers at the tower every month, from January 2004 up today. We used these records in the manuscript to define the seasonally flooded period, which usually extends from February to May. To address the concern regarding potential different flooding patterns between the analyzed MODIS pixels, we have extracted the minimum, maximum, mean and standard deviation (Std) from SRTM elevation data for each of the 7 analyzed MODIS pixels, excluding water channels. The topography is very similar among them (Table below). Therefore, we assume they have a similar flooding height and period.
Pixel ID | Min (m) | Máx (m) | Mean (m) | Std (m) |
1 | 170.0 | 188.0 | 178.5 | 2.8 |
2 | 172.0 | 184.0 | 177.9 | 2.2 |
3 | 172.0 | 182.0 | 176.6 | 2.1 |
4 | 168.0 | 184.0 | 178.6 | 1.9 |
5 | 173.0 | 184.0 | 179.1 | 2.0 |
6 | 170.0 | 184.0 | 178.2 | 2.4 |
7 | 172.0 | 184.0 | 178.0 | 1.9 |
Lines 189-193: For computing the Climatological Water Deficit (CWD), why do you use a fixed evapotranspiration ET value? As shown in Figure 2a, you have ET monthly data available from the tower. Such an approximation only show the rain variability, whose anomalies can be directly computed form TRMM data. Being ET influenced also by temperature dynamics, to avoid over/under estimation periods in CWD, I suggest to recalculate it with the correct monthly ET.
Response: At first, we used a fixed 100 mm evapotranspiration value which is an average value for the Amazon region (e.g., Aragão et al. 2007; Da Rocha et al., 2004; Von Randow et al., 2004), but also, we used this value due to gaps in our records from 2010 to 2016. However, following the reviewer suggestion, we recalculated the CWD based on our monthly ET estimation (Table S3), in order to better represent the CWD variability over the time series. Our results have shown that the CWD calculated with the tower ET is more negative than the previous CWD (with average ET) at the peak of the dry season (approximately -401 mm vs. -417 mm; Figure 2b) and it was far negative during the ENSO year (-430 mm vs -556 mm; Figure 7a). Although we were not able to calculate monthly ET for all years, this approach avoided under estimation periods in CWD. We updated all data, analysis, and figures containing the CWD values, such as figure 2b, figure 4g, figure 7a, table S3, table S4, and Table S6. We adjusted the sentence in the 2.4 methods section (Lines 213-215): “We calculated the monthly average ET from the tower, based on records with a good density of observations from 2004 to 2014 (Table S3), since there were gaps in measurements from 2010 to 2016.”
Aragão, L. E. O. C., Malhi, Y., Roman‐Cuesta, R. M., Saatchi, S., Anderson, L. O., and Shimabukuro, Y. E. (2007), Spatial patterns and fire response of recent Amazonian droughts, Geophys. Res. Lett., 34, L07701, doi:10.1029/2006GL028946.
Da Rocha, H. R., Goulden, M. L., Miller, S. D., Menton, M. C., Pinto, L. D. V. O., & Freitas, H. C. De., Figueira, A. A. M. E. S. (2004). Seasonality of Water and Heat Fluxes Over a Tropical Forest in Eastern Amazonia. Ecological Applications, 14(June 2000), 22–32. https://doi.org/10.1890/02-6001
Von Randow, C., Manzi, A. O., Kruijt, B., de Oliveira, P. J., Zanchi, F. B., Silva, R. L., et al. (2004). Comparative measurements and seasonal variations in energy and carbon exchange over forest and pasture in South West Amazonia. Theoretical and Applied Climatology, 78(1–3), 5–26. https://doi.org/10.1007/s00704-004-0041-z
Lines 213-214: What do you mean with “ The annual cycle is referred to as “inter-annual””? This means that correlations (e.g., in section 3.2) are based on the whole time series and not on the mean annual cycle (e.g., correlation EVI vs GPP for the period 2004-2015 is based on 120 monthly data). If it is right, as the correlation periods strongly differ among the variables, please add the number of correlation points jointly with rho/r and p into the results section.
Response: This statement was confusing. The correlations are based on measurements of climatological variables from 2011 to 2013, since the dependent variable (GPP) was only available for this period. We adjusted the sentence regarding the inter-annual analysis to (Lines 250-252): “We also performed an inter-annual analysis in order to evaluate anomalies in relationships from year to year between monthly EVI and the climatological variables described in table 1.”
Moreover, clarify if the correlation of spaceborne EVI is based on the mean of the seven MODIS pixels around the tower or only on the monthly mean of the pixel in which the tower is located.
Response: We added the following sentence on the 2.5 methods to clarify that, after the pixel filtering and selection, we have used the mean EVI of the seven MODIS pixels (Lines 243-244): “The analyzed EVI value was based on the mean of the seven pixels around the tower.”
Lines 226-230 : Comments on the bimodal temporal pattern of temperature seem to be too hasty. In Figure 2d, the total annual amplitude Tmax-Tmin seems to be about 2°C (annual range ~26-28°C), and the differences between May, June, and July are very subtle and surely in the measurement error of Ta. Thus, Ta in the period May-July is quite constant and the sentence “We observed a bimodal pattern of temperature .., with a pronounced decline in February and a secondary minimum occurring in July” have to be rewrite. Consequently, comments on Rn relationship have to be rearranged.
Response: We agree with the reviewer about the confusion of this sentence. Indeed there was no bimodal pattern on air temperature seasonality. We rewrite the entire section to make it clear and better describe the results presented in figure 2. We added two new paragraphs to introduce our analyses through the following sentences (Lines 266-278): “The seasonal variation of GPP and its interaction with environmental drivers (rainfall, evapotranspiration, cumulative water deficit, vapour pressure deficit, soil moisture, net radiation and temperature) and phenology (described through EVI), showed different patterns among periods of the year. We observed an extended dry season from May to September, when rainfall was below 100 mm.month-1, while ET remained high through all seasons (mean of 110 mm.month-1), especially during the flooding (Figure 2a). This high ET values indicated that soil moisture was sufficient to attend the atmospheric demands, which were described by the increased VPD (Figure 2c) and decreased CWD, from April to August (Figure 2b).”
“Although ET at this site exhibited a flat pattern, slight maximum values coincided with maximum GPP (Figure 2f), but the first and more pronounced peak occurred in March (123 mm), mid-flooding, in response to an increase of Rn at this month (Figure 2e). The second peak occurred in June / July (116 mm), mid-dry season, and a third in December, mid rainy season (122 mm), before the period of flooding.”
Figure 2 Caption: As the mean values of the reported annual cycles represent different periods depending on the variables, please add for each variable the relative years in brackets.
Response: Corrected
Line 274: (caption Figure 3) Please specify that the level 181 m refers to the height at the base of the tower and are meters a.s.l.
Response: Corrected
Lines 312-320: Since the study is based on only 7 MODIS pixels, it makes little sense to evaluate the canopy structure in a completely different area. Moreover, the evaluated LIDAR strip (500m width) represents half of a MODIS pixel.
Response: The main objective of using the LiDAR data was to assess the gap distribution within the forest canopy. This would give us an idea of how much ground-area was directly visible from above the canopy, which could represent water-covered areas during flooded periods. To more accurately represent the gap coverage within the forest canopy, we have revisited our gap analysis and filtered the occurrence of a small river. Hence we obtained a new forest gap cover of 0.51%, which was relatively smaller than the previous estimate of 0.73%. Regarding the canopy structure, field measurements of a forest inventory performed in 2015 reported by Tavares et al. (2019) support our assumptions that the forest structure near the tower is similar to that of the LiDAR transect. Tree heights of representative species were measured with a telescope pole of 15 m length in two experimental plots (BAN1 and BAN2). The two plots are located 200 (BAN1) and 100 meters (BAN2) from the tower, respectively. To address this point, we have added a table in the manuscript with the statistics of the tree heights from both field and LiDAR observations, and adjusted the sentence regarding the forest structure on the results section to (Lines 360-362): “Although the LiDAR transect is 7 km away from the tower (Figure 6a), field measurements of tree height support our assumptions that the forest structure around the tower is similar to those of the LiDAR transect, as described in table 2.” Moreover, we have changed the LiDAR map color (Figure 6, line 372) for a better vizualization with a color blind friendly pallet.
Table 2. Experimental plots and LiDAR statistics regarding tree height measurements
Plots | Number of individuals | Tree Mean height (m) | 5%Percentile (m) | 95%Percentile (m) | Maximum (m) |
BAN1 | 86 | 11.79 | 6.84 | 18.86 | 28.36 |
BAN2 | 84 | 12.48 | 4.5 | 19.77 | 38.89 |
LiDAR | - | 10.2 | 4.8 | 14.9 | 38.0 |
We also added this text on the methods section (Lines 232 - 236): “Regarding the canopy structure, as this transect is located 7 km distant from the LBA tower, we compared tree heights of 170 individuals distributed in two experimental plots located 200 (BAN1) and 100 meters (BAN2) from the tower (Tavares et al., 2019) with LiDAR measurements. A forest inventory was performed in 2015 in these plots, and tree heights were measured with a telescope pole of 15 m length (Table S2).”
Tavares, I. B.; Borma, L. S.; Fonseca, L. D. M.; Collicchio, E.; Domingues, T. F.; Rocha, H. R. The growth pattern of the forest located in a southeast Amazonian floodplain during the 2015/2016 ENSO year. Ecohydrology. (under review). 2019.
To evaluate the influence of water signal in analyzed MODIS pixels, the authors can use satellite SWIR data at higher spatial resolution (e.g. from Landsat series 30m, Sentinel2 20m) for a clear day in the flooding period.
We also evaluated the influence of water signal on the area covered by the MODIS pixels using Landsat-8 surface reflectance data. We used Landsat-8 data (30 m spatial resolution) instead of Sentinel-2 data (10 m) because Sentinel-2 records are still short for a time series analysis. We have masked the water-covered areas considering low values of the SWIR band (B7) from 2014 to 2019. We found that Landsat-8 EVI without and with water masking were similar. This means that exposed water channels did not influence on MODIS EVI signal. Although there are some cloudy scenes during the rainy period, the flooding period in this area extends for up to two months after the rain stops.
Although our number of observations are diminished filtering scenes that encompass the flooded period (from February to May; 73 scenes), we still have a high correlation between EVI with and without the water channel mask.
Moreover, another issue of the adopted approach, which correlate MODIS pixels on the LIDAR strip with the pixel of the tower, is represented by the values of correlations. Along the LIDAR transect, correlation seems to vary between 0.6-0.75. Even if the equations are not shown, such values suggest that EVI differences between these pixels and the tower pixel are higher than the annual amplitude ~ 0.06 (Figure 3). The authors can estimate the correlation by separating the dry and wet period. (By considering the water surface structure in the Planet and Bing imagery), it is expected that the correlation is higher during the dry period.
Response: The main point about this spatial correlation was to address that the forest has a similar canopy structure to that around the tower footprint, supporting that gaps found in the LiDAR transect are also representative for the pixels around the tower. However, we understood that our forest inventory supports these assumptions about the similarity of the canopy structure in both areas, suppressing this spatial correlation. Therefore, we kept this correlation analysis as complementary to the one regarding the LiDAR and field measurements (Lines 360 - 365).
Supplementary material
Table S1: specify in the caption mean annual data to which years it refers to.
Response: Corrected
Table S2: Data for 2013 are missing.
Response: Corrected
Author Response File: Author Response.docx
Round 2
Reviewer 3 Report
The revisions made by the authors largely improved the submitted manuscript and valorized the study. They have introduced additional verifications (EVI quality flags, MODIS-Landsat) and modified estimation methods of some indices (ET-CWD) to assure data consistency. Moreover, they have clarified the ambiguous points and better structured the large amount of data they used.
The paper is more clear and conclusions are well supported by the analyses.
It is ready to be published in Remote Sensing.
Suggestions
Line 180: correct (Feb-Mai) …May
Figure 4b: Please verify that this figure contains the new CWD data (It seems that only the axis are reversed).
