Next Article in Journal
Spatial and Temporal Monitoring of Pasture Ecological Quality: Sentinel-2-Based Estimation of Crude Protein and Neutral Detergent Fiber Contents
Previous Article in Journal
Predicting Forest Inventory Attributes Using Airborne Laser Scanning, Aerial Imagery, and Harvester Data
Open AccessArticle
Peer-Review Record

Spatio-Temporal Variability in Remotely Sensed Vegetation Greenness Across Yellowstone National Park

Remote Sens. 2019, 11(7), 798; https://doi.org/10.3390/rs11070798
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Remote Sens. 2019, 11(7), 798; https://doi.org/10.3390/rs11070798
Received: 25 January 2019 / Revised: 22 March 2019 / Accepted: 28 March 2019 / Published: 3 April 2019
(This article belongs to the Section Remote Sensing in Agriculture and Vegetation)

Round 1

Reviewer 1 Report

The authors are using a diverse combination of remote sensing data, climate, and disturbance data  to explore changes in greenness and productivity in the Yellowstone National Park over the past decades.

The introduction is very nice, flows well, and introduces goals of the study and state of the art, helping the reader to get through the description of the project.

I loved also the the way the methods are described. There is much detail that could potentially drift the reader's attention, and so I found very useful figure 3 and its summary of questions addressed and methods. 

The results section is probably a bit long, but the details are preparatory for the conclusions addressed in the discussion. I really enjoyed the organization of results and discussion into focal questions. 

In conclusion, despite the methods per se are not novel, they are solid and the authors succeeded in showing how useful can be the integration of remote sensing, climate, and disturbance information for monitoring phenology and more broadly ecological processes at intermediate scale (ecoregion). 

Nice job!

Author Response

Thank you for the positive review and encouraging comments. 


Reviewer 2 Report

Dear Authors,

thank you very much for the manuscript. The topic is important, but the quality of the manuscript has to be significantly improved. The most important problem which I can see is your negative correlation between vegetation indices and precipitation. It means if there is more rain, the condition is worst. It isn't acceptable.

The Introduction has to be rebuilt, the description of the research area has to be moved to another part of the Manuscript, but in the Introduction you have to present much deeper a theoretical background of your issue, to show how other researchers solve out this problem. There are a lot of indices, so please present the issue much deeper. I have been working on the national parks and I use completely different indices.

Methods should contain an accuracy assessment and validation of your results. It has to be described. Present version is a mix of your ideas and a theoretical background of indices.

Discussion doesn't present a real comparison of the best your achievements and results acquired by other researchers. It allow to select better methods.

All maps have to be improved (details in the manuscript), please add your interpretation of all your figures. There a lot of graphs, which should be discussed.

Much more details you can find in my attachment.

Best regards

Reviewer


Comments for author File: Comments.pdf

Author Response

We really appreciate all of the careful feedback provided by the reviewer.  We have substantially revised the manuscript and hope the current draft is much more satisfactory.

 

The reviewer objected to negative correlations between vegetation indices and precipitation.

 

Lines 368-372 now state, “Spatial correlations between VI and climatic maps indicate significant positive correlations between annual NDVI/EVI and air temperature (NDVI: 0.26, EVI: 0.31) and significant negative correlations (NDVI: -0.28, EVI: -0.28) between annual NDVI/EVI and precipitation, all at p<0.05.  This may be interpreted as climatologically warm, dry (cold, wet) regions corresponding to greater (reduced) annual greening.  More specifically, springtime (April-June) greening occurs later in cold, wet regions that are typically characterized by an extensive snowpack that must melt prior to vegetation emergence.”  Furthermore, we added on lines 497-500, “The negative correlation between late spring NDVI and both preceding and concurrent precipitation reflects delayed springtime growth due to (1) anomalously deep and persistent snowpack from cumulative cold-season precipitation and (2) lower air temperatures generally associated with wet conditions.”

 

The reviewer asked that we rebuild the introduction and provide a deeper, more theoretical background to the issue. 

 

On lines 48-52, we added, “The projected warming and drought-related water shortages for the western United States are expected to alter terrestrial ecosystems [Breshears et al. 2005; Williams et al. 2010; Williams et al. 2012] and induce pronounced shifts in species ranges and biodiversity patterns [Notaro et al. 2012], although with such studies placing a greater focus on the water-limited Southern Rockies than the temperature-limited Middle Rockies [Bassman et al. 2003; Nemani et al. 2003].”  On lines 63-65, we added, “According to the Green Wave Hypothesis, plant phenology critically shapes the resource landscape and migratory ungulates are found to track the green wave, with implications of climate change affecting the persistence of migratory taxa (Aikens et al. 2017).”

 

The reviewer asked that the description of the research area be moved to another part of the paper. 

 

The bulk of the description of Yellowstone National Park has been moved to section 2.1 Study area under Materials and Methods.

 

The reviewer asked for more details on the vegetation indices.

 

On lines 199-206, we added, “In an effort to assess robustness of results, this study considers two remotely sensed VIs, namely NDVI and Enhanced Vegetation Index (EVI), the former of which is the most commonly applied VI (with an extensive time duration, not requiring blue-band data as in EVI which was unavailable on earlier platforms) and the latter of which aims to reduce known deficiencies in NDVI related to atmospheric and background effects.  NDVI and EVI are commonly used as proxies [Garroutte 2016] for phenology [Zhang et al. 2003; Reed et al. 2009], aboveground biomass [Tucker et al. 1985; Cabrera-Bosquet et al. 2011], primary productivity [Tucker and Sellers 1986; Paruelo et al. 1997; Seaquist et al. 2003], forage quality [Hamel et al. 2009; Doiron et al. 2013], and chlorophyll content [Jones et al. 2007; Christianson and Creel 2009].”  On lines 226-228, we added, “Based on comparison of MODIS VIs with field observations of canopy reflectance, Miura et al. [2000] concluded that mean uncertainties are ±0.01 VI units for NDVI and ±0.02 VI units for EVI under typical atmospheric conditions with 2% reflectance calibration uncertainty.”

 

The reviewer asked that an accuracy assessment and validation section be added to the Methods.

 

In response to this request, we added Supplemental Figure 1 and section 2.4, “MODIS data reliability and quality.”

The seasonal cycle of quality assurance (QA) indices was examined for MODIS VIs across all land pixels in Yellowstone National Park during all 16-day periods within 2000-2017 (Supplemental Figure 1).  These indices include pixel reliability and multiple quality measures, including VI usefulness, MODIS land (MODLAND) QA, cloud shadow presence, atmospheric bidirectional reflectance distribution function (BRDF) correction application, snow/ice presence, adjacent cloud presence, mixed cloud presence, and aerosol quantity.  Pixel reliability was highest during June-September, averaging 85% under the category of “good data / use with confidence,” and lowest during November-April, averaging 71% under the category of “target covered with snow / ice.”  VI usefulness, based on aerosol quantity, atmospheric correction conditions, cloud cover, shadow, and sun-target viewing geometry, was categorized as “good quality” for 89% of the spatio-temporal data values, with only 1% of the data falling at or below intermediate quality level.  According to the MODLAND QA index, 50% of the data was produced with good quality, with no need to examine additional detailed QA flags, 41% has unquantifiable quality with the recommendation to check more detailed QA flags, 3% was produced but under cloudy conditions, and 6% was not produced due to reasons other than clouds.  Quality was potentially impacted by shadows (primarily in November-April), snow and ice (primarily in November-May), and mixed clouds for 48%, 16%, and 2% of the data, respectively.  The atmosphere-surface BRDF coupled correction and adjacency cloud correction were performed for 37% and 6% of the data, respectively, with the BRDF correction commonly applied in June-September.  The aerosol optical thickness was not known for 87% of the data, so aerosol climatology was used for atmospheric corrections, while low aerosol load was noted for 13% of the data. 

On lines 174-175, we added, “This study applies commonly used datasets that have been individually calibrated and validated, with details about quality control found in their respective cited literature.”

On lines 226-228, we added, “Based on comparison of MODIS VIs with field observations of canopy reflectance, Miura et al. [2000] concluded that mean uncertainties are ±0.01 VI units for NDVI and ±0.02 VI units for EVI under typical atmospheric conditions with 2% reflectance calibration uncertainty.”

 

The reviewer objected to theoretical backgrounds on methods/indices being included in the Methods section.  Specifically, the reviewer objected to the discussion of the theory behind EOF analysis being in the methods section and wanted it moved to the introduction.

 

We believe the most appropriate place for the description of EOF analysis is in the Methods section.  If that was moved to the Introduction, it would be extremely awkward and out of place as the Introduction focuses on motivating the study.

 

The reviewer asked that the Discussion section compare our results with past studies, perhaps as a table.  The reviewer felt the current Discussion was more of a literature review that did not discuss our most important achievements compared to other researchers. 

 

We added Table 1, which summarizes 9 papers on variability and change in Yellowstone’s vegetation and climate, including the time period, applied datasets, key results, and comparison of results with the current study.  On lines 665-668, we added, “A further comparison of the present study’s findings and those of past studies is presented in Table 1, while noting that few studies have extensively analyzed remotely sensed VIs across Yellowstone National Park.  Results are largely consistent, with disparities often associated with differences in analyzed time periods between studies.”  Much of the Discussion section is elaborating on our main findings for the two key questions and comparing with numerous related studies. 

 

The reviewer asked that we improve all maps.

 

We revised Figure 1 to include the latitude/longitude, compass, scale, and reversed color bar.  We revised Figure 2 to include the latitude/longitude, compass, scale, and 5th-95th percentiles of elevation, while removing the black triangle.  For Figure 8, we revised the caption with more details and added a more detailed description of how to interpret the figure, namely, “For example, in column 5 of panel (f), the park’s average NDVI anomalies in May exhibit a statistically significant negative correlation with snow-water equivalent from the preceding to concurrent period of January-May (as shown by the blue/purple boxes under lag 0 to lag -4).“

 

The reviewer asked that we discuss all figures and our opinions.

 

The following lines describe the results from each figure.

Figure 1…lines 125-134              Figure 2…lines 134-142              Figure 3…lines 266-309

Figure 4…lines 322-342              Figure 5…lines 352-375              Figure 6…lines 385-410

Figure 7…lines 435-453              Figure 8…lines 457-503              Figure 9…lines 508-535, 614-620

Figure 10…lines 548-553, 628-632

 

The reviewer asked for a shorter abstract.

 

The current edited abstract has 282 words, below the 300 word limit of the journal.  It presents the motivation, key questions, main findings, and implications.

 

The reviewer asked for more detail on the impact of references to our manuscript in the introduction and expressed concern over “citation trains.”

 

We reduced the number of references in “citation trains” to the most important ones.  We added more detailed discussion of the importance of their results. 

On lines 48-52, we added, “The projected warming and drought-related water shortages for the western United States are expected to alter terrestrial ecosystems [Breshears et al. 2005; Williams et al. 2010; Williams et al. 2012] and induce pronounced shifts in species ranges and biodiversity patterns [Notaro et al. 2012], although with such studies placing a greater focus on the water-limited Southern Rockies than the temperature-limited Middle Rockies [Bassman et al. 2003; Nemani et al. 2003].” 

On lines 63-65, we added, “According to the Green Wave Hypothesis, plant phenology critically shapes the resource landscape and migratory ungulates are found to track the green wave, with implications of climate change affecting the persistence of migratory taxa (Aikens et al. 2017).”

Section 5.3 now presents a comparison of results with past studies, including Table 1.

 

The reviewer asked why we discussed select studies in the Introduction under “There have been a limited…” given the 100s of papers each year using satellite VIs.

 

The sentence of concern is: “There have been a limited number of remote sensing-based vegetation monitoring studies specifically focused on western United States’ national parks, including Bandelier [Raul Rono Leon et al. 2012], Big Bend [White and Swint 2014], Crater Lake [O’Leary et al. 2018], Glacier [Brown 1994; Leon et al. 2012], Saguaro [Wallace et al. 2016], Yellowstone [Potter 2015; Frank et al. 2013; Garroutte et al. 2016; Zhao et al. 2016], and Yosemite [van Wagtendonk and Root 2003; Soulard et al. 2016].”  We agree that there are 100s of papers that apply remotely sensed vegetation indices, but we are listing some of the few available studies using VIs across western U.S. national parks.  Our point is that remote sensing has been largely underutilized to examine the health of vegetation across our valuable national parks.

 

The reviewer asked that we specify the spatial range of our research.

 

On lines 260-261, we added, “The spatial range of the study region is 44.13°N-45.11°N, 111.15°W-109.83°W.”

 

In Fig. 1, the reviewer wants us to reverse the colors, add blue to the legend, and add geographical coordinates.

 

As requested, we reversed the color scheme, mentioned blue in the caption, and added the geographical coordinates to Figure 1.

 

In Fig. 2, the reviewer felt that ranges of altitudes would be better than the mean.  The reviewer asked us to add the geographical grid and scale tab, and to remove the black on the right.

 

We followed your recommendations for Figure 2.  However, instead of the full range of altitudes, we listed the 5th and 95th percentiles to avoid confusion by extreme outliers.

 

The reviewer asked us to justify why we used NDVI and EVI specifically.

 

On lines 199-205, we added, “In an effort to assess robustness of results, this study considers two remotely sensed VIs, namely NDVI and Enhanced Vegetation Index (EVI), the former of which is the most commonly applied VI (with an extensive time duration, not requiring blue-band data as in EVI which was unavailable on earlier platforms) and the latter of which aims to reduce known deficiencies in NDVI related to atmospheric and background effects.  NDVI and EVI are commonly used as proxies [Garroutte 2016] for phenology [Zhang et al. 2003; Reed et al. 2009], aboveground biomass [Tucker et al. 1985; Cabrera-Bosquet et al. 2011], primary productivity [Tucker and Sellers 1986; Paruelo et al. 1997; Seaquist et al. 2003], forage quality [Hamel et al. 2009; Doiron et al. 2013], and chlorophyll content [Jones et al. 2007; Christianson and Creel 2009].”

 

The reviewer asked for details on the accuracy of NDVI and EVI.

 

On lines 226-228, we added, “Based on comparison of MODIS VIs with field observations of canopy reflectance, Miura et al. [2000] concluded that mean uncertainties are ±0.01 VI units for NDVI and ±0.02 VI units for EVI under typical atmospheric conditions with 2% reflectance calibration uncertainty.”

 

The reviewer wanted more detail on the methods, including dates of used images.  The reviewer wanted a research schema added.  The reviewer wants all statistical methods explained in the Methods section.

 

We expanded the text on lines 255-262 to, “All meteorological and topographic datasets were bilinearly regridded to the 250-m MODIS VI grid using the ESMF_regrid function in NCL, which applies the Earth System Modeling Framework software.  According to bilinear interpolation, each destination point is mapped to a location within the source grid, while the position of the destination point relative to the source points is used to compute the interpolation weights.  Further testing of the study’s results, instead using nearest neighbor regridding (not shown), reveals nearly identical findings.  The spatial range of the study region is 44.13°N-45.11°N, 111.15°W-109.83°W.  VI imagery was examined for 1982-2015 for AVHRR data and 2000-2017 for MODIS data.”

Likewise, we expanded the text on lines 273-285 to, “When computing spatial correlations among topographic, ecological, and climatic datasets, one must address the potential for large spatial autocorrelation that can artificially inflate the correlation coefficient.  As a consequence, semivariogram graphs are generated to quantify semi-variance, which measures the spatial dependence between observations as a function of the distance between them.  The semivariograms provide the effective radius of spatial autocorrelation, which here is 750 m.  Data is randomly sampled at a greater distance than that radius to remove the effects of spatial autocorrelation, leading to an adjusted sample size (N) of 17,481; this newly determined N is applied when assessing statistical significant for all spatial correlations  The seasonality and dominant modes of variability in vegetation phenological growth, and their sensitivity to climate and topography, are examined through EOF analysis of MODIS detrended NDVI data during the non-winter months of April-November.  EOF analysis permits a simplified interpretation of complex data across the space-time domain by decomposing a spatio-temporal field into dominant orthogonal spatial patterns and their associated uncorrelated time series, or principal components, while reducing noise in the data [Fukuoka 1951; Lorenz 1956; Kutzbach 1967; Hannachi et al. 2007; Wang et al. 2015].” 

We already included the research schema in Figure 3, as requested by the reviewer the last time.

 

The reviewer wanted the most important results outlined at the beginning of the results section.

 

            A review of many past publications in the Journal of Remote Sensing did not find that other authors have followed the recommended approach of outlining the key results at the top of the results section.  We feel that would be confusing and out of place.  Our results are listed in the order that they were investigated, per our research schema.

 

In the caption of Fig. 7, the reviewer wants us to include the source of the data for the % explained variances.

 

In the caption of Figure 7, we added, “The explained variance associated with each eigenvalue in EOF1-3 is 31.0%, 13.1%, and 7.7%, respectively, for the undetrended data and 31.8%, 9.3%, and 3.4%, respectively, for the detrended data.  The explained variance for each EOF mode is computed by dividing that mode’s eigenvalue by the sum of all modes’ eigenvalues and multiplying by 100%, where an eigenvalue expresses how much of the variance can be explained by its associated eigenvector.”

 

The reviewer wants a better explanation of the methods applied in Fig. 8 to make the figure more clear.

 

The caption now reads, “Figure 8.  Temporal correlations between detrended monthly Advanced Very High Resolution Radiometer (AVHRR) Normalized Difference Vegetation Index (NDVI) anomalies, averaged across Yellowstone National Park, and park-mean, antecedent atmospheric detrended monthly anomalies (leading NDVI by 0 to 7 months) from Daymet and Western Regional Climate Center (WRCC) during 1982-2015.  Only statistically significant correlations (p<0.05) are displayed.  Correlations are shown for (a) vapor pressure deficit, (b) Palmer Drought Severity Index (PDSI), (c) 6-month Standardized Precipitation Evapotranspiration Index (SPEI), (d) precipitation, (e) maximum daily temperature, (f) snow-water equivalent, (g) mean temperature, (h) minimum daily temperature, (i) vapor pressure, and (j) incoming surface shortwave radiation.  For each panel, the number of significant correlation boxes (out of 96) is provided.  For example, in column 5 of panel (f), the park’s average NDVI anomalies in May exhibit a statistically significant negative correlation with snow-water equivalent from the preceding to concurrent period of January-May (as shown by the blue/purple boxes under lag 0 to lag -4).“

 

The reviewer reminded us to add DOIs to the references.

 

We added all of the available DOIs to the references.

 

The reviewer asked us to add “interlines” between each reference.

 

We are unclear about this request, as the format of the journal does not include blank lines between each reference.


Round 2

Reviewer 2 Report

Dear Authors

thank you very much for your improvements. I can see some improvements, but the manuscript is still heavy to understand everything just reading because there isn't a clear structure, additionally a lot of details, please try to move some data to the Suplements.

The problem of the Introduction is still this same, You present a lot of well-known statements, the is a significant lack of details, how others solve out the issue, which is your topic? The present version is significantly too general and it hasn't be accepted.

Methods are very important, please add more details of your research algorith, how you acquired and processed the data? Why you selected such statistical methods? What about validation of the model? 

The Methods contain a lot of references to literature, you should present them in the Introduction or in a chapter before the Methods, It could allow to understand faster and easier your algorithm.

Maps are improved, but they have to be reformatted.

More comments  you can find in the attached manuscript.

With best regards

Reviewer

Comments for author File: Comments.pdf

Author Response

Dr. Li,

            We have revised the paper, “Spatio-Temporal Variability in Remotely Sensed Vegetation Greenness Across Yellowstone National Park,” based on comments from the anonymous reviewers.  Below, we respond to the reviewer’s concerns.

 

Take care,

Michael Notaro

Associate Director, Nelson Institute Center for Climatic Research

University of Wisconsin-Madison

[email protected]

 

The reviewer asked that some materials be moved to the Supplemental Materials.

            Due to its technical nature, we moved the section, “MODIS data reliability and quality,” to the Supplemental Materials.

 

The reviewer asked that we address details, how others solved the issue, and our topic in the introduction, so it is not so general.  The reviewer asked that we explain the impact of the national park remote sensing papers to our work, what is new, and what kind of methods were used.

            We modified this portion of the introduction on lines 89-136 as follows:

There have been a limited number of remote sensing-based vegetation monitoring studies specifically focused on western United States’ national parks, including Bandelier [28], Big Bend [29], Crater Lake [30], Glacier [31], Saguaro [32], Yellowstone [33-38], and Yosemite [39,40].  The Yellowstone studies of Franks et al. [33], Jakubauskas and Price [34], Zhao et al. [35], Emmett et al. [36], Potter [37], and Garroutte et al. [38] focused on relating remotely sensed VIs to observed leaf area index (LAI) and net primary productivity, forest structure, post-disturbance fire recovery, vegetation greening trends, analysis of vegetation cover changes, and relating remotely sensed VIs to grassland biomass and quality, respectively. Based on regression analysis, Franks et al. [33] determined that Landsat-derived Normalized Difference Vegetation Index (NDVI) and band 5 reflectance metrics could describe about 60-70% of the variability found in ground-based measurements of LAI and annual net primary productivity over a six-year fire recovery period; this led to the conclusion that spatial variability in regrowth rates can be inferred from temporal trends in satellite vegetation metrics.  Jakubauskas and Price [34] demonstrated the value of remote sensing for examining the biotic factors of gross physical structure for Yellowstone’s lodgepole pine forest canopies by regressing forest overstory and understory field data against Landsat radiance values and transformed data to quantify the reliability of the remote sensing.  By assessing long-term post-disturbance forest spectral recovery following disturbances in Landsat data via the application of the Landsat Time Series Stack-Vegetation Change Tracker algorithm, Zhao et al. [35] found that forest recovery is sensitive to forest type, soil type, and elevation.  Applying linear least squares regression and the Mann-Kendall significance test, Emmett et al. [36] focused on the greening trend across the Greater Yellowstone Ecosystem and the relative importance of disturbance and climate in explaining trends in United States Geological Survey (USGS) Earth Resources Observation and Science (EROS) NDVI during 1989-2014.  Consistent with Potter [37], the study concluded that disturbance history was a major driver of the park’s NDVI trends and that it is essential to analyze spatial patterns of trends in plant productivity using very high-resolution data.  The aforementioned studies proved that VIs can be effective measures of ecosystem attributes, such as LAI [33] and gross physical structure [34].  While these individual studies contribute to our understanding of discrete phenomena, NPS leadership has identified the need for synthesizing the complex interplay of multiple climate and vegetation signals in support of management decision-making [41].

Of the many previously-cited papers which use space-borne remote sensing to investigate ecological questions within individual western national parks, it is surprising that only two focus on vegetation phenology specifically: O’Leary et al. [30], which stressed the impact of snowmelt timing on greenup, and Wallace et al. [32], which mapped invasive buffelgrass and its phenology. This highlights an notable gap in the literature, considering that the importance of long-term, continuous remote sensing data was emphasized by White and Swint [29] and Soulard et al. [40], while most of the remote sensing studies of the western national parks have analyzed only a limited number of snapshot images [28,31,39] or 3-5 years of satellite data [29,32,37].  As pointed out by O’Leary et al. [30], conifer regions have especially received insufficient attention in remote sensing phenological studies.  Furthermore, such studies have often considered minimal or no climate datasets in their analysis of VI patterns, thereby neglecting the importance of spatio-temporal variations in meteorological conditions in driving phenological responses; these studies used topographic variations in elevation, slope, and aspect as proxies for temperature, precipitation, and solar radiation [28,30,31,40] and often concluded that elevation is the main driver of NDVI’s spatial heterogeneity [31].  Therefore, there is a need to quantify the complex interactions between the many important climate variables and wildfire disturbance as drivers of vegetation phenology and productivity, particularly within an important natural resource where changes in climate and disturbance patterns are expected into the future. This study aims to address this need by quantifying the spatio-temporal variability in the vegetation greenness of Yellowstone National Park in response to changing climatic and disturbance conditions across multiple spatio-temporal scales and ecosystems.”

 

The reviewer asked for more detail on the research algorithm, how we acquired and processed the data, why we chose specific statistical methods, and model validation.

We provided expanded details on the data sources, processing, and statistical methods.  As we did not apply a physical model or develop a new statistical model, we did not include a model validation section.  The Supplemental Materials address MODIS data reliability and quality.

On lines 205-215, the text was expanded to, “The primary plant type distribution was determined from the NPS’ YELL 1999 Cover Type dataset, as provided by the NPS, which consists of a geodatabase of land cover and habitat types layers based on 1:15,840 scale color aerial photography and field surveys [42].  The cover type dataset was processed into eight categories: aspen, Douglas-fir (combining three classifications, namely climax, post disturbance, and successional Douglas-fir), Engelmann spruce and subalpine fir, krummholz, lodgepole pine (combining four classifications, namely climax, post disturbance, successional, and pygmy lodgepole pine), nonforested, water, and whitebark pine (combining three classifications, namely climax, post disturbance, and successional).  A fire extent perimeter polygon dataset, containing shapefile perimeters of fires larger than 0.4 km2, for 1881-2016 was acquired from the Yellowstone National Park Spatial Analysis Center (provided by Alex Zaideman).” 

            On lines 216-228, the text was expanded to, “Monthly Palmer Drought Severity Indices (PDSI) [73,74] and Standardized Precipitation Evapotranspiration Indices (SPEI) [75] were obtained, averaged across Wyoming’s Park County, through the West Wide Drought Tracker website (https://wrcc.dri.edu/wwdt) of the Western Regional Climate Center [76].  SPEI combines the advantages of the Standardized Precipitation Index and PDSI, namely the consideration of different timescales and the simultaneous consideration of both temperature and precipitation, respectively, leading to a potentially more meaningful measure of drought impacts on vegetation [77,78].  Specifically, SPEI was considered here at time scales of one (SPEI1mn) and six months (SPEI6mn) to consider both short- and long-term drought and pluvial impacts.  Analyses applied air temperature, precipitation, incoming surface solar radiation, and snow-water equivalent (snow depth x density) on a 1 km x 1 km grid from the Daymet Daily Surface Weather and Climatological Summaries [79,80], which are based on the Global Historical Climatology Network (GHCN)-Daily dataset distributed by National Centers for Environmental Information (NCEI) [81].  The Daymet data was acquired through the Oak Ridge National Laboratory website (https://daymet.ornl.gov).”

            On lines 250-262, the text was expanded to, “The NDVI3g dataset, retrieved from the National Aeronautics and Space Administration (NASA) Monitoring, Modeling, and Forecasting Ecosystem Change website (https://ecocast.arc.nasa.gov), was generated by Zhu et al. [101] based on multiple AVHRR sensors using a Feed-Forward Neural Network algorithm, while accounting for calibration loss, orbital drift, and volcanic eruptions.  The Moderate Resolution Imaging Spectroradiometer (MODIS) MOD13Q1 Vegetation Indices level-3 dataset from the NASA Earth Observation System, with 250-m spatial resolution, covers the period of 2000-2017 [82,94,102]; this dataset was obtained through the Reverb website of the NASA Earth Observing System Data and Information System (https://earthdata.nasa.gov) for horizontal tile 10 and vertical tile 4.  Values of MODIS NDVI range from -2,000 to 10,000, with a scale conversion factor of 10,000.  For each 16-day period, the MOD13Q1 algorithm selects the most reliable pixel value from all data acquisitions based on low cloud abundance, aerosol loading, view angle, and peak NDVI value.  Values over water or outside of the shapefile boundaries of the park were masked.”   

            On lines 306-310, we added, “EOF analysis, also known as Principal Component Analysis, permits a simplified interpretation of complex data across the space-time domain (ideal for the vast, high-resolution Yellowstone datasets) by decomposing a spatio-temporal field into dominant orthogonal spatial patterns and their associated uncorrelated time series, or principal components, while reducing noise in the data [105-109].”  On lines 317-318, we added, “The amount of variance explained by the principal components is a metric for evaluating the efficacy of EOF analysis.”

            On lines 327-337, we added, “In order to identify potential climate change signals, linear trends are computed according to the non-parametric Theil-Sen slope estimator, both spatially by pixel and for the park average.  While least squares regression determines the slope based on the weighted mean, the Theil-Sen estimator determines the slope based on the median, thereby making the latter approach more robust against outliers.  Linear trends are assessed for significance via the two-sided Mann-Kendall test [36,110-112] for NDVI and EVI from AVHRR and MODIS (as proxies of productivity), WRCC PDSI and SPEI, and precipitation, snow-water equivalent, air temperature, vapor pressure deficit, incoming surface solar radiation, and vapor pressure from Daymet.  The Mann-Kendall Trend Test is a non-parametric test that assesses if there is a monotonic upward (consistently increasing, although not necessarily in a linear fashion) or downward trend over time.  The Theil-Sen slope estimator and Mann-Kendall test are flexible in that neither make assumptions about the distribution of the analyzed variable.”

 

The reviewer did not approve of the methods section containing background references/literature about methods, or “theoretical descriptions”.

            The description of the basics of EOF analysis, semivariograms, etc. belongs in the Method section.  If moved to the Introduction, it would be a large distraction and out of place.  This follows the guidelines of the “Research Manuscripts Sections” as described by the journal (https://www.mdpi.com/journal/remotesensing/instructions#manuscript).  We examined dozens of other Journal of Remote Sensing articles and all of them follow the approach taken in our paper.

 

The reviewer asked that the maps be reformatted, saying that we should look at the guidelines regarding font style, font size, and line width.

            The following text from the Journal of Remote Sensing website describes all of the guidelines for developing figures, which does not include any details regarding font styles or line widths in plots.  We have been advised by the editorial board to submit based on the guidelines below, and if the editorial team wants any further changes, they will notify us.

 

https://www.mdpi.com/journal/remotesensing/instructions#figures

File for      Figures and schemes must be provided during submission in a single zip      archive and at a sufficiently high resolution (minimum 1000 pixels      width/height, or a resolution of 300 dpi or higher). Common formats are      accepted, however, TIFF, JPEG, EPS and PDF are preferred.

Images      and maps must have a scale, a north arrow and coordinates (see an      example here).

Remote      Sensing can publish multimedia files in articles      or as supplementary materials. Please contact the editorial office for      further information.

All      Figures, Schemes and Tables should be inserted into the main text close to      their first citation and must be numbered following their number of      appearance (Figure 1, Scheme I, Figure 2, Scheme II, Table 1, etc.).

All      Figures, Schemes and Tables should have a short explanatory title and      caption.

All      table columns should have an explanatory heading. To facilitate the      copy-editing of larger tables, smaller fonts may be used, but no less than      8 pt. in size. Authors should use the Table option of Microsoft Word to      create tables.

Authors      are encouraged to prepare figures and schemes in color (RGB at 8-bit per      channel). There is no additional cost for publishing full color graphics.

 

The reviewer objected to “deepen the understanding” in the abstract since it “isn’t measurable.”

            On lines 17-19, we modified the text to state, “The study’s objective is to quantify the responses of vegetation greenness and productivity to climate variability and change across complex topographic, climatic, and ecological gradients in Yellowstone National Park through the use of remotely sensed data.”

 

The reviewer felt that Spearman was a better choice than Pearson.

            In response to the suggestion, all correlations and trend analyses were expanded to also include Spearman’s rho tests, noting that the results were largely consistent. 

On lines 278-281, we updated the text to state, “Applied statistical methods, as expanded upon below, include the Theil-Sen estimate of linear trend, Mann-Kendall test of trend significance, Pearson’s correlation, Spearman’s rho test, and unrotated empirical orthogonal function (EOF) analysis, using NCL’s trend_manken, rgrid2rcm, rtest, escorc, spcorr, and eofunc functions.”

            On lines 341-348, we added, “In order to ensure the robustness of the study’s findings, all correlations and trend analyses are repeated with the rank-based, non-parametric Spearman’s rho (r) test.  This test is advantageous over Pearson’s correlations as it operates on the data’s ranks, rather than the raw data, is unaffected by the population’s distribution, is mostly insensitive to outliers, and is not limited to linear dependencies.  However, it has certain disadvantages to the Pearson’s correlation, namely the loss of quantitative information when the data is converted to ranks and less power for normally distributed data.  In terms of trend analysis, a trend is considered significant with the Spearman’s rho test if there is a significant correlation between the ranks and time steps.”

            Spearman’s test results are included on lines 375, 403-409, 556-557, and 597-601.  Figure 8 was updated to also include Spearman’s rho results.

 

The reviewer wanted semivariograms explained more.

On lines, 291-303, we added, “As a consequence, semivariogram graphs are generated using the “gstat” R package [104] to quantify semi-variance, which measures the spatial dependence between observations as a function of the distance between them.  Specifically, plots of distance versus semivariance are generated for eight select comparisons: annual mean air temperature versus elevation, annual mean precipitation versus elevation, annual mean EVI versus annual mean NDI, annual mean NDVI versus annual mean air temperature, annual mean NDVI versus annual mean precipitation, annual mean EVI versus annual mean air temperature, annual mean EVI versus annual mean precipitation, and the annual seasonal maximum of EVI versus the annual seasonal maximum of NDVI.  The semivariograms from the residuals of these fitted linear models provide the effective radius of remaining spatial autocorrelation; for the eight aforementioned comparisons, the maximum radius is 750 m.  Data is randomly sampled at a greater distance than that radius to remove the effects of spatial autocorrelation, leading to an adjusted sample size (N) of 17,481; this newly determined N is applied when assessing statistical significant for all spatial correlations.”

 

With regards to “m-2) (Figure 4)”, the reviewer pointed out that there should not be brackets side by side.

            Lines 361-365 now state, “Four climatological variables with noteworthy spatial heterogeneity across Yellowstone National Park include summer (June-August) maximum temperature (ranging from 13.7°C to 28.2°C), winter (December-February) precipitation (ranging from 40 mm to 646 mm), annual incoming surface solar radiation (ranging from 307 W m-2 to 443 W m-2), and annual maximum snow water equivalent (ranging from 22 kg m-2 to 879 kg m-2), based on Daymet data for 1982-2015 (Figure 4).”


Back to TopTop