Next Article in Journal
Observations of Emissions and the Influence of Meteorological Conditions during Wildfires: A Case Study in the USA, Brazil, and Australia during the 2018/19 Period
Previous Article in Journal
Identifying Aerosol Subtypes from CALIPSO Lidar Profiles Using Deep Machine Learning
Previous Article in Special Issue
Effect of Various Types of ENSO Events on Moisture Conditions in the Humid and Subhumid Tropics
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Atmosphere
Volume 12
Issue 1
10.3390/atmos12010012
atmosphere-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Modeling the Radial Stem Growth of the Pine (Pinus sylvestris L.) Forests Using the Satellite-Derived NDVI and LST (MODIS/AQUA) Data

by
Yulia Ivanova
1,*,
Anton Kovalev
2 and
Vlad Soukhovolsky
3
1
Institute of Biophysics SB RAS, Federal Research Center “Krasnoyarsk Science Center SB RAS”, 660036 Krasnoyarsk, Russia
2
Federal Research Center “Krasnoyarsk Science Center SB RAS”, 660036 Krasnoyarsk, Russia
3
Sukachev Institute of Forest SB RAS, Federal Research Center “Krasnoyarsk Science Center SB RAS”, 660036 Krasnoyarsk, Russia
*
Author to whom correspondence should be addressed.
Atmosphere 2021, 12(1), 12; https://doi.org/10.3390/atmos12010012
Submission received: 6 November 2020 / Revised: 18 December 2020 / Accepted: 18 December 2020 / Published: 24 December 2020
(This article belongs to the Special Issue Response of Boreal and Tropical Forest Ecosystems to Extreme Weather Events)
Browse Figures
Versions Notes

Abstract

:
The paper considers a new approach to modeling the relationship between the increase in woody phytomass in the pine forest and satellite-derived Normalized Difference Vegetation Index (NDVI) and Land Surface Temperature (LST) (MODIS/AQUA) data. The developed model combines the phenological and forest growth processes. For the analysis, NDVI and LST (MODIS) satellite data were used together with the measurements of tree-ring widths (TRW). NDVI data contain features of each growing season. The models include parameters of parabolic approximation of NDVI and LST time series transformed using principal component analysis. The study shows that the current rate of TRW is determined by the total values of principal components of the satellite indices over the season and the rate of tree increment in the preceding year.

1. Introduction

Research addressing the growth and changes in productivity of boreal forests is useful for understanding and predicting sustainability of these ecosystems affected by diverse environmental stresses associated with extreme weather events and climate change [1,2,3,4,5,6,7]. Most studies dealing with this problem predict that, because of climate change, tree growth will be affected by more frequent and violent extreme events (ice storms, wetland formation, droughts, windblows, etc.). However, there is considerable uncertainty about the processes underlying tree response to the increase in the frequency of extreme events.
It is well known that tree growth results from complex physiological processes in tree organs, and availability of resources for tree growth is largely determined by climate [8]. Tree growth processes and weather events can be understood using methods of dendrochronology, which studies time series of tree ring widths (TRW) over many years [9,10,11,12].
Data on long-term TRW variations are used to determine how much wood is formed in the tree stem, which can be directly associated with the biomass increase and carbon consumption [13,14,15,16]. In recent decades, dendrochronology methods have been used to study interactions between species-specific growth and temperature and humidity variations at the local, regional, and continental scales [17,18,19,20,21].
Although the data on tree rings provide the basis for accurate and retrospective estimates of tree growth, to collect wood samples in the field and to measure them in the laboratory are painstaking tasks. Therefore, the use of this approach in real-time monitoring of forest growth at large spatial scales is limited [22]. Satellite remote sensing data, however, are successfully used to face these challenges. Over the past 30 years, extensive research efforts have been devoted to determining biophysical variables of the forest based on satellite measurements [23,24,25,26,27,28,29,30].
Normalized Difference Vegetation Index (NDVI) is the most commonly used vegetation index, which measures fractional absorbed photosynthetically active radiation [31]. The data derived from MODIS (Moderate-Resolution Imaging Spectroradiometer aboard the Aqua satellite) observations are used to produce NDVI time series at a spatial resolution of 0.25 km. However, the time period covered by these data is too short (since 2000) compared to the far longer series of the data obtained from tree rings.
Various vegetation indices determined using satellite data are effective indicators of tree photosynthetic activity and productivity, and a number of studies reported attempts to relate directly the data on TRW with the satellite data [32,33,34,35,36,37,38]. Those studies found a moderate positive correlation between the NDVI interannual variations and annual growth of the tree. There were other studies, though they did not find any significant positive correlation [35,36,39,40,41]. The relationship between the satellite vegetation indices and TRW is not obvious; it is determined by the species composition of the forest, landscape, and the choice of the proper initial satellite and meteorological data [42].
In most of the studies cited above, the authors analyzed correlations between the time series of TRW data and NDVI composite data for certain periods (e.g., 2 weeks or 1 month). The use of statistical methods, though, is not the best approach to revealing the relationship between NDVI and tree growth. For instance, annual wood formation is more strongly related to the net primary production (NPP) calculated over long periods, as the increase in tree biomass is not directly determined by the rate of photosynthesis but is rather associated with stems, branches, and roots growing and forming wood [43,44]. Thus, the storage of the long-lived phytomass in the woody parts of the trees is physically and temporally separated from the growth of leaves, identified by NDVI. Therefore, the different climatic conditions necessary for wood formation and for photosynthetic activity may disconnect the NDVI–growth relationship, as higher temperatures are needed for growth than for photosynthesis [45,46]. However, in boreal forests, NDVI indicates the physiological status of the trees, and, thus, the state of the tree crowns in summer (June and July) is a more important factor for tree growth than the length of the growing season [47].
As regards tree growth research, it can be conducted at a tree level, using, e.g., dendrochronology methods, or at a level of the tree stand or plot–to determine total biomass of trees or stands using various remote sensing techniques [48]. However, in the attempt to combine both approaches, researchers are faced with the challenge of differences in the time scales of crown formation processes, leaf photosynthesis (observed by satellites), and wood formation processes (integrated in tree rings). The response of photosynthesis to weather is relatively instantaneous, but there is a well-documented time lag between carbon absorption and increase in biomass of photosynthetic carbon [49]. Therefore, the TRW data and vegetation indices cannot be expected to be completely consistent with each other, as the space and time scales of TRW time series and satellite indices are different [46,50].
The reason for the spatial mismatch of the scales is that satellite data for NDVI (MODIS) are a pixel of land surface of an area of 6250 m2, whereas the TRW data are obtained for one tree with a crown projection area of up to 250 m2, and these data may vary considerably within the pixel. Another problem arises from the temporal mismatch between photosynthesis processes observed by satellite and wood formation processes integrated in TRW. For these reasons, no direct correspondence between TRW data and satellite vegetation indices can be expected. These processes will vary between ecosystems, tree species, and climate regions [2,3,4].
The search for interactions between the data on radial stem growth and the satellite-derived NDVI and LST (Land Surface Temperature) gives rise to the problem of selecting significant variables for analysis. For each year, radial stem growth is characterized by one value, while annual satellite data on NDVI and LST are time series. For instance, 8-day seasonal composites of NDVI and LST data consist of 40–50 points. It is rather difficult to find correlations between one number and function. Attempts to choose a certain number of variables in NDVI and LST time series and construct multiple regression equations of relationships between tree ring widths and the selected remote sensing data are hampered by the uncertainty about the choice of remote sensing variables for analysis. The problem is that for the time series of remote sensing data, one can choose from a great number of key variables or introduce additional variables such as the sums of temperatures for a certain period (but which?) of the season. The present study deals with the methods of constructing models of the relationship between the increase in woody phytomass and remote sensing indicators based on using integrated parameters of NDVI and LST time series transformed by principal component analysis.

2. Materials and Methods

Dynamics of radial stem growth was studied and phenological data were collected on sample plots in pine (Pinus sylvestris L.) forests located around the city of Krasnoyarsk. Trees aged about 100 years were selected to analyze radial stem growth on all sample plots. The trees had the stems of diameter at breast height ranging from 12 to 32 cm, and the average stand density was about 600–800 trees per ha. Time series of radial stem growth between 2003 and 2015 were analyzed. This time interval was chosen because the remote sensing data were only available from 2003. Coordinates of sample plots and characterization of tree stands are provided in Table 1.
An increment borer was used to extract core samples from all trees on the sample plots. The tree ring width d(t) in year t was measured on the polished surfaces of the core samples in the field of view of the microscope in the “Lintab 5 Tree-RingStation” (RINNTECH®) (version No. 5) with an accuracy of 0.1 mm. The dating of tree rings was carried out visually, using graphs in the TSAP-WinTM (RINNTECH®) software, and control of the dating was performed using the COFECHA program (version 6.0P) from the dendrochronology program library, DPL [51]. Figure 1 shows the positions of sample plots.
For the variables analyzed over the time interval selected in this study, no strong trend of the NDVI and LST data was observed, but there was a temporal trend of TRW. A typical TRW time series with trend is shown in Figure 2 (curve 1).
The procedure of estimating the relationships between the tree growth and NDVI and LST data consisted of several stages. In Stage 1, the temporal trend of TRW was excluded from the data on radial stem growth. Therefore, for each tree on each sample plot, time series of the TRW first differences (TRW FD) between years were calculated: TRW FD(t) = TRW(t) − TRW(t − 1). In subsequent calculations, the TRW FD time series, which were stationary time series, were used instead of the time series of tree ring widths (Figure 1, curve 2). The TRW FD time series was actually the rate of change of TRW. The time series of the first differences were calculated for each tree, and then average first differences were calculated for each sample plot for each year between 2003 and 2015. Table S1 provides data on the average radial stem growth for all sample plots during 1987–2015.
To estimate the relationships between the current value of radial stem growth and the corresponding values for the previous years, these relationships will be considered as the autoregressive equation:
T R W   F D ( t ) = z 0 + ∑ w k z w T R W   F D ( t − w )
where k is the order of the autocorrelation.
To estimate the order of the autocorrelation for the time series of the radial stem growth on the sample plots between 1987 and 2015, partial autocorrelation function (PACF) was calculated. Figure 3 shows the PACF for the TRW FD data on sample plot 1.
PACF is significant only at k = 1, i.e., the current value of TRW FD is only determined by its previous value (Figure 3). Calculations of PACF for other sample plots result in the same value of k = 1, i.e., the current value of TRW FD(t) correlates with the value of TRW FD(t − 1) in the previous year.
In Stage 2, integrated parameters characterizing NDVI and LST were considered. In analysis of remote sensing data, NDVI was determined using the standard formula as:
N D V I = N I R − Re d N I R + Re d ,
where NIR (0.620–0.670 µm) and Red (0.841–0.876 µm) are normalized values of the intensity of near-infrared and red light reflected by the land surface at a given point.
NDVI values were determined using the MYD09Q1 product of the MODIS (or Moderate Resolution Imaging Spectroradiometer), a key instrument aboard the Terra (originally known as EOS AM-1) and Aqua international research satellite. The MYD09Q1 product included values of the red and infrared components of the reflected radiation for a 250 m × 250 m pixel. For each pixel, an 8-day composite was used, taking into account atmospheric conditions. The data for 2003–2017 were obtained from the Earth Observing System Data and Information System EOSDIS resource [52]. The data on the land surface temperature (LST) were obtained for the same period (the MYD11A2 product has 8-day composite at a spatial resolution of 1 km).
Complete seasonal data of the MYD11A2, MYD09Q1, MOD11A2, and MOD09Q1 products on the Terra and Aqua satellites are available for the study area from 2003.
We assumed that NDVI data contain features of phenological dynamics of each growing season [53] and LST data characterize weather changes. NDVI and LST seasonal dynamics will be described using nonlinear regression equations, y ( n ) = a 1 n 2 + a 2 n + a 3 , where n is the first 8-day composite from the beginning of the year. This approximation is possible for boreal forests if the winter period is excluded. The annual curve of NDVI is used to find the n1 and n2 dates, which correspond to the beginning and end of the growing season determined for the study tree stands (when NDVI was stably higher than 0.35 of the NDVI maximum). Approximation of the NDVI and LST time series is done on the (n1 − n2) section. The NDVI time series was preliminarily smoothed using a rectangular filtering window of width of 3 composites to remove measurement artifacts.
A typical seasonal NDVI time series is shown in Figure 4 and an LST seasonal time series in Figure 5. Quadratic regression equations of the NDVI and LST time series for the growing season were calculated for each sample plot using the data for the 2003–2015 period. The accuracy of the parabolic approximation is described by the coefficient of determination, R2, which characterizes the proportion of the variance in the time series of remote sensing data accounted for by regression. Parameters of approximation of the NDVI and LST seasonal curves NDVI(k,j,n) = a1(k,j)n2 + a2(k,j)n − a3(k,j) and LST seasonal curves LST(k,n,j) = b1(k,j)n2 + b2(k,j)n − b3(k,j) for sample plot k (k = 1, … 4), in season j (j = 2003, …, 2015) and dates n can be used to estimate the relationships between the remote sensing data and the growth of the tree ring width. Typical time series of NDVI and LST are shown in Figure 4 and Figure 5.
Calculations of parameters of the NDVI and LST seasonal time series and validation of parabolic models using Fisher’s test for all study years and all sample plots are provided in Tables S2 and S3.
In Stage 3, the number of remote sensing variables was reduced. For this, parameters of NDVI and LST regression time series (coefficients of nonlinear regression equations and 1 − R2 values, where R2 is coefficient of determination of the regression equation for an individual season j (j = 2003,…2015) for 13 seasons for every sample plot k in the present study) were used to construct matrices A(k) = ‖a(i,j)‖ and B(k) = ‖b(i,j)‖. In these matrices, i = 1, …, 4 are numbers of parameters a1–a3 in equations of seasonal dynamics for NDVI and a4 for variable (1 − R N D V I 2 ( j , k ) ) ; b1 − b3 in equation of seasonal dynamics for LST; and b4 for variable (1 − R L S T 2 ( j , k ) ) ; j = 2003, …, 2015 is the year from 2003 to 2015. R2 and (1 − R2) are interrelated, and either can be used. Principal component analysis was used to make the transition from the variables of regression equations to the new independent variables. The new variables were constructed as the sum of initial variables with certain weights assigned to them. Principal component analysis is used to find eigenvalues characterizing the contribution of each new variable to the explanation of data variance. Further calculations may involve not all new variables but only those that make the greatest contribution to the explanation of data variance. Thus, the number of explanatory variables in the model of NDVI and LST dependence of tree ring width can be reduced. Table 2 provides characterization of principal components for sample plot No. 1.
Based on Table 2, the value of the first principal component, G1(t), can be expressed in the following equation:
G 1 ( t ) = 0.978 ⋅ a 1 ( t ) − 0.987 ⋅ a 2 ( t ) + 0.971 ⋅ a 3 ( t ) + 0.710 ⋅ ( 1 − R N D V I 2 ( t ) ) .
The values of other principal components can be determined in the same way from Table 2. Table 3 provides contributions of the first eigenvalues to the total NDVI and LST variance for all sample plots.
In all cases, the first principal components for NDVI and LST account for at least 72% of variance, and the first two principal components taken together account for at least 96% of variance (Table 3); thus, two variables, G1 and H1, or four variables, G1, G2, H1, and H2, can be used instead of eight. Figure 6 shows the curves indicating temporal dynamics of the first principal components G1(t) for the NDVI matrices (curve 1), H1(t) for LST matrices (curve 2), and the time series of the first differences of radial stem growth, TRW FD(t) (curve 3) for sample plot 1.
After such transformations, one can construct linear regression models of relationships between the first differences of radial stem growth and values of the principal components of matrices A and B. The quality of the proposed models can be assessed by t-test for coefficients of the model, coefficient of determination, R2, of the model series, and cross-correlation function of the TRW FD time series and the model series, characterizing synchrony of change of these series. Results of the calculations were used as the basis for selecting variables for constructing models of the relationship between TRW FD, NDVI, and LST.

3. Results

To construct the model of radial growth, we take into account the relationship of the current year growth, TRW(t), to the growth of the previous year, TRW(t − 1), determined by calculating PACF, and the effect of the weather will be characterized by the values of the first principal components, G1(t) and H1(t). This model was tested using the ADL (autoregressive distributed lag) model:
T R W   F D ( t ) = c 0 + c 1 G 1 ( t ) + c 2 H 1 ( t ) + c 3 T R W   F D ( t − 1 ) .
As model (1) had passed the test, it was used to make calculations for all sample plots over the period considered in this study. Results are given in Table 4.
Cross-correlation function, CCF, was constructed to test time synchrony of the field and model data (Figure 7). The CCF maximum was observed at the delay k = 0, suggesting the high degree of synchrony between the field and model data series.
Almost all coefficients for sample plots were significant (Table 5). The significance level of p < 0.1 was used, as some of the factors influencing phytomass increase (precipitation and soil conditions) cannot be estimated using remote sensing measurements, although they do affect the phytomass increase. Although satellite microwave data ASMR-E [54] can be used to estimate precipitation and soil moisture, the low spatial and temporal resolution and accuracies of such data are not appropriate for this research.
Susceptibility of the mean radial stem growth to the change of the first principal component for the NDVI data is positive, i.e., as G1 increases, TRW FD(t) increases as well. By contrast, susceptibility of TRW FD(t) to H1 is negative. For almost all sample plots, susceptibility of TRW FD(t) to the change of the mean radial growth in the preceding year is negative.
The LST and NDVI data for the period of tree dormancy were excluded from the time series of remote sensing data and were not used in calculations. Only the growing season data were taken into account. The reduction in NDVI and LST date is needed to make it possible to compare the time series of TRW with time series of remote sensing data.
Thus, when only satellite data were used to investigate growth processes in tree stands, no direct relationship was found between integrated parameters of NDVI and LST and radial stem growth rate.

4. Discussion

The signs of coefficients of Equation (2), which characterize susceptibility of the first differences of radial growth to changes in the values of independent variables in model (2), differ between sample plots. For example, for H1 on all sample plots (except No. 3), the variables have the negative sign, i.e., the larger G1, the lower the first differences of radial growth.
It could be assumed that TRW FD, D(t), might be related to not only the TRW FD value of the preceding year, D(t − 1), but also phenological and weather parameters of the preceding year G1(t − 1) and H1(t − 1). However, for model T R W   F D ( t ) = c 0 + c 1 G 1 ( t − 1 ) + c 2 H 1 ( t − 1 ) + c 3 T R W   F D ( t − 1 ) , coefficients c1 and c2 are nonsignificant even at p = 0.50, and the proportion of variance accounted by this model is only 0.30, i.e., the current TRW FD is not dependent on these parameters.
As the sample plots are located rather close to each other, the tree stands have comparable species compositions, and weather conditions are similar as well; therefore, the average values of G1, H1, and TRW FD(t) can be used in the model. Parameters of the averaged model (2) are listed in Table 5.
Coefficients of model (2) for averaged data are significant at p < 0.03, coefficient of determination reaches 0.70, and CCF(k = 0) = 0.84, i.e., the field and model data series are practically time synchronous (Table 5).
Both amplitudes and phases of the averaged field data and model data are in good agreement (Table 5). For some of the sample plots, model (2) accounts for more than 50% of the variance of long-term data (Table 4). For the data averaged over all sample plots, model (2) accounts for 70% of the data variance. Coefficients of Equation (2) are significant for both individual sample plots and the averaged data, characterizing susceptibility of growth processes to changes in independent variables of this equation.
To determine properties of tree assemblages on the sample plot, it is important to assess temporal synchrony between tree growth processes in the assemblage. The synchrony of changes in TRW FD values of individual trees could be assessed using the CCF(x,y) matrix of the series of radial growth first differences between trees x and y. Radial growth can be regarded as synchronous if the highest values of CCF in this matrix are close to 1 and are observed at delay k = 0. For a sample of 25–30 trees, it would be necessary to calculate about 400 paired CCF, and, thus, it may be more convenient to use the correlation matrix of the first differences of radial stem growth. The correlation coefficient r(x,y) of the TRW FD series for trees x and y corresponds to the value of CCF(x, y) at delay k = 0. If r(x,y) is close to 1, these time series can be considered as synchronous. The average value of correlation coefficients r(x,y) or the function of distribution f(r) of correlation coefficients for the tree assemblage can be used as an indicator of the synchrony of radial stem growth in the tree assemblage. For the tree assemblage with the high level of synchrony, the function of distribution of correlation coefficients will peak close to r = 1. Figure 8 shows the function of distribution f ( r ) of correlation coefficients for different sample plots.
The TRW FD time series on sample plots No. 3 (average correlation coefficient 0.578) and No. 1 (average correlation coefficient 0.571) are rather synchronous (Figure 8). The TRW FD time series on sample plots No. 4 (average correlation coefficient 0.232) and No. 2 (average correlation coefficient 0.448) are considerably less synchronous, and changes in some of the TRW FD time series on these sample plots are in antiphase, resulting in negative correlation coefficients. The possible desynchronization of radial stem growth on sample plots should be taken into account when determining the average values of TRW FD. Such phase shifts of the TRW FD time series of individual trees may cause a decrease in coefficients of determination in the models of current TRW FD.
Choosing relevant parameters for models connecting ecological characteristics (such as growth processes) with the environmental factors (such as weather) is always a challenge. The choice is difficult because of the diversity of the parameters, e.g., temperature and precipitation for a certain period of the season or the totals for the entire season. The approach to choosing independent model parameters proposed in this study makes it possible to limit the number of parameters to the values of two first principal components of approximated NDVI and LST series, considerably simplifying the analysis. The test of the accuracy of a model can be based on, first, coefficient of determination, R2, characterizing the similarity between the field and model data by amplitudes, and, second, cross-correlation function, characterizing the synchrony between the dynamics of the field and model time series.
The relationships between remote sensing data (NDVI) and radial stem growth were studied by a number of researchers [39,47]. For example, NOAA AVHRR PAL time series were used to analyze evolution of vegetation in the Komi Republic (the Northwest of Russia) between 1982 and 2001. The authors revealed statistically significant correlation (R2 = 0.44–0.59) between the Normalized Difference Vegetation Index (NDVI) and tree ring width [37]. At the same time, comparison of the chronology of tree ring width (TRW) and remote sensing data did not reveal any significant correlation between NDVI and TRW [35].
There may be various reasons for the disagreement between the conclusions reached by different researchers. For instance, most of the studies used the NDVI data obtained for a 64-km2 pixel, and a synchronous NDVI signal can hardly be detected on such an area. The forest vegetation on the study plot was not examined for uniformity, and the synchronicity of the time series of the radial stem growth was not estimated for individual trees. However, the present study demonstrates that as the level of synchronicity of the radial stem growth of individual trees decreases, correlation between the NDVI and TRW data becomes weaker. Moreover, analysis of the correlation between the time series of NDVI and TRW seasonal dynamics should take into consideration that if the dynamics are similar, correlation may be strong while there may be no functional relationship between the parameters.
The present study considered uniform tree stands on a 250 × 250 m2 pixel and estimated the synchronicity of the time series of radial stem growth in individual trees on the sample plot. Results of the study suggest that even on the small pixel area, the relationship between NDVI and TRW may depend on the synchronicity of the processes of radial stem growth, and both strong and rather weak correlations can be obtained between NDVI and TRW. Thus, research on relationships between woody plant growth and remote sensing data should take into account specific spatiotemporal dynamics of tree growth.

5. Conclusions

The model proposed in this study relates the average value of TRW FD for the tree stand to the current seasonal integrated remote sensing parameters NDVI and LST. Parameters NDVI and LST were determined from the curves of the temporal dynamics of time series using transformation of principal components. However, no direct relationship was revealed between TRW FD and remote sensing parameters for the tree stands on the study sample plots. The current modelled value of TRW FD was only determined by the total principal components of remote sensing parameters for the season and TRW FD of the preceding year. The first-order ADL-model, i.e., the model with the discrete-time Markov process, was used for analysis. This model is able to find the relationships between tree growth and NDVI and LST.
The present study shows that approximately 50–70% of the current radial stem growth is determined by the growth of the previous year (i.e., has Markov properties) and integrated properties of NDVI and LST. Both the current year growth and the growth of the previous year are certainly affected by weather factors, but the effect of the weather of previous years in the proposed ADL equation is actually represented by the parameter of radial stem growth of the previous year. Moreover, rather than determining the period in the season when the effect on the radial stem growth is the greatest (which may be a problem), one can use a very simple procedure of calculating principal components of the entire seasonal curve. The value of (1 − R2) can be used to estimate the contribution of other, unknown, components to the growth. In such calculations, it is important to take into account the synchrony of radial stem growth processes.

Supplementary Materials

The following are available online at https://www.mdpi.com/2073-4433/12/1/12/s1. Table S1 Average values of TRW FD (mm) for all sample plots, Table S2 Parameters of regression equation for seasonal dynamics of LST*, Table S3 Parameters of regression equation for seasonal dynamics of NDVI.

Author Contributions

Conceptualization, Y.I.; data curation, A.K.; investigation, Y.I. and A.K.; methodology, V.S.; software, A.K.; supervision, V.S.; writing–original draft, V.S.; writing–review and editing, Y.I. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by RFBR according to the research project number 18-04-00119-a.

Acknowledgments

The authors are grateful to Elena Krasova for translating the article.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Dobbertin, M. Tree growth as indicator of tree vitality and of tree reaction to environmental stress: A review. Eur. J. For. Res. 2005, 124, 319–333. [Google Scholar] [CrossRef]
  2. Barber, V.A.; Juday, G.P.; Finney, B.P. Reduced growth of Alaskan white spruce in the twentieth century from temperature-induced drought stress. Glob. Planet. Chang. 2008, 60, 289–305. [Google Scholar] [CrossRef]
  3. Lloyd, A.H.; Fastie, C.L. Spatial and temporal variability in the growth and climate response of treeline trees in Alaska. Clim. Chang. 2002, 52, 481–509. [Google Scholar] [CrossRef]
  4. Hirota, M.; Holmgren, M.; Van Nes, E.H.; Scheffer, M. Global resilience of tropical forest and savanna to critical transitions. Science 2011, 334, 232–235. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  5. Sarris, D.; Christodoulakis, D.; Körner, C. Recent decline in precipitation and tree growth in the eastern Mediterranean. Glob. Chang. Biol. 2007, 13, 1187–1200. [Google Scholar] [CrossRef]
  6. Williams, A.P.; Allen, C.D.; Macalady, A.K.; Griffin, D.; Woodhouse, C.A.; Meko, D.M.; Swetnam, T.W.; Rauscher, S.A.; Seager, R.; Grissino-Mayer, H.D.; et al. Temperature as a potent driver of regional forest drought stress and tree mortality. Nat. Clim. Chang. 2013, 3, 292–297. [Google Scholar] [CrossRef]
  7. Shestakova, T.A.; Gutiérrez, E.; Kirdyanov, A.V.; Camarero, J.J.; Génova, M.; Knorre, A.A.; Linares, J.C.; de Dios, V.R.; Sánchez-Salguero, R.; Voltas, J. Forests synchronize their growth in contrasting Eurasian regions in response to climate warming. Proc. Natl. Acad. Sci. USA 2016, 113, 662–667. [Google Scholar] [CrossRef] [Green Version]
  8. Sass-Klaassen, U.G.W.; Fonti, P.; Cherubini, P.; Gričar, J.; Robert, E.M.R.; Steppe, K.; Bräuning, A. A tree-centered approach to assess impacts of extreme climatic events on forests. Front. Plant Sci. 2016, 7, 1069. [Google Scholar] [CrossRef] [Green Version]
  9. Fritts, H.C. Tree Rings and Climate; Academic Press: London, UK, 1976; p. 567. [Google Scholar]
  10. Zuidema, P.A.; Frank, D. Forests: Tree rings track climate trade-offs. Nature 2015, 523, 531. [Google Scholar] [CrossRef] [Green Version]
  11. Babst, F.; Poulter, B.; Bodesheim, P.; Mahecha, M.D.; Frank, D.C. Improved tree-ring archives will support earth-system science. Nat. Ecol. Evol. 2017, 1, 1–2. [Google Scholar] [CrossRef]
  12. Gewehr, S.; Drobyshev, I.; Berninger, F.; Bergeron, Y. Soil characteristics mediate the distribution and response of boreal trees to climatic variability. Can. J. For. Res. 2014, 498, 487–498. [Google Scholar] [CrossRef] [Green Version]
  13. Vaganov, E.A.; Hughes, M.K.; Shashkin, A.V. Growth Dynamics of Conifer Tree Rings: Images of Past and Future Environments; Springer: New York, NY, USA, 2006; Volume 183, p. 354. [Google Scholar]
  14. Ketterings, Q.M.; Coe, R.; Van Noordwijk, M.; Ambagau, Y.; Palm, C.A. Reducing uncertainty in the use of allometric biomass equations for predicting above-ground tree biomass in mixed secondary forests. For. Ecol. Manag. 2001, 146, 199–209. [Google Scholar] [CrossRef]
  15. Van Breugel, M.; Ransijn, J.; Craven, D.; Bongers, F.; Hall, J.S. Estimating carbon stock in secondary forests: Decisions and uncertainties associated with allometric biomass models. For. Ecol. Manag. 2011, 262, 1648–1657. [Google Scholar] [CrossRef]
  16. Babst, F.; Bouriaud, O.; Alexander, R.; Trouet, V.; Frank, D. Toward consistent measurements of carbon accumulation: A multi-site assessment of biomass and basal area increment across Europe. Dendrochronologia 2014, 32, 153–161. [Google Scholar] [CrossRef]
  17. St. George, S. The global network of tree-ring widths and its applications to paleoclimatology. Pages 2014, 22, 16–17. [Google Scholar] [CrossRef] [Green Version]
  18. Babst, F.; Poulter, B.; Trouet, V.; Tan, K.; Neuwirth, B.; Wilson, R.; Carrer, M.; Grabner, M.; Tegel, W.; Levanic, T.; et al. Site- and species-specific responses of forest growth to climate across the European continent. Glob. Ecol. Biogeogr. 2013, 22, 706–717. [Google Scholar] [CrossRef]
  19. Baral, S.; Gaire, N.P.; Aryal, S.; Pandey, M.; Rayamajhi, S.; Vacik, H. Growth Ring Measurements of Shorea robusta Reveal Responses to Climatic Variation. Forests 2019, 10, 466. [Google Scholar] [CrossRef] [Green Version]
  20. Drobyshev, I.; Gewehr, S.; Berninger, F.; Bergeron, Y. Species specific growth responses of black spruce and trembling aspen may enhance resilience of boreal forest to climate change. J. Ecol. 2013, 101, 231–242. [Google Scholar] [CrossRef] [Green Version]
  21. Mérian, P.; Pierrat, J.-C.; Lebourgeois, F. Effect of sampling effort on the regional chronology statistics and climate-growth relationships estimation. Dendrochronologia 2013, 31, 58–67. [Google Scholar] [CrossRef]
  22. Camarero, J.J.; Franquesa, M.; Sangüesa-Barreda, G. Timing of drought triggers distinct growth responses in holm oak: Implications to predict warming-induced forest defoliation and growth decline. Forests 2015, 6, 1576–1597. [Google Scholar] [CrossRef] [Green Version]
  23. Vicente-Serrano, S.M. Evaluating the impact of drought using remote sensing in a Mediterranean, semi-arid region. Nat. Hazards 2007, 40, 173–208. [Google Scholar] [CrossRef]
  24. Büntgen, U.; Trouet, V.; Frank, D.; Leuschner, H.H.; Friedrichs, D.; Luterbacher, J.; Esper, J. Tree-ring indicators of German summer drought over the last millennium. Quat. Sci. Rev. 2010, 29, 1005–1016. [Google Scholar] [CrossRef]
  25. Hall, F.G.; Botkin, D.B.; Strebel, D.E.; Woods, K.D.; Goetz, S.J. Large-scale patterns of forest succession as determined by remote sensing. Ecology 1991, 72, 628–640. [Google Scholar] [CrossRef]
  26. Song, C.; Woodcook, C.E. Monitoring forest succession with multitemporal landsat images: Factors of uncertainty. IEEE Trans. Geosci. Remote Sens. 2003, 41, 2557–2567. [Google Scholar] [CrossRef]
  27. Tucker, C.J.; Sellers, P.J. Satellite remote sensing of primary production. Int. J. Remote Sens. 1986, 7, 1395–1416. [Google Scholar] [CrossRef]
  28. Lu, D. The potential and challenge of remote sensing-based biomass estimation. Int. J. Remote Sens. 2006, 27, 1297–1328. [Google Scholar] [CrossRef]
  29. Timothy, D.; Onisimo, M.; Cletah, S.; Adelabu, S.; Tsitsi, B. Remote sensing of above ground forest biomass: A review. Trop. Ecol. 2016, 57, 125–132. [Google Scholar]
  30. Crowther, T.W.; Glick, H.B.; Covey, K.R.; Bettigole, C.; Maynard, D.S.; Thomas, S.M.; Smith, J.R.; Hintler, G.; Duguid, M.C.; Amatulli, G.; et al. Mapping tree density at a global scale. Nature 2015, 525, 201–205. [Google Scholar] [CrossRef] [PubMed]
  31. Myneni, R.B.; Hall, F.G.; Sellers, P.J.; Marshak, A.L. The meaning of spectral vegetation indices. IEEE Trans. Geosci. Remote Sens. 1995, 33, 481–486. [Google Scholar] [CrossRef]
  32. Babst, F.; Bodesheim, P.; Charney, N.; Friend, A.D.; Girardin, M.P.; Klesse, S.; Moore, D.J.P.; Seftigen, K.; Björklund, J.; Bouriaud, O.; et al. When tree rings go global: Challenges and opportunities for retro- and prospective insight. Quat. Sci. Rev. 2018, 197, 1–20. [Google Scholar] [CrossRef]
  33. Bunn, A.G.; Hughes, M.K.; Kirdyanov, A.V.; Losleben, M.; Shishov, V.V.; Berner, L.T.; Oltchev, A.; Vaganov, E.A. Comparing forest measurements from tree rings and a space-based index of vegetation activity in Siberia. Environ. Res. Lett. 2013, 8, 035034. [Google Scholar] [CrossRef]
  34. Vicente-Serrano, S.M.; Camarero, J.J.; Olano, J.M.; Martín-Hernández, N.; Peña-Gallardo, M.; Tomás-Burguera, M.; Gazol, A.; Azorin-Molina, C.; Bhuyan, U.; Kenawy, A. Diverse relationships between forest growth and the Normalized Difference Vegetation Index at a global scale. Remote Sens. Environ. 2016, 187, 14–29. [Google Scholar] [CrossRef] [Green Version]
  35. Beck, P.S.A.; Andreu-Hayles, L.; D’Arrigo, R.D.; Anchukaitis, K.J.; Tucker, C.J.; Pinzón, J.E.; Goetz, S.J. A large-scale coherent signal of canopy status in maximum latewood density of tree rings at arctic treeline in North America. Glob. Planet. Chang. 2013, 100, 109–118. [Google Scholar] [CrossRef]
  36. Berner, L.T.; Beck, P.S.A.; Bunn, A.G.; Goetz, S.J. Plant response to climate change along the forest-tundra ecotone in northeastern Siberia. Glob. Chang. Biol. 2013, 19, 3449–3462. [Google Scholar] [CrossRef] [PubMed]
  37. Lopatin, E.; Kolström, T.; Spiecker, H. Determination of forest growth trends in Komi Republic (northwestern Russia): Combination of tree-ring analysis and remote sensing data. Boreal Environ. Res. 2006, 11, 341–353. [Google Scholar]
  38. Decuyper, M.; Chávez, R.O.; Copini, P.; Sass-Klaassen, U. A multi-scale approach to assess the effect of groundwater extraction on Prosopis tamarugo in the Atacama Desert. J. Arid. Environ. 2016, 131, 25–34. [Google Scholar] [CrossRef]
  39. D’Arrigo, R.D.; Malmstrom, C.M.; Jacoby, G.C.; Los, S.O.; Bunker, D.E. Correlation between maximum latewood density of annual tree rings and NDVI based estimates of forest productivity. Int. J. Remote Sens. 2000, 21, 2329–2336. [Google Scholar] [CrossRef]
  40. Beck, P.S.A.; Goetz, S.J. Satellite observations of high northern latitude vegetation productivity changes between 1982 and 2008: Ecological variability and regional differences. Environ. Res. Lett. 2011, 6, 045501. [Google Scholar] [CrossRef]
  41. Pasho, E.; Alla, A.Q. Climate impacts on radial growth and vegetation activity of two co-existing Mediterranean pine species. Can. J. For. Res. 2015, 45, 1748–1756. [Google Scholar] [CrossRef] [Green Version]
  42. Kaufmann, R.K.; D’Arrigo, R.D.; Paletta, L.F.; Tian, H.Q.; Jolly, W.M.; Myneni, R.B. Identifying climatic controls on ring width: The timing of correlations between tree rings and NDVI. Earth Interact. 2008, 12, 1–14. [Google Scholar] [CrossRef] [Green Version]
  43. Huston, M.A.; Wolverton, S. The global distribution of net primary production: Resolving the paradox. Ecol. Monogr. 2009, 79, 343–377. [Google Scholar] [CrossRef]
  44. Soukhovolsky, V.; Ivanova, Y. Modeling production processes in forest stands: An adaptation of the Solow growth model. Forests 2018, 9, 391. [Google Scholar] [CrossRef] [Green Version]
  45. Suni, T.; Rinne, J.; Reissell, A.; Altimir, N.; Keronen, P.; Rannik, Ü.; Dal Maso, M.; Kulmala, M.; Vesala, T. Long-term measurements of surface fluxes above a Scots pine forest in Hyytiälä, southern Finland, 1996–2001. Boreal Environ. Res. 2003, 8, 287–301. [Google Scholar]
  46. Rossi, S.; Morin, H.; Deslauriers, A.; Plourde, P.Y. Predicting xylem phenology in black spruce under climate warming. Glob. Chang. Biol. 2011, 17, 614–625. [Google Scholar] [CrossRef]
  47. Kaufmann, R.K.; D’Arrigo, R.D.; Laskowski, C.; Myneni, R.B.; Zhou, L.; Davi, N.K. The effect of growing season and summer greenness on northern forests. Geophys. Res. Lett. 2004, 31, 3–6. [Google Scholar] [CrossRef]
  48. Avitabile, V.; Herold, M.; Heuvelink, G.B.M.; Lewis, S.L.; Phillips, O.L.; Asner, G.P.; Armston, J.; Ashton, P.S.; Banin, L.; Bayol, N.; et al. An integrated pan-tropical biomass map using multiple reference datasets. Glob. Chang. Biol. 2016, 22, 1406–1420. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  49. Cuny, H.E.; Rathgeber, C.B.K.; Frank, D.; Fonti, P.; Makinen, H.; Prislan, P.; Rossi, S.; Martinez Del Castillo, E.; Campelo, F.; Vavrčik, H.; et al. Woody biomass production lags stem-girth increase by over one month in coniferous forests. Nat. Plants 2015, 1, 6. [Google Scholar] [CrossRef]
  50. Bhuyan, U.; Zang, C.; Vicente-Serrano, S.M.; Menzel, A. Exploring Relationships among Tree-Ring Growth, Climate Variability, and Seasonal Leaf Activity on Varying Timescales and Spatial Resolutions. Remote Sens. 2017, 9, 526. [Google Scholar] [CrossRef] [Green Version]
  51. Available online: http://web.utk.edu/~grissino/software.htm (accessed on 10 July 2020).
  52. Available online: https://earthdata.nasa.gov (accessed on 20 June 2020).
  53. Ivanova, Y.; Kovalev, A.; Yakubailik, O.; Soukhovolsky, V. The Use of Satellite Information (MODIS/Aqua) for Phenological and Classification Analysis of Plant Communities. Forests 2019, 10, 561. [Google Scholar] [CrossRef] [Green Version]
  54. Available online: https://aqua.nasa.gov/amsr-e (accessed on 10 July 2020).
Atmosphere 12 00012 g001 550
Figure 1. Positions of sample plots No. 1–4 around the city of Krasnoyarsk.
Figure 1. Positions of sample plots No. 1–4 around the city of Krasnoyarsk.
Atmosphere 12 00012 g001
Atmosphere 12 00012 g002 550
Figure 2. Temporal dynamics of tree-ring widths (TRW) (curve 1) and TRW first differences (FD) (curve 2), sample plot No. 1.
Figure 2. Temporal dynamics of tree-ring widths (TRW) (curve 1) and TRW first differences (FD) (curve 2), sample plot No. 1.
Atmosphere 12 00012 g002
Atmosphere 12 00012 g003 550
Figure 3. Partial autocorrelation function of the time series (TRW FD) of the average radial stem growth for the trees on sample plot 1. Here, 1–partial autocorrelation function (PACF), 2–standard error of PACF.
Figure 3. Partial autocorrelation function of the time series (TRW FD) of the average radial stem growth for the trees on sample plot 1. Here, 1–partial autocorrelation function (PACF), 2–standard error of PACF.
Atmosphere 12 00012 g003
Atmosphere 12 00012 g004 550
Figure 4. A typical Normalized Difference Vegetation Index (NDVI)(n) time series, sample plot No. 2 (2014) and processing procedure. 1–a series of 8-day NDVI time composites; 2–a filtered NDVI time series; 3–series 2 clipped at critical values NDVIcrit = 0.35(NDVImax–NDVImin); 4–parabolic approximation of NDVI series; n1, n2-dates of the beginning and end of the growing season, when NDVI(n) > 0.5 NDVImax(n).
Figure 4. A typical Normalized Difference Vegetation Index (NDVI)(n) time series, sample plot No. 2 (2014) and processing procedure. 1–a series of 8-day NDVI time composites; 2–a filtered NDVI time series; 3–series 2 clipped at critical values NDVIcrit = 0.35(NDVImax–NDVImin); 4–parabolic approximation of NDVI series; n1, n2-dates of the beginning and end of the growing season, when NDVI(n) > 0.5 NDVImax(n).
Atmosphere 12 00012 g004
Atmosphere 12 00012 g005 550
Figure 5. A curve of Land Surface Temperature (LST)(n) seasonal dynamics, sample plot No. 2 (2014) and the series data processing procedure. 1–a series of 8-day LST time composites; 2–series 1 clipped at the dates of the beginning and end of the growing season, n1 and n2, determined for the NDVI(n) time series; 3–parabolic approximation of LST series.
Figure 5. A curve of Land Surface Temperature (LST)(n) seasonal dynamics, sample plot No. 2 (2014) and the series data processing procedure. 1–a series of 8-day LST time composites; 2–series 1 clipped at the dates of the beginning and end of the growing season, n1 and n2, determined for the NDVI(n) time series; 3–parabolic approximation of LST series.
Atmosphere 12 00012 g005
Atmosphere 12 00012 g006 550
Figure 6. Time series of the first principal components of NDVI G1(t) (curve 1), the first principal components of LST H1(t) (curve 2), and the first differences of the radial stem growth time series TRW FD(t) (curve 3) for sample plot 1.
Figure 6. Time series of the first principal components of NDVI G1(t) (curve 1), the first principal components of LST H1(t) (curve 2), and the first differences of the radial stem growth time series TRW FD(t) (curve 3) for sample plot 1.
Atmosphere 12 00012 g006
Atmosphere 12 00012 g007 550
Figure 7. Cross-correlation function (CCF) of the field and model data series for sample plot No. 1. 1–CCF; 2–Standard Error of CCF.
Figure 7. Cross-correlation function (CCF) of the field and model data series for sample plot No. 1. 1–CCF; 2–Standard Error of CCF.
Atmosphere 12 00012 g007
Atmosphere 12 00012 g008 550
Figure 8. The function of distribution f ( r ) of correlation coefficients of the TRW FD time series for different sample plots. Curve 1–sample plot No. 4; curve 2–sample plot No. 2; curve 3–sample plot No. 3; curve 4–sample plot No. 1.
Figure 8. The function of distribution f ( r ) of correlation coefficients of the TRW FD time series for different sample plots. Curve 1–sample plot No. 4; curve 2–sample plot No. 2; curve 3–sample plot No. 3; curve 4–sample plot No. 1.
Atmosphere 12 00012 g008
Table
Table 1. Coordinates of sample plots and characterization of tree stands.
Table 1. Coordinates of sample plots and characterization of tree stands.
Plot No.CoordinatesNumber of Trees Mean Annual TRW, mmPhotograph
156.043725° N
93.161901° E		300.95 ± 0.076 Atmosphere 12 00012 i001
256.222678° N
92.990675° E		291.92 ± 0.13 Atmosphere 12 00012 i002
355.994809° N
92.735070° E		321.31 ± 0.07 Atmosphere 12 00012 i003
455.963131° N
92.854473° E		290.98 ± 0.08 Atmosphere 12 00012 i004
Table
Table 2. Weights for principal components G1–G4 and H1–H4 and eigenvalues (sample plot No. 1).
Table 2. Weights for principal components G1–G4 and H1–H4 and eigenvalues (sample plot No. 1).
NDVI MatrixG1G2G3G4
a10.978−0.1530.1440.013
a2−0.9870.1550.0110.031
a30.971−0.203−0.1280.017
1−R2a0.7100.704−0.0080.001
Eigenvalue3.3770.5850.0370.001
LST matrixH1H2H3H4
b10.985−0.0090.1730.012
b2−0.9950.097−0.0170.022
b30.957−0.226−0.1810.010
1−R2b0.3430.938−0.0400.000
Eigenvalue2.9930.9420.0640.001
Table
Table 3. Contributions of the first two principal components, G1 and G2 for NDVI and H1 and H2 for LST, to the total NDVI and LST variance for different sample plots.
Table 3. Contributions of the first two principal components, G1 and G2 for NDVI and H1 and H2 for LST, to the total NDVI and LST variance for different sample plots.
NDVILST
Sample PlotG1G1 + G2H1H1 + H2
No. 184.4299.0474.83 98.37
No. 272.0796.0274.5299.07
No. 381.3799.0677.2299.83
No. 488.4698.9980.7899.72
Table
Table 4. Parameters of model (2) for sample plots.
Table 4. Parameters of model (2) for sample plots.
Coefficients Sample Plot
No. 1No. 2No. 3No. 4
c00.050–0.0300.0130.015
c10.049 *–0.0040.059 *0.052 *
c2–0.048 *0.052 *–0.089 *–0.069 *
c3–1.058 *–0.535 *–0.342 *0.445 *
R20.5100.4300.5000.580
CCF (k = 0)0.720.740.780.76
* coefficient is significant at p ≤ 0.10.
Table
Table 5. Parameters of model (2) based on the data averaged over the sample plots.
Table 5. Parameters of model (2) based on the data averaged over the sample plots.
VariableCoefficientStandard Errort(9)p-Level
c00.0100.0350.2920.777
G10.0890.0312.8190.020
H1–0.1420.053–2.6700.026
TRW FD(t − 1)–0.5100.183–2.7870.021
R20.70
p0.005
CCF(k = 0)0.84
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Ivanova, Y.; Kovalev, A.; Soukhovolsky, V. Modeling the Radial Stem Growth of the Pine (Pinus sylvestris L.) Forests Using the Satellite-Derived NDVI and LST (MODIS/AQUA) Data. Atmosphere 2021, 12, 12. https://doi.org/10.3390/atmos12010012

AMA Style

Ivanova Y, Kovalev A, Soukhovolsky V. Modeling the Radial Stem Growth of the Pine (Pinus sylvestris L.) Forests Using the Satellite-Derived NDVI and LST (MODIS/AQUA) Data. Atmosphere. 2021; 12(1):12. https://doi.org/10.3390/atmos12010012

Chicago/Turabian Style

Ivanova, Yulia, Anton Kovalev, and Vlad Soukhovolsky. 2021. "Modeling the Radial Stem Growth of the Pine (Pinus sylvestris L.) Forests Using the Satellite-Derived NDVI and LST (MODIS/AQUA) Data" Atmosphere 12, no. 1: 12. https://doi.org/10.3390/atmos12010012

APA Style

Ivanova, Y., Kovalev, A., & Soukhovolsky, V. (2021). Modeling the Radial Stem Growth of the Pine (Pinus sylvestris L.) Forests Using the Satellite-Derived NDVI and LST (MODIS/AQUA) Data. Atmosphere, 12(1), 12. https://doi.org/10.3390/atmos12010012

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop