Remote Sensing Indicators of Spongy Moth (Lymantria dispar L.) Damage to Birch Stands in Western Siberia

Abstract

The goal of this study is to detect indicators of damage to birch stands in western Siberia by spongy moth (Lymantria dispar L.) using remote sensing methods. The need for such indicators is due to the fact that the size of the study area is about 1 million square kilometers, and ground methods are too laborintensive. It is crucial for these indicators to differentiate the effects of insects from other destructive factors like fires and droughts. During the 2021–2022 fieldwork, we identified 18 areas for trial (intensive damage due to caterpillars) and control. For each area, we obtained seasonal time-series data of vegetation index NDVI mean values within its boundaries. We acquired the data from a Sentinel-2 satellite with a spatial resolution of 10 m. Relative reduction indices of NDVI during the season were introduced for pairs of damaged–control plots. We also considered the effect of foliage regeneration on damaged trees. The obtained indicators demonstrate increased analytical significance in identifying areas affected by pests compared to the simple reduction in vegetative indices.

1. Introduction

Outbreaks of forest pest species are the serious problem in forestry of many countries including Russia. Leaves feeding insects and pests leads to the severe defoliation of trees during population outbreaks, which is followed by a decrease in wood productivity, increase in infestation by secondary pests, and the death of trees. Moreover, a loss of green mass during outbreak will affect the speed of carbon accumulation from gas to solid state in soil that will change the biosphere function of forests [1,2]. This problem increased in the last half century because the warming of the climate may increase the frequency of pest outbreaks [3,4,5,6], and there is a risk to the secondary tundra in the boreal forests areas, as it was registered in Fennoscandia in the last century [7].Thus, the management of forest pests is aglobal problem whose intensity is increasing simultaneously alongside the increase in the speed of climate warming.

Spongy moth (former gypsy moth) Lymantria dispar L. is a notorious defoliator that is well known to the forest service staff of many countries in the Holarctic, being in the top list of dangerous forest pests. In recent decades, countless hectares of forests have been affected by severe defoliation. Millions of dollars have been spent on managing pests and huge efforts have been undertaken by quarantine services in logistical centers, such as harbor cities, to contain the spread of the Asian subspecies. Transport companies have incurred significant losses if this subspecies entered through the quarantine filters. These are just a few of the issues relating to this species. Spongy moth is a classical eruptive species with a wide amplitude of population fluctuations [8,9,10,11]; its pest density ranges from less than one caterpillar per tree at low population densities to several thousand individuals at breeding outbreak peaks.

Severe defoliation by spongy moth larvae occurs in areas greater than one million hectares, but it does not always occur following the complete drying of a forest. However, it does result in significant transformation of the lower vegetation layer due to increased sunlight from defoliated trees. This might be the reason for forest drying and subsequently fires in some regions. Moreover, trees weakened by defoliation are more susceptible to secondary pests and diseases. Also, flying adults are very allergenic for humans because of their abdominal fluff, and this is a serious problem for Asian subspecies because females are attracted to electric light at nighttime. Thus, the regulation of L. dispar populations (especially Far East subspecies) is an economically and socially important problem in the countries of the northern hemisphere, especially in Far Eastern and North American countries containing different subspecies of L. dispar, which significantly differ regarding their potential spreading activity.

The assessment of forest stands in vast territories, like taiga regions, can only be completed through remote sensing data. At present, these studies are mainly used for the identification of damage caused by insects. The measurement of different versions of the vegetation index by distinguishing red and near-infrared reflectance [12] is one of the main methods used in this analysis. The vegetation index NDVI allows the assessment of productivity and physiological elements within the vegetation of an ecosystem [13,14,15] in addition to serving as a spectral indicator of photosynthesis and plant metabolic intensity [16,17]. The study of vegetation index dynamics accounts for both intra- and inter-annual changes linked to climate variability [18]. Remote sensing data have found extensive use in mapping the spatial dynamics of insect outbreak hotspots in recent years [19,20,21,22,23]. However, the primary aim of such studies is to estimate the damage caused by outbreaks and calculate the area of loss. Such an index is reliable and operational, depicting the degradation of tree crowns during the development of an insect population outbreak. However, it is important to note that NDVI can estimate the photosynthesis of herbaceous vegetation after insect damage to tree crowns, but the results may not be recorded due to shading caused by the tree crowns. In our study, we develop a remote method that enables separating spongy moth-induced severe defoliation from other types of tree damaging associated with loss of foliage.

When analyzing the interactions between trees and phyllophagous insects, specifically the spongy moth, there are variations in the nature of the observed processes. For instance, if most of the leaves in the crown of trees are removed by insects in a given season on one sample area, then crown damage is not observed on another sample area located a hundred meters away from the first one. Furthermore, during the outbreak on various sample areas damaged by insects, a non-monotonic response of trees to damage is observed, consisting of two phases: the phase of leaf removal caused by pests and the phase of refoliation, when leaf regeneration takes place. Additionally, the intensity of these phases varies among trees in different sample plots, leading to the following inquiries:

-

What factors contribute to varying levels of insect attacks in nearby plantations?

-

How do the phases of leaf defoliation and refoliation relate to one another?

-

Does the condition of trees in sample plots with different levels of insect damage to their leaf apparatus correlate with the level of damage caused by phyllophagous insects?

Previously, the authors discovered links between the characteristics of regulatory processes in tree growth prior to outbreaks and the level of insect attacks on trees while investigating the first question [24]. The characteristics of amplitudes and frequency of oscillations in the series of first differences of radial growth of tree rings during the 15–20 years before the outbreak of insect attack were found to be the indicators of attractiveness of trees for insect attack. The first differences were chosen to emphasize the age trend of annual ring growth. Trees in uninfested stands exhibited high-frequency fluctuations and moderate amplitude in the series of first differences of annual ring widths. In contrast, trees in pest-damaged stands were characterized by a lower frequency of oscillations and either a small or excessive amplitude of these oscillations. In this study, we continue our research on the interaction between trees and insect pests and investigate the processes involved in the development of trees by pests and in the response of trees to insect attack. Specifically, we investigated the dynamics of leaf apparatus development and refoliation during the L. dispar outbreak in birch stands, analyzing the relationship between foliage damage and tree condition.

2. Materials and Methods

2.1. Satellite Data Acquisition and Processing

The birch stands of western Siberia examined in this study represent their own peculiarities [11]. Firstly, these territories cover vast areas of approximately one million hectares and are covered with small birch groves. Secondly, spongy moths periodically attack most of these forest stands during population outbreaks, resulting in defoliation rates of 80%–100%. This leads to the weakening and, in some cases, the death of trees. It is important to understand and address these peculiarities to preserve the health of the forests. Although the partial refoliation of tree crowns often occurs during the latter half of the growing season, weakened trees exhibit decreased productivity and increased susceptibility. Additionally, adult spongy moths will migrate to neighboring stands. All of this leads to the need for regular pest control.

The operative detection of pest outbreaks in large areas of western Siberia is possible using satellite data and the NDVI vegetation index. Additionally, it is necessary to differentiate stands that have been subjected to insect attacks from those affected by fire or soaking, which are common in this area. These distinctions can be made by analyzing seasonal changes in tree crowns during defoliation and refoliation. Figure 1 shows the change that occurred in tree crowns during the first half of the growing season, when the trees were attacked by spongy moth caterpillars, and at the end of the season when the caterpillars had finished feeding and pupated.

During the field work carried out in 2021 in the central and western regions of Novosibirsk, we established 9 experimental sample areas that were intensively damaged by spongy moth caterpillars and their corresponding 9 control areas that were either undamaged or had a low degree of damage. The study area is located in the eastern part of the distribution area of the spongy moth in western Siberia, covering an area of 300 × 250 km (Figure 2).

To minimize the variability of environmental factors, control sites were established in maximum proximity to insect-damaged stands. If possible, a separate birch clump was selected (Figure 3a), and in the absence of such a clump, an undamaged part of the studied clump was selected (Figure 3b).

Initially, the presence of pest damage was recorded visually. The boundaries of the stands and damaged areas were clarified using a DJI Phantom dron. Vector coordinate boundaries of the sample plots were delineated using the obtained images. The average size of sample plots was about 1–2 ha.

Data from the European Space Agency’s Sentinel−2 Earth remote sensing satellite system were used to assess the vegetative state. The choice of this system was determined by several factors:

-

Availability of the calculated Normalized Difference Vegetative Index (NDVI) for all studied territories. In this paper, the NDVI is calculated using the standard formula:

$$NDVI=\frac{NIR-Red}{NIR+Red}$$

(1)

where NIR and Red are the normalized values of the reflection intensity in the near-infrared and red ranges of the spectrum for a given point on the Earth’s surface [13].

-

Open access to the NDVI indicator time series via EO-browser analysis system of Sentinel-2 satellite data (https://apps.sentinel-hub.com/eo-browser/, accessed on 15 September 2023).

-

High spatial resolution up to 10 m per pixel.

-

Possibility of obtaining a sample of averaged NDVI time series for selected contours of the study sites directly through the EO-browser system.

-

In order to improve the accuracy of the obtained observations, we used images with the cloudiness level not exceeding 5%.

For damage assessment, the obtained contours were loaded into the EObrowser. The calculated NDVI layer with a spatial resolution of 10 m was used for analysis. For each sample area, a time series of the seasonal dynamics of the mean NDVI value within its boundaries for 2020–2022 was obtained. Some of the observations were considered unreliable due to heavy cloudiness and thus excluded from the analyzed sample. The time intervals between measurements were not constant and varied in the range of 2–10 days. On average, the series of annual dynamics of NDVI contains 50–60 consecutive observations.

Figure 4 shows a typical view of NDVI dynamics for the sample area compared to the control. It is possible to note the similar behavior of the two curves except for several moments: a strong dip of the index for PP06A in the period 07.06.21–05.07.21 associated with leaf eating by spongy moth caterpillars and a subsequent rise of the index 05.07.21–14.08.21 due to the refoliation of eaten leaves. No such effects were observed for the control area.

To obtain integral stand damage indices, the time series of control and experimental points was synchronized on a time scale. A normalized difference parameter of stand weakening was introduced:

DF(t) = (NDVI_A(t)−NDVI_C(t))/NDVI_C(t)

(2)

where NDVI_A(t) is the value of the vegetative index of the damaged site at the moment t, and NDVI_C(t) is the value of vegetative index of the control site at the moment t. This index characterizes the weakening of the trial site relative to the control site at the current moment. The dynamics of the DF index for the same PP06 is shown in Figure 5.

Defoliation parameters were introduced for the annual dynamics of the stand weakening index DF for the active part of the vegetative period (June–August). The period of sharp rise (spring) and decline (fall) of NDVI was excluded from consideration, since outbreaks of spongy moths at this time are impossible. The time range used is typical for the studied area of western Siberia. The following defoliation indices were used:

• MinDF—minimum DF value for the period: maximum stand damage.
• TMinDF—MinDF time point, which defines the moment of maximum stand damage.
• Integ(<MinDF)—integral sum of indicator during defoliation: defines the defoliation intensity for the whole time up to the moment of MinDF.
• Integ(>MinDF)—integral sum of the indicator during refoliation after MinDF has been reached.

2.2. Assessing Tree Condition Using Biophysical Methods

To evaluate the present condition of trees, we used dielectric indices of trunk cambium, which characterize the state of cell membranes and conductivity of plant tissues [25,26,27,28]. To measure these indices, we used an original designed Dielectric Pulse Fourier Spectrometer (DIFS) connected to a portable computer, which is used to control and power the device [29]. The device’s design enables the rapid acquisition of information about the dielectric properties of tree trunk tissues within the 1–100 kHz frequency range, with results processed, visualized, and saved on a computer in the field.

The electrical characteristics of tissues vary in response to the frequency of the transmitted electric current, which can be illustrated using impedancehodographs. These curves characterize the relationship between the values of ohmic and capacitive conductivity at different frequencies [30,31,32].

The impedance-hodograph plot takes the form of a half-circle (Debye diagram) in the Re(ω) and Im(ω) components of impedance (Figure 6).

The impedance-hodograph curve can be described by studying the parameters R₀ and R_∞, which represent the ohmic and capacitive components of the object’s impedance. Pathological processes in tree tissues modify these parameters. Specifically, the R₀ value [29,33] decreases as the quality of the cell membrane properties degrades. Later in the study, this index was employed to evaluate the condition of woody plant tissues in field conditions.

Compared to the subjective, qualitative visual indicators used to assess tree conditions, dielectric spectroscopy data are quantitative and objective. This method allows for a faster and less labor-intensive assessment compared to using radial growth measurements.

To compare the present condition of trees in the outbreak area with those in undamaged stands, we measured the dielectric properties of trees in both types of sample plots. Data were collected in early July of the outbreak year on 10–15 trees in each sample plot. The average characteristics R₀ and R_∞ of the impedance hodograph were measured along with the value of the variable

$$d{R}_0=\frac{R_0(A)-R_0(C)}{R_0(C)}$$

(3)

which represents the deviations between the average characteristics R₀(A) in damaged stands and the average characteristics R₀(C) in control stands.

To assess the dielectric indices of tree conditions, 10 trees were measured at each sample site. Since the indicators are Gaussian distributed for the same type of tree groups, averaged indicators for each site were used. Linear regression analysis methods were used to evaluate the relationship between remote sensing and ground-based indicators of stand condition; the obtained models were evaluated by the determination coefficient R².

3. Results

Estimated indicators for the period 2020–2022 are given in

Table 1. Estimated stand-weakening indicators at sample sites PP03–PP11 for the period 2020–2022.

Year	Plot	MInDF	Date MInDF	Integ(<MinDF)	Integ(>MinDF)	Insect Attack
2020	PP03	−0.181	14.07.2020	2.898	4.162	Possible
2020	PP04	−0.157	22.07.2020	3.618	2.362	Impossible
2020	PP05	−0.021	27.07.2020	2.866	1.947	Impossible
2020	PP06	−0.009	24.08.2020	4.907	0.467	Impossible
2020	PP07	−0.014	30.07.2020	1.408	2.373	Impossible
2020	PP08	−0.152	09.08.2020	1.716	1.059	Impossible
2020	PP09	−0.037	04.08.2020	1.677	0.083	Impossible
2020	PP10	−0.081	16.08.2020	3.120	0.827	Impossible
2020	PP11	−0.031	15.06.2020	0.232	0.295	Impossible
2021	PP03	−0.145	02.07.2021	5.784	3.987	Possible
2021	PP04	−0.272	02.07.2021	5.150	5.050	Possible
2021	PP05	−0.363	02.07.2021	5.557	5.676	Possible
2021	PP06	−0.187	05.07.2021	3.260	5.055	Possible
2021	PP07	−0.208	03.07.2021	3.394	4.171	Possible
2021	PP08	−0.198	30.06.2021	3.821	4.216	Possible
2021	PP09	−0.058	05.07.2021	0.800	0.628	Impossible
2021	PP10	−0.031	20.07.2021	0.681	0.801	Impossible
2021	PP11	−0.073	27.07.2021	2.141	1.168	Impossible
2022	PP03	−0.143	31.08.2022	5.610	0.000	Impossible
2022	PP04	−0.022	10.07.2022	1.041	2.903	Impossible
2022	PP05	0.150	09.08.2022	19.204	2.415	Impossible
2022	PP06	0.023	22.07.2022	6.193	3.110	Impossible
2022	PP07	−0.129	04.08.2022	3.625	1.495	Impossible
2022	PP08	0.004	31.08.2022	4.344	0.000	Impossible
2022	PP09	−0.069	17.07.2022	1.640	0.860	Impossible
2022	PP10	−0.039	20.06.2022	0.408	4.517	Impossible
2022	PP11	−0.146	09.08.2022	4.367	0.460	Impossible

. Indicators not corresponding to the timeframe of the insect outbreak (maximum defoliation occurs in the second half of June–early July) or having too low intensity of damage (NDVI reduction less than 10% of the control level) are highlighted in bold.

It can be seen that almost all the estimated values for 2020 and 2021 do not meet these conditions. This is consistent with ground observations that no significant damage to sample plots was observed in these years. At the same time, the low MinDF values for sample plots PP09–PP11 in the year of defoliation (2021) should be noted. Although these areas were included in the experiment, ground surveys show a low intensity of damage and the influence of weather factors. Therefore, these values of the index are explainable. For the remaining areas in the year of the insect outbreak, there is good temporal consistency in the peak of leaf dieback (30.06–5.07) and consistency in the parameters Integ(<MinDF) and Integ(>MinDF), i.e., a leaf being eaten is followed by intense refoliation, which is reflected in the satellite observations.

Next, we looked at how Integ(<MinDF) and Integ(>MinDF) related to each other within the six trial sites (excluding PP09–PP11) for the period 2020–2021(Figure 7).

As can be seen, in plantations at the initial stages of outbreak development (in 2020–2021), the Integ(>MinDF) response to impacts is almost identical to the Integ(<MinDF) impact characteristics. The response’s susceptibility to exposures was quantified through a regression equation based on the relationship Integ(>MinDF) = b Integ(<MinDF).The coefficient b is very close to 1; i.e., the response of the stand is equal to the impact level.

In general, the Integ(<MinDF) and Integ(>MinDF) indices increase significantly when insects massively devastate the stand. Figure 8 shows that Integ(<MinDF) and Integ(>MinDF) for the six sample sites for 2021 are higher than both 2020 (pre-outbreak) and 2022 (post-outbreak). Interestingly, the observed post-outbreak (2022) refoliation rates are lower than the corresponding defoliation rates. Most likely, this is due to the weakening of trees after the massive devastation of the previous year and the inability to restore phytomass in full.

In Figure 8, the Integ(<MinDF) parameters in the year of the stands devastation are statistically significantly different from those of the previous and following year. By discriminant analysis [34], the parameters Integ(<MinDF) and Integ(>MinDF) in the year of outbreak differ from those for 2020 and 2022 with an accuracy of 88% at a significance level of p < 0.05.

The measurements of tree state dielectric, Integ(<=MinDF) and Integ(>MinDF) indices, characterizing the level of crown defoliation and refoliation in the sample plots, are given in

Table 2. Characteristics of tree condition in the outbreak area and in control undamaged nearby stands.

Plots	Control		Damage		dR0	Integ(<=MinDF)	Integ(>MinDF)
	R0(C)	R∞ (C)	R0(A)	R∞ (A)
pp03	4.50	1.61	4.50	1.98	−0.13	5.78	3.99
pp04	4.73	1.77	4.67	2.01	−0.10	5.15	5.05
pp05	4.75	1.73	4.67	1.61	0.01	5.56	5.68
pp06	4.73	1.73	4.58	1.87	−0.10	3.26	5.05
pp07	4.52	1.80	4.32	1.94	−0.12	3.39	4.17
pp08	4.58	1.92	4.60	2.03	−0.03	3.82	4.22
pp09	4.41	1.93	4.49	2.15	−0.05	0.80	0.63
pp10	4.54	1.98	4.56	1.86	0.06	0.68	0.80
pp11	4.52	2.07	4.54	2.19	−0.04	2.14	1.17

.

Next, we look at how the characteristics of current condition and the level of tree development in the sample plots relate to each other. Figure 9 shows the relationship between these indicators.

As can be seen from Figure 9, there is a negative linear relationship $Integ(<MinDF)=A_d-B_d\cdot d{R}_0$, where A_d and B_d are coefficients, and $B_d=\frac{\partial (Integ(<MinDF))}{\partial (d{R}_0)}$ indicates the susceptibility of trees to insect attack depending on their relative condition dR₀. Experimental site PP05 differs significantly from the other trial sites. The dielectric values indicate a further decline in tree tissue functioning, which is likely due to drought conditions during the time of measurement, as observed at this trial site. The parameters of the equations of the relationship between Integ(<MinDF) and dR₀, and between Integ(>MinDF) and dR₀, are given in

Table 3. Parameters of the coupling equations between Integ(<MinDF) and dR₀ and between Integ(>MinDF) and dR₀.

Parameters	Values	Std.Err.	t-Test	p-Value
dR0, Integ(<MinDF)
Ad	1.68	0.74	2.28	0.06
Bd	−22.09	8.44	−2.62	0.04
R2	0.53
adjR2	0.46
F-test	6.86
dR0, Integ(>MinDF)
Ar	1.72	0.82	2.11	0.08
Br	−21.46	9.34	−2.30	0.06
R2	0.47
adjR2	0.38
F-test	5.28

.

As can be seen from Table 3, the coefficient B_d of susceptibility of trees to insect damage depending on their condition is significant at the level of p < 0.05, and the coefficient B_r of susceptibility of trees to refoliation is significant at the level of p < 0.1.

Thus, according to the results of remote sensing and biophysical studies, it is shown that the susceptibility of trees to insect attack depends linearly on the characteristics of tree condition. The worse the condition of trees, the stronger the intensity of insect attacks and the weaker the refoliation of tree foliage after the end of insect attack.

Next, we looked at the temporal dynamics of defoliation and refoliation.

In addition to characterizing the relative changes in NDVI values during insect defoliation of trees and leaf refoliation after the end of damage, as presented in Table 1 and Figure 8, important characteristics of tree–insect interactions are seen in the temporal dynamics of relative NDVI changes during the outbreak season. To characterize the temporal dynamics of NDVI, the following features were introduced: D₁ is the date of damage onset, when the NDVI value in the outbreak becomes lower than this value in the control, D₂ is the date when the maximum crown damage is reached, and D₃ is the date when the NDVI value peaks during stand refoliation. By utilizing these dates, the following characteristics of the temporal dynamics of changes in NDVI were calculated: T₁ = D₂−D₁—period from the beginning to the maximum damage, and T₂ = D₃− D₂—period from the date of maximum damage to the maximum reforestation. For the specific values of D₁ to D₃, T₁, and T₂, see

Table 4. Temporal dynamics of tree defoliation and refoliation during the L.dispar outbreak in the Novosibirsk region in 2021.

Plot	D1, Date	D2, Date	D3, Date	T1, Days	T2, Days
PP03	30.05.2021	09.06.2021	02.07.2021	10	23
PP04	31.05.2021	07.06.2021	05.07.2021	7	28
PP05	05.06.2021	07.06.2021	02.07.2021	2	25
PP06	07.06.2021	15.06.2021	02.07.2021	8	17
PP07	08.06.2021	15.06.2021	28.06.2021	7	13
PP08	05.06.2021	15.06.2021	30.06.2021	10	15
PP09	05.06.2021	12.06.2021	02.07.2021	7	20
PP10	05.06.2021	15.06.2021	25.06.2021	10	10
PP11	01.06.2021	06.06.2021	20.06.2021	5	14

.

The temporal dynamics of insect impacts and the subsequent tree responses are not synchronized among different sample plots; however, the phase shifts are not large. Thus, the onset phase of visible insect damage to trees varies by a week at most. As a result, visible damage at the experimental sites PP03 and PP04, located in the south of the Novosibirsk region, was observed at the end of May 2021. Damage was observed on trial plots PP06 and PP07, located in the central area of the Novosibirsk region, by the end of the first week of June 2021. The earlier the damage became visible, the quicker the peak of foliage removal was reached. For trial plots PP03 and PP04, the foliage removal peak was already observed by the end of the first week of June, while for the other trial plots, it was recorded a week later in mid-June. The defoliation peak occurred between late June and early July 2021 for most sample plots with the exception of sample PP11, which saw the peak almost 10 days earlier than the other plots.

The question arises—what is the relationship between the dynamics of characteristic times of different phases of tree response to insect attack? To estimate the relationship of T₁ and T₂ with the current state of trees in the stands, the same dielectric characteristics dR0 that were used to estimate the response of tree response amplitudes were used. Figure 10 and Figure 11 show the dependence of characteristic times T₁ and T₂ on the dR₀ state of trees in the outbreak area relative to the control.

As can be seen from Figure 10 and Figure 11, there is no strict connection between the condition and reactions of trees to insect impacts (

Table 5. Parameters of regression equations of the relationships between T₁, T₂ and dR_0, and between T₁ and T₂.

Parameters	Value	Std.Err.	t-Test	p-Value
T1
Intercept	3.16	0.90	3.53	0.02
Slope	−44.27	9.93	−4.46	0.01
R2	0.80
F-test	19.90
T2
Intercept	13.76	2.75	5.00	0.00
Slope	−56.81	31.53	−1.80	0.12
R2	0.35
F-test	3.2
T1/T2
Intercept	27.88	5.30	5.26	0.00
Slope	−1.17	0.59	−1.97	0.10
R2	0.39
F-test	3.9

).

However, Table 5 indicates a trend of decreasing characteristic times T₁ and T₂ with increasing dR₀ values. The convergence of tree conditions in the outbreak zone and control leads to faster defoliation and refoliation processes, suggesting a correlation between tree condition and the rate of these processes. At the same time, for outbreak zones, the change in the duration of leaf removal period as a function of state change dR₀$\frac{\partial T_1}{\partial (d{R}_0)}\approx -44$ is smaller in absolute value than the same value for refoliation $\frac{\partial T_1}{\partial (d{R}_0)}\approx -57$ (Table 5).

The relationship between the values of characteristic times T₁ and T₂ also exists, but this relationship is not strong (coefficient of determination R² = 0.39 (Figure 12 and Table 5).

4. Discussion

Two objectives arise when analyzing forest insect damage, particularly spongy moth, using remote sensing methods. The initial objective is the detection of stands that have suffered from pest attacks. The second goal includes not only identifying the area of damage but also assessing the level of damage. For the first task, one-time remote sensing data [35] can be used. To accurately assess not only the presence but also the level of damage, it is necessary to use seasonal remote sensing data in dynamics, which we consider in this paper. By comparing seasonal fluctuations of NDVI indices between the experimental and control plots, it is possible to identify characteristic signs of pest outbreaks. The resulting indices provide greater analytical value than the simple decrease in vegetative indices observed by remote sensing methods in the study area. Single-step assessments of forest condition based on remote indicators are subject to observation noise, e.g., due to high cloud cover during the measurement period, and may change under the influence of a large number of external factors [36].

Ground surveys can be avoided in order to reduce the effort required and increase the speed of the analysis. For this purpose, instead of defining a control stand for each object of study, it is possible to take the vegetative index dynamics averaged for a group of stands without damage in the first half of the growing season. Separate control plots should be selected for local areas 30–50 km in size with similar growing conditions and climate. The time series of relative NDVI values for the damaged stands will exhibit characteristic peaks determined by the durations of the phases of insect leaf defoliation and regrowth (as shown in this paper). The power of these peaks will be directly proportional to the volume of phytomass removed by pests and the intensity of refoliation processes.

In the current study, the dynamics of the NDVI indicator was used to assess the condition and level of stand damage. The ground-based field observations (dielectric properties of precambial stem tissues of damaged and control trees) were compared with remotely sensed indicators of relative defoliation and refoliation (Integ(<MinDF) and Integ(>MinDF)). The measurements taken by both ground and satellite means showed high consistency. In assessing the damage possibility, a measure of exceeding the calculated reference values at the level of 10% was introduced. It seems to us that this is sufficient accuracy, since ground-based methods estimate damage intensity at the level of 10%–20%.

This indicator is suitable for implementation in Siberia’s sharply continental climate zone, where cloudiness in the summer season is low, and the use of the 8-day NDVI composite allows filtering out the disturbances caused by clouds during the survey period. In different climatic conditions, when days with high cloudiness are frequent enough, it is suggested to use the NDWI index constructed from the data of two bands of infrared study or radar imaging, which allows eliminating the influence of cloudiness [37,38]. Most contemporary studies on the use of remote sensing techniques to detect forest damage do not include the “control-experiment” pair and therefore include short-term fluctuations of vegetative indices in the method error [22,23,36,37].At the same time, the method proposed by the authors allows separating tree damage from local fluctuations of environmental factors.

Of particular interest is the comparison between the integral defoliation and refoliation parameters Integ(<MinDF) and Integ(>MinDF). During the initial stage of damage, there is foliage regeneration, but the following year, the refoliation parameter is noticeably lower than the defoliation parameter for nearly all sampling plots. This could result in stand weakening, thereby necessitating forest protection measures.

The duration of defoliation and refoliation periods for each sample area was calculated from the processed satellite data of NDVI dynamics. Figure 10, Figure 11 and Figure 12 show that the relationship between defoliation times T₁ and refoliation times T₂ is not clear, but it still exists. The chosen indicator to determine tree condition in the stands significantly influences the trees’ response to insect impacts. At high values of the condition index, the duration of both leaf defoliation and refoliation in trees decreases, resulting in decreased levels of insect impact on trees. Therefore, larger values of the tree state characteristic are correlated with decreased impact by insects. Additionally, there is an inverse relationship between the characteristic times T₁ and T₂, where the duration of refoliation decreases as defoliation duration increases.

5. Conclusions

The application of remote sensing methods allows the identification of stages associated with pest-induced tree damage and the subsequent recovery of the leaf apparatus. The damage–response relationship and its variation throughout the outbreak can be evaluated for the stands under study.

The resulting indices provide greater analytical significance than simple remotely sensed declines in vegetation indices in the study area. Single-time NDVI estimates at a point are sensitive to observational noise and can be altered by a variety of external factors. Examining seasonal dynamics of indicators relative to control sites aids in detecting characteristic signs of an insect pest outbreak.

Taking into account the fact that the studied birch stands are exposed to different types of negative impacts (drought, diseases, etc.), the identification of precise signs of weakening due to insect outbreaks and the assessment of damage intensity seems promising. The suggested technique for identifying outbreak zones could be incorporated into a remote sensing-based automated forest management system.