1. Introduction
European forests have been monitored according to common standards within the International Cooperative Program on Assessment and Monitoring of Air Pollution Effects on Forests (ICP Forests) since 1985. In the 1980s and partially in the 1990s, forest degradation was mainly due to air pollution [1]. Recently, however, it has been caused mostly by weather extremes like storms and drought, along with bark beetle infestations [2,3]. The occurrence and severity of drought events have been intensifying, which is generally interpreted as a key phenomenon of climate change [4]. The most noticeable indicators of declining tree health, stemming from various stresses such as drought, are the loss and/or discolouration in trees’ assimilation organs [5]. Discolouration is typically manifested as the chlorosis (a decrease in chlorophyll content and yellowing) or necrosis (tissue death and browning) of leaves. It is primarily described using qualitative indicators of nutrient deficiency and imbalance [6] or reduced water availability [7,8].
During the summer of 2022, a significant decline in the health of deciduous forests was observed in Slovakia. The main symptoms were discolouration (mainly browning) and defoliation, which were accompanied by abundant fruiting in European beech and common hornbeam stands [9]. According to a study [10], the proportion of damaged stands gradually decreased with increasing forest vegetation zones (FVZs); i.e., the drought mainly damaged deciduous stands in lower FVZs, where the average annual temperature is the highest and the total precipitation is the lowest.
Within the current millennium, the drought of 2022 was not the only episode affecting the forests in Central Europe. Extraordinary droughts also occurred in 2003 [11,12], in 2018 [13,14], and in 2019 [14]. Harmful impacts of droughts on forests can lead to several repercussions, such as crown diebacks and premature fruit abortions [15], increased tree mortality rates in stands [3], early leaf senescence and premature yellowing, i.e., the termination of assimilation [16], fine root mortality and reduced biomass stock [17], diminished growth [18] and overall tree biomass production [19], and carbon stock loss [20]. Moreover, drought stress leads to the systematically induced susceptibility of forest trees to a variety of pests, especially fungal pathogens and bark beetles [21,22].
The frequent occurrence of storms and droughts, combined with pest infestations, has re-stimulated the importance of implementing surveys at the Pan-European level that link terrestrial and satellite monitoring using a standardised methodology [23]. According to the UN ECE ICP Forests [24], tree crown condition surveys are conducted terrestrially at permanent sampling points by visual assessments of tree crown damage, including evaluations of defoliation and discolouration. Monitoring is conducted annually in a network of 5628 Level I monitoring plots systematically distributed in a 16 × 16 km grid across all Europe [25]. These ground-based tree crown condition assessments are time-demanding and based on sample surveys. In contrast, satellite Earth remote sensing (ERS) can provide extensive and up-to-date data with a high resolution that are suitable for assessing the health status of forests [26].
Long-term research and outputs from local studies create prerequisites for the wider application of satellite ERS in Pan-European forest health assessments (FHA). Defoliation and discoloration in forest stands can result from a variety of factors which can be monitored through remote sensing, including water deficiency and stress [27], nutrient deficiencies [28], pest infestations and diseases [29,30,31], change in pigment contents [32,33,34,35], induced physiological stress such as photochemical oxidant exposure [36], and surface temperature [37]. These relationships between forest damage symptoms and reflected radiance are applied to various sensors, primarily in the optical domain, and are regularly updated, e.g., in the database of remote sensing indices [38]. The potential of Sentinel-1 SAR to assess damage in drought-affected forests was investigated with the results indicating further research is required [39]. An improved radiometric resolution of satellite data in combination with advanced algorithms allow the continuous identification of forest damage along the gradient from healthy to dead stands. Techniques used to estimate defoliation include temporal spectral trajectories, time series analyses [40,41], vegetation indices [42], differences in leaf area index or disturbance index between damaged and reference years [43,44,45], the orthogonal transformation of Landsat or Sentinel-2 bands [46], and machine learning methods [47]. To capture the complexity of forest damage, a complex indicator of spruce decline (CI4), which combines the number of dead trees, discoloration, the number of dry treetops, and the IUFRO vitality index, was proposed [31].
The aforementioned FH estimation techniques and methods provide critical insights to help identify vulnerable areas and specific stressors affecting forest vitality. Understanding the underlying causes of defoliation and discoloration is essential for effective forest management. Recent approaches that enhance the resilience and adaptability of forest ecosystems to climate change while maximizing their role in climate change mitigation include adaptive silviculture [48,49] and climate-smart forestry [50,51].
Despite a significant potential to integrate terrestrial and satellite monitoring at both national and pan-European scales, only a few studies focus on harmonizing field-based FHAs with satellite ERS [46,52]. According to the study [52], the use of satellite data for forest health assessments is influenced by several stand characteristics, such as the species mixture, crown coverage, and age classes. It means that damage classification is restricted to certain stand types due to the spectral similarity of different stand characteristics like needle loss, admixture of broadleaved species, and low crown closure. Existing solutions are based on forest stratification according to the mentioned parameters, deriving new or using existing forest mask layers [53], incorporating a priori knowledge about the parameters into defoliation models [52], or applying hybrid classifications. For example, distinguishing damaged stands from stands with a reduced crown canopy after thinning requires the use of forest management records to filter out the parts where thinning was carried out. The study [44] presents a methodology in which health status is not assessed based on the absolute amount of leaf biomass, with the leaf area index (LAI) used as a proxy for leaf biomass, but rather on its change over time. The basic premise is that the health status can be objectively determined only by observing the relative change in the LAI over time. A current research challenge is to eliminate the impact of mentioned factors (tree composition, canopy closure, and age) on FHAs without the need for forest stratification or the use of additional information about forest stands.
The primary objective of this study is to investigate the spectral response of forest stands to drought based on defoliation and discoloration symptoms and to propose an integrated damage indicator combining both factors. We hypothesize that the combined indicator provides a more accurate assessment of forest damage than separate defoliation and discoloration indicators. Scholarly sources support this assumption, noting that premature yellowing can halt assimilation and reduce overall productivity. Therefore, relying solely on defoliation to assess forest health might underestimate the true severity of the damage.
The second objective is to develop a method for operationally assessment of forest health at the national and Pan-European levels using a standardized methodology that integrates terrestrial and satellite monitoring. Since defoliation and discoloration are quantitatively assessed during fieldwork, we propose a modified two-phase sampling method with a regression analysis. We hypothesize that in the first phase, this method will allow a quick and cost-effective estimation of forest damage using only satellite images, without the need for forest stratification or additional stand information. The approach should also significantly reduce the impact of factors such as tree composition and age on damage classification. The second phase will refine these estimates with field-based assessments of defoliation and discoloration, thereby improving the overall accuracy of the damage evaluation.
2. Materials and Methods
2.1. Study Area
The area of interest was the entire territory of Slovakia, located in Central Europe. The territory is geographically defined by the coordinates of 47°43′′–49°37′′ north latitude and 16°50′′–22°35′′ east longitude. For this study, the terrestrial assessment of tree crowns was conducted in the Javorie Mountains (Figure 1).
Figure 1(Left): Area of interest (Slovakia) and the location of the survey area (black rectangle). (Right): The field survey area with the plot IDs. The black polygon represents the photo in Figure 2.
[Figure omitted. See PDF]
Figure 2(a) Browned crowns of beech–oak stand (depicted in Figure 1) due to drought. (b) A detailed view of partly necrotic beech leaves, reported under ‘red to brown discolouration’ according to the ICP Forests methodology. Photo: T. Bucha, 8 August 2022.
[Figure omitted. See PDF]
The territory belongs geomorphologically to the sub-system of the Carpathians and the Pannonian Basin. The geological and terrain diversity of the territory is also reflected in the soil conditions in terms of their chemical and physical properties. The most common soils are from the Cambisols group (nearly 70% of the forest area). Moreover, Leptosols, Luvisols, Umbrisols, Podzols, Arenosols, and Planosols, according to the World Reference Base international system for classification of soils [54], are also significantly represented. The climate has a transitional character between oceanic and continental climate with more pronounced features of continental climate. However, the altitude has the greatest influence on climatic conditions [55].
The subject of this research is the health condition of forest stands. According to the Green Report [56], forests cover 2.24 million hectares; i.e., the country’s forest cover is 45.7%. Regarding macroclimatic conditions, they belong to the temperate zone. Vertically, forests occur from about 100 m to 1800 m above sea level and are divided into eight forest vegetation zones (FVZs): oak, beech–oak, oak–beech, beech, fir–beech, fir–beech–spruce, spruce, and dwarf pine. Common beech (34.6%), Norway spruce (21.8%), and Sessile oak and Pedunculate oak (10.4%) have the highest representation. Category 9 represents all azonal forest communities, with a dominant occurrence of poplar, willow, and alder species [56].
A map of mid-summer 2022 drought [57] and a discussion of its impact on forest health by FVZs, soil properties, and exposure were published in our previous study [10]. Drought phenomenon was attributed to very low atmospheric precipitation and extremely high air temperatures from June to early August.
2.2. Satellite Images, Mosaicking, Calibration, and Topographic Normalization
Sentinel-2A and Sentinel-2B satellite images were used to classify forest damage in 2022 using L1C processing-level products. These are radiometrically and geometrically corrected images without atmospheric correction. Atmospheric corrections were not necessary, as the Sentinel scenes were either cloud-free or only locally affected by clouds and unaffected by haze. In addition, we preferred the Sentinel L1C product over the L2A product due to our detection of atmospheric overcorrection of DN values in some forest stands and the insufficient removal of pixels contaminated by clouds and shadows on the edges of the masks, which we found in the L2A products. The products were downloaded from the ESA Copernicus Open Access Hub.
Since it was not possible to obtain cloud-free Sentinel scenes for a part of Slovakia from the end of July to the end of August, one scene from the Landsat 9 satellite was downloaded from the USGS EarthExplorer platform. The scene was influenced by haze. Manual masking was problematic, therefore, the atmospherically corrected LC09_L2SP product was preferred. Table 1 provides an overview of the satellite images used for analyses.
Sentinel-2A and Sentinel-2B images were combined to achieve higher temporal resolution. This combination is possible as both satellites provide identical spectral and spatial characteristics (Table 2). The aim was to obtain satellite data from the period related to the worst drought conditions according to drought map [57].
Sentinel-2 bands 4, 8, 11, and 12 and corresponding Landsat 9 OLI-2 bands 4/5/6/7 (red/near infrared/shortwave infrared 1/shortwave infrared 2) were used for the classification of FH due to their higher resolution (in case of Sentinel bands), lower levels of sensitivity to cirrus and cirrostratus clouds, and significant spectral differences between healthy and damaged stands [47]. The inclusion of SWIR in analyses is supported by studies in which SWIR-based VIs are better for defoliation monitoring than the NDVI using red and NIR bands (e.g., [43,52]).
Sentinel-2 bands 11 and 12 were resampled from 20 m to 10 m using the nearest neighbour method. In the same way, the Landsat 9 bands were resampled from 30 m to 10 m.
Based on the visual interpretation, we manually masked the areas affected by clouds and shadows and then replaced them with cloud-free images from the nearest period listed in Table 1. The remaining gaps, 1.3% of the forest area, were filled with cloud-free parts of Sentinel-2 scenes from 21 June and 27 August 2022.
The countrywide mosaic was created by combining individual satellite images. Before making the mosaic, we used the relative calibration method [58] to remove differences in spectral reflectance between adjacent images (Sentinel vs Sentinel or Sentinel vs Landsat). The method is based on selecting a reference image to which other images were calibrated using a linear regression analysis. Regression equations were derived for each band of satellite images. We set paired values from the overlapping parts of the images classified as a forest. The scenes from 5 August 2022 were selected as the reference because they were cloud-free and covered the whole western and central part of Slovakia.
The topographic effect (slopes facing away from the sun appear darker than sun-facing slopes) was eliminated by the topographical normalization of the images using a simple statistical-empirical correction method [59]. The shading model (for the azimuth and elevation angle of the reference scenes) was derived from the digital terrain model DMR3.5 with a pixel resolution of 25 × 25 m. DMR3.5 is freely available on the Geoportal of the Geodetic and Cartographic Institute Bratislava.
We performed mosaicking using ERDAS IMAGINE 2014 (version 14.00, Hexagon Geospatial, Norcross, GA, USA, 2014) setting the overlay function in the MOSAIC module. Sentinel images were overlaid on Landsat images. As a result, 93.8% of the study area was covered by Sentinel and 6.2% by Landsat.
2.3. Visual Assessment of Crown Condition According to the ICP Forests Manual
Defoliation and discolouration are considered key indicators of forest health. According to the ICP Forests Manual for Visual Assessment of Crown Condition and Damaging Agents [24], defoliation is defined as needle/leaf loss in the assessable crown as compared to a reference tree. Defoliation is assessed in 5% steps in range from 0% (full-foliage trees) to 100% (standing dead trees).
Discolouration is defined as any deviation from the usual colour of the living foliage for the assessed tree species. Totally brown or necrotic leaves are considered dead; hence, ‘discolouration’ does not apply here since this symptom is restricted to living foliage.
Fruiting is defined as annual seed production of trees. Only the fruiting that occurs in the specific year of assessment is considered.
Coupling of Defoliation and Discolouration
Even though defoliation and discolouration refer to distinct phenomena, to our knowledge, no studies have used both symptoms for satellite-based forest damage estimation. If forests are affected, e.g., by insect herbivores, they defoliate, because leaves/needles are consumed. However, the reflectance of spectral bands in the trees affected by drought is affected by the change in the colour and structure of leaves, i.e., changes in concentrations of chlorophylls, carotenoids, and anthocyanins. Therefore, we merged both symptoms to describe the spectral changes in forests and estimate forest damage. To couple defoliation and discolouration into one indicator, we propose the following formula:
DEF-DIS = DEF + (1 – (DEF/100)) × DIS (1)
where DEF—defoliation in %; DIS—discolouration in %; and DEF-DIS—coupled DEF and DIS in %.When applying Formula (1), DEF-DIS values range from 0 to 100%. The formula consists of two components: (1) defoliation and (2) discoloration. The discoloration (DIS) of fallen leaves on the ground is not considered, so DIS only applies to leaves still on the tree. Therefore, when calculating DEF-DIS, the DIS value, estimated in the field on a scale from 0 to 100%, is adjusted by the coefficient (1 – DEF/100). The DEF-DIS value is then determined by adding the adjusted DIS value to the DEF value. This approach ensures that DEF-DIS = 0% corresponds to fully leafed tree crowns without discolouration and DEF-DIS = 100% to dead-standing trees with 100% defoliation. In both cases, DIS = 0%, and therefore DEF-DIS = DEF. In all other cases, when DIS > 0%, the DEF-DIS value exceeds DEF. The philosophy behind the indicator is that the individual partial indicators of defoliation and discolouration could underestimate the forest damage level when used separately.
2.4. Two-Phase Sampling Assessment of Forest Damage
2.4.1. The Principles of Satellite Two-Phased Sampling
The classification of forest condition in 2022 was based on the principle of two-phase regression sampling [60]. According to study [61], satellite two-phased sampling in comparison with other two-phased sampling designs is different in the following ways:
-. In the first phase, estimated sample units are usually not targeted values (e.g., defoliation) but auxiliary variables—spectral signatures (expressed as DN value)—which are utilised as independent variables in a reflectance regression model.
-. Alternatively, data from previous surveys can advantageously be used as data for damage estimation in the first phase of sampling.
-. Generally, the sizes (number) of the first (n1) and the second (n2) sample are the subjects for optimisation. This is based on the cost ratio between the second and the first sample units, the correlation between the target and auxiliary variables, and the required level of precision. When employing satellite data, the optimisation of n1 is not important. Data can be collected from the whole population, and the cost is not a problem.
In this study, we implemented a two-phase sampling method combined with regression, following the flowchart outlined in Figure 3. The goal of the first phase was to estimate forest damage quickly and cheaply from satellite images, i.e., to reduce the overall cost of data collection. The second phase’s aim was to refine the estimation using the field-based defoliation assessment on monitoring plots, i.e., to improve its accuracy. The values from the refinement of forest damage (DEF, DIS, and DEF-DIS) obtained from regression analyses were rescaled to the range from 0% (healthy stands) to 100% (dead stands).
2.4.2. The First Phase—Image Sampling
Components derived from satellite mosaic bands using the Gram–Schmidt transformation (GST) were used for forest damage classification. The mathematical background was described in study [62], and the biophysical significance of the components and the calculation procedure optimised for FHA were described in study [46]. The first component (NSC1—New Synthetic Component) represents the difference between coniferous and broadleaf trees. Marginal values for the digital numbers (DN) of the first component (NSC1) were defined using a set of plots that included fully foliated beech trees (bright objects) and spruce stands (dark objects). Transformation coefficients were then calculated. The second component (NSC2) was optimized specifically for estimating damage, with transformation coefficients derived to ensure NSC2 was orthogonal to NSC1. To determine the marginal DN values for NSC2, which indicate maximum possible damage, we used a set of dead spruce stands with 100% defoliation. The calculated values of NSC2 represent a perpendicular distance from the line of DN values of the NSC1 component defined by fully foliated pure stands of beech and spruce on its border. The advantage of this method is that the distance from the line (i.e., the NSC2 value) is proportional to the level of tree damage and can be used to estimate the damage for any pixel.
Another advantage of NSC approach is the elimination of the differences between broadleaf and coniferous trees. When Sentinel/Landsat bands were used directly for damage classification, the influence of species composition on the classification was evident from the visual interpretation. Coniferous stands had lower defoliation than broadleaf stands. Both components (NSC1 and NSC2) were derived by the linear combination of 4 satellite mosaic bands according to coefficients listed in Table 3.
2.4.3. The Second Phase—Field Sampling
The field campaign was carried out at 31 field plots dominated by broadleaf trees on 25–26 August 2022. Plots were selected purposefully from the entire range of damage, from undamaged to heavily damaged stands, and were located southwest of the town of Zvolen, specifically in the forest complex in the Javorie Mountains. When selecting plots, there was an emphasis on ensuring that the spectral and textural characteristics in the satellite images were consistent across a larger area surrounding the chosen centre of each plot. This is important due to the positional inaccuracy of satellite images (±12 m) and the determination of the centre by the GPS. The expected similar correlation between DN values and damage indicators (DEF and DIS) within and around the monitoring plot can reduce the impact of positional discrepancies on the accuracy of the FHA analysis. Such a selection was made based on the visual assessment of the spectral manifestations of forest stands in the composition of the VNIR/SWIR1/RED satellite bands. In this composition, stands with greater defoliation and discolouration (browning) appear in shades of turquoise, depending on the intensity of the damage.
At each plot, defoliation, discolouration, and fructification of 15 dominant or codominant trees closest to the centre of the plot were evaluated. Therefore, the sample size was variable, with a plot diameter from 10 m to 25 m depending on stand density and age. Defoliation was assessed in accordance with the ICP Forests methodology [24] in 5% steps. Discolouration was assessed as % of discoloured assimilatory organs including necrosis in 5% steps. Fruiting was categorized in accordance with the ICP Forests methodology into 4 classes: 0—Absent, 1—Scarce, 2—Common, and 3—Abundant.
Crown canopy cover varied from 50% to 100%, with an average of 76%. Regarding tree species occurrence, beech dominated (47%), followed by oak (41%) and hornbeam (6.9%). Coniferous trees were only individually admixed with 0.43% proportion. The set was extended by three plots with standing dead spruce trees located in the High Tatra Mountains.
The evaluation was preceded by a one-day training of evaluators, consisting of a theoretical part and a practical evaluation of defoliation, changes in colour and fruiting in the monitored area according to the ICP Forests methodology. The Sanasilva manual [63] was used to objectify and facilitate the defoliation assessment.
2.5. Statistical Evaluation of Data
The data (defoliation, discolouration, and fructification) assessed at a tree level were used to calculate the arithmetic means per plot. Calculated means were stored in the database and linked to the field plot vector layer. This vector layer was used for the extraction of the plot’s statistical characteristics from the satellite data using zonal statistics. The data taken from the forest damage field assessment and linked with satellite data (arithmetic mean of DN values by bands and NSC components) at a plot level were used for statistical analyses in Statistica (version 14.0.0.15, TIBCO Software Inc., Palo Alto, CA, USA). Two NSC components were calculated using the Gram–Schmidt transformation (GST) of four satellite mosaic bands. The components were optimised for damage classification (NSC2) and to significantly reduce the influence of species composition on damage classification (NSC1). Details are provided in Section 2.4.2.
Basic statistics, represented by a correlation matrix among defoliation, discolouration, fruiting, and derived NSCs, were calculated. To obtain full overview, we included satellite mosaic bands into the matrix. Further analyses were guided according to the revealed interrelationships among variables.
NSC2, having the highest correlation coefficient from predictors, was selected in the second-phase sampling stage to improve the first-phase forest damage estimates. Simple linear regression analyses were used to evaluate the dependence between defoliation, discolouration, and their combination as response variables and NSC2 as a predictor in the second-phase sampling stage. The accuracy of the model was assessed using the following regression analysis parameters: the correlation coefficient, the standard error of the estimate (SEE), and the residuals (i.e., the differences between the ground-truth values and the values predicted by the model). The model was then applied to classify FH on the satellite mosaic of the entire country.
We reclassified continuous DEF, DIS, and DEF-DIS data into categorical data. The aim was to provide easily interpretable information about FH status in Slovakia and to simplify the further interpretation of the results. The results were summarized in maps and in a table with 11 damage classes ranging from 0 to 100% in 10% steps. If the percentage of damage exceeded a threshold of 110%, pixels were assigned to class 11, representing mainly logging. The accuracy was expressed by the confusion matrix only for the combined defoliation and discoloration symptoms (DEF-DIS). The matrix is based on the number of monitoring plots classified into individual categories according to the average DEF-DIS calculated from the evaluations at the plots (ground truth) and according to the regression model (classification). The overall accuracy was calculated as the ratio of the correctly assigned plots in individual categories to the total number of plots. To calculate the adjusted overall accuracy (OA-adj), we determined the acceptable range for correct predictions as ±1 category. In the case of ordinal data, for which adjacent misclassifications are acceptable and expected (according to SEE), a shift of plus or minus one category can provide a more accurate view of model performance. The user accuracy (precision), producer accuracy (recall), and their adjusted versions (±1 category) were calculated for each category individually by procedures explained in [64].
3. Results
3.1. Damage Symptoms and Spectral Response of Forest
Defoliation and discolouration were used as the two basic symptoms of forest damage. Both traits were evaluated during the terrestrial survey in 2022. The combined DEF-DIS symptom was calculated from their individual values according to Formula (1). In addition, fruiting, which generally affects leaf size and defoliation, was also assessed. The relationships among the mentioned symptoms and the spectral responses given by reflectance in individual bands or derived components were the subject of our basic correlation statistics (Table 4).
The results clearly show high correlations between defoliation (DEF), discolouration (DIS), and their combination (DEF-DIS) and the spectral response across the four satellite bands and the derived NSC2 component. We achieved the strongest correlation coefficients with the NSC2 component for all the damage symptoms. This indicates the suitability of the GST approach for satellite-based FHAs. Fructification did not significantly affect defoliation or discolouration, so we did not include this variable in our further analyses.
3.2. Classification of Forest Damage
The classification of forest conditions in 2022 was based on the principle of two-phase regression sampling [51]. The GST was used to derive NSC2 to estimate forest damage in the first phase and NSC1 to eliminate the influence of wood composition on the damage classification. Figure 4a–c shows the distribution of defoliation, discoloration, and their coupled symptoms (DEF-DIS) at monitoring plots against NSC1 and NSC2 in three-dimensional contour plots. The three-dimensional contour plots were created using linear fitting for the depicted variables. It is clearly visible that in the NSC2 direction, the percentage of damage on the monitoring plots gradually increases for all three of the damage symptoms. Similar NSC2 values correspond to similar percentages of damage on the monitoring plots. In contrast, similar NSC1 values correspond to different damage levels, indicating that the NSC1 component is much less sensitive to damage than NSC2. The slope of the contour lines in the three-dimensional contour plots indicates that NSC1 influences DIS least, DEF-DIS moderately, and DEF most. The distribution of DEF and DEF-DIS (shown as the average % at the monitoring plots) in Figure 4a,c suggests that the impact of the NSC1 component is primarily influenced by three of the monitoring plots with 100% defoliation. When NSC1 was included in the multiple linear regression models together with NSC2, it was found to be statistically insignificant for DEF (p = 0.062), DIS (p = 0.762), and DEF-DIS (p = 0.133). Therefore, NSC1 is not included in the models presented in Table 5.
The second phase aimed to refine the estimations using field-based FH assessments with the simple linear regression (Figure 5a–c). The obtained values were then rescaled to the defoliation range from 0% (healthy stands) to 100% (dead stands).
Table 5 summarizes the regression model parameters for both phases, including the precision achieved, expressed as the standard error of the estimate.
To fully assess the model’s accuracy for DEF-DIS, the plots of the raw residuals from the second phase of sampling were added to the Appendix A (Figure A1).
3.3. Large-Scale Forest Damage Assessment in Slovakia
For the evaluation of the proposed approach across the entire area of Slovakia, the derived reflectance models (Table 5) were applied to all pixels classified as forests according to the study [65]. This forest mask was created using satellite data from 1995 to 1996. Since forests change over time, the mask includes all stages of forest’s life, such as logging, natural disasters, and regrowth. The mask is consistently used in our forest health (FH) analyses, allowing us to perform reliable and comparable FH analyses on the same acreage. The analysed area was 2,160,264 ha (Table 6). The classification accuracy can be assessed by regression analysis parameters (the correlation coefficient and the standard error of the estimate). Utilizing field assessments of defoliation in 2022, we obtained the correlation coefficient between the first and second phases of 0.87. The standard error of the estimate (SEE) was ± 10.7%. This implies that if the determined defoliation is 30%, the actual value can range from 19.3% to 41.7% with 68% confidence. The standard errors of the estimate for discolouration and the coupled symptoms can be interpreted similarly. In both cases, the error was lower than that of defoliation, namely ±8.4% for discolouration and ±9.7% for the coupled symptoms (Table 5).
Defoliation, discoloration, and their combination were reclassified into categorical data to simplify the further interpretation of the results. Figure 6 presents the national forest condition classification results for 2022, according to DEF, DIS, and DEF-DIS, classified into 11 damage classes. Figure 7 illustrates the combined effect of defoliation and discolouration on the resulting damage in the region of the field survey.
A confusion matrix (Table A1 in Appendix A) revealed an overall accuracy of 44.1% for the DEF-DIS map. In ordinal data, for which adjacent misclassifications are acceptable and expected (SEE = ±9.7%), a shift of plus or minus one category can provide a more accurate view of the model’s performance. We achieved an adjusted overall accuracy of 88.2%.
Table 6 shows the results of the 2022 classifications according to defoliation, discolouration, and their coupling in 11 damage classes. It enables a comparison of the differences in the representation of individual categories. In addition to defoliation and discolouration, forest canopies also affect the classification of forest damage. There is a mixture of spectral properties for the different categories (forest, undergrowth, herbal community, and soil) in stands with an open canopy cover. For this reason, stands with different damage levels can be included in the same damage class. The verbal description of the category expresses their potential difference. The logging category in the legend represents pixels classified as having a level of defoliation > 100%.
The results of the FHA revealed that approximately 49.2% of the forests occurred in damage classes ranging from 31% to 100% according to the DEF indicator. This share increases to 67.3% when using the DEF-DIS indicator (Table 6). The percentage of severely damaged and dead stands, represented by classes of damage > 60%, increased from 10.8% by the DEF indicator to 14.7% by the DEF-DIS indicator. These increases in the shares of damaged classes indicate the substantial extent of forest areas exposed to the drought stress in 2022 when discolouration is taken into account. This highlights the importance of including this indicator in forest health assessments. We note that a logging class was not considered in the above calculation.
4. Discussion
4.1. Spectral Response of Damaged Forest Stands
The examination of defoliation and spectral response relationships in drought-affected forest stands showed the positive correlations for the RED (B4) and SWIR2 (B12) bands, an ambiguous effect for the SWIR1 (B11) band, and a negative correlation for the VNIR (B8) band (Table 3). The best correlations were found for the VNIR (r = −0.77) and SWIR2 bands (r = 0.78). These results align with those of most studies since the 1990s, which have found that as the defoliation severity increases, the amount of reflectance increases in the visible bands, decreases in the NIR band, and increases in the SWIR band (e.g., [66,67]). The highest correlation coefficient, r = 0.87, was achieved with NSC2, which confirms the suitability of the Gram–Schmidt transformation for forest damage classification. NSC2 is optimized specifically for defoliation assessments, while simultaneously eliminating the effect of tree species composition on the spectral response of forest stands. The study [68] reported that a 20% hardwood component in Norway spruce stands completely neutralised the effect of a 20% needle loss when using NIR and SWIR bands. The correction model was based on a priori knowledge of the percentage of hardwood component in the compartments. Our approach does not require any a priori knowledge of forest stands.
Moderate and high correlation coefficients were also found between discolouration and the DN values of all four of the analysed satellite mosaic bands. Although the reflectance can vary depending on the nature and extent of the damage, the detected positive correlations for RED, SWIR1, and SWIR2 confirm the prevailing findings from the spectrometric measurements. A change in the colour of the assimilation organs leads to the increased reflectance in the red part of the spectrum. This is partly due to the decrease in chlorophyll content, which absorbs the red light in healthy leaves. In the SWIR1 and SWIR2 wavelengths, reflectance is influenced by strong water absorption in vegetation. Totally brown or necrotic leaves are considered dead, i.e., their amount of water is reduced, leading to a higher reflectance compared to healthy vegetation. The VNIR part of the spectrum shows high reflectance and minimal absorption. There is a strong link between reflectance and the amount of green, healthy vegetation. If the leaves are brown or are lost, the amount of reflectance decreases, leading to negative correlation coefficients in both cases. The highest correlation coefficient, r = 0.91, was achieved with NSC2, similarly to that for defoliation. This is expected because we found a strong positive correlation (r = 0.82) between defoliation and discolouration; i.e., as defoliation increases, the extent and intensity of colour changes also increase (Figure 5d). This correlation pattern could be derived from the data published in the study [69]. However, this relationship was weaker and was based on a limited sample of only eight observations, all exhibiting low levels of defoliation and discolouration.
Some authors [68,70] have pointed out the mutual influence of defoliation and colouring (yellowing and browning) on the resulting reflection spectra, e. g., there is no apparent increase in the reflection of yellow spruce needles, although brownish pine needles exhibit an increase in the SWIR region. As leaf browning was dominant in our case, it could explain the correlation coefficients of 0.60 and 0.79 between discolouration and the SWIR1 and SWIR2 bands (Table 4). The study [68] stated that in the case of increased SWIR and decreased NIR bands, the ratio of SWIR1/NIR led to favourable results in the defoliation assessment. However, this is valid only when chlorosis and defoliation occur simultaneously and not in regions where slight-to-moderate defoliation is the only symptom. The results (Table 4) from our dataset confirm an increase in SWIR2 (a positive correlation) and a decrease in NIR reflectance (a negative correlation) in cases of the simultaneous occurrence of both symptoms.
The suitability of NSC2 for determining defoliation as well as discolouration can be explained by the similarity of the spectral properties of spruce dead trees with rest of brown needles (used for deriving NSC2) and the brown necrotic leaves of broadleaved species. In both cases, spectroradiometric measurements revealed a continuous increase in reflectivity in the range of 400–1100 nm [67].
4.2. Precision of Two-Phased Sampling with Regression Estimator
The second objective of our study to propose a simple, fast, and cost-effective estimate of forest damage was achieved. In the first phase, the estimation was based solely on satellite data, and the NSC2 component was proven to be the most suitable for this purpose. The main advantage of our method compared to other studies [44,52,71] is that it provided accurate damage estimates without the need for forest stratification or additional stand information. The GS transformation substantially reduced the impact of variables such as tree composition and age on damage classification. NSC1 was not significant predictor of DEF (p = 0.062), DIS (p = 0.762), or DEF-DIS (p = 0.133) in the refinement of the first phase of classification. Therefore, NSC1 was not included in the model presented in Table 5.
The classification accuracy for defoliation, with the standard error of the estimate (SEE) of ±10.7%, matched our previous results. Specifically, in the period from 1990 to 2021, the accuracy ranged from ±5.9% to 12.5% [72]. Our results also correspond with similar studies focusing on the estimation of defoliation. For instance, the study [71] reported an error of ±10%, the errors presented in the study [52] ranged from ±3% to 11%, and the study [43] achieved an RMSE estimation of defoliation equal to ±14.9%, a mean absolute error of 10.8%, and a cross-validation R2 of 0.805.
Although the NDVI is the most widely used and tested index to assess defoliation [73], it was not found to be robust in the study [47]. Instead, the Green Normalized Difference Vegetation Index (GNDVI) and MERIS Terrestrial Chlorophyll Index (MTCI) were found to be more suitable, assuming that with increasing defoliation, the total amount of chlorophyll in the canopy decreases. Furthermore, the study [47] investigated different ML (machine learning) methods (SVM, RF, and KNN) and stated that MTCI and GNDVI are robust regardless of the ML approach. All regression models predicted stand defoliation with a moderate level of accuracy. The nRMSE value was the lowest for the SVM model, 11.6%, followed by 11.9% for the RF model, and 12.2% for the kNN model. The R2 varied from 0.53 (the kNN model) to 0.57 (the RF and SVM models). These results are comparable to ours. The potential contribution of the green band is a challenge, and future research on discolouration and the assessment of coupled symptoms should investigate this.
Concerning discolouration, the study [30] shows an overall accuracy of 84% for the four discolouration levels in pine forests. Our result, with an SEE of ±8.4%, demonstrates the potential of satellite-based discolouration assessment not only for these four levels but also quantitatively across the entire range, from 0% to 100%. We did not find any works in the literature that presented the errors of the estimate when determining the coupled DEF-DIS symptoms.
In addition to having a comparable classification accuracy with that of other studies, the advantage of our approach is its consistency, simplicity, and cost-effectiveness based on the two-phase sampling principle. The benefits arise in the first phase during the damage estimation from the unique use of the GS transformation. Alternatively, the classification from the previous year can be used to estimate forest damage in the first phase instead of the GST [46]. In the second phase, annually acquired forest monitoring data or our own ground-based assessments of defoliation are used to refine the classification. However, it can be expected that the model based on ICP Forests monitoring plots will lack data with defoliation ranging from 50% to 95%. The average defoliation in the monitoring plots in the regular 16 × 16 km network of the ICP Forests program ranges from 10 to 50% in ‘normal’ years when it is not influenced by the severe impact of damaging agents. Even in 2022, when the summer in Europe was characterized by extreme drought and widespread dry conditions, two-thirds (65.9%) of 5,453 plots had a mean defoliation up to 25%, and only 1.4% of the plots showed severe defoliation of more than 60% [25]. This problem of insufficient data can be partially overcome by including plots with standing dead trees (100% defoliation) in the model, as these are easily identifiable on satellite images. Furthermore, we recommend to extend annual national monitoring surveys by at least five additional plots with expected defoliation between 50 and 95%, a level intentionally pre-selected based on the visual inspection of satellite images on a PC monitor to refine the estimation of damage in the first phase.
The potential misalignment between the sample plots and satellite pixels can influence the accuracy. The geolocation accuracy for Sentinel-2 Level-1C products (orthorectified and resampled to a global UTM/WGS84 grid) is typically within 12 m with the 95% confidence level without the use of ground control points (GCPs). To set the plot centres in the terrain, we applied an RTK GNSS method using a u-Blox F9P receiver and an ANN-MB dual-frequency antenna. The national SKPOS service provided corrections. Such a combination of our method and the equipment provides centimetre-level accuracy under optimal conditions; even in forests, the accuracy is well within the accuracy declared for the Sentinel-2 Level-1C product. Therefore, we consider the main source of ambiguity to be the accuracy of the Sentinel data. The possible discrepancy between satellite and field-evaluated data was addressed by selecting plots ranging from healthy to strongly defoliated or discoloured forests, with the homogeneity of spectral or textural expression in a wider area around the plots.
4.3. Analysing Spatial Distribution of Drought-Induced Forest Damage in Slovakia
The stands that were most significantly affected by the drought in 2022, according to the three analysed symptoms of DEF, DIS, and DEF-DIS, are depicted in Figure 6. From the maps, it is evident that the influence of the drought was not spatially uniform for any of the symptoms. The defoliation caused by the drought (Figure 6 top) was observed in the western part of Slovakia and the southern part of central Slovakia. Stands in Eastern Slovakia were less affected, although drought manifestations were observed in the southern part of the Volovské, Slanské, and Vihorlat Mountains, as well as in the Slovak Karst. We did not detect a clustered occurrence of pixels with defoliated stands in the mountainous regions of Northern Slovakia. These results are consistent with the spatial distribution of drought-damaged stands identified on the basis of the ground monitoring of forests in a 16 × 16 km network in 2022 [9].
The discolouration distribution map (Figure 6 middle) reflects the defoliation spatial distribution, as the regression coefficients are similar for the second phase of sampling (Table 5). The intensity of the damage is lower, as the discolouration categories are about three categories shifted against the defoliation categories (Figure A2 in Appendix A). This is due to the difference in the absolute coefficients of the regression models that is lower for discolouration (−40.1) than for defoliation (−14.9). As a consequence, defoliation classes 1–4 correspond to discolouration class 1, defoliation class 10 (dying and dead trees) corresponds to discolouration classes 7–9, and defoliation class 11 corresponds to discolouration classes 9–11 (Table A2 in Appendix A).
The resulting damage based on the combination of both symptoms is exhibited in Figure 6 bottom. The spatial distribution of the damaged stands is similar to the damage map derived from the defoliation, differing mainly in areas where a higher level of discolouration was recorded. In these areas, the overall level of damage is higher than that derived from defoliation alone.
In the ICP Forests program, defoliation from 0% to 25% is considered within the bounds of normal variability. Normal defoliation refers to the extent of leaf loss that does not indicate significant stress or damage. Defoliation between 26% and 60% is classified as moderate, indicating an impact of stress factors such as pests, diseases, environmental stress, or pollution. Defoliation between 61% and 99% is considered severe, representing significant stress and damage to the tree, which may lead to reduced growth and an increased risk of mortality. A value of 100% indicates a dead tree. Our results presented in Table 6 show that the damage of about 49.2% of forests ranged from 31% to 100% according to DEF indicator. According to the DEF-DIS indicator, the share of forests increased to 67.3%. This 18.1% difference indicates the significant extent of forest areas exposed to the drought stress in 2022 when discoloration is taken into account. The percentage of severely damaged and dead stands, represented by classes with damage levels > 60%, was 10.8% (~227 ths. hectares) using the DEF indicator and 14.7% (~308 ths. hectares) using the DEF-DIS indicator. This 3.9% difference in the share of severely damaged stands means that the area of direct economic damage due to the drought in 2022 is greater by approximately 81 ths. hectares, when discoloration is taken into account in the FHA.
It is beneficial to point out that the severity of the impact of droughts on forests in Slovakia is generally underestimated, and the relevant data on the drought-induced damage (or mortality) of forest stands are partially missing from the current forestry evidence. This is related to the fact that drought-weakened trees are usually attacked by biotic pests, especially insects, which are easy to identify. Hence, evaluators can visually recognize and record them, which overestimates the significance of biotic (secondary) pests and, at the same time, underestimates the severity of the impact of droughts [74]. This is despite the generally known findings that droughts and their negative consequences on temperate forests are, without any doubt, the most relevant accompanying phenomena of climate change [75].
Some limitations of the present study should be acknowledged. Our terrestrial survey focused exclusively on broadleaved species (99.6%), prevailingly beech and oak forests with a negligible admixture of conifers (0.4%), growing within relatively small areas with rather uniform site conditions. Therefore, the influence of the soil, ground vegetation, crown canopy disruption, and discolouration heterogeneity on the reflected radiance was limited and did not include all possible variabilities. Different results may be achieved for stands in distinct site types and with abundant understories.
The proposed method employs widely accepted health indicators, such as defoliation and colour change, and is independent of tree species. The method was tested in the Carpathian-Pannonian region of Slovakia; therefore, its potential to evaluate forest health across temperate deciduous, mixed, and coniferous forests throughout Europe should be verified.
The methodology for defoliation evaluation has been elaborated and standardised within the framework of the ICP Forests program. In contrast to defoliation, the evaluation of discolouration is not methodologically fixed. The current ICP Forests manual states discolouration is any deviation from the usual colour of the living foliage for the assessed tree species. The problem is that the ICP Forests manual does not quantify discoloration but instead uses a qualitative classification into four categories, described by the following codes: 02 (light green to yellow discoloration), 03 (red to brown discoloration, including necrosis), 04 (bronzing), and 05 (other colours). According to Sanasilva’s inventory approach [76], discolouration is recorded as the distance of the average tree crown colour from a reference crown colour for this species. It has been assessed using Munsell colour charts for plants [77]. This allows us to quantify discolouration in terms of the percentage of affected assimilatory organs and the intensity of discolouration as well. Our approach, described in the Methodology section, used only the percentage of affected assimilatory organs without considering the intensity of discolouration. The standardization of this parameter, such as through an update to the ICP Forests manual, is needed to fully harmonize the Forest Health Assessment (FHA) survey at the European level.
The study of [70] pointed out that the influence of leaf parameters on the reflection spectra are minor compared to the external parameters like illumination, shadow, and background and that the NIR region, in particular, is extremely sensitive. Our previous study [10] showed that a change detection method between two years significantly reduces the potential error of classifying pixels with crown canopy disruption when it occurs in both years. More research covering all these factors is still needed.
4.4. Implications for Forest Management and Policy
The combination of defoliation and discolouration into final damage classes is listed in the FAO manual [78] and in older ICP Forests manuals. A disadvantage is that this combination is only considered at the level of four defoliation classes and three discolouration classes. Moreover, older ICP manuals consider the combined indicator as an optional parameter, and since 2006, it has not even been included. Repeated instances of drought-induced discolouration over the last two decades have made it necessary to include this parameter in the FHA at a more precise level. The new combined indicator allowed us to quantitatively assess damage continuously (the SEE ± 9.7%; see Table 5), which we consider the most significant progress compared to approaches presented in the ICP Forests or FAO manuals. This confirmed that the indicator can be effectively used at an operational level for a country-wide forest health assessment. It helps identify the areas most affected by damage, enabling timely and proactive management. The map outputs (Figure 7), with their 10 m spatial resolution, allow the identification of forest damage within individual stands. This is crucial for implementing necessary forestry measures, as the stand compartment is the fundamental unit of forestry planning, control, and the record-keeping of actions taken. In Slovakia, the practical usability of forest damage maps is ensured online through the STALES web map application [46]. Besides forestry practice, the STALES service is also used by the state forestry administration to control the legitimacy and timeliness of implemented measures. Online forest health maps in the Czech Republic continuously monitor bark beetle damage using satellite data. This helps the government to compensate forest owners for damages and potential losses through direct subsidies and tax relief. Moreover, the legislative instrument (the public decree) that defines the list of cadastral units selected according to the actual FHA helps forest owners by reducing the regulation of their obligations under the national forest law, enabling them to manage the bark beetle calamity in the most affected regions [44].
Understanding the causes of defoliation and discoloration is crucial for selecting the best forest management strategies. In the Carpathian-Pannonian region, a strong relationship was found between leaf loss and different meteorological parameters such as maximum monthly temperatures, the monthly number of summer and hot days. The correlations were stronger in arid sites, indicating that specifically beech forests may suffer heavy damage if climate change continues as projected [79]. Based on changes in defoliation between dry and normal years, the study [10] identified the forest areas in Slovakia most affected by drought in 2022. Their analysis of site, soil, morphometric, and altitudinal factors updated our existing knowledge, revealing that the damage was significant up to an elevation of about 800 m, compared to the previously reported 500 m [80]. The results presented in Figure 7 more precisely identify the most drought-vulnerable areas in the Carpathian-Pannonian region of Slovakia. It is reasonable to assume that future drought episodes are likely to have the most severe impact in these areas. Identifying these areas accurately can help direct forestry support measures aimed at mitigating the negative effects of climate change, in line with the European Union’s Common Agricultural Policy (CAP). Financial instruments such as the Rural Development Program provide funds to implement tailored measures defined by member states according to local needs and conditions. The urgent need to focus on finer-scale and more intensive management options was emphasized in the study [48]. Key practices include selective thinning, using forward-thinking seed mixtures, site contouring, managing pests, and fires. According to the study [20], combating climate-change-induced mortality may require using more drought-tolerant species for stand regeneration rather than relying solely on native species. Satellite assessments of forest health can provide essential information for selecting and monitoring forest management measures. Although it can deliver detailed data at the compartment level while covering large territorial units, its potential remains underutilized in forestry practice.
5. Conclusions
This study addresses the previously overlooked symptom of leaf discoloration when assessing forest health during the summer of 2022, a period marked by a significant deterioration in the health of deciduous forests in Slovakia. In addition to increased defoliation, colour changes, especially various forms of browning, were observed in a large complex of mainly broadleaved stands at lower altitudes. These phenomena were closely associated with drought conditions, characterized by significantly low levels of atmospheric precipitation and exceptionally high air temperatures from June to early August.
Scholarly sources emphasize that premature yellowing and browning lead to the cessation of photosynthesis and a decrease in overall productivity. Relying solely on defoliation as a metric for assessing forest health may thus result in an underestimation of the actual severity. To address this concern, we first proposed a formula to effectively integrate defoliation and discoloration. Subsequently, regression models were developed using a two-phase sampling approach, allowing the reliable estimation of defoliation, discoloration, and their combined impact. In the first phase, we utilized the Gram–Schmidt transformation (GST) of four satellite mosaic bands to derive two components optimized for damage classification (NSC2) and to significantly reduce the influence of species composition on damage classification (NSC1).
This analytical framework provides a comprehensive understanding of forest health, encompassing both visible signs of distress. The proposed method could be applied nationally and across the temperate forest zone in Europe using a harmonized approach with the ICP Forests, integrating both terrestrial and satellite monitoring. While the concept was validated on a limited set of plots, further validation is needed, such as using data from the pan-European monitoring network of ICP Forests. Currently, the methodology of the program records four types of colour changes qualitatively without quantification. We recommend quantifying discoloration similarly to defoliation, using 5% or 10% intervals.
Finally, we emphasize that it is reasonable to assume that the areas with the most severe forest damage in 2022 are also the most drought-sensitive regions within Slovakia. Future drought episodes are likely to have the most extensive impacts in these areas. Our approach can help precisely locate vulnerable stands, improve forest management strategies, and implement practical measures to protect forests from the negative effects of droughts and the subsequent spread of secondary (biotic) factors.
Conceptualization, T.B. and P.S.; Methodology, T.B. and P.S.; Field campaign design, J.T. and J.C.; Resources, T.B and J.C.; Investigation, T.B., P.P. and B.K.; Data curation, T.B.; Writing—original draft preparation, T.B., B.K., P.P. and P.S.; Writing—review and editing, T.B.; Project administration, T.B.; Funding acquisition, B.K. All authors have read and agreed to the published version of the manuscript.
The original contributions presented in the study are included in the article, further inquiries can be directed to the corresponding author.
The authors declare no conflicts of interest.
Footnotes
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Enlarge this image.Figure 3. Flowchart of country-wide forest health assessment from satellite data (Slovakia).
Enlarge this image.Figure 4. Three-dimensional surface plot of defoliation (a), discolouration (b), and coupled defoliation-discolouration (c) against NSC1 and NSC2. Values on y-axis represent the average defoliation (%), discolouration (%), and their combination (%) at the monitoring plots.
Enlarge this image.Figure 5. Refinement of forest damage estimation in the 2nd phase according to field assessments in Slovakia. Spectral reflectance models for (a) defoliation; (b) discolouration; and (c) defoliation- discolouration. X-axis: NSC2 component. (d) Relationship between defoliation and discolouration. The shown confidence intervals are at the 95% level.
Enlarge this image.Figure 6. Classification of forest conditions in Slovakia in 2022 from satellite images: defoliation (Top), discolouration (Middle), and coupled defoliation and discolouration (Bottom).
Enlarge this image.Figure 6. Classification of forest conditions in Slovakia in 2022 from satellite images: defoliation (Top), discolouration (Middle), and coupled defoliation and discolouration (Bottom).
Enlarge this image.Figure 7. Comparison of classification of forest conditions expressed by defoliation (a), discolouration (b), and coupled defoliation and discolouration (c). The selected window covers the area of the field survey in 2022. The black dots indicate the plots examined. The legend is identical to that of Figure 6.
Overview of satellite images used for classification of forest damage in 2022.
| Satellite Platform | Sensing Date | Sensing Area and Satellite Product Type | Scene Cloudiness|Covered Area of Mosaic |
|---|---|---|---|
| Sentinel-2B | 5 August 2022 | Western and Central Slovakia: S2B_MSIL1C_20220805T094549_N0400_R079 | Cloud-free|64% |
| Sentinel-2A | 4 August 2022 | Eastern Slovakia: S2A_MSIL1C_20220804T093051_N0400_R136 | Cloud-free|17.3% |
| Sentinel-2B | 12 August 2022 | Central and Eastern Slovakia: S2B_MSIL1C_20220812T093549_N0400_R036 | Partly cloud-free|11.2% |
| Landsat 9 | 1 August 2022 | Central and Eastern Slovakia: LC09_L2SP_187026_20220801_20220803_02_T1 | Partly cloud-free|6.2% |
Note: When creating the 2022 mosaic, Sentinel-2 images from 21 July 2022 and 27 August 2022 were used to fill gaps caused by clouds in the images from 1 July and 12 August 2022.
Sentinel-2 bands selected for forest damage classification in 2022.
| Sentinel-2 Band | Central Wavelength (nm) | Band Widths (nm) | Resolution (m) |
|---|---|---|---|
| B4—Red | 664.6/665.0 | 31/31 | 10 × 10 |
| B8—Visible and near infrared (VNIR) | 832.8/833.0 | 106/106 | 10 × 10 |
| B11—Shortwave infrared (SWIR1) | 1613.7/1610.4 | 91/94 | 20 × 20 |
| B12—Shortwave infrared (SWIR2) | 2202.4/2185.7 | 175/185 | 20 × 20 |
DN input values and Gram–Schmidt coefficients of satellite mosaic bands optimized for forest damage classification.
| RED | VNIR | SWIR1 | SWIR2 | |
|---|---|---|---|---|
| DN Values Used for Gram–Schmidt Transformation (GST) | ||||
| Fully foliated beech stands | 95.67 | 247.30 | 131.47 | 107.00 |
| Fully foliated spruce stands | 91.32 | 61.88 | 55.00 | 77.61 |
| Dead spruce stands | 112.16 | 82.78 | 112.06 | 116.79 |
| Calculated transformation coefficients for 1st and 2nd components according to the GST | ||||
| NSC1 | 0.0215 | 0.9145 | 0.3771 | 0.1449 |
| NSC2 | 0.3366 | −0.3708 | 0.6687 | 0.5496 |
Correlation coefficients among symptoms of forest damage (defoliation, discolouration, their combination, and fructification) and satellite bands of 2022 mosaic.
| Variable | Defoliation | Defol-Discol | Fruct. | RED | VNIR | SWIR1 | SWIR2 | NSC2 | NSC1 |
|---|---|---|---|---|---|---|---|---|---|
| ** DISCOLOURATION | 0.82 | 0.91 | −0.07 | 0.90 | −0.62 | 0.60 | 0.79 | 0.91 | −0.47 |
| * DEFOLIATION | 0.98 | −0.21 | 0.53 | −0.77 | 0.02 | 0.78 | 0.87 | −0.71 | |
| DEFOL-DISCOL | −0.23 | 0.65 | −0.74 | 0.20 | 0.81 | 0.91 | −0.66 | ||
| ** FRUCTIFICATION | −0.15 | −0.13 | −0.21 | −0.04 | −0.04 | −0.17 |
Correlations in bold are significant at p < 0.05000. N = 34 (all plots included); * N = 33 (1 plot excluded as an outlier); ** N = 31 (3 cases excluded with defoliation = 100%, as the colouring and fructification could not be evaluated).
Spectral reflectance models for the classification of forest damage in 2022 by two-phase regression sampling.
| Model | r | SEE | Sample Size |
|---|---|---|---|
| First phase: Estimation of damage using Gram–Schmidt transformation | |||
| NSC2 = 0.337 × R2022 − 0.371 × IR2022 + 0.669 × SWIR12022 + 0.550 × SWIR22022 | - | - | n1: all pixels |
| Second phase: Refinement of classification based on terrestrial FHA | |||
| Defoliation: DEF2022 = −14.9224 + 1.2874 × NSC2 | 0.87 | ±10.7 | n2: 33 plots * |
| Discolouration: DIS2022 = −40.106 + 1.2479 × NSC2 | 0.91 | ±8.4 | n2: 31 plots ** |
| Defoliation-Discolouration: DEF-DIS2022 = −9.389 + 1.2975 × NSC2 | 0.91 | ±9.7 | n2: 34 plots |
Note: r—correlation coefficient; SEE—standard error of the regression (estimate); * 1 plot excluded as an outlier; ** 3 plots excluded (DEF = 100%); FHA—Forest health assessment.
Classification of forest damage in Slovakia in 2022 according to defoliation (DEF), discolouration (DIS), and their coupling (DEF-DIS).
| Classes of Damage | DEF | DIS | DEF-DIS | Description of Damage Level in Pixels for DEF and DEF-DIS | |
|---|---|---|---|---|---|
| % of Pixels | % of Pixels | % of Pixels | Area (ha) | ||
| 0%–10% | 5.33 | 58.50 | 2.14 | 46,313 | No or only negligible damage |
| 11%–20% | 14.38 | 18.14 | 7.66 | 165,466 | Stands with slight damage |
| 21%–30% | 28.82 | 9.07 | 20.12 | 434,678 | |
| 31%–40% | 20.64 | 4.95 | 25.46 | 549,923 | Stands with moderate damage |
| 41%–50% | 11.62 | 3.27 | 17.87 | 386,100 | |
| 51%–60% | 6.44 | 1.61 | 9.75 | 210,606 | Stands with severe damage, disturbed areas, and forest stands in regeneration |
| 61%–70% | 3.90 | 1.25 | 6.11 | 132,023 | |
| 71%–80% | 2.81 | 0.86 | 3.58 | 77,245 | |
| 81%–90% | 1.79 | 0.61 | 2.24 | 48,297 | Dying and dead stands, disturbed areas without forest regeneration, logging |
| 91%–100% | 2.01 | 0.75 | 2.33 | 50,417 | |
| Logging | 2.26 | 1.00 | 2.74 | 59,195 | |
| Total: | 100 | 100 | 100 | 2,160,264 | |
Appendix A
Enlarge this image.Figure A1. Distribution of raw residuals for DEF-DIS model on monitoring plots. X-axis: differences between the observed (ground-truth) values of DEF-DIS and the values predicted by a model. Y-axis: number of plots (n = 34).
Enlarge this image.Figure A2. Scatterplot of defoliation versus discoloration after classification into 11 categories, as described in Table 6. Blue dots represent the number of pixels for a given combination of categories. The area of 1 pixel = 0.09 hectare. The red dashed line represents the relationship where y = x.
A confusion matrix for forest health classification using combined defoliation and discolouration symptoms (DEF-DIS). The matrix is based on the classification of the 34 plots where the field survey was conducted. UA—User Accuracy; PA—Producer Accuracy.
| DEF-DIS (%) | 21–30 | 31–40 | 41–50 | 51–60 | 61–70 | 71–80 | 81–90 | 91–100 | Logging | ||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DEF-DIS | Ground Truth (Row)\Classified (Column) | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | Row Total | Producer Accuracy (%) | PA Adjusted |
| 21–30 | 3 | 0 | 1 | 1 | 2 | 0.0 | 50.0 | ||||||
| 31–40 | 4 | 4 | 1 | 5 | 80.0 | 100.0 | |||||||
| 41–50 | 5 | 1 | 1 | 1 | 3 | 33.3 | 66.7 | ||||||
| 51–60 | 6 | 3 | 1 | 1 | 2 | 7 | 14.3 | 71.4 | |||||
| 61–70 | 7 | 1 | 3 | 1 | 5 | 60.0 | 100.0 | ||||||
| 71–80 | 8 | 2 | 2 | 100.0 | 100.0 | ||||||||
| 81–90 | 9 | 4 | 1 | 5 | 20.0 | 100.0 | |||||||
| 91–100 | 10 | 1 | 3 | 1 | 5 | 60.0 | 100.0 | ||||||
| Logging | 11 | 0 | 0 | - | - | ||||||||
| Column Total | 0 | 6 | 6 | 2 | 5 | 9 | 2 | 3 | 1 | 34 | |||
| User Accuracy (%) | - | 66.7 | 16.7 | 50.0 | 60.0 | 22.2 | 50.0 | 100.0 | 0 | 44.1 | |||
| UA adjusted ± 1 category (%) | - | 100.0 | 83.3 | 100.0 | 80.0 | 77.8 | 100.0 | 100.0 | 100.0 | 88.2 |
Pixel layout matrix of defoliation (columns) against discolouration (rows). The damage categories are described in
| DEFOL | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | ||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DISCOL | 0%–10% | 11%–20% | 11%–20% | 31%–40% | 41%–50% | 51%–60% | 61%–70% | 71%–80% | 81%–90% | 91%–100% | Logging | Total | |
| 1 | 0%–10% | 1,249,729 | 3,451,358 | 6,916,861 | 2,407,198 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 140,25,146 |
| 2 | 11%–20% | 0 | 0 | 0 | 2,547,361 | 2,160,393 | 0 | 0 | 0 | 0 | 0 | 0 | 4,707,754 |
| 3 | 21%–30% | 0 | 0 | 0 | 0 | 629,610 | 1,371,503 | 0 | 0 | 0 | 0 | 0 | 2,001,113 |
| 4 | 31%–40% | 0 | 0 | 0 | 0 | 0 | 173,562 | 935,354 | 0 | 0 | 0 | 0 | 1,108,916 |
| 5 | 41%–50% | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 674,188 | 0 | 0 | 0 | 674,188 |
| 6 | 51%–60% | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 386,430 | 0 | 0 | 386,430 |
| 7 | 61%–70% | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 44,309 | 256,185 | 0 | 300,494 |
| 8 | 71%–80% | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 205,987 | 0 | 205,987 |
| 9 | 81%–90% | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 21,135 | 124,476 | 145,611 |
| 10 | 91%–100% | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 179,320 | 179,320 |
| 11 | Logging | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 238,824 | 238,824 |
| Total | 1,249,729 | 3,451,358 | 6,916,861 | 4,954,559 | 2,790,003 | 1,545,065 | 935,354 | 674,188 | 430,739 | 483,307 | 542,620 | 23,973,783 |
Note: 1 pixel represents an area of 0.09 ha.
References
1. Eggleston, S.; Goodwin, J.; Pulles, T.; Visschedijk, A.; Bakker, J.; Ritter, M.; Koch, D.; Pazdan, W. Emissions of Atmospheric Pollutants in Europe, 1980–1996; Topic report 9 European Environment Agency: Copenhagen, Denmark, 2000; 115p.
2. Seidl, R.; Thom, D.; Kautz, M.; Martin-Benito, D.; Peltoniemi, M.; Vacchiano, G.; Wild, J.; Ascoli, D.; Petr, M.; Honkaniemi, J. et al. Forest disturbances under climate change. Nat. Clim. Chang.; 2017; 7, pp. 395-402. [DOI: https://dx.doi.org/10.1038/nclimate3303] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/28861124]
3. Senf, C.; Buras, A.; Zang, C.S.; Ramming, A.; Seild, R. Excess forest mortality is consistently linked to drought across Europe. Nat. Commun.; 2020; 11, 6200. [DOI: https://dx.doi.org/10.1038/s41467-020-19924-1]
4. Dai, A. Increasing drought under global warming in observations and models. Nat. Clim. Chang.; 2012; 3, pp. 52-58. [DOI: https://dx.doi.org/10.1038/nclimate1633]
5. Česljar, G.; Jovanovič, F.; Brašanec-Bosanac, L.; Dordevič, I.; Mitrovič, S.; Eremija, S.; Ćirković-Mitrović, T.; Lučić, A. Impact of an Extremely Dry Period on Tree Defoliation and Tree Mortality in Serbia. Plants; 2022; 11, 1286. [DOI: https://dx.doi.org/10.3390/plants11101286] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/35631711]
6. Gregory, S.C.; Redfern, D.B. Diseases and Disorders of Forest Trees. A Guide to Indentifying Causes of Ill-Health in Woods and Plantations; Forestry Commission, The Stationary Office: Edinburgh, UK, 1998; 136p.
7. Flexas, J.; Bota, J.; Loreto, F.; Cornic, G.; Sharkey, T.D. Diffusive and metabolic limitations to photosynthesis under drought and salinity in C(3) plants. Plant Biol.; 2004; 6, pp. 269-279. [DOI: https://dx.doi.org/10.1055/s-2004-820867] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/15143435]
8. Verslues, P.E.; Agarwal, M.; Katiyar-Agarwal, S.; Zhu, J.; Zhu, J.K. Methods and concepts in quantifying resistance to drought, salt and freezing, abiotic stresses that affect plant water status. Plant J.; 2006; 45, pp. 523-539. Erratum in Plant J. 2006, 46, 1092 [DOI: https://dx.doi.org/10.1111/j.1365-313X.2005.02593.x] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/16441347]
9. Pavlenda, P.; Pajtík, P.; Sitková, Z.; Priwitzer, T.; Pavlendová, P. Manifestations of extreme drought on forest trees species in permanent monitoring plots of PMS Forests. APOL; 2022; 3, pp. 94-101.
10. Bucha, T.; Pavlenda, P.; Konôpka, B.; Tomaštík, J.; Chudá, J.; Surový, P. Identification of drought-induced forest damage in 2022 and of its key site condition drivers through satellite imagery. Cent. Eur. For. J.; 2024; 70, pp. 156-175.
11. Rebetez, M.; Mayer, H.; Dupont, O.; Schindler, D.; Gartner, K.; Kropp, J.P.; Menzel, A. Heat and drought 2003 in Europe: A climate synthesis. Ann. For. Sci.; 2006; 63, pp. 569-577. [DOI: https://dx.doi.org/10.1051/forest:2006043]
12. Buras, A.; Ramming, A.; Zang, C.S. Quantifying impacts of the 2018 drought on European ecosystems in comparison to 2003. Biogeosciences; 2020; 17, pp. 1655-1672. [DOI: https://dx.doi.org/10.5194/bg-17-1655-2020]
13. Brun, P.; Psomas, A.; Ginzler, C.; Thuiller, W.; Zappa, M.; Zimmermann, N.E. Large-Scale Early-Wilting Response of Central European Forests to the 2018 Extreme Drought. Glob. Chang. Biol.; 2020; 26, pp. 7021-7035. [DOI: https://dx.doi.org/10.1111/gcb.15360]
14. Rukh, S.; Sanders, T.G.M.; Krüger, I.; Schad, T.; Bolte, A. Distinct responses of European beech (Fagus sylvatica L.) to drought intensity and length—A review of the impacts of the 2003 and 2018–2019 drought events in Central Europe. Forests; 2023; 14, 248. [DOI: https://dx.doi.org/10.3390/f14020248]
15. Nussbaumer, A.; Meusburger, K.; Schmitt, M.; Waldner, P.; Gehrig, R.; Haeni, M.; Rigling, A.; Brunner, I.; Thimonier, A. Extreme Summer Heat and Drought Lead to Early Fruit Abortion in European Beech. Sci. Rep.; 2020; 10, 5334. [DOI: https://dx.doi.org/10.1038/s41598-020-62073-0] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/32210278]
16. Rohner, B.; Kumar, S.; Liechti, K.; Gessler, A.; Ferretti, M. Tree Vitality Indicators Revealed a Rapid Response of Beech Forests to the 2018 Drought. Ecol. Indic.; 2021; 120, 106903. [DOI: https://dx.doi.org/10.1016/j.ecolind.2020.106903]
17. Konôpka, B.; Lukac, M. Moderate drought alters biomass and depth distribution of fine roots in Norway spruce. For. Pathol.; 2013; 43, pp. 115-123. [DOI: https://dx.doi.org/10.1111/efp.12005]
18. Scharnweber, T.; Smiljanic, M.; Cruz-García, R.; Manthey, M.; Wilmking, M. Tree Growth at the End of the 21st Century—The Extreme Years 2018/19 as Template for Future Growth Conditions. Environ. Res. Lett.; 2020; 15, 074022. [DOI: https://dx.doi.org/10.1088/1748-9326/ab865d]
19. Colangelo, M.; Camarero, J.J.; Rippulone, F.; Gazol, A.; Sánchez-Salguero, R.; Oliva, J.; Redondo, M.A. Drought Decreases Growth and Increases Mortality of Coexisting Native and Introduced Tree Species in a Temperate Floodplain Forest. Forests; 2018; 9, 205. [DOI: https://dx.doi.org/10.3390/f9040205]
20. Somogyi, Z. Projected effects of climate change on the carbon stocks of European beech (Fagus sylvatica L.) forests in Zala County, Hungary. Lesn. Časopis–For. J.; 2016; 62, pp. 3-14. [DOI: https://dx.doi.org/10.1515/forj-2016-0001]
21. Klutsch, J.G.; Shamoun, S.F.; Erbilgin, N. Drought stress leads to systemic induced susceptibility to a nectrotrophic fungus associated with mountain pine beetle in Pinus banksiana seedlings. PLoS ONE; 2017; 12, e0189203. [DOI: https://dx.doi.org/10.1371/journal.pone.0189203]
22. Netherer, S.; Ehn, M.; Blackwell, E.; Kirisits, T. Defence reaction of mature Norway spruce (Picea abies) before and after inoculation of the blue-stain fungus Endoconidiophora polonica in a drought stress experiment. Lesn. Časopis–For. J.; 2016; 62, pp. 169-177. [DOI: https://dx.doi.org/10.1515/forj-2016-0014]
23. Drechsel, J.; Forkel, M. Remote sensing forest health assessment—A literature review on a European level. Cent. Eur. For. J.; 2024; in print
24. Eichhorn, J.; Roskams, P.; Potočić, N.; Timmermann, V.; Ferretti, M.; Mues, V.; Szepesi, A.; Durrant, D.; Seletković, I.; Schröck, H.-W. et al. Part IV: Visual Assessment of Crown Condition and Damaging Agents; Thünen Institute of Forest Ecosystems: Eberswalde, Germany, 2020; 49p.
25. Michel, A.; Kirchner, T.; Prescher, A.-K.; Schwärzel, K. Forest Condition in Europe: The 2023 Assessment. ICP Forests Technical Report under the UNECE Convention on Long-range Transboundary Air Pollution; Thünen Institute: Eberswalde, Germany, 2023.
26. Torres, P.; Rodes-Blanco, M.; Viana-Soto, A.; Nieto, H.; Garcia, M. The Role of Remote Sensing for the Assessment and Monitoring of Forest Health: A Systematic Evidence Synthesis. Forests; 2021; 12, 1134. [DOI: https://dx.doi.org/10.3390/f12081134]
27. Ceccato, P.; Flasse, S.; Tarantola, S.; Jacquemoud, S.; Grégoire, J.M. Detecting vegetation leaf water content using reflectance in the optical domain. Remote Sens. Environ.; 2001; 77, pp. 22-33. [DOI: https://dx.doi.org/10.1016/S0034-4257(01)00191-2]
28. Walshe, D.; McInerney, D.; Van De Kerchove, R.; Goyens, C.; Balaji, P.; Byrne, K.A. Detecting nutrient deficiency in spruce forests using multispectral satellite imagery. Int. J. Appl. Earth Obs. Geoinf.; 2020; 86, 101975. [DOI: https://dx.doi.org/10.1016/j.jag.2019.101975]
29. Franklin, S.E.; Wulder, M.A.; Skakun, R.S.; Carroll, A.L. Mountain pine beetle red-attack forest damage classification using stratified Landsat TM data in British Columbia, Canada. Photogramm. Eng. Remote Sens.; 2003; 69, pp. 283-288. [DOI: https://dx.doi.org/10.14358/PERS.69.3.283]
30. Leckie, D.G.; Cloney, E.; Joyce, S.P. Automated detection and mapping of crown discolouration caused by jack pine budworm with 2.5m resolution multispectral imagery. Int. J. Appl. Earth Obs. Geoinf.; 2005; 7, pp. 61-77.
31. Brovkina, O.; Cienciala, E.; Zemek, F.; Lukeš, P.; Fabianek, T.; Russ, R. Composite indicator for monitoring of Norway spruce stand decline. Eur. J. Remote Sens.; 2017; 50, pp. 550-563. [DOI: https://dx.doi.org/10.1080/22797254.2017.1372697]
32. Albrechtová, J.; Rock, B.N.; Soukupová, J.; Entcheva, P.; Šolcová, B.; Polák, T. Biochemical, histochemical, structural and reflectance markers of damage in Norway spruce from the Krušné hory Mts. used for interpretation of remote sensing data. J. For. Sci.; 2001; 47, pp. 26-33.
33. Zarco-Tejada, P.J.; Miller, J.R.; Noland, T.L.; Mohammed, G.H.; Sampson, P. Scalling-up and Model Inversion Methods with Narrow-band Optical Indices for Chlorophyll Content Estimation in Closed Forest Canopies with Hyperspectral Data. IEEE Trans. Geosci. Remote Sens.; 2001; 39, pp. 1491-1507. [DOI: https://dx.doi.org/10.1109/36.934080]
34. Panigada, C.; Rossini, M.; Busetto, L.; Meroni, M.; Fava, F.; Colombo, R. Chlorophyll concentration mapping with MIVIS data to assess crown discoloration in the Ticino Park oak forest. Int. J. Remote Sens.; 2010; 31, pp. 3307-3332. [DOI: https://dx.doi.org/10.1080/01431160903193497]
35. Darvishzadeh, R.; Skidmore, A.; Abdullah, H.; Cherenet, E.; Ali, A.; Wang, T.; Nieuwenhuis, W.; Heurich, M.; Vrieling, A.; O’Connor, B. et al. Mapping leaf chlorophyll content from Sentinel-2 and RapidEye data in spruce stands using the invertible forest reflectance model. Int. J. Appl. Earth Obs. Geoinf.; 2019; 79, pp. 58-70. [DOI: https://dx.doi.org/10.1016/j.jag.2019.03.003]
36. De Marco, A.; Vitale, M.; Popa, I.; Anav, A.; Badea, O.; Silaghi, D.; Leca, S.; Screpanti, A.; Paoletti, E. Ozone exposure affects tree defoliation in a continental climate. Sci. Total Environ.; 2017; 596–597, pp. 396-404. [DOI: https://dx.doi.org/10.1016/j.scitotenv.2017.03.135]
37. Farella, M.M.; Fisher, J.B.; Jiao, W.; Key, K.B.; Barnes, M.L. Thermal remote sensing for plant ecology from leaf to globe. J. Ecol.; 2022; 110, pp. 1996-2014. [DOI: https://dx.doi.org/10.1111/1365-2745.13957]
38. Henrich, V.; Jung, A.; Götze, C.; Sandow, C.; Thürkow, D.; Gläßer, C. Development of an online indices database: Motivation, concept and implementation. Proceedings of the 6th EARSeL SIG IS Workshop Imaging Spectroscopy: Innovative Tool for Scientific and Commercial Environment Applications; Tel Aviv, Israel, 16–18 March 2009.
39. Schellenberg, K.; Jagdhuber, T.; Zehner, M.; Hese, S.; Urban, M.; Urbazaev, M.; Hartmann, H.; Schmullius, C.; Dubois, C. Potential of Sentinel-1 SAR to Assess Damage in Drought-Affected Temperate Deciduous Broadleaf Forests. Remote Sens.; 2023; 15, 1004. [DOI: https://dx.doi.org/10.3390/rs15041004]
40. Joyce, S.; Olsson, H. Long-term forest monitoring with temporal-spectral trajectories from Landsat TM data. Proceedings of the Conference on Remote Sensing and Forest Monitoring; Rogow, Poland, 1–3 June 1999; pp. 68-81.
41. Wulder, M.A.; White, J.C.; Coops, N.C.; Butson, C.R. Multi-temporal analysis of high spatial resolution imagery for disturbance monitoring. Remote Sens. Environ.; 2008; 112, 2729. [DOI: https://dx.doi.org/10.1016/j.rse.2008.01.010]
42. Entcheva-Campbell, P.; Rock, B.N.; Martin, M.E.; Neefus, C.D.; Irons, J.R.; Middleton, E.M.; Albrechtová, J. Detection of initial damage in Norway spruce canopies using hyperspectral airborne data. Int. J. Remote Sens.; 2004; 25, pp. 5557-5583. [DOI: https://dx.doi.org/10.1080/01431160410001726058]
43. Townsend, P.A.; Singh, A.; Foster, J.R.; Rehberg, N.J.; Kingdon, C.C.; Eshleman, K.N.; Seagle, S.W. A general Landsat model to predict canopy defoliation in broadleaf deciduous forests. Remote Sens. Environ.; 2012; 119, pp. 255-265. [DOI: https://dx.doi.org/10.1016/j.rse.2011.12.023]
44. Lukeš, P. Monitoring of bark beetle forest damage. Big Data in Bioeconomy; Ödergård, M.; Habyarimana, B.; Fernandes, Z.-W. Springer International Publishing: Cham, Switzerland, 2021; pp. 351-361.
45. Thonfeld, F.; Gessner, U.; Holzwarth, S.; Kriese, J.; da Ponte, E.; Huth, J.; Kuenzer, C. A First Assessment of Canopy Cover Loss in Germany’s Forests after the 2018–2020 Drought Years. Remote Sens.; 2022; 14, 562. [DOI: https://dx.doi.org/10.3390/rs14030562]
46. Barka, I.; Lukeš, P.; Bucha, T.; Hlásny, T.; Strejček, R.; Mlčoušek, M.; Křístek, S. Remote sensing-based forest health monitoring systems—Case studies from Czechia and Slovakia. Cent. Eur. For. J.; 2018; 64, pp. 259-275.
47. Hawryło, P.; Bednarz, B.; Wężyk, P.; Szostak, M. Estimating defoliation of Scots pine stands using machine learning methods and vegetation indices of Sentinel-2. Eur. J. Remote Sens.; 2018; 51, pp. 194-204. [DOI: https://dx.doi.org/10.1080/22797254.2017.1417745]
48. Field, J.P.; Breshears, D.D.; Bradford, J.B.; Law, D.J.; Feng, X.; Allen, C.D. Forest Management Under Megadroughts: Urgent Needs at Finer Scale and High Intensity. Front. For. Glob. Chang.; 2020; 3, 502669. [DOI: https://dx.doi.org/10.3389/ffgc.2020.502669]
49. Manrique-Alba, A.; Begueria, S.; Camarero, J.J. Long-term effects of forest management on post-drought growth resilience: An analytical framework. Sci. Total Environ.; 2022; 810, 152374. [DOI: https://dx.doi.org/10.1016/j.scitotenv.2021.152374]
50. Bowditch, E.; Santopuoli, G.; Binder, F.; del Rio, M.; Kluvankova, T.; Lesinski, J.; Lesinski, J.; Motta, R.; Pach, M.; Panzacchi, P. et al. What is Climate-Smart Forestry? A definition from a multinational collaborative process focused on mountain regions of Europe. Ecosyst. Serv.; 2020; 43, 101113. [DOI: https://dx.doi.org/10.1016/j.ecoser.2020.101113]
51. Gregor, K.; Knoke, T.; Krause, A.; Reyer, C.P.O.; Lindeskog, M.; Papastefanou, P.; Smith, B.; Lansø, A.S.; Rammig, A. Trade-Offs for Climate-Smart Forestry in Europe Under Uncertain Future. Clim. Earth’s Future; 2022; 10, e2022EF002796. [DOI: https://dx.doi.org/10.1029/2022EF002796]
52. Ekstrand, S.; Schardt, M.; Granica, K.; Koch, B.; Kahabka, H.; Carnemolla, S.; Bottai, L.; Häusler, T. SEMEFOR. Satellite based environmental monitoring of European forests; EC Community Research, EUR 19435 Office for Official Publications of the European Communities: Luxembourg, 2001; 103p.
53. Vogelmann, J.E.; Rock, B.N. Assessing forest damage in high-elevation coniferous forests in Vermont and New Hampshire using Thematic mapper dat. Remote Sens. Environ.; 1988; 24, pp. 227-246. [DOI: https://dx.doi.org/10.1016/0034-4257(88)90027-2]
54. WRB. World Reference Base for Soil Resources. 2015; Available online: https://www.isric.org/explore/wrb (accessed on 24 June 2023).
55. Climate Atlas of Slovakia; Slovak Hydrometeorological Institute: Bratislava, Slovakia, 2015; 132p.
56. MPRV SR. Správa o lesnom hospodárstve v Slovenskej republike za rok 2021—Zelená správa. (In Slovak) Available online: https://www.mpsr.sk/zelena-sprava-2022/123---18463/ (accessed on 24 June 2023).
57. Intersucho. Drought Map for 32nd Week of 2022. Available online: https://www.intersucho.sk/en/?mapcountry=sk&from=2022-08-01&to=2022-08-10 (accessed on 27 December 2023).
58. Olsson, H. Regression functions for multitemporal relative calibration of thematic mapper data over Boreal forest. Remote Sens. Environ.; 1993; 46, pp. 89-102. [DOI: https://dx.doi.org/10.1016/0034-4257(93)90034-U]
59. Teillet, P.M.; Guindon, B.; Goodenough, D.G. On the slope-aspect correction of multispectral scanner data. Can. J. Remote Sens.; 1982; 8, pp. 84-106. [DOI: https://dx.doi.org/10.1080/07038992.1982.10855028]
60. Cochran, W.G. Sampling Techniques; Wiley & Sons: New York, NY, USA, 1977; 412p.
61. Scheer, Ľ. Assessment of forest conditions employing two-phased satellite remote sensing. Proceedings of the International Workshop: Application of Remote Sensing in European Forest Monitoring; Vienna, Austria, 14–16 October 1996; pp. 337-346.
62. Jackson, R.D. Spectral indices in N-Space. Remote Sens. Environ.; 1983; 13, pp. 409-421. [DOI: https://dx.doi.org/10.1016/0034-4257(83)90010-X]
63. Innes, J.L. The Sanasilva Manual: A Field Manual for Assessing Forest Damage; Swiss Federal Institute for Forest, Snow and Landscape Research: Birmensdorf, Switzerland, 1990.
64. Congalton, R.G.; Green, K. Assessing the Accuracy of Remotely Sensed Data: Principles and Practices; 3rd ed. CRC Press: Boca Raton, FL, USA, 2019.
65. Bucha, T. Classification of tree species composition in Slovakia from satellite images as a part of monitoring forest ecosystems biodiversity. Acta Instituti For. Zvolen; 1999; 9, pp. 65-84.
66. Joria, E.P.; Ahearn, S.A.; Connor, M. A comparison of the SPOT and LANDSAT Thematic Mapper 816 satellite systems for detecting gypsy moth defoliation in Michigan. Photogramm. Eng. Remote Sens.; 1991; 57, pp. 1605-1612.
67. Bucha, T.; Raši, R.; Vladovič, J. Metódy monitoringu zdravotného stavu lesov prostriedkami DPZ; Technical Report Lesnícky výskumný ústav vo Zvolene: Zvolen, Slovakia, 2002; 91p. (In Slovak) Available online: https://www.researchgate.net/publication/353761076_Metody_monitoringu_zdravotneho_stavu_lesov_prostriedkami_DPZ (accessed on 15 May 2024).
68. Ekstrand, S. Assessment of forest damage with Landsat TM: Correction for varying forest stand characteristics. Remote Sens. Environ.; 1994; 47, pp. 291-302. [DOI: https://dx.doi.org/10.1016/0034-4257(94)90097-3]
69. Maresi, G.; Salvadori, C. Crown conditions and damages in two forest ecosystems in Trentino (Italy). Studi Trent. Sci. Nat. Acta Biol.; 2005; 81, (Suppl. 1), pp. 253-260.
70. Koch, B.; Ammer, U.; Schneider, T.; Wittmeier, H. Spectroradiometer measurements in the laboratory and in the field to analyse the influence of different damage symptoms on the reflection spectra of forest trees. Int. J. Remote Sens.; 1990; 11, pp. 1145-1163. [DOI: https://dx.doi.org/10.1080/01431169008955085]
71. Stoklasa, M.; Fabiánek, P. Dálkový průzkum země a pozemní šetření zdravotního stavu lesů ČR. Lesn. Práce; 1998; 77, pp. 368-370. (In Czech)
72. Barka, I.; Bucha, T. Satelitné monitorovanie zdravotného stavu lesov Slovenska. Zborník odborných prác z konferencie LignoSilva 2021; Tóthová, S.; Gergeľ, T. NLC: Zvolen, Slovakia, 28 September 2021; pp. 84-92. (In Slovak)
73. Rullan-Silva, C.D.; Olthoff, A.E.; Delgado de la Mata, J.A.; Pajares-Alonso, J.A. Remote monitoring of forest insect defoliation. A review. For. Syst.; 2013; 22, pp. 377-391. [DOI: https://dx.doi.org/10.5424/fs/2013223-04417]
74. Kunca, A.; Zúbrik, M.; Galko, J.; Vakula, J.; Leontovyč, R.; Konôpka, B.; Nikolov, C.; Gubka, A.; Longauerová, V.; Maľová, M. et al. Salvage felling in the Slovak Republic’s forests during the last twenty years (1998–2017). Cent. Eur. For. J.; 2019; 65, pp. 3-11. [DOI: https://dx.doi.org/10.2478/forj-2019-0007]
75. Pretzsch, H.; del Rio, M.; Grote, H.J.; Klemmt, H.J.; Ordonez, C.; Oviedo, F.B. Tracing drought effects from the tree to the stand growth in temperate and Mediterranean forests: Insights and consequences for forest ecology and management. Eur. J. For. Res.; 2022; 14, pp. 727-751. [DOI: https://dx.doi.org/10.1007/s10342-022-01451-x]
76. WSL. Methods of the Sanasilva Inventory. Available online: https://www.wsl.ch/en/forest/forest-development-and-monitoring/sanasilva-forest-health-inventory/methods-of-the-sanasilva-inventory/ (accessed on 29 July 2024).
77. Munsell Color (firm). Munsell Plant Tissue Color Book; 2nd ed. re. Munsell Color: Baltimore, MD, USA, 1977.
78. Lakatos, F.; Mirtchev, S.; Mehmeti, A.; Shabanaj, H. Manual for Visual Assessment of Forest Crown Condition; Food and Agriculture Organization of the United Nations (FAO): Pristina, Kosovo, 2014; ISBN 978-92-5-108641-4
79. Janik, G.; Pödör, Z.; Koltay, A.; Hirka, A.; Juhász, J.; Kovacs, G.; Csóka, G. Effects of Meteorological and Site Parameters on the Health Status of Beech (Fagus sylvatica L.) Forests in Hungary. Acta Silv. Lignaria Hung.; 2020; 16, pp. 67-78. [DOI: https://dx.doi.org/10.37045/aslh-2020-0005]
80. Kodrík, H.; Hlaváč, P. Integrovaná ochrana lesa; Technická univerzita vo Zvolene: Zvolen, Slovakia, 2013; 328p. (In Slovak)
© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.
Details
; Pavlenda, Pavel 1 ; Konôpka, Bohdan 2
; Tomaštík, Julián 3
; Chudá, Juliána 3 ; Surový, Peter 4
1 National Forest Centre, Forest Research Institute Zvolen, T.G. Masaryka 22, SK-960 01 Zvolen, Slovakia;
2 National Forest Centre, Forest Research Institute Zvolen, T.G. Masaryka 22, SK-960 01 Zvolen, Slovakia;
3 Forestry Faculty, Technical University in Zvolen, T.G. Masaryka 24, SK-960 01 Zvolen, Slovakia;
4 Faculty of Forestry and Wood Sciences, Czech University of Life Sciences Prague, Kamýcká 129, CZ-165 00 Praha-Suchdol, Czech Republic;