Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Radar Satellite Imagery for Detecting Bark Beetle Outbreaks in Forests

Radar Satellite Imagery for Detecting Bark Beetle Outbreaks in Forests Purpose of Review The overall objective of this paper is to review the state of knowledge on the application of radar data for detecting bark beetle attacks in forests. Due to the increased availability of high spatial and temporal resolution radar data (e.g. Sentinel-1 (S1)), the question is how this time series data can support operational forest management with respect to forest insect damage prevention. Furthermore, available radar systems will be listed and their potential for detecting bark beetle attacks will be discussed. To increase the understanding of the potential of radar time series for detecting bark beetle outbreaks, a theoretical background about the interaction of the radar signals with the forest canopy is given. Finally, gaps in the available knowledge are identified and future research questions are formulated which could advance our understanding of using radar data for detecting forest bark beetle attacks. Recent Findings Few studies already demonstrate the high potential of S1 time series data for forest disturbance mapping in general. It was demonstrated that multi-temporal S1 data provide an excellent data source of describing the phenological characteristics of forests, which provide the basic knowledge for detecting bark beetle induced forest damages. It has been found that the optimal time for data acquisition is April to June for the pre-event and August to October for the post-event acquisitions. Summary For detecting bark beetle induced forest damages, the literature review shows that mono-temporal radar data are of limited use, that shorter wavelength (e.g. C-band; X-band) have a higher potential than longer wavelength such as L-band and that the current S1 time series data have a high potential for operational applications. . . . . . Keywords Forest SAR Time series Damages Monitoring Bark beetle Introduction an important economic aspect which need to be considered in operational forest management [7, 8]. Due to ongoing climate Forest disturbances, mainly caused by storm events, ice, snow, change and the fact that there is a synchronisation of forest fire, and insects, disrupt ecosystem dynamics and significantly disturbance rates from forest mortality patterns with climate influence ecosystem goods and services [1, 2] and strongly change, the topic of forest disturbance becomes an increasing- impact forest carbon budgets [3�� , 4]. Especially, in mountain- ly important issue. For example, Senf et al. [9]found arelation ous regions, the dynamics of forest disturbances are influenc- between satellite-derived estimates of forest disturbance rates ing the protective function of forests [5, 6] and, therefore, are and climate-related events since 1985 in forest landscapes of Central Europe. Furthermore, in Seidl et al. [10], an overview is given how climate change may affect disturbance regimes via direct, indirect and interaction effects. They summarized This article is part of the Topical Collection on Remote Sensing that warmer and drier conditions facilitate fire, drought and insect disturbances, while warmer and wetter conditions in- * Markus Hollaus crease disturbances from winds and pathogens. Furthermore, Markus.Hollaus@geo.tuwien.ac.at in the review paper of Pureswaran et al. [3�� ], the connection between forest insects and climate change is described. They Mariette Vreugdenhil Mariette.Vreugdenhil@geo.tuwien.ac.at concluded that the relationships between climate change and forest insects are clear in some species but cannot be general- Department of Geodesy and Geoinformation, TU Wien, Wiedner ized to all species. Finally, they concluded that process-based Hauptstraße 8-10, 1040 Vienna, Austria Curr Forestry Rep (2019) 5:240–250 241 phenology models permit predictions of population dynamics several spectral bands provide the potential for mapping forest if input data are appropriate. disturbances as summarized in the above cited review papers. Apart from climate-disturbance relationships, contempo- The majority of the applied algorithms for forest disturbance rary frequent and severe abiotic and biotic damage incidents mapping is based on the monitoring of forest dynamics and similarly affecting primary and managed conifer forests are thus a high temporal resolution is essential. However, for op- due to increasingly high disturbance susceptibility with regard tical satellite data, this requirement can limit the application to stand history, development stage and structure [11, 12]. for up-to-date monitoring due to the sensitivity of optical re- Species composition, stand age and density in combination mote sensing to clouds, aerosols and smoke. The advantage of with increased temperature and drought conditions were iden- radar is that it is not hindered by cloud cover, smoke, aerosol tified as the most relevant parameters for explaining/ contamination and low solar illumination. Due to the longer predicting salvage cuttings in Austrian Norway spruce forests wavelength of microwaves compared to optical remote sens- following attacks of Europe’s most important forest insect ing, the penetration depth in vegetation is larger, making it pest, the Eurasian spruce bark beetle, Ips typographus, during sensitive to the canopy and woody parts of the vegetation [19]. the period 2014–2016 [13]. Documentation of forest damage Nonetheless, until recently, the temporal and spatial reso- [14] shows that for example in Austria a strong increase of lutions of publicly available radar data were not sufficient for bark beetle induced timber is recorded since 1990 and reaches detecting forest insect attacks. With the launch of ESA’s approximately 20% and 28% of the total harvested timber for Sentinel-1 satellite constellation carrying C-band synthetic ap- 2017 and 2018, respectively. Similar situations are observed erture radar instruments, cross- and co-polarized backscatter for other European and North American countries [15]. data are now available for the first time at the spatial and Based on the increased frequency of forest insect attacks temporal resolution needed. Hence, the overall objective of and the attendant increase of ecological and economic im- this paper—to review the state of knowledge on the applica- pacts, there is a strong demand for remote sensing–based tion of radar data for detecting bark beetle attacks in forests— monitoring approaches. Several publications can be found that is formulated. Based on a systematic literature review of pa- use different remote sensing data for detecting forest insect pers published since 2000, different applications of radar data attacks. Most studies use optical remote sensing data, and only for detecting forest disturbances due to bark beetle attacks few studies use radar data. Optical remote sensing makes use including their applied methods are summarized. of the large absorption of light in the visible spectrum and Additionally, an overview of available types of forest insect reflection in the near infrared and short wave infrared spec- infestations and the related physiological changes in the can- trum during photosynthesis to derive information about the opy with respect to remotely sensed properties is given. This forest canopy. Radar systems transmit pulses of electromag- review paper focuses on bark beetle induced forest damages as these insect attacks lead to extensive calamities in coniferous netic energy and record the reflected signal at the sensor from the area of interest. Most systems, which aim at monitoring the forests. As the interaction of the radar signal with forest can- Earth’s surface, operate in the microwave domain, with fre- opies is different from the spectral properties that are impor- quencies between 3 and 30 GHz. In the microwave domain, tant in optical remote sensing, an introduction to radar and the backscattered signal is sensitive to a number of land sur- microwave remote sensing is given, including a theoretical face parameters, including, but not limited to, water content in background about the interaction of microwaves with vegeta- soil and vegetation and vegetation structure. However, most of tion. Finally, gaps in the available knowledge are identified the investigated publications use optical satellite data as has and future research questions are formulated which could ad- been summarized in the recent review paper about remote vance our understanding of using radar data for detecting for- sensing of forest insect disturbance mapping from Senf et al. est bark beetle attacks in an operational sense. [16], which gives an extensive overview of different insect types and remote sensing data and methods for mapping dis- turbances. In this review paper, only one paper was mentioned Types of Bark Beetle Infestations and Impact that uses TerraSAR-X radar data in combination with optical on the Canopy Properties for Optical high-resolution RapidEye data for the early detection of bark and Radar Data beetle green attacks [17]. Furthermore, the review paper of Stone and Mohammed [18] gives an overview of applications The most important forest insects can be grouped into of remote sensing technologies for assessing planted forests xylophagous and folivorous insects. As summarized in Seidl damaged by insect pests and fungal pathogens. They conclude et al. [10], bark beetle species are the most prominent that the simultaneous acquisition of spectral and 3D point data xylophagous insects, whereas defoliators are the most impor- will have the highest benefit for assessing tree health in a cost- tant species in the group of folivorous insects. The insects of effective manner. From an operational point of view, the spec- these two groups affect trees in a different manner. Bark beetle tral properties of optical satellite data have the benefit that species bore into the phloem of trees, copulate, mine galleries 242 Curr Forestry Rep (2019) 5:240–250 and oviposit [20]. Furthermore, they introduce several species Characteristics of Radar Backscatter of fungi that colonize the phloem and vascular tissue. Finally, from Forests the translation of water and nutrients within the tree is getting interrupted leading to the death of the tree. In contrast to bark Radar systems transmit and receive pulses of electromagnetic beetles, defoliating species feed on leaves and needles and energy. For most radar systems, two polarizations are com- thus influence the photosynthetic activity of a tree. This can mon, vertical polarization (V) which is perpendicular to the lead to growth reduction, deformation or to tree mortality. surface and horizontal polarization (H) which is parallel to the Depending on the infestation intensity and the consequent surface. The amount of electromagnetic energy intercepted degree of defoliation and the surrounding environmental and re-radiated is a complex combination of multiple factors stress factors (e.g. drought), trees can regenerate in the follow- including electromagnetic and geometric properties of the tar- ing year or in the case of some broadleaved tree species even get and sensor specifics such as wavelength and polarization in the same year [10]. of the transmitted and received signal. The energy intercepted For mapping infestations with remote sensing techniques, and reflected is described by the so-called radar cross section. knowledge about the physiological effects of the attacks in the In remote sensing of the Earth’s surface, the target is an area in forest canopy is essential. Niemann and Visintini [21� ] de- which it is assumed that no single scatterer dominates. To scribe three stages that occur during a bark beetle attack. compare measurements from different sensors with different The green attack stage occurs during the host colonization footprints, i.e. target areas, the radar cross section can be nor- and the development of the first bark beetle population. malized by the area, resulting in the backscatter coefficient. During this phase, the physiology of the tree is not influenced The most important electromagnetic properties of the me- and no visual change in the canopy can be observed, even dia for microwave remote sensing are the electric permittivity. though the tree is getting stressed due to the increased produc- It is a complex number with a real and imaginary part, where tion of resin and due to decreased water availability. The de- the real part is often referred to as the dielectric constant. The tection of this very early infestation stage would have the dielectric constant (ε) of the medium with which the wave highest benefit for supporting operational forest management, interacts plays an important role in the amount of but remote sensing has been of limited success until now backscattered energy. In the microwave domain, electromag- [21� ]. A reliable detection of infested trees can only be done netic waves incident on the Earth’s surface excite water mol- with ground surveys by searching for entry and exit holes on ecules. The dipole character of water causes the water mole- the bole, boring dust and pitch tubes, which are all indicators cules to continuously reorient in the electromagnetic radia- of beetle infestations [22]. However, such field work can only tion’s oscillating electric field. This leads to high ε values of be done for sample plots due to the high demand of time and water (ε = 80), compared to dry vegetation matter and soil consequently costs for such in situ measurements. (1.5 < ε <2) [25, 26]. The red attack stage includes the period where the colour of The geometric properties of the target area influence the needles turns to yellow and brown and where the tree starts to type of scattering. If a medium is homogenous, most of the lose needles. Additional to the changing colour of the canopy, electromagnetic wave is scattered at the boundary of the sur- the needle water content is the most important physiological face. The amount of energy reflected can be described by the change during the red attack stage [23, 24]. The final grey attack Fresnel coefficient which is a function of incidence angle and stage describes the period where the tree will lose all of its ε. This mechanism is called surface scattering or specular foliage. From a remote sensing perspective, short-wave infrared reflection. Volume scattering occurs if a medium is inhomo- wavelengths are most sensitive to changes in the needle water geneous and the signal penetrates the lower medium. Within content [10]. The majority of the studies investigated in the the medium, randomly distributed dielectric inhomogeneities review of Seidl el al. [10] used the normalized difference veg- scatter the incoming electromagnetic waves in all directions etation index (NDVI) for mapping broadleaved defoliations regardless of incident angle [26]. The amount of energy while different indices were used for mapping coniferous defo- returned to the sensor is proportional to the number of dielec- liation (NDVI, Normalized Burn Ratio, Moisture Stress Index, tric inhomogeneities and the difference in ε between them and Leave Area Index). For mapping bark beetle infestations, the the medium. most frequently used index was the difference in the Tasseled The total backscatter of a vegetated surface can be de- Cap Wetness component. However, when doing time series scribed by scattering from the soil, the vegetation and an in- analyses based on spectral images for detecting bark beetle teraction between the two. Soil and vegetation produce two infestations, one has to consider that there is a natural decrease different scattering mechanisms, namely surface scattering in the chlorophyll content of needles during fall and winter and volume scattering. Vegetation is often described as a seasons. This decline in chlorophyll content is due to dormancy cloud of water droplets held in place by dry matter: an inho- and frost hardening of the needles which is required in order to mogeneous medium with randomly distributed dielectric in- protect foliage against photo-oxidative damages [25]. homogeneities. Since vegetation consists for a large part of Curr Forestry Rep (2019) 5:240–250 243 water, ε of the total vegetation increases rapidly with increas- and low backscatter was observed in winter and high back- ing above ground biomass and moisture status. From this, it scatter in summer [34� , 35, 37� ]. A recent study on backscatter can be inferred that backscatter increases with increasing from Sentinel-1 over a deciduous forest in France showed that above ground biomass and moisture status. the ratio of co- and cross-polarized backscatter coincides tem- The dielectric inhomogeneities can also change the polari- porally with NDVI, increasing in spring and decreasing in zation of the incident wave. Although this does not happen to autumn. However, no conclusions were given to what drives a large degree, a change in polarization is more likely to occur the variations in the so-called cross ratio [35]. The potential of with volume scattering or with interaction between the soil radar for monitoring changes in forest structure and moisture and the vegetation. Because of the low occurrence, cross- content was also demonstrated by studies of drought effects in polarized backscatter (HV or VH transmit-receive) is usually the Amazon. Saatchi et al. [38] showed the complementary lower than co-polarized backscatter (HH or VV transmit-re- information which can be gained next to optical observations ceive). It does, however, show a stronger increase with vol- when investigating the effects of the Amazonian drought of ume and double-bounce scattering than co-polarized back- 2005. A decrease in backscatter was observed after the 2005 scatter. Thus, when describing vegetation as a water cloud, it drought, which took place over a long time period, and was can be inferred that cross-polarized backscatter is more sensi- attributed to a reduction in canopy water and a change in tive to changes in vegetation. structure. In order to use radar data for forest monitoring, it needs to be taken into account that, depending on the density of the vegetation canopy, wavelength of the signal and the incidence Radar Data for Bark Beetle Disturbance angle of the observations, microwaves penetrate the vegeta- Mapping tion and scattering from the soil or from soil-vegetation inter- action could contribute to the total backscatter signal [27]. This review paper was based on a screen of the literature Most studies agree that the contribution of the soil and soil- between 2000 and 2019 using the ISI Web of Science database vegetation interaction is small in co-polarized backscatter (http://www.webofknowledge.com/) and Scopus (https:// from forests, but not negligible. Especially for longer wave- www.scopus.com) with the general search terms focussing lengths, low biomass forests, high soil moisture conditions or on synthetic aperture radar, forests and insect disturbances. for lower incidence angles the total backscatter signal is sen- In detail, the following search string was used: ALL sitive to the soil surface [28–31]. In addition, due to its sensi- FIELDS: (sar OR radar OR microwave OR sentinel*1 OR tivity to volume scattering, studies have found cross-polarized ers* OR tandem-x* OR l-band) AND ALL FIELDS: backscatter to be most sensitive to changes in forest above (disturbance OR insect OR attack OR beetle) AND ALL ground biomass and that the contribution of soil and soil- FIELDS: (satellite OR airborne) AND ALL FIELDS: (forest vegetation interaction was found to be negligible [31–33]. OR tree). Within the time period from 2000 to 2019, 105 hits Apart from the above ground biomass and moisture con- appear containing these keywords in all fields. After a tent, the structure of the vegetation can play a large role on the screening of the results, only four publications could be temporal backscatter signal. This is especially the case in de- identified as being relevant. In the following paragraphs, an ciduous forests where major changes in structure occur due to overview of these publications is given. the development of the foliage. One of the first studies ad- Ranson et al. [39] explored the use of combined use of dressing the effect of structural changes of vegetation on the mono-temporal JERS and Radarsat data for detecting insect backscatter signal was performed by [27] when defoliating damages in Siberian forests. They used co-polarized (HH) corn. The effect of each major plant constituent, i.e. leaves, JERS and Radarsat data from 2 years. After the pre- stalks and fruits, on backscatter at 5.1 GHz was investigated. processing steps, they applied a 3 × 3 Frost filter to each For high incidence angles, backscatter was higher from SAR data set to reduce the speckle. For both data sets, the defoliated corn, i.e. stalks and cobs, than from the whole plant. backscatter coefficients and the standard deviations showed The backscatter from stalks or interaction between soil and low differences across insect damaged and healthy coniferous stalks is very significant but is strongly attenuated by the pres- forests. As a reason for that, they argued that for JERS the ence of leaves. In deciduous forests, similar behaviour was damaged foliage have a negligible influence on the backscat- observed where higher backscatter was often found in winter ter signal because they are too small. For Radarsat, only the followed by a drop in backscatter in spring and summer and an separability between coniferous forest and clear-cuts was increase in autumn. This was attributed to the change in scat- high. The reason for this can be found in the occurring volume tering mechanism from volume to surface scattering and the scattering within the tree canopies, whereas volume scattering contribution of soil and soil-vegetation scattering to the total from grassy clear-cuts is small. The combined use of Radarsat signal when no foliage is present [28, 34� , 35, 36, 37� ]. For and JERS slightly increased the classification accuracies and coniferous forests, this temporal signature was not observed, 29% of the severely insect damaged classes and 46% of the 244 Curr Forestry Rep (2019) 5:240–250 moderately insect damaged area could be classified. Finally, coefficient was normalized with a DTM, and a multi- they concluded that the results were limited when using any temporal filtering was applied to decrease speckle. The single-channel radar data. multi-temporal filtering was based on five pre-event and six Kaasalainen et al. [40] studied the potential of ERS-2 SAR post-event SAR data sets. Radar change ratios were computed data for detecting defoliation of Scots pine dominated forest using pre- and post-disturbance averaged backscatter coeffi- located in eastern Finland caused by the European pine saw- cients to better represent the forest condition by removing the fly. They used terrestrial laser scanner (TLS) to examine the stochastic part of the signal. They found out that a change of defoliation during the active period of the pine sawfly hazard. the backscatter coefficient of − 1.0 dB is a consistent indicator The first measurement was undertaken during the early phase of forest structural changes induced by bark beetle attacks. of defoliation and the second following the defoliation period. They also discuss the influence of weather conditions to the For the study area, seven ERS-2 images were available from discrimination capacity. For example, the drier conditions of the same orbit and cover the time period from before and after the reference data set and/or wetter conditions for the remain- the defoliation. For all images, only the amplitude information ing images lead to nearly no changes of the backscatter coef- of backscattering signal was available. For their analyses, the ficients. The classification of the areas affected by bark beetle ERS-2 backscatter values were averaged using a circle with a outbreaks was done using a threshold approach and support 50-m radius to reduce speckle. The results show only a slight vector machines based on the backscatter change ratios. They change in the averaged SAR backscatter for those plots where reached overall accuracies ranging from 74 to 91% depending defoliation could be observed in the TLS data. For plots where on the selected image pair. They conclude that the classifica- signs of defoliation were not present, little or no changes in the tion accuracy for windthrow affected areas was high and low- backscatter were observed. As no weather data was available er for bark beetle–induced damages due to the fact that chang- for the acquisition times, the influence of the soil surface and es are more abrupt for windthrow than for insect outbreaks. vegetation moisture to the backscatter signal could not be They also state that the interpretation of the backscatter signal investigated in this study. can be complex because of the multitude of factors affecting Ortiz et al. [17] evaluated the capabilities of RapidEye and them, the limitations due to the density of the radar observa- TerraSAR-X imagery for detecting areas affected by bark bee- tions and that it is difficult to extract the specific stress factor. tle green attacks. They investigated generalized linear models, In a more general sense, few studies are available that use maximum entropy and random forest for detecting bark beetle SAR data for detecting forest disturbances. Even though these infested areas in a flat study area in south-west Germany. The studies are not focussing on bark beetle disturbance mapping, RapidEye and the TerraSAR-X data were acquired in two their applied methods could be useable for it. For example, De consecutive days at the end of May. Both data sets were Grandi [42� ] used spatial wavelet statistics of ENVISAT ASAR and ALOS PALSAR backscatter to describe the struc- orthorectified using a high-resolution digital surface model with a resolution of 2 m derived from airborne laser scanning ture of degraded forests in Cameroon, Central Africa. They data. To investigate the potential of both data sets for bark found out that analytic parameters describing the functional beetle green attack detection, explanatory variables were de- form (i.e. third-degree polynomial function) of the scaling rived from circular plots with a radius of 5 m covering either signatures show statistically significant differences between only attacked or vital trees. The explanatory variables from the signatures of intact and degraded forests for ENVISAR TerraSAR-X were the standard deviation, max, min, mean, ASAR data. For the ALOS PALSAR, no significant differ- median and the first and third quartiles of the backscatter dis- ences could be found. As a reason for that the authors argue tribution. The analyses have shown that the third quartiles and that L-band penetrates more into the canopy, and therefore, the the median of the backscatter were higher for attacked plots observed backscatter texture is influenced more by the distri- than for vital plots. They achieved an overall accuracy of 87% bution of large scattering elements and by the ground return, (kappa 0.23) for the TerraSAR-X data and 94% (kappa 0.51) whereas the shorter wavelength of the C-band ENVISAT for the RapidEye data. Finally, they conclude that improve- ASAR is able to describe the canopy much better. Lei et al. ments could be expected if multi-temporal SAR data and/or [43] used spaceborne repeat-pass InSAR correlation measure- varying polarizations are available. ments in combination with a physical InSAR scattering model Tanase et al. [41] used ALOS L-band spaceborne synthetic to detect forest disturbances over selective logging sites in aperture radar (SAR) within a change detection framework to Queensland, Australia. They used dual-pol ALOS PALSAR delineate forested areas affected by wind and insect distur- InSAR coherence data with a temporal baseline of 92 days. bances. Several SAR images from the same orbit before and The basic idea of their approach is to compensate for the after the bark beetle attack were selected as a basis for a pixel- scene-wide mean behaviour of the volumetric and temporal based comparison. The cross-polarized (HV) channel was decorrelation effects and then use the residual decorrelation as mainly investigated because of its sensitivity to forest struc- an indicator of forest disturbance. The detected forest distur- tural characteristics and their change. The backscatter bance areas (i.e. forest and non-forest areas) were compared to Curr Forestry Rep (2019) 5:240–250 245 results from Landsat time series with a relative RMSE of 13% where cloud cover is commonly the limiting factor for over 0.81 ha forest stands. One limitation of this approach is using optical data only. The Sentinel-1 data consists of that the disturbance affected area cannot be large compared ten acquisitions with single polarisation (VV) from the with the spaceborne InSAR image itself because the mean same orbit acquired within 9 months. In a first step, the behaviour of the volumetric and temporal decorrelation effect forest disturbance mapping was done in parallel for the is estimated from the scene-wide forest coherence distribution. radar and the optical data. For the combination of both results, the Bayes’ theorem was used. The results showed the highest detection accuracies for the combined ap- Discussion and Outlook proach (83.7% for the area-based and 97.1% for the plot-based validation). Furthermore, the results showed The reviewed papers show the limits of mono-temporal radar for plot areas < 1 ha lower detection accuracies for radar images for forest disturbance mapping. Since the availability data than for optical data. of Sentinel-1 data, the popularity of radar images for forest Furthermore, Sentinel-1 time series were used for disturbance mapping has increased. In addition to the freely distinguishing forest from non-forest areas or for tree species available Sentinel-1 data, other radar sensors exist but often classifications. For example Dostálová et al. [47]used their data are not freely available, especially radar data with X- Sentinel-1 time series data for forest area delineation with band are often not freely available. An overview of available a spatial resolution of 10 m in the eastern part of Austria and and planned satellite radar sensor is given in Table 1.An achieved accuracies of 88% compared to a forest mask de- extensive overview of satellite missions can be found in the rived from airborne laser scanning data. A similar result can eoPortal Directory [44]. also be found in the study of Dostálová et al. [34� ], in which For Europe, the Sentinel-1 satellites offer high tempo- dense Sentinel-1 time series backscatter signals were used ral (3–6 days) and spatial (5 × 20 m) resolution, which are for forest area delineation and classification into tree species independent of the cloud clover. Time series from groups. For three European study sites, they achieved over- Sentinel-1 are promising not only for bark beetle distur- all accuraciesof86–91% for the forest area delineation and bance mapping but also for other forest disturbance map- 65–85% for the classification into broadleaf forests, conifer- ping applications such as windthrow or clear-cut detec- ous forests and non-forest classes. A similar result is also tion. For example Rüetschi et al. [45]used Sentinel-1C- reported in Rüetschi et al. [37� ]. They achieved an overall band VV and VH polarisation data for a rapid windthrow accuracy of 86% for the classification into coniferous and detection in mixed temperate forests in Switzerland and deciduous forests and 72% for the classification of three northern Germany. They used five Sentinel-1 acquisitions different tree species. These studies used the differences in the backscatter seasonality between different tree species. before the event and ten after the storm event including ascending and descending data. The backscatter values of Backscatter variations were further found to hint at differ- Sentinel-1 were normalised using the modelled local illu- ences in the forest structure (mainly tree heights) and vege- minated area based on a DTM rather than a modelled tation cover; however, differences were less pronounced local incident angle based on an ellipsoid. In this way, than between forest types [47]. the influence of the topography on the SAR acquisition Also, Rauste et al. [48] used Sentinel-1 time series data is mitigated and the modelled illuminated area was over a period of 1 year to describe the temporal behaviour of exported as an additional input layer for the further forest the C-band backscatter for areas representing areas clear-cut disturbance mapping. For the detection of disturbed forest during the Sentinel-1 acquisition period, unchanged forest areas, the change detection method “image differencing” areas and areas clear-cut before the Sentinel-1 acquisition was applied for VV and VH polarisations. The mean and started. The analyses were done for a boreal forest site in the standard deviation were calculated for the difference Finland. They could not only show the potential for detecting images. By summing up the differences of VV and VH logged forest areas despite high sensitivity to seasonal and polarised backscatter values, a heuristic windthrow index weather conditions but also mention that the detection of log- was calculated, which was the basis for a decision tree ging areas was only successful for good scenes. Consequently, classification. The results show that the backscatter from they suggest the use of Sentinel-1 data from two winters for windthrown areas was typically about 0.5 dB higher than the clear-cut mapping. before the storm event. The detected windthrown areas In addition to these summarized studies, Hollaus et al. had a minimum extend of 0.5 ha and a producer accuracy [49�� ] analysed the potential of Sentinel-1 time series data of 0.88 and a user accuracy of 0.85. Also, Hirschmugl for recognizing the beginning of bark beetle infestations. et al. [46] showed the potential of Sentinel-1 time series Within their study, homogenous areas with vital and data in combination with optical Sentinel-2 and Landsat-8 infested forest stands were analysed separately for a study data for forest disturbance mapping in tropical forests area in the north-eastern part of Austria. These forest 246 Curr Forestry Rep (2019) 5:240–250 Table 1 Overview of operating and planned radar sensors Sensor Launch date Band Spatial resolution Orbital repeat (days) Polarisation Owner/operator Scientific data access (m) ALOS-2 PalSAR 2014 L 10–100 14 Fully Japan Aerospace Exploration freely available polarimetric Agency NISAR Scheduled L3–10 12 Fully NASA ? 2021 polarimetric RADARSAT-2 2007 C 3–100 24 Fully Canadian Space agency (CSA) Restricted polarimetric Sentinel-1 2013 C 10 12 VV, VH (over European Space Agency freely available land) RADARSAT-constellation Scheduled C1–100 12 Fully Canadian Space Agency ? (3-satellites) 2019 polarimetric Risat-1 2013 C 1–50 25 Fully Indian Space Research Available to Copernicus polarimetric Organisation users PAZ/PAZ 2 SAR 2001/2018 X 0.25–40 > 2 Fully Spanish Space Agency Restricted polarimetric TerraSAR-X 2007 X 1–16 11 Fully German Aerospace Centre Restricted polarimetric (DLR) Tandem-X 2010 X 1–16 11 Fully German Aerospace Centre Restricted polarimetric Cosmos Skymed-1/2 2007/2010 X 1–100 16; revisit time for the full constellation is in the range Fully Italian Space Agency Restricted of few hours polarimetric TecSAR 2008 X 1–814 Fully Israel Aerospace Industries Restricted polarimetric Kompsat-5 2013 X 1–20 28 Fully Korean Space Agency Restricted polarimetric Kompsat-6 Scheduled X0.5–20 11 Fully Korean Space Agency ? 2020 polarimetric Curr Forestry Rep (2019) 5:240–250 247 Fig. 1 Monthly averaged Sentinel-1 backscatter ratios for bark beetle infested and vital forest patches. a The ratios of the monthly averaged backscatter values for vital and infested forest stands. b The differences of the ratio values stands were derived from Sentinel-2 time series by apply- stand mortality may be detected only months later. Due ing a change detection approach. As the backscatter signal to missing reference data with detailed information about underlies a natural variation due to phenological changes the degree of infestation, it could not be analysed which of the canopy and the moisture condition in the canopy degree of bark beetle related damage can be recognized in and the terrain [34� ] monthly ratios were calculated for Sentinel-1 time series data. In addition to the used differ- vital and infested forest stands from backscatter coeffi- ences in the backscatter values for vital and infested for- cients from the year before the attack and the year of ests, one has to consider annual differences in the back- the bark beetle attack. In Fig. 1a, the ratios of the back- scatter values. For example Rüetschi et al. [37� ] showed scatter coefficients for vital and infested forest stands are inter-annual differences of the backscatter from coniferous given, whereas in Fig. 1b, the difference of the ratios is study areas. The seasonal variations are minimized by shown for the VH polarized backscatter values. calculating the ratios between the backscatter signals. For this analysis, Sentinel-1 data from the year 2015, However, the influence of the moisture conditions in the where for this investigation area no bark beetle attacks canopy as well as in the terrain surface was not consid- took place, and Sentinel-1 data from the year 2017, where ered in this study yet. The consideration of local varia- several bark beetle attacks occur, were selected. The tions in the backscatter behaviour could improve the de- Sentinel-1 time series analyses showed a clear difference tection accuracy by applying local adaptive thresholds in of the backscatter coefficients between vital and infested the change detection workflow. forest stands of ~ 1 dB for both VV- and VH-polarized Sentinel-1 backscatter values. The differences can also be seen in the ratio values as shown in Fig. 1 and start ap- proximately in July to August, which corresponds well Conclusions with a terminated development of the first filial generation and starting of tree dieback, corresponding to the grey The conclusion gives a short summary of the main findings of stage. However, crowns of Norway spruce often remain the reviewed publications for forest disturbance mapping with green long after a bark beetle attack was initiated, and the focus on insect induced damages. The potential and the 248 Curr Forestry Rep (2019) 5:240–250 limitations for operational applications are summarized, and The following open research questions are formulated: future research questions are formulated. The following conclusions are drawn: 1. Commonly speckle reduction is done by applying filters with a fixed kernel size and shape. As insect-induced for- 1. Defoliators and bark- and wood-boring insects have the est damage starts within small forest patches that are cov- most potential to cause damage in forests. The most prom- ered by only a few radar pixels, this type of speckle re- inent insects are bark beetles and after an attack trees duction leads to an averaging and consequently to commonly die within few months. In Niemann and “mixed” pixels. Therefore, research is needed around Visintini [21� ], three stages are defined that occur during how a land cover–dependent and/or time-dependent a bark beetle attack, the green, the red and the grey attack speckle reduction can improve the detectability of forest stage. From a remote sensing perspective, the detection of disturbances. the green attack stage has had limited success as only 2. The reviewed studies have shown that the low temporal slight changes are ongoing in the foliage during this stage. resolution is the most limiting factor in using radar data 2. Until the start of the Copernicus Sentinel-1 satellites, very for the detection of insect induced forest damages. The few radar data sets were used for detecting forest distur- new Sentinel-1 time series data allows the description of bances due to insect infestations. The available sensors phenological characteristics with a high level of detail, (i.e. JERS, ERS-1, Radarsat, TerraSAR-X, ALOS which is the basis for forest disturbance mapping. As PALSAR) operate in different wavelengths (X-, C-, L- shown in the studies of Dostalova et al. [34� ]and band) and have either limited spatial or temporal resolu- Rüetschi et al. [37� ], the backscatter signal has an annual tion. Most of the studies used the backscatter signal from but also inter-annual variability. Furthermore, local varia- acquisitions from the same orbit (either ascending or de- tions can occur. Therefore, research is needed around how scending), and only in the study of Lei et al. [43], ALOS these small scale variations (i.e. in time and space) in the PALSAR InSAR coherence data were used. backscatter signal can be considered. This research also 3. Concerning the processing steps and the applied methods needs to consider influences from varying satellite orbits, for disturbance mapping, it can be concluded that after i.e. ascending and descending. topographic correction, speckle reduction is undertaken 3. From an operational forest management point of view, it is by applying various filters or by averaging the backscatter important to know the degree of damage (i.e. green, red, signal for plots or forest stands. grey attack) or the minimum mapping unit that can be 4. Several studies have shown that shorter wavelengths (e.g. derived from radar data. To answer this research question, C-band; X-band) have a higher potential than longer appropriate reference data are necessary. wavelengths, such as L-band radar data, for detecting in- 4. In the reviewed papers, only one [43] used coherence sect induced forest damages. For example De Grandi data. Due to the relatively long temporal baseline of [42� ] argued that L-band penetrates more into the canopy, 92 days and the fact that they used L-band data, the results and therefore, the observed backscatter texture is influ- for detecting insect infested forest areas were not satisfac- enced more by the distribution of large scattering ele- tory. Due to the availability of C-band, Sentinel-1 data ments and by the ground return, whereas the shorter research is needed to analyse the potential of the coher- wavelengths of, e.g. the C-band are able to describe the ence information for forest disturbance mapping canopy much better. applications. 5. The high temporal and spatial resolutions of Sentinel-1 5. As already shown in several studies [37� , 50, 51], time series have shown a high potential for forest distur- Sentinel-1 time series data in combination with optical bance mapping. Until now, the Sentinel-1 data have been data from, e.g. Sentinel-2, have a high potential for for- mainly used for windthrow mapping (e.g. [45]). Based on estry applications. For insect-induced forest disturbance, deviations from the natural variability of the backscatter mapping further research is needed to explore the full signal during the phenological cycle ,Hollaus et al. [49 ] potential of multi-sensor data. demonstrated for the first time the potential of this data Funding Information Open access funding provided by TU Wien source for detecting bark beetle infestations in coniferous (TUW). Mariette Vreugdenhil was supported by a Living Planet forest. Fellowship from the European Space Agency (4000125441/18/I-NS). 6. Although though there is still a lack of knowledge about the applicability of Sentinel-1 time series for insect distur- Compliance with Ethical Standards bance mapping, the study of Hollaus et al. [49�� ] sug- gested that the optimal time for data acquisition is April Conflict of Interest Markus Hollaus and Mariette Vreugdenhil declare to June for the pre-event and August to October for the that they have no conflict of interest. post-event acquisitions. �� Curr Forestry Rep (2019) 5:240–250 249 Human and Animal Rights and Informed Consent This article does not 12. Schurman JS, Trotsiuk V, BačeR, Čada V, Fraver S, Janda P, et al. contain any studies with human or animal subjects performed by any of Large-scale disturbance legacies and the climate sensitivity of pri- the authors. mary Picea abies forests. Glob Chang Biol. 2018;24(5):2169–81. 13. Netherer S, Pennerstorfer BMJ. Trockenstress von Open Access This article is distributed under the terms of the Creative Fichtenbeständen fördert den Schadholzanfall durch Commons Attribution 4.0 International License (http:// Buchdrucker. Forstschutz Aktuell. 2018;65:1–9. creativecommons.org/licenses/by/4.0/), which permits unrestricted use, 14. Hoch G, Chrisitian L. Documentation of forest damaging factors. distribution, and reproduction in any medium, provided you give appro- BFW. 2019. http://bfw.ac.at/rz/bfwcms2.web?dok=9756. Accessed priate credit to the original author(s) and the source, provide a link to the 24. 03. 2019. Creative Commons license, and indicate if changes were made. 15. Katz C. Small Pests, Big problems: the global spread of bark bee- tles. Yale School of Forestry & Environmental Studies. 2017. https://e360.yale.edu/features/small-pests-big-problems-the-global- spread-of-bark-beetles. Accessed March 2019. 16. Senf C, Seidl R, Hostert P. Remote sensing of forest insect distur- References bances: current state and future directions. Int J Appl Earth Obs Geoinf. 2017;60:49–60. https://doi.org/10.1016/j.jag.2017.04.004. 17. Ortiz S, Breidenbach J, Kändler G. Early detection of bark beetle Papers of particular interest, published recently, have been green attack using TerraSAR-X and RapidEye data. Remote Sens. 2013;5(4):1912–31. https://doi.org/10.3390/rs5041912. highlighted as: 18. Stone C, Mohammed C. Application of remote sensing technolo- � Of importance gies for assessing planted forests damaged by insect pests and fun- �� Of major importance gal pathogens: a review. Curr For Rep. 2017;3(2):75–92. https:// doi.org/10.1007/s40725-017-0056-1. 1. Maroschek M, Seidl R, Netherer S, Lexer MJ, et al. Climate change 19. Liu YY, De Jeu RAM, McCabe MF, Evans JP, Van Dijk AIJM. impacts on goods and services of European mountain forests. Global long-term passive microwave satellite-based retrievals of Unasylva. 2009;60(231/232):76–80. vegetation optical depth. Geophys Res Lett. 2011;38(18). https:// 2. Thom D, Seidl R. Natural disturbance impacts on ecosystem ser- doi.org/10.1029/2011GL048684. vices and biodiversity in temperate and boreal forests. Biol Rev. 20. Raffa KF, Aukema BH, Bentz BJ, Carroll AL, Hicke JA, Turner 2016;91(3):760–81. MG, et al. Cross-scale drivers of natural disturbances prone to an- 3.�� Pureswaran DS, Roques A, Battisti A. Forest insects and climate thropogenic amplification: the dynamics of bark beetle eruptions. change. Curr For Rep. 2018;4(2):35–50. https://doi.org/10.1007/ BioScience. 2008;58(6):501–17. https://doi.org/10.1641/B580607. s40725-018-0075-6. This paper gives a review of effects of 21.� Niemann K, Visintini F. Assessment of potential for remote sensing climate change on forest insects. detection of bark beetle-infested areas during green attack: a litera- 4. Seidl R, Schelhaas M-J, Rammer W, Verkerk PJ. Increasing forest ture review. Victoria, BC, Canada: Natural Resources Canada, disturbances in Europe and their impact on carbon storage. Nat Canadian Forest Service; 2005. http://cfs.nrcan.gc.ca/ Clim Chang. 2014;4(9):806–10. https://doi.org/10.1038/ pubwarehouse/pdfs/25269.pdf. This working paper explains the nclimate2318. different stages of insect infestations and how the individual 5. Brang P. Resistance and elasticity: promising concepts for the man- stages can be detected with remote sensing data. agement of protection forests in the European Alps. For Ecol 22. Robertson C, Farmer CJQ, Nelson TA, Mackenzie IK, Wulder MA, Manag. 2001;145(1–2):107–19. https://doi.org/10.1016/s0378- White JC. Determination of the compositional change (1999–2006) 1127(00)00578-8. in the pine forests of British Columbia due to mountain pine beetle 6. Bebi P, Kienast F, Schönenberger W. Assessing structures in moun- infestation. Environ Monit Assess. 2008;158(1–4):593–608. tain forests as a basis for investigating the forests’ dynamics and https://doi.org/10.1007/s10661-008-0607-9. protective function. For Ecol Manag. 2001;145(1–2):3–14. https:// 23. Skakun RS, Wulder MA, Franklin SE. Sensitivity of the thematic doi.org/10.1016/s0378-1127(00)00570-3. mapper enhanced wetness difference index to detect mountain pine 7. Notaro S, Paletto A. The economic valuation of natural hazards in beetle red-attack damage. Remote Sens Environ. 2003;86(4):433– mountain forests: an approach based on the replacement cost meth- 43. https://doi.org/10.1016/s0034-4257(03)00112-3. od. J For Econ. 2012;18(4):318–28. https://doi.org/10.1016/j.jfe. 24. Wulder MA, Dymond CC, White JC, Leckie DG, Carroll AL. 2012.06.002. Surveying mountain pine beetle damage of forests: a review of 8. Berger F, Rey F. Mountain protection forests against natural hazards remote sensing opportunities. For Ecol Manag. 2006;221(1–3): and risks: new French developments by integrating forests in risk 27–41. https://doi.org/10.1016/j.foreco.2005.09.021. zoning. Nat Hazards. 2004;33(3):395–404. https://doi.org/10.1023/ 25. Oquist G. Photosynthesis of overwintering evergreen plants. Annu b:nhaz.0000048468.67886.e5. Rev Plant Biol. 2003;54(1):329–55. https://doi.org/10.1146/ annurev.arplant.54.072402. 9. Senf C, Pflugmacher D, Hostert P, Seidl R. Using Landsat time series for characterizing forest disturbance dynamics in the coupled 26. Attema EPW, Ulaby FT. Vegetation modeled as a water cloud. human and natural systems of Central Europe. ISPRS J Radio Sci. 1978;13(2):357–64. https://doi.org/10.1029/ Photogramm Remote Sens. 2017;130:453–63. https://doi.org/10. RS013i002p00357. 1016/j.isprsjprs.2017.07.004. 27. Ulaby FT, Moore RK, Fung AK. Microwave remote sensing active and passive-volume III: from theory to applications. 1986. 10. Seidl R, Thom D, Kautz M, Martin-Benito D, Peltoniemi M, Vacchiano G, et al. Forest disturbances under climate change. Nat 28. Ahern FJ, Leckie DJ, Drieman JA. Seasonal changes in relative C- Clim Chang. 2017;7(6):395–402. https://doi.org/10.1038/ band backscatter of northern forest cover types. IEEE Trans Geosci nclimate3303. Remote Sens. 1993;31(3):668–80. https://doi.org/10.1109/36. 11. Schelhaas M-J, Nabuurs G-J, Schuck A. Natural disturbances in the European forests in the 19th and 20th centuries. Glob Chang Biol. 29. Pulliainen JT, Kurvonen L, Hallikainen MT. Multitemporal behav- 2003;9(11):1620–33. ior of L- and C-band SAR observations of boreal forests. IEEE 250 Curr Forestry Rep (2019) 5:240–250 Trans Geosci Remote Sens. 1999;37(2):927–37. https://doi.org/10. 41. Tanase MA, Aponte C, Mermoz S, Bouvet A, Le Toan T, Heurich 1109/36.752211. M. Detection of windthrows and insect outbreaks by L-band SAR: a case study in the Bavarian Forest National Park. Remote Sens 30. Pulliainen JT, Mikhela PJ, Hallikainen MT, Ikonen J. Seasonal Environ. 2018;209:700–11. dynamics of C-band backscatter of boreal forests with applications 42.� De Grandi EC, Mitchard E, Woodhouse IH, De Grandi GD. Spatial to biomass and soil moisture estimation. IEEE Trans Geosci wavelet statistics of SAR backscatter for characterizing degraded Remote Sens. 1996;34(3):758–70. https://doi.org/10.1109/36. Forest: a case study from Cameroon. IEEE J Sel Top Appl Earth Obs Remote Sens. 2015;8(7):3572–84. https://doi.org/10.1109/ 31. Wang Y, Day JL, Davis FW. Sensitivity of modeled C- and L-band jstars.2015.2420596. This paper describes spatial wavelet radar backscatter to ground surface parameters in loblolly pine for- statistics of SAR backscatter, which could be used for est. Remote Sens Environ. 1998;66(3):331–42. https://doi.org/10. detecting forest insect infestions. 1016/S0034-4257(98)00074-1. 43. Lei Y, Lucas R, Siqueira P, Schmidt M, Treuhaft R. Detection of 32. Santoro M, Fransson JES, Eriksson LEB, Magnusson M, Ulander forest disturbance with spaceborne repeat-pass SAR interferometry. LMH, Olsson H. Signatures of ALOS PALSAR L-band backscatter IEEE Trans Geosci Remote Sens. 2018;56(4):2424–39. https://doi. in Swedish forest. IEEE Trans Geosci Remote Sens. 2009;47(12): org/10.1109/tgrs.2017.2780158. 4001–19. https://doi.org/10.1109/TGRS.2009.2023906. 44. eoPortal [database on the Internet]2019. Available from: https:// 33. Toan TL, Beaudoin A, Riom J, Guyon D. Relating forest biomass to directory.eoportal.org/web/eoportal/satellite-missions. Accessed: SAR data. IEEE Trans Geosci Remote Sens. 1992;30(2):403–11. March 2019. https://doi.org/10.1109/36.134089. 45. Rüetschi M, Small D, Waser LT. Rapid detection of windthrows 34.� Dostálová A, Wagner W, Milenković M, Hollaus M. Annual sea- using Sentinel-1 C-band SAR data. Remote Sens. 2019;11(2):115. sonality in Sentinel-1 signal for forest mapping and forest type 46. Hirschmugl M, Deutscher J, Gutjahr K-H, Sobe C, Schardt M. classification. Int J Remote Sens. 2018;39(21):7738–60. https:// Combined use of SAR and optical time series data for near real- doi.org/10.1080/01431161.2018.1479788. In this paper, the time forest disturbance mapping. 2017 9th International Workshop seasonal changes of Sentinel-1 backscatter information are on the Analysis of Multitemporal Remote Sensing Images analysed for different forest types. (MultiTemp), 2017/06. IEEE; 2017. 35. Frison P-L, Fruneau B, Kmiha S, Soudani K, Dufrêne E, Le Toan T, 47. Dostálová A, Hollaus M, Milenković M, Wagner W. Forest area et al. Potential of Sentinel-1 data for monitoring temperate mixed derivation from Sentinel-1 data. ISPRS Ann Photogramm Remote forest phenology. Remote Sens. 2018;10(12):2049. https://doi.org/ Sens Spat Inf Sci. 2016;III-7:227–33. 10.3390/rs10122049. 48. Rauste Y, Antropov O, Mutanen T, Häme T. On clear-cut mapping 36. Proisy C, Mougin E, Dufrene E, Dantec VL. Monitoring seasonal with time-series of Sentinel-1 data in boreal forest. Living Planet changes of a mixed temperate forest using ERS SAR observations. Symposium 2016; Prague, Czech Republic; 2016. IEEE Trans Geosci Remote Sens. 2000;38(1):540–52. https://doi. 49.�� Hollaus M, Bauer-Marschallinger B, Löw M, Schadauer K, Wagner org/10.1109/36.823949. W. Potential of Sentinel-1 time series to detect bark beetle out- 37.� Rüetschi M, Schaepman ME, Small D. Using multitemporal breaks. ForestSAT; 2018-10-01 - 2018-10-05; College Park, Sentinel-1 C-band backscatter to monitor phenology and classify Maryland, USA; 2018. p. 1. This conference paper describes deciduous and coniferous forests in northern Switzerland. Remote the influence of bark beetle infestations on Sentinel-1 backscat- Sens. 2018;10(1):55. https://doi.org/10.3390/rs10010055. This ter information. paper investigate the phenological changes of Sentinel-1 C- 50. Liu Y, Gong W, Hu X, Gong J. Forest type identification with band backscatter for different forest types. random forest using Sentinel-1A, Sentinel-2A, multi-temporal 38. Saatchi S, Asefi-Najafabady S, Malhi Y, Aragão LEOC, Anderson Landsat-8 and DEM data. Remote Sens. 2018;10(6):946. https:// LO, Myneni RB, et al. Persistent effects of a severe drought on doi.org/10.3390/rs10060946. Amazonian forest canopy. PNAS. 2013;110(2):565–70. https:// 51. Verhegghen A, Eva H, Ceccherini G, Achard F, Gond V, Gourlet- doi.org/10.1073/pnas.1204651110. Fleury S, et al. The potential of sentinel satellites for burnt area 39. Ranson KJ, Kovacs K, Sun G, Kharuk VI. Disturbance recognition mapping and monitoring in the Congo Basin forests. Remote in the boreal forest using radar and Landsat-7. Can J Remote Sens. Sens. 2016;8(12):22. https://doi.org/10.3390/rs8120986. 2003;29(2):271–85. https://doi.org/10.5589/m02-096. 40. Kaasalainen S, Hyyppä J, Karjalainen M, Krooks A, Paivi L-S, Holopainen M et al. Comparison of terrestrial laser scanner and Publisher’sNote Springer Nature remains neutral with regard to synthetic aperture radar data in the study of forest defoliation. 2010. jurisdictional claims in published maps and institutional affiliations. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Current Forestry Reports Springer Journals

Radar Satellite Imagery for Detecting Bark Beetle Outbreaks in Forests

Download PDF
11 pages

Loading next page...
 
/lp/springer-journals/radar-satellite-imagery-for-detecting-bark-beetle-outbreaks-in-forests-2bsYKmsL0M

References (53)

Publisher
Springer Journals
Copyright
Copyright © 2019 by The Author(s)
Subject
Environment; Sustainable Development; Environmental Management; Nature Conservation; Forestry; Forestry Management; Ecology
eISSN
2198-6436
DOI
10.1007/s40725-019-00098-z
Publisher site
See Article on Publisher Site

Abstract

Purpose of Review The overall objective of this paper is to review the state of knowledge on the application of radar data for detecting bark beetle attacks in forests. Due to the increased availability of high spatial and temporal resolution radar data (e.g. Sentinel-1 (S1)), the question is how this time series data can support operational forest management with respect to forest insect damage prevention. Furthermore, available radar systems will be listed and their potential for detecting bark beetle attacks will be discussed. To increase the understanding of the potential of radar time series for detecting bark beetle outbreaks, a theoretical background about the interaction of the radar signals with the forest canopy is given. Finally, gaps in the available knowledge are identified and future research questions are formulated which could advance our understanding of using radar data for detecting forest bark beetle attacks. Recent Findings Few studies already demonstrate the high potential of S1 time series data for forest disturbance mapping in general. It was demonstrated that multi-temporal S1 data provide an excellent data source of describing the phenological characteristics of forests, which provide the basic knowledge for detecting bark beetle induced forest damages. It has been found that the optimal time for data acquisition is April to June for the pre-event and August to October for the post-event acquisitions. Summary For detecting bark beetle induced forest damages, the literature review shows that mono-temporal radar data are of limited use, that shorter wavelength (e.g. C-band; X-band) have a higher potential than longer wavelength such as L-band and that the current S1 time series data have a high potential for operational applications. . . . . . Keywords Forest SAR Time series Damages Monitoring Bark beetle Introduction an important economic aspect which need to be considered in operational forest management [7, 8]. Due to ongoing climate Forest disturbances, mainly caused by storm events, ice, snow, change and the fact that there is a synchronisation of forest fire, and insects, disrupt ecosystem dynamics and significantly disturbance rates from forest mortality patterns with climate influence ecosystem goods and services [1, 2] and strongly change, the topic of forest disturbance becomes an increasing- impact forest carbon budgets [3�� , 4]. Especially, in mountain- ly important issue. For example, Senf et al. [9]found arelation ous regions, the dynamics of forest disturbances are influenc- between satellite-derived estimates of forest disturbance rates ing the protective function of forests [5, 6] and, therefore, are and climate-related events since 1985 in forest landscapes of Central Europe. Furthermore, in Seidl et al. [10], an overview is given how climate change may affect disturbance regimes via direct, indirect and interaction effects. They summarized This article is part of the Topical Collection on Remote Sensing that warmer and drier conditions facilitate fire, drought and insect disturbances, while warmer and wetter conditions in- * Markus Hollaus crease disturbances from winds and pathogens. Furthermore, Markus.Hollaus@geo.tuwien.ac.at in the review paper of Pureswaran et al. [3�� ], the connection between forest insects and climate change is described. They Mariette Vreugdenhil Mariette.Vreugdenhil@geo.tuwien.ac.at concluded that the relationships between climate change and forest insects are clear in some species but cannot be general- Department of Geodesy and Geoinformation, TU Wien, Wiedner ized to all species. Finally, they concluded that process-based Hauptstraße 8-10, 1040 Vienna, Austria Curr Forestry Rep (2019) 5:240–250 241 phenology models permit predictions of population dynamics several spectral bands provide the potential for mapping forest if input data are appropriate. disturbances as summarized in the above cited review papers. Apart from climate-disturbance relationships, contempo- The majority of the applied algorithms for forest disturbance rary frequent and severe abiotic and biotic damage incidents mapping is based on the monitoring of forest dynamics and similarly affecting primary and managed conifer forests are thus a high temporal resolution is essential. However, for op- due to increasingly high disturbance susceptibility with regard tical satellite data, this requirement can limit the application to stand history, development stage and structure [11, 12]. for up-to-date monitoring due to the sensitivity of optical re- Species composition, stand age and density in combination mote sensing to clouds, aerosols and smoke. The advantage of with increased temperature and drought conditions were iden- radar is that it is not hindered by cloud cover, smoke, aerosol tified as the most relevant parameters for explaining/ contamination and low solar illumination. Due to the longer predicting salvage cuttings in Austrian Norway spruce forests wavelength of microwaves compared to optical remote sens- following attacks of Europe’s most important forest insect ing, the penetration depth in vegetation is larger, making it pest, the Eurasian spruce bark beetle, Ips typographus, during sensitive to the canopy and woody parts of the vegetation [19]. the period 2014–2016 [13]. Documentation of forest damage Nonetheless, until recently, the temporal and spatial reso- [14] shows that for example in Austria a strong increase of lutions of publicly available radar data were not sufficient for bark beetle induced timber is recorded since 1990 and reaches detecting forest insect attacks. With the launch of ESA’s approximately 20% and 28% of the total harvested timber for Sentinel-1 satellite constellation carrying C-band synthetic ap- 2017 and 2018, respectively. Similar situations are observed erture radar instruments, cross- and co-polarized backscatter for other European and North American countries [15]. data are now available for the first time at the spatial and Based on the increased frequency of forest insect attacks temporal resolution needed. Hence, the overall objective of and the attendant increase of ecological and economic im- this paper—to review the state of knowledge on the applica- pacts, there is a strong demand for remote sensing–based tion of radar data for detecting bark beetle attacks in forests— monitoring approaches. Several publications can be found that is formulated. Based on a systematic literature review of pa- use different remote sensing data for detecting forest insect pers published since 2000, different applications of radar data attacks. Most studies use optical remote sensing data, and only for detecting forest disturbances due to bark beetle attacks few studies use radar data. Optical remote sensing makes use including their applied methods are summarized. of the large absorption of light in the visible spectrum and Additionally, an overview of available types of forest insect reflection in the near infrared and short wave infrared spec- infestations and the related physiological changes in the can- trum during photosynthesis to derive information about the opy with respect to remotely sensed properties is given. This forest canopy. Radar systems transmit pulses of electromag- review paper focuses on bark beetle induced forest damages as these insect attacks lead to extensive calamities in coniferous netic energy and record the reflected signal at the sensor from the area of interest. Most systems, which aim at monitoring the forests. As the interaction of the radar signal with forest can- Earth’s surface, operate in the microwave domain, with fre- opies is different from the spectral properties that are impor- quencies between 3 and 30 GHz. In the microwave domain, tant in optical remote sensing, an introduction to radar and the backscattered signal is sensitive to a number of land sur- microwave remote sensing is given, including a theoretical face parameters, including, but not limited to, water content in background about the interaction of microwaves with vegeta- soil and vegetation and vegetation structure. However, most of tion. Finally, gaps in the available knowledge are identified the investigated publications use optical satellite data as has and future research questions are formulated which could ad- been summarized in the recent review paper about remote vance our understanding of using radar data for detecting for- sensing of forest insect disturbance mapping from Senf et al. est bark beetle attacks in an operational sense. [16], which gives an extensive overview of different insect types and remote sensing data and methods for mapping dis- turbances. In this review paper, only one paper was mentioned Types of Bark Beetle Infestations and Impact that uses TerraSAR-X radar data in combination with optical on the Canopy Properties for Optical high-resolution RapidEye data for the early detection of bark and Radar Data beetle green attacks [17]. Furthermore, the review paper of Stone and Mohammed [18] gives an overview of applications The most important forest insects can be grouped into of remote sensing technologies for assessing planted forests xylophagous and folivorous insects. As summarized in Seidl damaged by insect pests and fungal pathogens. They conclude et al. [10], bark beetle species are the most prominent that the simultaneous acquisition of spectral and 3D point data xylophagous insects, whereas defoliators are the most impor- will have the highest benefit for assessing tree health in a cost- tant species in the group of folivorous insects. The insects of effective manner. From an operational point of view, the spec- these two groups affect trees in a different manner. Bark beetle tral properties of optical satellite data have the benefit that species bore into the phloem of trees, copulate, mine galleries 242 Curr Forestry Rep (2019) 5:240–250 and oviposit [20]. Furthermore, they introduce several species Characteristics of Radar Backscatter of fungi that colonize the phloem and vascular tissue. Finally, from Forests the translation of water and nutrients within the tree is getting interrupted leading to the death of the tree. In contrast to bark Radar systems transmit and receive pulses of electromagnetic beetles, defoliating species feed on leaves and needles and energy. For most radar systems, two polarizations are com- thus influence the photosynthetic activity of a tree. This can mon, vertical polarization (V) which is perpendicular to the lead to growth reduction, deformation or to tree mortality. surface and horizontal polarization (H) which is parallel to the Depending on the infestation intensity and the consequent surface. The amount of electromagnetic energy intercepted degree of defoliation and the surrounding environmental and re-radiated is a complex combination of multiple factors stress factors (e.g. drought), trees can regenerate in the follow- including electromagnetic and geometric properties of the tar- ing year or in the case of some broadleaved tree species even get and sensor specifics such as wavelength and polarization in the same year [10]. of the transmitted and received signal. The energy intercepted For mapping infestations with remote sensing techniques, and reflected is described by the so-called radar cross section. knowledge about the physiological effects of the attacks in the In remote sensing of the Earth’s surface, the target is an area in forest canopy is essential. Niemann and Visintini [21� ] de- which it is assumed that no single scatterer dominates. To scribe three stages that occur during a bark beetle attack. compare measurements from different sensors with different The green attack stage occurs during the host colonization footprints, i.e. target areas, the radar cross section can be nor- and the development of the first bark beetle population. malized by the area, resulting in the backscatter coefficient. During this phase, the physiology of the tree is not influenced The most important electromagnetic properties of the me- and no visual change in the canopy can be observed, even dia for microwave remote sensing are the electric permittivity. though the tree is getting stressed due to the increased produc- It is a complex number with a real and imaginary part, where tion of resin and due to decreased water availability. The de- the real part is often referred to as the dielectric constant. The tection of this very early infestation stage would have the dielectric constant (ε) of the medium with which the wave highest benefit for supporting operational forest management, interacts plays an important role in the amount of but remote sensing has been of limited success until now backscattered energy. In the microwave domain, electromag- [21� ]. A reliable detection of infested trees can only be done netic waves incident on the Earth’s surface excite water mol- with ground surveys by searching for entry and exit holes on ecules. The dipole character of water causes the water mole- the bole, boring dust and pitch tubes, which are all indicators cules to continuously reorient in the electromagnetic radia- of beetle infestations [22]. However, such field work can only tion’s oscillating electric field. This leads to high ε values of be done for sample plots due to the high demand of time and water (ε = 80), compared to dry vegetation matter and soil consequently costs for such in situ measurements. (1.5 < ε <2) [25, 26]. The red attack stage includes the period where the colour of The geometric properties of the target area influence the needles turns to yellow and brown and where the tree starts to type of scattering. If a medium is homogenous, most of the lose needles. Additional to the changing colour of the canopy, electromagnetic wave is scattered at the boundary of the sur- the needle water content is the most important physiological face. The amount of energy reflected can be described by the change during the red attack stage [23, 24]. The final grey attack Fresnel coefficient which is a function of incidence angle and stage describes the period where the tree will lose all of its ε. This mechanism is called surface scattering or specular foliage. From a remote sensing perspective, short-wave infrared reflection. Volume scattering occurs if a medium is inhomo- wavelengths are most sensitive to changes in the needle water geneous and the signal penetrates the lower medium. Within content [10]. The majority of the studies investigated in the the medium, randomly distributed dielectric inhomogeneities review of Seidl el al. [10] used the normalized difference veg- scatter the incoming electromagnetic waves in all directions etation index (NDVI) for mapping broadleaved defoliations regardless of incident angle [26]. The amount of energy while different indices were used for mapping coniferous defo- returned to the sensor is proportional to the number of dielec- liation (NDVI, Normalized Burn Ratio, Moisture Stress Index, tric inhomogeneities and the difference in ε between them and Leave Area Index). For mapping bark beetle infestations, the the medium. most frequently used index was the difference in the Tasseled The total backscatter of a vegetated surface can be de- Cap Wetness component. However, when doing time series scribed by scattering from the soil, the vegetation and an in- analyses based on spectral images for detecting bark beetle teraction between the two. Soil and vegetation produce two infestations, one has to consider that there is a natural decrease different scattering mechanisms, namely surface scattering in the chlorophyll content of needles during fall and winter and volume scattering. Vegetation is often described as a seasons. This decline in chlorophyll content is due to dormancy cloud of water droplets held in place by dry matter: an inho- and frost hardening of the needles which is required in order to mogeneous medium with randomly distributed dielectric in- protect foliage against photo-oxidative damages [25]. homogeneities. Since vegetation consists for a large part of Curr Forestry Rep (2019) 5:240–250 243 water, ε of the total vegetation increases rapidly with increas- and low backscatter was observed in winter and high back- ing above ground biomass and moisture status. From this, it scatter in summer [34� , 35, 37� ]. A recent study on backscatter can be inferred that backscatter increases with increasing from Sentinel-1 over a deciduous forest in France showed that above ground biomass and moisture status. the ratio of co- and cross-polarized backscatter coincides tem- The dielectric inhomogeneities can also change the polari- porally with NDVI, increasing in spring and decreasing in zation of the incident wave. Although this does not happen to autumn. However, no conclusions were given to what drives a large degree, a change in polarization is more likely to occur the variations in the so-called cross ratio [35]. The potential of with volume scattering or with interaction between the soil radar for monitoring changes in forest structure and moisture and the vegetation. Because of the low occurrence, cross- content was also demonstrated by studies of drought effects in polarized backscatter (HV or VH transmit-receive) is usually the Amazon. Saatchi et al. [38] showed the complementary lower than co-polarized backscatter (HH or VV transmit-re- information which can be gained next to optical observations ceive). It does, however, show a stronger increase with vol- when investigating the effects of the Amazonian drought of ume and double-bounce scattering than co-polarized back- 2005. A decrease in backscatter was observed after the 2005 scatter. Thus, when describing vegetation as a water cloud, it drought, which took place over a long time period, and was can be inferred that cross-polarized backscatter is more sensi- attributed to a reduction in canopy water and a change in tive to changes in vegetation. structure. In order to use radar data for forest monitoring, it needs to be taken into account that, depending on the density of the vegetation canopy, wavelength of the signal and the incidence Radar Data for Bark Beetle Disturbance angle of the observations, microwaves penetrate the vegeta- Mapping tion and scattering from the soil or from soil-vegetation inter- action could contribute to the total backscatter signal [27]. This review paper was based on a screen of the literature Most studies agree that the contribution of the soil and soil- between 2000 and 2019 using the ISI Web of Science database vegetation interaction is small in co-polarized backscatter (http://www.webofknowledge.com/) and Scopus (https:// from forests, but not negligible. Especially for longer wave- www.scopus.com) with the general search terms focussing lengths, low biomass forests, high soil moisture conditions or on synthetic aperture radar, forests and insect disturbances. for lower incidence angles the total backscatter signal is sen- In detail, the following search string was used: ALL sitive to the soil surface [28–31]. In addition, due to its sensi- FIELDS: (sar OR radar OR microwave OR sentinel*1 OR tivity to volume scattering, studies have found cross-polarized ers* OR tandem-x* OR l-band) AND ALL FIELDS: backscatter to be most sensitive to changes in forest above (disturbance OR insect OR attack OR beetle) AND ALL ground biomass and that the contribution of soil and soil- FIELDS: (satellite OR airborne) AND ALL FIELDS: (forest vegetation interaction was found to be negligible [31–33]. OR tree). Within the time period from 2000 to 2019, 105 hits Apart from the above ground biomass and moisture con- appear containing these keywords in all fields. After a tent, the structure of the vegetation can play a large role on the screening of the results, only four publications could be temporal backscatter signal. This is especially the case in de- identified as being relevant. In the following paragraphs, an ciduous forests where major changes in structure occur due to overview of these publications is given. the development of the foliage. One of the first studies ad- Ranson et al. [39] explored the use of combined use of dressing the effect of structural changes of vegetation on the mono-temporal JERS and Radarsat data for detecting insect backscatter signal was performed by [27] when defoliating damages in Siberian forests. They used co-polarized (HH) corn. The effect of each major plant constituent, i.e. leaves, JERS and Radarsat data from 2 years. After the pre- stalks and fruits, on backscatter at 5.1 GHz was investigated. processing steps, they applied a 3 × 3 Frost filter to each For high incidence angles, backscatter was higher from SAR data set to reduce the speckle. For both data sets, the defoliated corn, i.e. stalks and cobs, than from the whole plant. backscatter coefficients and the standard deviations showed The backscatter from stalks or interaction between soil and low differences across insect damaged and healthy coniferous stalks is very significant but is strongly attenuated by the pres- forests. As a reason for that, they argued that for JERS the ence of leaves. In deciduous forests, similar behaviour was damaged foliage have a negligible influence on the backscat- observed where higher backscatter was often found in winter ter signal because they are too small. For Radarsat, only the followed by a drop in backscatter in spring and summer and an separability between coniferous forest and clear-cuts was increase in autumn. This was attributed to the change in scat- high. The reason for this can be found in the occurring volume tering mechanism from volume to surface scattering and the scattering within the tree canopies, whereas volume scattering contribution of soil and soil-vegetation scattering to the total from grassy clear-cuts is small. The combined use of Radarsat signal when no foliage is present [28, 34� , 35, 36, 37� ]. For and JERS slightly increased the classification accuracies and coniferous forests, this temporal signature was not observed, 29% of the severely insect damaged classes and 46% of the 244 Curr Forestry Rep (2019) 5:240–250 moderately insect damaged area could be classified. Finally, coefficient was normalized with a DTM, and a multi- they concluded that the results were limited when using any temporal filtering was applied to decrease speckle. The single-channel radar data. multi-temporal filtering was based on five pre-event and six Kaasalainen et al. [40] studied the potential of ERS-2 SAR post-event SAR data sets. Radar change ratios were computed data for detecting defoliation of Scots pine dominated forest using pre- and post-disturbance averaged backscatter coeffi- located in eastern Finland caused by the European pine saw- cients to better represent the forest condition by removing the fly. They used terrestrial laser scanner (TLS) to examine the stochastic part of the signal. They found out that a change of defoliation during the active period of the pine sawfly hazard. the backscatter coefficient of − 1.0 dB is a consistent indicator The first measurement was undertaken during the early phase of forest structural changes induced by bark beetle attacks. of defoliation and the second following the defoliation period. They also discuss the influence of weather conditions to the For the study area, seven ERS-2 images were available from discrimination capacity. For example, the drier conditions of the same orbit and cover the time period from before and after the reference data set and/or wetter conditions for the remain- the defoliation. For all images, only the amplitude information ing images lead to nearly no changes of the backscatter coef- of backscattering signal was available. For their analyses, the ficients. The classification of the areas affected by bark beetle ERS-2 backscatter values were averaged using a circle with a outbreaks was done using a threshold approach and support 50-m radius to reduce speckle. The results show only a slight vector machines based on the backscatter change ratios. They change in the averaged SAR backscatter for those plots where reached overall accuracies ranging from 74 to 91% depending defoliation could be observed in the TLS data. For plots where on the selected image pair. They conclude that the classifica- signs of defoliation were not present, little or no changes in the tion accuracy for windthrow affected areas was high and low- backscatter were observed. As no weather data was available er for bark beetle–induced damages due to the fact that chang- for the acquisition times, the influence of the soil surface and es are more abrupt for windthrow than for insect outbreaks. vegetation moisture to the backscatter signal could not be They also state that the interpretation of the backscatter signal investigated in this study. can be complex because of the multitude of factors affecting Ortiz et al. [17] evaluated the capabilities of RapidEye and them, the limitations due to the density of the radar observa- TerraSAR-X imagery for detecting areas affected by bark bee- tions and that it is difficult to extract the specific stress factor. tle green attacks. They investigated generalized linear models, In a more general sense, few studies are available that use maximum entropy and random forest for detecting bark beetle SAR data for detecting forest disturbances. Even though these infested areas in a flat study area in south-west Germany. The studies are not focussing on bark beetle disturbance mapping, RapidEye and the TerraSAR-X data were acquired in two their applied methods could be useable for it. For example, De consecutive days at the end of May. Both data sets were Grandi [42� ] used spatial wavelet statistics of ENVISAT ASAR and ALOS PALSAR backscatter to describe the struc- orthorectified using a high-resolution digital surface model with a resolution of 2 m derived from airborne laser scanning ture of degraded forests in Cameroon, Central Africa. They data. To investigate the potential of both data sets for bark found out that analytic parameters describing the functional beetle green attack detection, explanatory variables were de- form (i.e. third-degree polynomial function) of the scaling rived from circular plots with a radius of 5 m covering either signatures show statistically significant differences between only attacked or vital trees. The explanatory variables from the signatures of intact and degraded forests for ENVISAR TerraSAR-X were the standard deviation, max, min, mean, ASAR data. For the ALOS PALSAR, no significant differ- median and the first and third quartiles of the backscatter dis- ences could be found. As a reason for that the authors argue tribution. The analyses have shown that the third quartiles and that L-band penetrates more into the canopy, and therefore, the the median of the backscatter were higher for attacked plots observed backscatter texture is influenced more by the distri- than for vital plots. They achieved an overall accuracy of 87% bution of large scattering elements and by the ground return, (kappa 0.23) for the TerraSAR-X data and 94% (kappa 0.51) whereas the shorter wavelength of the C-band ENVISAT for the RapidEye data. Finally, they conclude that improve- ASAR is able to describe the canopy much better. Lei et al. ments could be expected if multi-temporal SAR data and/or [43] used spaceborne repeat-pass InSAR correlation measure- varying polarizations are available. ments in combination with a physical InSAR scattering model Tanase et al. [41] used ALOS L-band spaceborne synthetic to detect forest disturbances over selective logging sites in aperture radar (SAR) within a change detection framework to Queensland, Australia. They used dual-pol ALOS PALSAR delineate forested areas affected by wind and insect distur- InSAR coherence data with a temporal baseline of 92 days. bances. Several SAR images from the same orbit before and The basic idea of their approach is to compensate for the after the bark beetle attack were selected as a basis for a pixel- scene-wide mean behaviour of the volumetric and temporal based comparison. The cross-polarized (HV) channel was decorrelation effects and then use the residual decorrelation as mainly investigated because of its sensitivity to forest struc- an indicator of forest disturbance. The detected forest distur- tural characteristics and their change. The backscatter bance areas (i.e. forest and non-forest areas) were compared to Curr Forestry Rep (2019) 5:240–250 245 results from Landsat time series with a relative RMSE of 13% where cloud cover is commonly the limiting factor for over 0.81 ha forest stands. One limitation of this approach is using optical data only. The Sentinel-1 data consists of that the disturbance affected area cannot be large compared ten acquisitions with single polarisation (VV) from the with the spaceborne InSAR image itself because the mean same orbit acquired within 9 months. In a first step, the behaviour of the volumetric and temporal decorrelation effect forest disturbance mapping was done in parallel for the is estimated from the scene-wide forest coherence distribution. radar and the optical data. For the combination of both results, the Bayes’ theorem was used. The results showed the highest detection accuracies for the combined ap- Discussion and Outlook proach (83.7% for the area-based and 97.1% for the plot-based validation). Furthermore, the results showed The reviewed papers show the limits of mono-temporal radar for plot areas < 1 ha lower detection accuracies for radar images for forest disturbance mapping. Since the availability data than for optical data. of Sentinel-1 data, the popularity of radar images for forest Furthermore, Sentinel-1 time series were used for disturbance mapping has increased. In addition to the freely distinguishing forest from non-forest areas or for tree species available Sentinel-1 data, other radar sensors exist but often classifications. For example Dostálová et al. [47]used their data are not freely available, especially radar data with X- Sentinel-1 time series data for forest area delineation with band are often not freely available. An overview of available a spatial resolution of 10 m in the eastern part of Austria and and planned satellite radar sensor is given in Table 1.An achieved accuracies of 88% compared to a forest mask de- extensive overview of satellite missions can be found in the rived from airborne laser scanning data. A similar result can eoPortal Directory [44]. also be found in the study of Dostálová et al. [34� ], in which For Europe, the Sentinel-1 satellites offer high tempo- dense Sentinel-1 time series backscatter signals were used ral (3–6 days) and spatial (5 × 20 m) resolution, which are for forest area delineation and classification into tree species independent of the cloud clover. Time series from groups. For three European study sites, they achieved over- Sentinel-1 are promising not only for bark beetle distur- all accuraciesof86–91% for the forest area delineation and bance mapping but also for other forest disturbance map- 65–85% for the classification into broadleaf forests, conifer- ping applications such as windthrow or clear-cut detec- ous forests and non-forest classes. A similar result is also tion. For example Rüetschi et al. [45]used Sentinel-1C- reported in Rüetschi et al. [37� ]. They achieved an overall band VV and VH polarisation data for a rapid windthrow accuracy of 86% for the classification into coniferous and detection in mixed temperate forests in Switzerland and deciduous forests and 72% for the classification of three northern Germany. They used five Sentinel-1 acquisitions different tree species. These studies used the differences in the backscatter seasonality between different tree species. before the event and ten after the storm event including ascending and descending data. The backscatter values of Backscatter variations were further found to hint at differ- Sentinel-1 were normalised using the modelled local illu- ences in the forest structure (mainly tree heights) and vege- minated area based on a DTM rather than a modelled tation cover; however, differences were less pronounced local incident angle based on an ellipsoid. In this way, than between forest types [47]. the influence of the topography on the SAR acquisition Also, Rauste et al. [48] used Sentinel-1 time series data is mitigated and the modelled illuminated area was over a period of 1 year to describe the temporal behaviour of exported as an additional input layer for the further forest the C-band backscatter for areas representing areas clear-cut disturbance mapping. For the detection of disturbed forest during the Sentinel-1 acquisition period, unchanged forest areas, the change detection method “image differencing” areas and areas clear-cut before the Sentinel-1 acquisition was applied for VV and VH polarisations. The mean and started. The analyses were done for a boreal forest site in the standard deviation were calculated for the difference Finland. They could not only show the potential for detecting images. By summing up the differences of VV and VH logged forest areas despite high sensitivity to seasonal and polarised backscatter values, a heuristic windthrow index weather conditions but also mention that the detection of log- was calculated, which was the basis for a decision tree ging areas was only successful for good scenes. Consequently, classification. The results show that the backscatter from they suggest the use of Sentinel-1 data from two winters for windthrown areas was typically about 0.5 dB higher than the clear-cut mapping. before the storm event. The detected windthrown areas In addition to these summarized studies, Hollaus et al. had a minimum extend of 0.5 ha and a producer accuracy [49�� ] analysed the potential of Sentinel-1 time series data of 0.88 and a user accuracy of 0.85. Also, Hirschmugl for recognizing the beginning of bark beetle infestations. et al. [46] showed the potential of Sentinel-1 time series Within their study, homogenous areas with vital and data in combination with optical Sentinel-2 and Landsat-8 infested forest stands were analysed separately for a study data for forest disturbance mapping in tropical forests area in the north-eastern part of Austria. These forest 246 Curr Forestry Rep (2019) 5:240–250 Table 1 Overview of operating and planned radar sensors Sensor Launch date Band Spatial resolution Orbital repeat (days) Polarisation Owner/operator Scientific data access (m) ALOS-2 PalSAR 2014 L 10–100 14 Fully Japan Aerospace Exploration freely available polarimetric Agency NISAR Scheduled L3–10 12 Fully NASA ? 2021 polarimetric RADARSAT-2 2007 C 3–100 24 Fully Canadian Space agency (CSA) Restricted polarimetric Sentinel-1 2013 C 10 12 VV, VH (over European Space Agency freely available land) RADARSAT-constellation Scheduled C1–100 12 Fully Canadian Space Agency ? (3-satellites) 2019 polarimetric Risat-1 2013 C 1–50 25 Fully Indian Space Research Available to Copernicus polarimetric Organisation users PAZ/PAZ 2 SAR 2001/2018 X 0.25–40 > 2 Fully Spanish Space Agency Restricted polarimetric TerraSAR-X 2007 X 1–16 11 Fully German Aerospace Centre Restricted polarimetric (DLR) Tandem-X 2010 X 1–16 11 Fully German Aerospace Centre Restricted polarimetric Cosmos Skymed-1/2 2007/2010 X 1–100 16; revisit time for the full constellation is in the range Fully Italian Space Agency Restricted of few hours polarimetric TecSAR 2008 X 1–814 Fully Israel Aerospace Industries Restricted polarimetric Kompsat-5 2013 X 1–20 28 Fully Korean Space Agency Restricted polarimetric Kompsat-6 Scheduled X0.5–20 11 Fully Korean Space Agency ? 2020 polarimetric Curr Forestry Rep (2019) 5:240–250 247 Fig. 1 Monthly averaged Sentinel-1 backscatter ratios for bark beetle infested and vital forest patches. a The ratios of the monthly averaged backscatter values for vital and infested forest stands. b The differences of the ratio values stands were derived from Sentinel-2 time series by apply- stand mortality may be detected only months later. Due ing a change detection approach. As the backscatter signal to missing reference data with detailed information about underlies a natural variation due to phenological changes the degree of infestation, it could not be analysed which of the canopy and the moisture condition in the canopy degree of bark beetle related damage can be recognized in and the terrain [34� ] monthly ratios were calculated for Sentinel-1 time series data. In addition to the used differ- vital and infested forest stands from backscatter coeffi- ences in the backscatter values for vital and infested for- cients from the year before the attack and the year of ests, one has to consider annual differences in the back- the bark beetle attack. In Fig. 1a, the ratios of the back- scatter values. For example Rüetschi et al. [37� ] showed scatter coefficients for vital and infested forest stands are inter-annual differences of the backscatter from coniferous given, whereas in Fig. 1b, the difference of the ratios is study areas. The seasonal variations are minimized by shown for the VH polarized backscatter values. calculating the ratios between the backscatter signals. For this analysis, Sentinel-1 data from the year 2015, However, the influence of the moisture conditions in the where for this investigation area no bark beetle attacks canopy as well as in the terrain surface was not consid- took place, and Sentinel-1 data from the year 2017, where ered in this study yet. The consideration of local varia- several bark beetle attacks occur, were selected. The tions in the backscatter behaviour could improve the de- Sentinel-1 time series analyses showed a clear difference tection accuracy by applying local adaptive thresholds in of the backscatter coefficients between vital and infested the change detection workflow. forest stands of ~ 1 dB for both VV- and VH-polarized Sentinel-1 backscatter values. The differences can also be seen in the ratio values as shown in Fig. 1 and start ap- proximately in July to August, which corresponds well Conclusions with a terminated development of the first filial generation and starting of tree dieback, corresponding to the grey The conclusion gives a short summary of the main findings of stage. However, crowns of Norway spruce often remain the reviewed publications for forest disturbance mapping with green long after a bark beetle attack was initiated, and the focus on insect induced damages. The potential and the 248 Curr Forestry Rep (2019) 5:240–250 limitations for operational applications are summarized, and The following open research questions are formulated: future research questions are formulated. The following conclusions are drawn: 1. Commonly speckle reduction is done by applying filters with a fixed kernel size and shape. As insect-induced for- 1. Defoliators and bark- and wood-boring insects have the est damage starts within small forest patches that are cov- most potential to cause damage in forests. The most prom- ered by only a few radar pixels, this type of speckle re- inent insects are bark beetles and after an attack trees duction leads to an averaging and consequently to commonly die within few months. In Niemann and “mixed” pixels. Therefore, research is needed around Visintini [21� ], three stages are defined that occur during how a land cover–dependent and/or time-dependent a bark beetle attack, the green, the red and the grey attack speckle reduction can improve the detectability of forest stage. From a remote sensing perspective, the detection of disturbances. the green attack stage has had limited success as only 2. The reviewed studies have shown that the low temporal slight changes are ongoing in the foliage during this stage. resolution is the most limiting factor in using radar data 2. Until the start of the Copernicus Sentinel-1 satellites, very for the detection of insect induced forest damages. The few radar data sets were used for detecting forest distur- new Sentinel-1 time series data allows the description of bances due to insect infestations. The available sensors phenological characteristics with a high level of detail, (i.e. JERS, ERS-1, Radarsat, TerraSAR-X, ALOS which is the basis for forest disturbance mapping. As PALSAR) operate in different wavelengths (X-, C-, L- shown in the studies of Dostalova et al. [34� ]and band) and have either limited spatial or temporal resolu- Rüetschi et al. [37� ], the backscatter signal has an annual tion. Most of the studies used the backscatter signal from but also inter-annual variability. Furthermore, local varia- acquisitions from the same orbit (either ascending or de- tions can occur. Therefore, research is needed around how scending), and only in the study of Lei et al. [43], ALOS these small scale variations (i.e. in time and space) in the PALSAR InSAR coherence data were used. backscatter signal can be considered. This research also 3. Concerning the processing steps and the applied methods needs to consider influences from varying satellite orbits, for disturbance mapping, it can be concluded that after i.e. ascending and descending. topographic correction, speckle reduction is undertaken 3. From an operational forest management point of view, it is by applying various filters or by averaging the backscatter important to know the degree of damage (i.e. green, red, signal for plots or forest stands. grey attack) or the minimum mapping unit that can be 4. Several studies have shown that shorter wavelengths (e.g. derived from radar data. To answer this research question, C-band; X-band) have a higher potential than longer appropriate reference data are necessary. wavelengths, such as L-band radar data, for detecting in- 4. In the reviewed papers, only one [43] used coherence sect induced forest damages. For example De Grandi data. Due to the relatively long temporal baseline of [42� ] argued that L-band penetrates more into the canopy, 92 days and the fact that they used L-band data, the results and therefore, the observed backscatter texture is influ- for detecting insect infested forest areas were not satisfac- enced more by the distribution of large scattering ele- tory. Due to the availability of C-band, Sentinel-1 data ments and by the ground return, whereas the shorter research is needed to analyse the potential of the coher- wavelengths of, e.g. the C-band are able to describe the ence information for forest disturbance mapping canopy much better. applications. 5. The high temporal and spatial resolutions of Sentinel-1 5. As already shown in several studies [37� , 50, 51], time series have shown a high potential for forest distur- Sentinel-1 time series data in combination with optical bance mapping. Until now, the Sentinel-1 data have been data from, e.g. Sentinel-2, have a high potential for for- mainly used for windthrow mapping (e.g. [45]). Based on estry applications. For insect-induced forest disturbance, deviations from the natural variability of the backscatter mapping further research is needed to explore the full signal during the phenological cycle ,Hollaus et al. [49 ] potential of multi-sensor data. demonstrated for the first time the potential of this data Funding Information Open access funding provided by TU Wien source for detecting bark beetle infestations in coniferous (TUW). Mariette Vreugdenhil was supported by a Living Planet forest. Fellowship from the European Space Agency (4000125441/18/I-NS). 6. Although though there is still a lack of knowledge about the applicability of Sentinel-1 time series for insect distur- Compliance with Ethical Standards bance mapping, the study of Hollaus et al. [49�� ] sug- gested that the optimal time for data acquisition is April Conflict of Interest Markus Hollaus and Mariette Vreugdenhil declare to June for the pre-event and August to October for the that they have no conflict of interest. post-event acquisitions. �� Curr Forestry Rep (2019) 5:240–250 249 Human and Animal Rights and Informed Consent This article does not 12. Schurman JS, Trotsiuk V, BačeR, Čada V, Fraver S, Janda P, et al. contain any studies with human or animal subjects performed by any of Large-scale disturbance legacies and the climate sensitivity of pri- the authors. mary Picea abies forests. Glob Chang Biol. 2018;24(5):2169–81. 13. Netherer S, Pennerstorfer BMJ. Trockenstress von Open Access This article is distributed under the terms of the Creative Fichtenbeständen fördert den Schadholzanfall durch Commons Attribution 4.0 International License (http:// Buchdrucker. Forstschutz Aktuell. 2018;65:1–9. creativecommons.org/licenses/by/4.0/), which permits unrestricted use, 14. Hoch G, Chrisitian L. Documentation of forest damaging factors. distribution, and reproduction in any medium, provided you give appro- BFW. 2019. http://bfw.ac.at/rz/bfwcms2.web?dok=9756. Accessed priate credit to the original author(s) and the source, provide a link to the 24. 03. 2019. Creative Commons license, and indicate if changes were made. 15. Katz C. Small Pests, Big problems: the global spread of bark bee- tles. Yale School of Forestry & Environmental Studies. 2017. https://e360.yale.edu/features/small-pests-big-problems-the-global- spread-of-bark-beetles. Accessed March 2019. 16. Senf C, Seidl R, Hostert P. Remote sensing of forest insect distur- References bances: current state and future directions. Int J Appl Earth Obs Geoinf. 2017;60:49–60. https://doi.org/10.1016/j.jag.2017.04.004. 17. Ortiz S, Breidenbach J, Kändler G. Early detection of bark beetle Papers of particular interest, published recently, have been green attack using TerraSAR-X and RapidEye data. Remote Sens. 2013;5(4):1912–31. https://doi.org/10.3390/rs5041912. highlighted as: 18. Stone C, Mohammed C. Application of remote sensing technolo- � Of importance gies for assessing planted forests damaged by insect pests and fun- �� Of major importance gal pathogens: a review. Curr For Rep. 2017;3(2):75–92. https:// doi.org/10.1007/s40725-017-0056-1. 1. Maroschek M, Seidl R, Netherer S, Lexer MJ, et al. Climate change 19. Liu YY, De Jeu RAM, McCabe MF, Evans JP, Van Dijk AIJM. impacts on goods and services of European mountain forests. Global long-term passive microwave satellite-based retrievals of Unasylva. 2009;60(231/232):76–80. vegetation optical depth. Geophys Res Lett. 2011;38(18). https:// 2. Thom D, Seidl R. Natural disturbance impacts on ecosystem ser- doi.org/10.1029/2011GL048684. vices and biodiversity in temperate and boreal forests. Biol Rev. 20. Raffa KF, Aukema BH, Bentz BJ, Carroll AL, Hicke JA, Turner 2016;91(3):760–81. MG, et al. Cross-scale drivers of natural disturbances prone to an- 3.�� Pureswaran DS, Roques A, Battisti A. Forest insects and climate thropogenic amplification: the dynamics of bark beetle eruptions. change. Curr For Rep. 2018;4(2):35–50. https://doi.org/10.1007/ BioScience. 2008;58(6):501–17. https://doi.org/10.1641/B580607. s40725-018-0075-6. This paper gives a review of effects of 21.� Niemann K, Visintini F. Assessment of potential for remote sensing climate change on forest insects. detection of bark beetle-infested areas during green attack: a litera- 4. Seidl R, Schelhaas M-J, Rammer W, Verkerk PJ. Increasing forest ture review. Victoria, BC, Canada: Natural Resources Canada, disturbances in Europe and their impact on carbon storage. Nat Canadian Forest Service; 2005. http://cfs.nrcan.gc.ca/ Clim Chang. 2014;4(9):806–10. https://doi.org/10.1038/ pubwarehouse/pdfs/25269.pdf. This working paper explains the nclimate2318. different stages of insect infestations and how the individual 5. Brang P. Resistance and elasticity: promising concepts for the man- stages can be detected with remote sensing data. agement of protection forests in the European Alps. For Ecol 22. Robertson C, Farmer CJQ, Nelson TA, Mackenzie IK, Wulder MA, Manag. 2001;145(1–2):107–19. https://doi.org/10.1016/s0378- White JC. Determination of the compositional change (1999–2006) 1127(00)00578-8. in the pine forests of British Columbia due to mountain pine beetle 6. Bebi P, Kienast F, Schönenberger W. Assessing structures in moun- infestation. Environ Monit Assess. 2008;158(1–4):593–608. tain forests as a basis for investigating the forests’ dynamics and https://doi.org/10.1007/s10661-008-0607-9. protective function. For Ecol Manag. 2001;145(1–2):3–14. https:// 23. Skakun RS, Wulder MA, Franklin SE. Sensitivity of the thematic doi.org/10.1016/s0378-1127(00)00570-3. mapper enhanced wetness difference index to detect mountain pine 7. Notaro S, Paletto A. The economic valuation of natural hazards in beetle red-attack damage. Remote Sens Environ. 2003;86(4):433– mountain forests: an approach based on the replacement cost meth- 43. https://doi.org/10.1016/s0034-4257(03)00112-3. od. J For Econ. 2012;18(4):318–28. https://doi.org/10.1016/j.jfe. 24. Wulder MA, Dymond CC, White JC, Leckie DG, Carroll AL. 2012.06.002. Surveying mountain pine beetle damage of forests: a review of 8. Berger F, Rey F. Mountain protection forests against natural hazards remote sensing opportunities. For Ecol Manag. 2006;221(1–3): and risks: new French developments by integrating forests in risk 27–41. https://doi.org/10.1016/j.foreco.2005.09.021. zoning. Nat Hazards. 2004;33(3):395–404. https://doi.org/10.1023/ 25. Oquist G. Photosynthesis of overwintering evergreen plants. Annu b:nhaz.0000048468.67886.e5. Rev Plant Biol. 2003;54(1):329–55. https://doi.org/10.1146/ annurev.arplant.54.072402. 9. Senf C, Pflugmacher D, Hostert P, Seidl R. Using Landsat time series for characterizing forest disturbance dynamics in the coupled 26. Attema EPW, Ulaby FT. Vegetation modeled as a water cloud. human and natural systems of Central Europe. ISPRS J Radio Sci. 1978;13(2):357–64. https://doi.org/10.1029/ Photogramm Remote Sens. 2017;130:453–63. https://doi.org/10. RS013i002p00357. 1016/j.isprsjprs.2017.07.004. 27. Ulaby FT, Moore RK, Fung AK. Microwave remote sensing active and passive-volume III: from theory to applications. 1986. 10. Seidl R, Thom D, Kautz M, Martin-Benito D, Peltoniemi M, Vacchiano G, et al. Forest disturbances under climate change. Nat 28. Ahern FJ, Leckie DJ, Drieman JA. Seasonal changes in relative C- Clim Chang. 2017;7(6):395–402. https://doi.org/10.1038/ band backscatter of northern forest cover types. IEEE Trans Geosci nclimate3303. Remote Sens. 1993;31(3):668–80. https://doi.org/10.1109/36. 11. Schelhaas M-J, Nabuurs G-J, Schuck A. Natural disturbances in the European forests in the 19th and 20th centuries. Glob Chang Biol. 29. Pulliainen JT, Kurvonen L, Hallikainen MT. Multitemporal behav- 2003;9(11):1620–33. ior of L- and C-band SAR observations of boreal forests. IEEE 250 Curr Forestry Rep (2019) 5:240–250 Trans Geosci Remote Sens. 1999;37(2):927–37. https://doi.org/10. 41. Tanase MA, Aponte C, Mermoz S, Bouvet A, Le Toan T, Heurich 1109/36.752211. M. Detection of windthrows and insect outbreaks by L-band SAR: a case study in the Bavarian Forest National Park. Remote Sens 30. Pulliainen JT, Mikhela PJ, Hallikainen MT, Ikonen J. Seasonal Environ. 2018;209:700–11. dynamics of C-band backscatter of boreal forests with applications 42.� De Grandi EC, Mitchard E, Woodhouse IH, De Grandi GD. Spatial to biomass and soil moisture estimation. IEEE Trans Geosci wavelet statistics of SAR backscatter for characterizing degraded Remote Sens. 1996;34(3):758–70. https://doi.org/10.1109/36. Forest: a case study from Cameroon. IEEE J Sel Top Appl Earth Obs Remote Sens. 2015;8(7):3572–84. https://doi.org/10.1109/ 31. Wang Y, Day JL, Davis FW. Sensitivity of modeled C- and L-band jstars.2015.2420596. This paper describes spatial wavelet radar backscatter to ground surface parameters in loblolly pine for- statistics of SAR backscatter, which could be used for est. Remote Sens Environ. 1998;66(3):331–42. https://doi.org/10. detecting forest insect infestions. 1016/S0034-4257(98)00074-1. 43. Lei Y, Lucas R, Siqueira P, Schmidt M, Treuhaft R. Detection of 32. Santoro M, Fransson JES, Eriksson LEB, Magnusson M, Ulander forest disturbance with spaceborne repeat-pass SAR interferometry. LMH, Olsson H. Signatures of ALOS PALSAR L-band backscatter IEEE Trans Geosci Remote Sens. 2018;56(4):2424–39. https://doi. in Swedish forest. IEEE Trans Geosci Remote Sens. 2009;47(12): org/10.1109/tgrs.2017.2780158. 4001–19. https://doi.org/10.1109/TGRS.2009.2023906. 44. eoPortal [database on the Internet]2019. Available from: https:// 33. Toan TL, Beaudoin A, Riom J, Guyon D. Relating forest biomass to directory.eoportal.org/web/eoportal/satellite-missions. Accessed: SAR data. IEEE Trans Geosci Remote Sens. 1992;30(2):403–11. March 2019. https://doi.org/10.1109/36.134089. 45. Rüetschi M, Small D, Waser LT. Rapid detection of windthrows 34.� Dostálová A, Wagner W, Milenković M, Hollaus M. Annual sea- using Sentinel-1 C-band SAR data. Remote Sens. 2019;11(2):115. sonality in Sentinel-1 signal for forest mapping and forest type 46. Hirschmugl M, Deutscher J, Gutjahr K-H, Sobe C, Schardt M. classification. Int J Remote Sens. 2018;39(21):7738–60. https:// Combined use of SAR and optical time series data for near real- doi.org/10.1080/01431161.2018.1479788. In this paper, the time forest disturbance mapping. 2017 9th International Workshop seasonal changes of Sentinel-1 backscatter information are on the Analysis of Multitemporal Remote Sensing Images analysed for different forest types. (MultiTemp), 2017/06. IEEE; 2017. 35. Frison P-L, Fruneau B, Kmiha S, Soudani K, Dufrêne E, Le Toan T, 47. Dostálová A, Hollaus M, Milenković M, Wagner W. Forest area et al. Potential of Sentinel-1 data for monitoring temperate mixed derivation from Sentinel-1 data. ISPRS Ann Photogramm Remote forest phenology. Remote Sens. 2018;10(12):2049. https://doi.org/ Sens Spat Inf Sci. 2016;III-7:227–33. 10.3390/rs10122049. 48. Rauste Y, Antropov O, Mutanen T, Häme T. On clear-cut mapping 36. Proisy C, Mougin E, Dufrene E, Dantec VL. Monitoring seasonal with time-series of Sentinel-1 data in boreal forest. Living Planet changes of a mixed temperate forest using ERS SAR observations. Symposium 2016; Prague, Czech Republic; 2016. IEEE Trans Geosci Remote Sens. 2000;38(1):540–52. https://doi. 49.�� Hollaus M, Bauer-Marschallinger B, Löw M, Schadauer K, Wagner org/10.1109/36.823949. W. Potential of Sentinel-1 time series to detect bark beetle out- 37.� Rüetschi M, Schaepman ME, Small D. Using multitemporal breaks. ForestSAT; 2018-10-01 - 2018-10-05; College Park, Sentinel-1 C-band backscatter to monitor phenology and classify Maryland, USA; 2018. p. 1. This conference paper describes deciduous and coniferous forests in northern Switzerland. Remote the influence of bark beetle infestations on Sentinel-1 backscat- Sens. 2018;10(1):55. https://doi.org/10.3390/rs10010055. This ter information. paper investigate the phenological changes of Sentinel-1 C- 50. Liu Y, Gong W, Hu X, Gong J. Forest type identification with band backscatter for different forest types. random forest using Sentinel-1A, Sentinel-2A, multi-temporal 38. Saatchi S, Asefi-Najafabady S, Malhi Y, Aragão LEOC, Anderson Landsat-8 and DEM data. Remote Sens. 2018;10(6):946. https:// LO, Myneni RB, et al. Persistent effects of a severe drought on doi.org/10.3390/rs10060946. Amazonian forest canopy. PNAS. 2013;110(2):565–70. https:// 51. Verhegghen A, Eva H, Ceccherini G, Achard F, Gond V, Gourlet- doi.org/10.1073/pnas.1204651110. Fleury S, et al. The potential of sentinel satellites for burnt area 39. Ranson KJ, Kovacs K, Sun G, Kharuk VI. Disturbance recognition mapping and monitoring in the Congo Basin forests. Remote in the boreal forest using radar and Landsat-7. Can J Remote Sens. Sens. 2016;8(12):22. https://doi.org/10.3390/rs8120986. 2003;29(2):271–85. https://doi.org/10.5589/m02-096. 40. Kaasalainen S, Hyyppä J, Karjalainen M, Krooks A, Paivi L-S, Holopainen M et al. Comparison of terrestrial laser scanner and Publisher’sNote Springer Nature remains neutral with regard to synthetic aperture radar data in the study of forest defoliation. 2010. jurisdictional claims in published maps and institutional affiliations.

Journal

Current Forestry Reports – Springer Journals

Published: Nov 8, 2019

There are no references for this article.