Assessment of post-fire vegetation state dynamics in Ivan-Arakhley natural Park (Zabaikalsky Krai) using radar Sentinel 1 and optical Sentinel 2 data Natalia Rodionova Irina Vakhnina Tatiana Zhelibo Kotel’nikov FIRE RAS lInstitute of Natural Resouces, Ecology lInstitute of Natural Resouces, Ecology Fryazino, Russia and Cryology SB RAS and Cryology SB RAS rnv@sunclass.ire.rssi.ru Chita, Russia Chita, Russia vahnina_il@mail.ru Zhelibo@mail.ru Abstract—The results of the analysis of multi - temporal This paper traces the dynamics of vegetation restoration satellite monitoring of the post-fire vegetation state dynamics in in the 4 years since the fire in 2015 on the territory of the the territory of Ivan - Arakhley natural Park (Zabaikalsky Krai) Ivan-Arakhley nature Park using radar and optical data after the fire in 2015 is presented using Sentinel 1 (S1) radar Sentinel 1/2 satellites. The task is to find out whether the data and Sentinel 2 (S2) optical data. To assess the dynamics of vegetation is being restored and how this process changes revegetation affected by natural fire, spectral vegetation indices over the years. (VI) NDVI, ARVI, NBR, NDMI and radar vegetation index RVI are used. A positive trend has been revealed in the restoration of II. STUDY AREA vegetation in the test areas of the natural Park by both optical and radar indices. In the mountain-taiga larch landscapes of the Kondinsky district, test areas are selected for studying the dynamics of Keywords—C-band radar data, multispectral optical data, changes in plant communities (Fig. 1 (b)). The area on the vegetation cover, wildfires, vegetation indices Aspen ridge located in the basin of the river Osinovka, conventionally called Osinovka, the area on the Yablonovy I. INTRODUCTION ridge, in the basin of the river Rasmalai called “Rasmalai”. In Siberia, one of the highest levels of wildfire activity is Detailed descriptions of 12 test sites in Osinovka with a observed in Zabaikalsky Krai. Vegetation renewal in this photo of general view of the sample areas where employees area is significantly difficult due to the arid climate. There is of the Institute of Natural Resources, Ecology and Cryology little precipitation, 90% of it (within 300 mm) fall during the SB RAS worked in 2018 are given in the table. warm period, mainly in July and August. Winters are The climate in the study area is sharply continental, a snowless, and the soil is not moistened by snow. This leads characteristic phenomenon should be noted the presence of to the fact that the region is very hot in the spring. Experts permafrost. The ground freezes deep in winter at 1-1.5 [1] found that the successful renewal of the forest is hindered meters and thaws slowly. by a number of reasons: 1) the high temperature of the soil on the burning, leading to the death of undergrowth, 2) lack of moisture and nutrients, leading to severe competition between plants and the grass grow, 3) repeated fires. Remote sensing data and ground data are used to monitor forest development after a fire. The paper [2] describes the ground - based studies of vegetation conditions on the territory of Ivan-Arakhley nature Park conducted in 2013-2014 after the grass-roots fires of 2000, 2001, 2003 and 2010. The test sites laid on the South-Eastern slopes of the Aspen ridge have characteristic forest types: rhododendron, cranberry and yernikovye leaf forests. These stands were affected by grass-roots fires of (a) (b) varying intensity. As a result of the study, it is shown that the Fig. 1. Fire map for 14.04.2015 according to ScanEx operational natural renewal of wood species is characterized as monitoring system [3] (a), the location of test areas for study the dynamics unsatisfactory. of changes in plant communities (b). In mid-April 2015, severe forest fires were observed in III. INPUT DATA AND RESEARCH METHODS the TRANS-Baikal region near the Beklemishev lake system. Fig. 1 (a) shows a map of fires on the territory of Osinovka A. Sentinel 1 radar data. Radar vegetation index for April 14, 2015 according to the ScanEx operational The work uses open-access Sentinel 1 radar data, IW monitoring system, the service “Kosmosnimki - Fires” [3]. In (interferometric wide swath) mode, VV and VH [4], the radar images of the Sentinel 1 satellite were used to polarizations, spatial resolution 10 m. All the images were determine the ashes in this area based on the use of pre-processed by the Sentinel-1 Toolbox and later SNAP [5]. amplitude and texture information. Pre-processing included the selection of a fragment with the After 2015, the territories of the Ivan-Arakhley Park were study area and radiometric calibration. not exposed to fire for four years. Copyright © 2020 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0) Image Processing and Earth Remote Sensing TABLE I. DESCRIPTION OF OSINOVKA TEST SITES № Сoordinates Height, m Type of community Violation of the territory Photo 1 52.20855° N, 112.56738°E 1070 burn on cutting down 2 52.20631°, 112.56490° 1029 the ernika cereal ashes 3 52.19739°, 112.57023° 996 the listvyaga forb gorelnik, windfall the listvyaga 4 52°12´24.9´´, 112°32´40.3´´ 1053 ashes, windfall Brusnichnoye the listvyaga 5 52°12´42.0´´, 112°32´02.5´´ 1099 ashes, windfall Brusnichnoye 6 52.18975°, 112.54653° 1051 the listvyaga forb ashes, windfall 52.19275°, 7 1021 the listvyaga forb ashes 112.54384° the listvyaga 8 52.19133°, 112.54929° 1026 ashes, windfall omnicopy the listvyaga 9 52.21100°, 112.53900° 1068 ashes, windfall rhododendron 10 52.21421°, 112.53182° 1071 the listvyaga forb gorelnik, windfall the listvyaga 11 52.20604°, 112.55075° 1044 ashes, windfall Brusnichnoye the listvyaga 12 52.21030°, 112.53596° 1102 ashes, windfall Brusnichnoye S1 sessions for 26.07.2017, 02.08.2018 and 28.07.2019 sites studied. For comparison with the sites with burning, a were taken to determine the average value of the background site (№13) was selected, where there was no fire backscattering coefficient (BC) for the profiles for the 12 test in 2015, in the Northern part of lake Shakshinsky, the profile VI International Conference on "Information Technology and Nanotechnology" (ITNT-2020) 55 Image Processing and Earth Remote Sensing coordinates are 52.2037° N, 112.7229° E. Fig. 2 (a) shows a survey dates of 26.7.17, 2.8.18, and 28.7.19. Note the graph of the BC change in dB for 12 sites plus the significant growth of vegetation for site №5, where the RVI background for 2017-2019. The BC values increased for changed from 0.385 in 2017 to 0.99 in 2018. For the same both polarizations over the period 2017-2019. The greatest site, an increase in the BC of 6.6 dB for VH polarization is changes in the BC were observed for VH cross-polarization. noted above. A slightly smaller increase in RVI is obtained For example, for site №5, the changes are 6.6 dB over two from site №9 with an increase in RVI of 0.5 and from site years, i.e. a significant increase in volume scattering №7 with an increase in RVI of 0.35. The decrease in RVI is associated with vegetation growth, whereas for VV noted for site №8. polarization is 1.8 dB. The smallest changes in the BC VH Fig. 3 shows images of the study area for both polarization – for site №8-are less than 2 dB, this value is polarizations in RGB encoding: red-26.07.2017, green – even less than the changes for the background profile. 02.08.2018, blue-28.07.2019. All changes took place over The backscattering coefficient is an absolute polarimetric the past years 2017-2019 on the territory of burn after the parameter, whereas the radar vegetation index (RVI) [6] is a fire in 2015. This territory is allocated in figure 3 by white relative parameter that is not very sensitive to the view angle line. Multi-temporal radar images revealed areas of ashes. and natural conditions. RVI is used to monitor vegetation The rest of the territory has changed slightly. growth using multi-temporal radar data: 8𝜎𝐻𝑉 B. Sentinel 2 multispectral data. Vegetation indices 𝑅𝑉𝐼 = (1) The ESA Sentinel 2A satellite was launched in June 𝜎𝐻𝐻 +𝜎𝑉𝑉 +2𝜎𝐻𝑉 2015, and the second Sentinel 2B in March 2017. The RVI changes from 0 (smooth bare soil) to 1 as vegetation multispectral camera has 13 spectral bands spanning from grows and is a measure of volume scattering. Sentinel 1 IW the visible and near infrared to the short wave infrared. The GRD mode has two polarizations VV and VH. Then under spatial resolution varies from 10 m to 60 m depending on the the assumption [7], spectral band. The temporal resolution one of S2 is 10 days, 𝜎𝐻𝐻 ≈ 𝜎𝑉𝑉 (2) and two satellites – 5 days. Image processing was performed by SNAP. equation (1) can be represented as: 4𝜎𝑉𝐻 The S2 sessions for 31.07.2016, 05.08.2017, 31.07.2018, 𝑅𝑉𝐼 = and 26.07.2019 were taken to determine the VI by profiles 𝜎𝑉𝑉 + 𝜎𝑉𝐻 for the 12 sites studied. The choice of images was determined primarily by the lack of clouds for the month of Assumption (2) is valid for negligible interaction July and the proximity of the dates to the dates of the radar between soil and vegetation [8]. RVI correlates with VWC survey. Since the territories were covered by clouds or their (Volumetric Water Content), LAI (Leaf Area Index) and shadows for a number of sites on 26.07.2019, the data for the NDVI (Normalized Difference Vegetation Index) and is sites №5, №6, №7, №10 and №12 were replaced with data poorly sensitive to natural conditions [9]. Fig. 2 (b) shows a for 06.07.2019. graph of RVI changes for the test sites under study with (a) (b) Fig. 2. Change in the backscatter coefficient (a), change in the RVI (b) for test sites over three years. Spectral indices obtained from remote sensing optical NDVI and ARVI data, such as the Normalized Difference Vegetation Index (NDVI) [10], Normalized Burn Ratio (NBR) [11] and NDVI- VI showing the presence and state of vegetation Normalized Difference Moisture Index (NDMI) [12] and (relative biomass): their modifications dNDVI, dNBR, dNDMI, which 𝑁𝐼𝑅 − 𝑅 𝑁𝐷𝑉𝐼 = determine the difference of indices before and after a fire, 𝑁𝐼𝑅 + 𝑅 give good results in identifying areas with damage to where NIR and R are the values of the reflection coefficient vegetation. Graphs based on the NBR and dNBR reflect the of the earth's surface in the NIR and Red bands. NDVI dynamics and nature of vegetation cover restoration in changes in the range from -1 to 1. Changes in reflectivity in burned areas [13]. the visible and NIR bands are associated with a decrease in VI International Conference on "Information Technology and Nanotechnology" (ITNT-2020) 56 Image Processing and Earth Remote Sensing the content of chlorophyll in the vegetative organs of following disadvantages were noted: non-linearity, influence shrinking trees. The absorption zone of chlorophyll in Red of the atmosphere (water vapor and aerosols), saturation at band of the spectrum determines the low level of vegetation high biomass, sensitivity to the presence of clouds, influence reflection in the visible spectral band. Under stress, the of soil, object geometry, influence of spectral effects formation of chlorophyll in plants decreases, which leads to (various tools). The main limitation of NDVI and similar a decrease in its absorption in the visible range and, indices is that optical sensors can only monitor a very thin consequently, an increase in reflectivity. In the NIR band, layer of vegetation, and cannot provide information about the reflection coefficient of green vegetation increases woody vegetation. noticeably, reaching 45-50% [14]. One of the modifications of NDVI to account for the Fig. 4 (a) shows graphs of NDVI changes for the test influence of the atmosphere is the atmospheric resistant VI sites under study from 2016 to 2019. The NDVI values for ARVI (Atmospheric Resistant Vegetation Index) [16]: 2019 were unexpectedly lower than for 2018 and even for 𝑁𝐼𝑅 − 𝑅𝐵 𝐴𝑅𝑉𝐼 = 2016 for sites №5, №10, and №12. A possible reason is a 𝑁𝐼𝑅 + 𝑅𝐵 number of disadvantages of the NDVI, which lead to uncertainties in its quantitative assessment. In [15], the Fig. 3. Images of the study area in RGB encoding: red-26.07.17, green-2.08.18, blue-28.07.19 for VV and VH polarizations. NDVI 31.07.2016 05.08.2017 ARVI 31.07.2016 31.07.2018 6-26.07.2019 0,8 31.07.2018 11-26.07.2019 1,0 0,7 0,9 0,6 0,8 0,5 0,7 0,4 0,6 1 2 3 4 5 6 7 8 9 10 11 12 13 1 2 3 4 5 6 7 8 9 10 11 12 13 (a) (b) Fig. 4. Changes NDVI (a) and ARVI (b) for test sites in 2016-2019. depending on the type of aerosol, is γ=1 [16]. ARVI is 4 where RB=R- γ (B-R), B is the value of the reflection times less sensitive to atmospheric effects (aerosol) than coefficient in the blue range of the spectrum. This index NDVI [14], and its dynamic range is the same as that of replaces the Red band in NDVI with a combination of Red NDVI. The greatest effect of using ARVI, instead of NDVI, and Blue bands. This combination has self-correcting is achieved for surfaces with vegetation rather than for soil, properties for atmospheric effects. ARVI variations with and for particle sizes in the atmosphere from medium to atmospheric opacity variations are significantly smaller than small, rather than for large particles (marine aerosols or NDVI variations. The optimal value of the coefficient γ, dust). It should be noted that this index was proposed for the VI International Conference on "Information Technology and Nanotechnology" (ITNT-2020) 57 Image Processing and Earth Remote Sensing MODIS sensor with bands: Blue (0.47±0.01 µm), Red the NIR and SWIR1 bands to create a coefficient designed (0.66±0.025 µm) and NIR (0.865±0.02 µm). Having S2 to dim lighting and atmospheric effects: bands with wavelengths very close to the corresponding 𝑁𝐼𝑅 − 𝑆𝑊𝐼𝑅1 values for MODIS, namely, Blue with a central wavelength 𝑁𝐷𝑀𝐼 = of 0.4966 µm - S2A and 0.4921 µm - S2B, Red - 0.6645 and 𝑁𝐼𝑅 + 𝑆𝑊𝐼𝑅1 0.665 µm and NIR-0.835 and 0.833 µm, you can use the It should be noted that the author of [18] showed that a ARVI formula obtained for MODIS also for S2. Fig. 4 (b) change in SWIR1 spectral band makes the largest shows ARVI change graphs for three dates. The use of the contribution to the separation of disturbed and undisturbed ARVI VI showed 1) an increase in biomass over the years forest ecosystems. since the fire for almost all test sites, 2) an increase in the Fig. 5 (b) shows graphs of NDMI changes for the test saturation threshold. sites in 2016-2019. The graphs in Fig. 5 (a) and (b) are very NBR and NDMI similar in the nature of changes in values. For all test sites, the vegetation humidity level increased in 2016-2019, with VI NBR is determined by the equation [11]: the least for the background site. NIR − SWIR2 𝑁𝐵𝑅 = Based on the considered optical vegetation indices NIR + SWIR2 NDVI, ARVI, NBR and NDMI, we can make a general where SWIR2 is the value of the reflection coefficient of the conclusion about the positive dynamics of vegetation earth's surface in the Shortwave infrared spectral band. This recovery after the fire in 2015 at 12 sites under spectral band reflects changes in the moisture content of consideration in the territory of the Ivan-Arakhley nature plants, as well as changes in the structure of the canopy and Park. Specifically, the values of the ARVI vegetation index, the structure of leaves. The combined use of SWIR2 with which shows relative biomass, have increased for all test NIR, which does not depend on the moisture saturation of sites since the fire, except for site 10. The values of the NBR the plant, but depends on the leaf structure, increases the and NDMI indices, which reflect the presence of moisture in accuracy of estimating the moisture content in the plant vegetation, increased over the post-fire years for all test regardless of the leaf structure [17]. sites. Fig. 5 (a) shows the NBR change graphs for the test sites IV. CONCLUSION for 2016-2019. Based on changes in NBR values for 2016- This paper assesses the dynamics of vegetation 2019, the moisture content for all test sites, exposed to fire, restoration in the territory of Ivan-Arakhley natural Park increased, and remained almost the same for the background after wildfires in April 2015 using radar and optical data area where there was no fire. from Sentinel 1/2 satellites. The radar (RVI) and optical VI NDMI is sensitive to the level of humidity in vegetation indices (NDVI, ARVI, NBR, NDMI) showed vegetation. It is used to track droughts and also indicates the positive dynamics in the state of vegetation growth in 12 test level of combustible materials in fire-prone areas. VI uses sites in the post-fire years. Only for test site №10 the ARVI value decreased in the post-fire years. (a) (b) Fig. 5. NBR and NDMI changes for test sites in 2016-2019. ACKNOWLEDGMENT using remote sensing observations,” Environ. Res. Let., vol. 14, pp. 1- 10, 2019. DOI: 10.1088/1748-9326/ab083d. The authors thank Vladimir Petrovich Makarov, Ph. D., [2] I.V. Gorbunov, V.P. Makarov and O.F. Malyh, “Postfire vegetation Institute of Natural Resources, Ecology and Cryology SB state in territory Ivan-Arakhley natural park (Zabaikalsky krai),” Uspekhi sovremennogo estestvoznaniya, vol. 7, pp. 54-59, 2015. RAS, for the organization of field work in the area of the [3] URL: http://fires.kosmosnimki.ru/. Beklemishevskaya depression. [4] N.V. Rodionova, “Evaluation of SENTINEL 1 imagery for burned area detection in southern Siberia in spring and summer 2015,” REFERENCES Sovremennye problemy distancionnogo zondirovaniya Zemli iz [1] E.G. Shvetsov, E.A. Kukavskaya, L.V. Buryak and K. Barrett, kosmosa, vol. 13, no. 2, pp. 164-175, 2016. “Assessment of post-fire vegetation recovery in Southern Siberia [5] URL: https://sentinel.esa.int/web/sentinel/toolboxes/sentinel-1. VI International Conference on "Information Technology and Nanotechnology" (ITNT-2020) 58 Image Processing and Earth Remote Sensing [6] Y. Kim and J. van Zyl, “Vegetation effects on soil moisture [13] V.G. Bondur, M.N. Cidilina and E.V. Cherepanova, “Satellite estimation,” Proc. IGARSS, vol. 2, pp. 800-802, 2004. monitoring of wildfire impacts on the conditions of various types of [7] F. Charbonneau, M. Trudel and R. Fernandes, “Use of dual vegetation cover in the federal districts of the Russian Federation,” polarization and multi-incidence SAR for soil permeability mapping,” Issledovanie Zemli iz kosmosa, vol. 3, pp. 13-32, 2019. Advanced Synthetic Aperture Radar (ASAR), St-Hubert Canada, [14] S.A. Bartalev, V.A. Egorov, A.M. Krylov, F.V. Stycenko and T.S. 2005. Hovratovich, “The evaluation of possibilities to assess forest burnt [8] M. Trudel, F. Charbonneau and R. Leconte, “Using RADARSAT-2 severity using multi-spectral satellite data,” Sovrem. problemy DZZ polarimetric and ENVISAT-ASAR dual-polarization data for iz kosmosa, vol. 7, pp. 215-225, 2010. estimating soil moisture over agricultural fields,” Can. J. Remote [15] A. Huete, K. Didan and T. Miura, “Overview of the radiometric and Sensing, vol. 38, no. 4, pp. 514-527, 2012. biophysical performance of the modis vegetation indices,” Remote [9] Y. Kim and J. van Zyl, “A time-series approach to estimate soil Sens. Environ., vol. 83, no. 195, pp. 213-221, 2002. moisture using polarimetric radar data,” IEEE Trans. GRS, vol. 47, [16] Y.J. Kaufman and D. Tanré, “Atmospherically Resistant Vegetation no. 8, pp. 2519-2527, 2009. Index (ARVI) for EOS-MODIS,” IEEE Trans. GRS, vol. 30, no. 2, [10] J.W. Rouse, R.H. Haas, J.A. Schell and D.W. Deering, “Monitoring pp. 261-270, 1992. DOI: 10.1109/36.134076. vegetation systems in the Great Plains with ERTS,” Third ERTS [17] P. Ceccato, S. Flasse, S. Tarantola, S. Jacquemond and J. Gregoire, Symposium NASA SP-351 I , pp. 309-317, 1973. “Detecting vegetation water content using reflectance in the optical [11] C.H. Key and N.C. Benson, “Landscape assessment: ground measure domain,” Remote Sensing of Environment, vol. 77, pp. 22-33, 2001. of severity, the Composite Burn Index; and remote sensing of [18] E.A. Terekhin, “Detection of disturbed forest ecosystems in the severity, the Normalized Burn Ratio,” FIREMON: Fire Effects forest-steppe zone using reflectance values,” Computer Optics, vol. Monitoring and Inventory System, USDA Forest Service, Rocky 43, no. 3, pp. 412-418, 2019. DOI: 10.18287/0134-2452-2019-43-3- Mountain Research Station, General Technical Report RMRS-GTR- 412-418. 164-CD: LA, pp. 1-15, 2005. [12] E.H. Wilson and S.A. Sader, “Detection of forest harvest type using multiple dates of Landsat TM imagery,” Remote Sensing of Environment, vol. 80, pp. 385-396, 2002. VI International Conference on "Information Technology and Nanotechnology" (ITNT-2020) 59