Assessment of forest recovery at Wu-Ling fire scars in Taiwan using multi-temporal Landsat imagery

Ecological Indicators 79 (2017) 196–206

Contents lists available at ScienceDirect

Ecological Indicators journal homepage: www.elsevier.com/locate/ecolind

Original Articles

Assessment of forest recovery at Wu-Ling fire scars in Taiwan using multitemporal Landsat imagery Chuphan Chompuchana,b, Chao-Yuan Lina, a b

MARK

⁎

Department of Soil and Water Conservation, National Chung Hsing University, 145 Xingda Rd., South Dist., Taichung 402, Taiwan Department of Irrigation Engineering, Kasetsart University (Kamphaeng Saen Campus), 1 Malaiman Rd., Kamphaeng Saen Dist., Nakhon Pathom 73140, Thailand

A R T I C L E I N F O

A B S T R A C T

Keyword: Fire scars Forest recovery index Time of complete recovery Landsat

Normalized Burn Ratio (NBR), a satellite-derived index widely used to map the burned area and to assess burn severity level, was reconceptualized to propose the indices of post-fire recovery condition and resilience. Time series Landsat imagery during 1994–2015 were used to observe the forest recovery of Wu-Ling fire scars in Taiwan. Burn Recovery Ratio (BRR) was newly developed as the indicator to better clarify the forest recovery status. Results show that BRR coupled with dNBR (bi-temporal NBR) could quantitatively describe the level of forest recovery through the heterogeneity of forest landscape which is confirmed by field investigation. Time of complete recovery (tc), indicator of post-fire resilience, were predicted using curve-fitting of forest recovery trajectories to the exponential decay function. The spatial distribution of tc could reveal the patterns of post-fire recovery across the fire scars. For wildfire prevention, the issue of fire recurrence should be concerned at the areas of fire-adapted species with low tc value. For areas of deterioration sites with high tc value, the rehabilitation project should be implemented to accelerate forest restoration.

1. Introduction Wildfire is an important disturbance on the forest ecosystem and causes a lot of impact on both global and local scales. In recent years, there are several studies showing major concerns about wildfire impacts throughout short and long-term (Shakesby, 2011; Shakesby and Doerr, 2006; van Wagtendonk et al., 2004). For the post-fire forest recovery context, it is normally monitored for a longer period for understanding the ecosystem response to fire or the rehabilitated ecosystems developed through post-fire management (Ireland and Petropoulos, 2015), and it is also involved in the restoration of forest biomass or canopy structure, soil properties and variety of species etc. Remote sensing has been highly effective to explore large coverage of fire-disturbance in remote terrain and to monitor post-fire response (Spasojevic et al., 2016). From a remote sensing perspective on ecological application, post-fire forest recovery often relates to the changes of spectral values and their temporal dynamics which could then describe recovery conditions as well as ecology and successional patterns (Bartels et al., 2016; Chu et al., 2016; Cohen and Goward, 2004). The Normalized Burn Ratio (NBR) is one of the most effective satellite-driven indices for post-fire environment studies. NBR consists of the near infrared (NIR) spectral region at 0.76–0.90 μm which is

⁎

Corresponding author. E-mail address: [email protected] (C.-Y. Lin).

http://dx.doi.org/10.1016/j.ecolind.2017.04.038 Received 2 November 2016; Received in revised form 10 April 2017; Accepted 14 April 2017 1470-160X/ © 2017 Elsevier Ltd. All rights reserved.

particularly sensitive to the changes in the chlorophyll content of live vegetation, and short wave infrared (SWIR) spectral region at 2.08–2.35 μm which is sensitive to water content in both vegetation and soils, and some soil conditions (Miller and Thode, 2007). The difference Normalized Burn Ratio (dNBR), bi-temporal NBR between pre- and post-fire, is broadly used to detect burned area and interpret burn severity level (Chafer, 2008; De Santis and Chuvieco, 2007; Epting et al., 2005; Escuin et al., 2008; Lee et al., 2008; Vlassova et al., 2014). Since NBR is mainly developed for assessing burn severity, several studies are concerned that NBR does not adequately convey information about ecosystem responses in an early succession (Keeley, 2009; Lentile et al., 2007; Murphy et al., 2008). Nonetheless, many studies confirmed a reasonably good performance of NBR for quantifying the long-term vegetation regeneration on fire scars (Chen et al., 2011; Epting and Verbyla, 2005; García and Caselles, 1991). A time series of satellite-derived burn severity index acquiring over a relatively long period of time enables the generation of the disturbance-recovery trajectory as well as the characterization of ecosystem pattern and process (Gómez et al., 2011; Kennedy et al., 2014; Pasquarella et al., 2016; Walker et al., 2010). Conceptually, at any point during the post-burn, recovery should be the deduction of severity as shown in Fig. 1. It could be alternately defined that the long-term severity is the residual suffer level referencing with the pre-fire

Ecological Indicators 79 (2017) 196–206

C. Chompuchan, C.-Y. Lin

Fig. 1. Changes of residual severity and the correspondent recovery ratio (time axis is not scale, modified from Key, 2006).

Both fire scars were approximately 350 ha with similar altitude range of 1600–2500 m a.s.l. and average slope of 28°. The main vegetation type was pine plantation forest. Due to the implementation of “Stand Conversion Project” from the Forestry Bureau during 1967–1975, Taiwan red pine was selected as the major plantation species with the purpose of improving poorly stocked natural forests through reforestation (Lin et al., 2005; Lo and Feng, 1987). The overstories were dominated by Pinus taiwanensis (Taiwan red pine), Picea asperata (dragon spruce), Pinus armandii (Chinese white pine), Alnus formosana (Formosan alder) and the understory was mainly covered with Miscanthus sinensis (Chinese silver grass). The whole Huan-Shan area and partial Sheng-Guang area are located in Shei-Pa national park which was established in 1992. The fires not only destroyed forests but also consequently affected the habitat of Formosan landlocked salmon. The rise in conservation awareness has focused attention on forest post-fire restoration work (Chen and Chen, 2015).

condition, whereas the long-term recovery is the degree of regeneration from the greatest severity level. Therefore, forest recovery should be considered with 3 reference points in the consequent scale i.e. the time of pre-fire condition, the time of highest severity and the time of assessment. Generally, the highest severity level may not be reached few weeks/months after being burned due to the delayed mortality effect: the foliage that appears green and outwardly healthy soon after burning (Key, 2006), especially in a high dense canopy forest. Two conceptual definitions of resilience have been widely used in the dynamics of ecological systems, generally referred to as ecological resilience and engineering resilience (Todman et al., 2016). Ecological resilience has been defined as the persistence of systems and their capacity to resist the amount of disturbance and still retain the main functions and processes (Holling, 1973). Engineering resilience has been defined as the time to return the disturbed ecosystems back to stable state (Pimm, 1984). Much of the current research shows that engineering resilience is potentially applicable to long-term ecosystem recovery following disturbance, through analysis of recovery trajectories (Díaz-Delgado et al., 2002; Newton and Cantarello, 2015; Nolan et al., 2015). From the trajectory of the forest recovery as shown in Fig. 1, time of complete recovery (tc) could then be predicted. In this study, engineering resilience here is defined as tc, and stable state is assumed as the level of pre-disturbed or undisturbed reference area. This study aimed to develop the forest recovery index and predict the time of complete recovery considering the concept mentioned above, using multi-temporal Landsat imagery. Post-fire forest recovery was explored in Wu-Ling fire scars in central Taiwan. Aerial photos, the images extracted from Google Earth®, and field investigation were the supplementary data to clearly describe the variations in forest recovery.

2.2. Data set Cloud-free Landsat imagery (Path: 117, Row: 043), with spatial resolution 30 m, during years 1994–2015 were obtained for free from the United States Geological Survey (USGS) Earth Explorer website (URL http://earthexplorer.usgs.gov/) as listed in Table 1. Landsat 8 Operational Land Imager (OLI) surface reflectance scenes were preprocessed from Landsat 8 Surface Reflectance (L8SR) algorithm, however Landsat 5 Thematic Mapper (TM) and Landsat 7 Enhanced Thematic Mapper Plus (ETM+) data need to correct the surface reflectance using Landsat Ecosystem Disturbance Adaptive Processing System (LEDAPS) algorithm (U.S. Geological Survey, 2015a,b). Due to a slight difference between OLI and ETM+ reflectance, a simple empirical procedure developed by Flood (2014) was applied to remove between-sensor difference. Multi-temporal datasets were acquired in the same period in order to minimize the between-scene misinterpretation caused by seasonal variation on crop phenology. Topographic effects due to low sun elevation angle in the winter image could degrade the surface reflectance value (Key, 2006; Picotte and Robertson, 2011; van Wagtendonk et al., 2004). In Taiwan, the precipitation is generally high during summer, resulting in less opportunity to retrieve cloud-free Landsat scenes. Therefore, in this study autumn series images (October–December scenes) were selected.

2. Materials and methods 2.1. Study area Wu-Ling farm is located at the south-east of Shei-Pa national park, Taiwan (Fig. 2). The main river is Qijiawan Creek which is the habitat of Formosan landlocked salmon (Oncorhynchus masou formosanus), a critical endangered species. During the 1960′s, a group of Taiwan’s veterans resettled there as farmers and started the agricultural development of the mountain area. The area was named as “Wu-Ling farm” on 10th May 1963 (Veterans Affairs Commission R.O.C., 2011). In the late 1990s, the government’s policy shifted towards environmental protection as well as sustainable development, and the agricultural activities therefore shifted to ecotourism eventually. There are 2 fire scars, Huan-Shan and Sheng-Guang, located at the south-west and the east of Wu-Ling farm respectively. Historical records show that Huan-Shan site was burned on 3rd December 1995 and 11th February 2001; and Sheng-Guang site was burned on 11th May 2002.

2.3. Methodology 2.3.1. Burn perimeter extraction and severity level There were several studies in burned area delineation that use various spectral indices e.g. Normalized Difference Vegetation Index (NDVI) (Fernández et al., 1997), Global Environmental Index (GEMI) 197

Ecological Indicators 79 (2017) 196–206

C. Chompuchan, C.-Y. Lin

Fig. 2. Study area.

The dNBR value varies between −2 to 2. Key and Benson (2006) proposed the practical data range −0.5 to 1.3 where dNBR less than −0.5 or greater than 1.3 may occur but are usually not considered as burnt area. Rather, they are likely anomalies caused by miss-registration, clouds, or other factors not related to real land cover differences (Ireland and Petropoulos, 2015). Ordinal severity level and example range of dNBR are shown in Table 2. The threshold value of dNBR to distinguish unburned and low severity is 0.10, which is close to the values derived from Epting et al. (2005), Escuin et al. (2008) and Cansler and McKenzie (2012), respectively. Initially, the average NBR of burned and unburned pixels were sampled and plotted as shown in Fig. 3. Fire events on December 1995 and February 2001 caused the dramatic decline of NBR value at HuanShan site, as well as the fire event on May 2002 at Sheng-Guang site. After that, those two variables continuously dropped in the following year due to delay mortality. For the burn boundary extraction considering the lag mortality effect at Sheng-Guang site, pre-fire and

(Pereira, 1999), Burnt Area Index (BAI) (Chuvieco et al., 2002; Liu et al., 2015), Mid-Infrared Bi-spectral Index (MIRBI) (Smith et al., 2007), Normalized Difference of Disturbance Index (NDDI) (He et al., 2011). NBR outperformed the other spectral indices in burned area delineation (Schepers et al., 2014; Veraverbeke et al., 2011) and also provided a flexible, robust, and analytical simple approach (Brewer et al., 2005). NBR is defined as:

NBR =

ρNIR − ρSWIR2 ρNIR + ρSWIR2

(1)

where ρNIR and ρSWIR2 are the reflectance of near infrared wavelength, Landsat TM/ETM+ band 4 and shortwave infrared wavelength, Landsat TM/ETM+ band 7 respectively. The range of NBR values is between −1 to 1. The dNBR is used in this study to extract burned area and to assess burn severity level. The dNBR could be expressed as:

dNBR = NBR pre − NBR post

(2)

Table 1 Landsat scenes selected. Date (dd/mm/yyyy) 06/11/1994 24/10/1995 14/11/1997 12/11/1999 30/11/2000 25/11/2001 06/12/2002 15/11/2003 01/11/2004 04/11/2005 22/10/2006 28/11/2008 30/10/2009 05/11/2011 26/11/2013 29/11/2014 16/11/2015

Satellite/Sensor

Huan-Shan site

Sheng-Guang site st

392 days before 1 burn 40 days before 1st burn* 712 days after 1st burn 1440 days after 1st burn 1824 days after 1st burn 280 days after 2nd burn 656 days after 2nd burn** 1000 days after 2nd burn 1352 days after 2nd burn 1720 days after 2nd burn 2072 days after 2nd burn 2840 days after 2nd burn 3176 days after 2nd burn 3912 days after 2nd burn 4664 days after 2nd burn 5032 days after 2nd burn 5384 days after 2nd burn

Landsat-5/TM Landsat-5/TM Landsat-5/TM Landsat-7/ETM+ Landsat-7/ETM+ Landsat-5/TM Landsat-7/ETM+ Landsat-5/TM Landsat-5/TM Landsat-5/TM Landsat-5/TM Landsat-5/TM Landsat-5/TM Landsat-5/TM Landsat-8/OLI Landsat-8/OLI Landsat-8/OLI

* Pre-fire scenes. ** Post-fire scenes which showed the highest magnitude of fire damage.

198

2743 days before burn 2391 days before burn 1639 days before burn 911 days before burn 527 days before burn 167 days before burn* 209 days after burn 553 days after burn** 905 days after burn 1273 days after burn 1625 days after burn 2393 days after burn 2729 days after burn 3465 days after burn 4217 days after burn 4585 days after burn 4937 days after burn

Ecological Indicators 79 (2017) 196–206

C. Chompuchan, C.-Y. Lin

plants occurred (Franklin et al., 2006), and could contribute to the decrease of severity level. At the final year of the assessment, about 76.24% of Huan-Shan site and 57.33% of Sheng-Guang site were classified into unburned level (Table 3). Although the pixels of unburned level could consist of “recovered pixels” and “unburned pixels”, they cannot be further classified using dNBR at this stage (Fig. 4d and h). The “unburned pixels” have no significance in recovery context and should be masked at the delay mortality stage. The severity level interpreted using dNBR is the residual severity. From a recovery perspective, such descriptions are insufficient to clarify recovery implication. Therefore a new index to describe the recovery condition in response to the post-burn magnitude of damage is essential.

Table 2 Ordinal severity level and dNBR range. Severity Level

dNBR Range

Enhance regrowth, high Enhance regrowth, low Unburned Low Moderate-low Moderate-high High

−0.500 to −0.251 −0.250 to −0.101 −0.100 to 0.099 0.100 to 0.269 0.270 to 0.439 0.440 to 0.659 0.660 to 1.300

Key and Benson (2006).

post-fire images were taken on November 25th, 2001 and November 15th, 2003, respectively. The image acquired on December 6th, 2002 was selected to be the post-fire condition (after 2nd burn) whereas the image taken on October 24th, 1995 (before 1st burn) was selected as pre-fire condition at Huan-Shan site due to twice fire-event occurrence. Fig. 4a–h shows the spatial distribution of burn severity based on dNBR. Initially, delay mortality and recovery stages (Fig. 1) were categorized to quantitatively describe the burn severity. At the initial stage, about 29.26% of Huan-Shan site and 44.90% of Sheng-Guang site can be classified into unburned level according to dNBR threshold (Table 3). The relative low-intensity ground fires may successfully damage roots and cambium of conifers, but the canopies still apparently seem to be healthy after a certain period of time (Key, 2006). Satellite image could not detect bole and crown scorch from the top layer, and the sites where ground fires affected could be classified as unburned severity level. Therefore, the classification of burn severity from dNBR during the early stage could not signify the actual magnitude of forest damage. At the delay mortality stage, several unburned pixels of Huan-Shan site classified by dNBR have been changed to burned severity levels from 29.26% in 2001–11.92% in 2002 (Table 3). As Huan-Shan site, the effects of delay mortality in Sheng-Guang site could greatly decrease the number of unburned pixels from 44.90% in 2002–9.11% in 2003 (Table 3). This phenomenon is consistent with previous studies which found that delay mortality occurred a few years after burning and its consequence could increase the number of pixels rise severity level in the following year (Kane et al., 2013; Whittier and Gray, 2016). Therefore, the burn severity assessment according to dNBR should consider the phenomenon of lag mortality. During the recovery stage, burn severity assessment would be more complicated at high severity level sites due to the occurrence of several processes (Veraverbeke et al., 2010). For example, most pixels with high severity level are recovering when considering the comparison between high severity level pixels derived from dNBR2002 (Fig. 4e) and dNBR2003 (Fig. 4f). This recovering phenomenon explained that the extreme damage of combustion yields a large amount of ash and charred matter which could be detected by satellite images and classified into high severity level at the initial stage (Kokaly et al., 2007), and then the ash and charred matter could be easily eroded by rainfall and/or wind. At the same time, the succession of invasion

2.3.2. Burn Recovery Ratio (BRR) The spectral index for the forest recovery here is defined as the magnitude of recovery at the time of assessment divided by the magnitude of fire damage. This study adopted the previous concept of Vegetation Recovery Ratio (VRR) (Chou et al., 2009; Lin et al., 2004). Based on NBR, the Burn Recovery Ratio (BRR) for the forest recovery can be expressed as:

BRR(%) =

NBR(t a) − NBR(t d) × 100 NBR(t o) − NBR(t d)

(3)

where to, td and ta mean the index at pre-fire event, the index at the time when delay mortality existed and the index at the time of assessment, respectively. The values of BRR were categorized into 6 levels i.e. very poor (< 0%), poor (0%–25%) average (25%–50%), good (50%–75%), very good (75%–100%) and excellent (> 100%). Originally, VRR was used to monitor vegetation recovery of a landslide (Chou et al., 2009; Lin et al., 2006). Our modified index aims to explore post-fire recovery, thus the appropriate post-fire images concerning the effect of delay mortality were selected instead of the images immediately after fire events. In addition, the unburned islands within burn boundaries were subtracted using dNBR threshold value of 0.1 at delay mortality stage. 2.3.3. Prediction time of complete recovery Several studies have shown that forest regeneration after fire events could be characterized based on a modified Beer’s law by fitting time series of spectral indices to the single-exponential equation (Baret and Guyot, 1991; Gouveia et al., 2010; Viedma et al., 1997). This study adopted Beer’s law by assigning the time series of NBR as a function of time. Let NBR(t) be defined as post-fire NBR at any time of assessment and then y(t) can be expressed as follows:

y(t) = NBR(t) − NBR α = − NBR oe−k.t

(4)

where t is time after fire (year), k is the decay constant and it is expressed in the unit of year−1, NBRα is the reference NBR obtained from the nearby area which is undisturbed forest with healthy state of vegetation (Riaño et al., 2002) and NBRo is interception of the curve which corresponds to the theoretical lowest NBR value at initial stage of

Fig. 3. NBR trajectories of burned and unburned area at Huan-Shan site (left) and Sheng-Guang site (right).

199

Ecological Indicators 79 (2017) 196–206

C. Chompuchan, C.-Y. Lin

Fig. 4. Severity level of Huan-Shan site (a–d) and Sheng-Guang site (e–h) interpreted from dNBR.

greater tc estimation (Olson, 1963), therefore the pixels in which k values were lower than 0.01 were removed and then Kriging interpolation was employed to replace the tc value for the eliminated pixels.

the burned (Viedma et al., 1997). Gouveia et al. (2010) described the meaning of k and NBRo parameters characterized as the rate of recovery and the indicator of fire damage, respectively. To minimize phenological variations among the compared areas (Díaz-Delgado et al., 2003), NBRα value of 0.72 was determined from the average NBR of unburned reference sites which are assumed to be stable stage throughout the assessment duration. We plotted y(t) versus t for each pixel to estimate NBRo and k parameters using curve fitting. NBR values during delay mortality duration were ignored to avoid the anomalies on parameters estimation. The ideal complete recovery can be defined as y(t) at the time of complete recovery (t = tc) is equal to zero. The time of complete recovery (tc) could be expressed as shown in Eq. (5) and applied to individual burned pixels throughout the study areas.

tc=

⎛ NBR(t) − NBR ⎞ ln⎜ −NBR α ⎟ o ⎠ ⎝ −k

≈

⎛ 0.95NBR α − NBR α ⎞ ln⎜ ⎟ −NBR o ⎠ ⎝ −k

=

3. Results and discussions 3.1. Forest recovery conditions Fig. 5 shows the spatial distribution of forest recovery level of both fire scars based on BRR. The excellent recovery level (BRR > 100%) means that the recovery condition is superior to the pre-disturbance condition. The recovery condition of the year 2015 was divided into 3 groups for quantitatively describing the levels of complete, partial complete and incomplete recovery (Table 4) from the overlay of BRR and dNBR. The results showed that 22.96% of Huan-Shan site and 14.54% of Sheng-Guang site could reach the complete recovery condition for the final year of the assessment (Table 4). The heterogeneity of the forest landscape at Huan-Shan site could qualitatively demonstrate the levels of recovery condition (Fig. 6). The interested quadrat marked in Fig. 6a shows that the burned areas in 2006 have reached the level of partial complete recovery condition in

⎛ 0.05NBR ⎞ ln⎜ NBR α ⎟ o ⎠ ⎝ (5)

−k

The above logarithmic equation is applied under the condition of y(t) greater than zero. Thus, we defined the level of complete recovery as 95% of NBRα. The very low k parameter could mathematically result in

Table 3 Percentage of area of difference severity level interpreted from dNBR at Huan-Shan and Sheng-Guang fire scars. Site

Huan Shan

Sheng Guang

Severity level

Unburned Low Mod-low Mod-high High

Year 2001

2002

2003

2004

2005

2006

2008

2009

2011

2013

2014

2015

29.26 40.80 19.65 10.11 0.18

11.92 43.28 30.10 14.36 0.34

21.12 53.65 19.70 5.53 0.00

36.55 42.39 16.87 4.19 0.00

43.20 40.56 15.38 0.86 0.00

50.95 35.47 11.06 2.52 0.00

60.29 31.20 6.44 2.04 0.03

74.06 21.54 3.51 0.89 0.00

74.78 20.72 3.14 1.36 0.00

68.17 28.40 3.25 0.18 0.00

61.07 35.00 3.56 0.37 0.00

76.24 21.27 2.46 0.03 0.00

Total

100.00

100.00

100.00

100.00

100.00

100.00

100.00

100.00

100.00

100.00

100.00

100.00

Unburned Low Mod-low Mod-high High

– – – – –

44.90 30.57 11.89 7.51 5.13

9.12 32.47 31.85 26.15 0.41

15.53 30.11 32.06 21.96 0.34

21.75 42.62 33.25 2.38 0.00

34.20 53.50 11.89 0.41 0.00

44.49 36.61 15.64 3.26 0.00

50.13 32.68 13.08 3.44 0.67

43.11 33.95 18.02 4.87 0.05

37.44 42.80 16.86 2.90 0.00

48.32 33.64 15.15 2.90 0.00

57.33 31.28 11.06 0.34 0.00

Total

–

100.00

100.00

100.00

100.00

100.00

100.00

100.00

100.00

100.00

100.00

100.00

200

Ecological Indicators 79 (2017) 196–206

C. Chompuchan, C.-Y. Lin

Fig. 5. Forest recovery status at (a–d) Huan-Shan site, (e–h) Sheng-Guang site interpreted from BRR.

no salvage logging practice after wildfires. The exposed remnants sparely distribute on the east-aspect of Sheng-Guang. Logging fraction in burned pixels could result in low NBR value and therefore degrade the level of recovery. We proposed BRR, the relative index, to measure the degree of return to prior status (pre-fire) which might be more robust than dNBR. Interpreting the effect of fire from thresholding the absolute values of bi-temporal NBR has a certain limit on the timing of the assessment (Lhermitte et al., 2011). Considering post-fire dynamics in Huan-Shan as shown in Fig. 8, dNBR trends of the recovering forests with the condition of complete and partial complete recovery were found to decrease gradually to less than 0.1 at the final year of assessment, and could be distinguished from that of the firebreaks with the condition of incomplete recovery. As such, the severity levels of the recovering forests become similar to that of the unburned forests. The comparison of BRR at the final year indicates that the spectral index of the sites with incomplete and partial complete recovery still has not equivalently return to the pre-fire status. A fluctuation in BRR of the firebreaks is associated with regularly operating practices. The use of time series was recommended to monitor multiple disturbances on the post-fire dynamics. For example, at the east-aspect of Sheng-Guang outside Shei-Pa national park boundary, Forest Conservation and Management Administration, Veterans Affairs Commission applied the conservation methods over the area that experienced high burn severity. In addition to the staking and wattling method, the Rehabilitation Projects (RP) have been continuously implemented to accelerate forest restoration since 2005. The species of native broadleaved tree, e.g. Cyclobalanopsis stenophylloides Hayata, Schima superba Gardn. et Champ, and Quercus glauca were planted to create a conifer-hardwood mixed forest in the areas because the rapid invasion of Miscanthus sinensis dominated grass communities could inhibit tree succession at the initial stage of recovery. Fig. 8 shows the effect of difference post-fire managements on the recovery dynamics. Considering the final year of assessment, these 3 sites have similar levels of recovery, but that actually is quite unlike their recovering processes. BRR pattern of the site without RP shows the progress of recovery at the beginning stage but still does not reach the level of complete recovery as a result of grass succession, whereas the site with RP show the disturbance of natural recovery which result in high

Table 4 Forest recovery conditions of Huan-Shan and Sheng-Guang sites in 2015. Category of recovery conditions

Incomplete Partial complete Complete

Conditions

Huan-Shan site

Sheng-Guang site

Area (ha)

%

Area (ha)

%

BRR ≤ 100%, dNBR > 0.1 BRR ≤ 100%, dNBR ≤ 0.1 BRR > 100%, dNBR ≤ 0.1

81.54

26.95

148.23

46.96

151.56

50.09

121.50

38.49

69.48

22.96

45.90

14.54

Total

302.58

100.00

315.63

100.00

2015 except the location of the firebreak belt (Fig. 6b). On-site investigation also revealed the same levels of recovery condition derived from the recovery index (Fig. 6c). Theoretically, NBR would vary by canopy moisture (Murphy et al., 2008). The areas with partial complete recovery show that the vegetation community is still suffering from physiological stress, in which the vegetation is vulnerable to drought and could wither in dry seasons (Keeling and Sala, 2012). The areas of man-made restoration occupied most percentages of the incomplete recovery category on both fire scars. At Huan-Shan site, the pixels of incomplete recovery were mostly located on the ridges where the firebreaks and erosion control practices were implemented. At Sheng-Guang site, there were erosion control and firebreak treatments which obviously resulted in incomplete recovery. Fig. 7a and b shows the heterogeneity of the experimental site; unburned area with dense vegetation and burned area with sparse vegetation. Aerial photos and field surveys reveal that the staking and wattling method were implemented at the burned area (Fig. 7c and d). The method is one of the common ecological engineering practices in Taiwan, and is usually adopted by the Soil and Water Conservation Bureau to reduce soil erosion and to enhance revegetation in the initial stage (Wu and Feng, 2006). Due to the sensitivity of NBR on the non-photosynthetic vegetation (NPV) (Souza et al., 2005), burned forested stands generally have charred boles and large branches that are retained and could influence spectral response (Epting et al., 2005). At Wu-Ling fire scars, there are 201

Ecological Indicators 79 (2017) 196–206

C. Chompuchan, C.-Y. Lin

Fig. 6. Landscape-scale heterogeneity at Huan-Shan site with difference recovery condition (a) image from Google Earth® on 1st Feb. 2006, (b) image from Google Earth® on 14th Oct. 2015 and (c) photo from on-site investigation on 28th Nov. 2015.

predicted tc of Huan-Shan site requires shorter recovery time than that of Sheng-Guang site, and k value indicates that Huan-Shan site has a faster recovery rate than that of Sheng-Guang site. Fire regime does affect tc and k values in both sites due to fire occurring frequency. Taiwan red pines, a fire-adapted species, were introduced under the implementation of Stand Conversion Project during 1967–1975, and the document of fire history reveals that wildfire occurred in HuanShan site with a return period of 7-year whereas only a single event occurred in Sheng-Guang site for the time being (Lin et al., 2005; Lin and Qiu, 2002). Fire recurrence could help the removal of accumulated fuel and allow the burned pine community to recover rapidly. On the contrary, frequent fire could cause drought vulnerability in the burned

magnitude of change in BBR. The year of implementing RP could be identified on the segmented trajectories. Thus, the post-fire dynamic from time series is crucial to examine the operations following fire disturbance. 3.2. Time of complete recovery Fig. 9 shows a significant relationship between y(t) and t (p < 0.001). The recovery trajectory shows a good fit to the exponential decay curve, and then the decay curve model could be used to predict the time of complete recovery (tc). The estimated tc and its upper and lower confidence intervals were shown in Table 5. The

Fig. 7. Landscape-scale heterogeneity at Sheng-Guang site in which implemented staking and wattling method for soil erosion control (a) aerial photo, (b) photo from on-site investigation on 28th Nov. 2015, (c) example of staking and wattling (Soil and Water Conservation Bureau, 2007) and (d) staking and wattling method.

202

Ecological Indicators 79 (2017) 196–206

C. Chompuchan, C.-Y. Lin

Fig. 8. BRR and dNBR of Huan-Shan site (left) and Sheng-Guang site (right), dot lines are 0.1 dNBR threshold value and long dash lines are 100% of BRR.

sites (Lippitt et al., 2013; Meng et al., 2015) and interfere the following vegetation succession. The spatial distribution of tc in Huan-Shan and Sheng-Guang sites were shown in Fig. 10. The recovery pattern could provide several practical implications for forest managers and stakeholders. Due to the sensitivity of NBR on the canopy moisture content, the locations which are vulnerable to continuous artificial and natural interference could result in a prolong tc prediction such as the firebreak treatment implemented at Huan-Shan site, erosion control at the burn severe areas at Sheng-Guang site and the landslides at both sites (Fig. 10a–d). To prevent the spread of wildfire, firebreaks located at the ridge of Huan-Shan site (Fig. 10a) have been operated and stayed in a clear-cut status, which could lower the time of complete recovery of the burned site. However, the firebreaks could also decrease the risk of wildfire hazards for the burned site with fire-adapted species. Landslides located on the east aspect of Sheng-Guang site (Fig. 10d) and the area of rivercut cliff in Huan-Shan site (Fig. 10b) have been suffering from extreme soil loss and have shallow soil layers with the properties of low soil nutrients and moisture contents (Milodowski et al., 2015) which could result in poor revegetation and a high tc value would be expected. The spatial distribution of tc in Sheng-Guang site could be roughly classified into two categories of aspect-based delineation using the boundary of national park. Classified tc values in areas of the east-

Table 5 Parameters of the exponential model and the time of complete recovery (tc). Site

NBRo

k (year−1)

tc (year)

Lower C.I. of tc (year)

Upper C.I. of tc (year)

Huan-Shan Sheng-Guang

0.3599 0.3587

0.0985 0.0843

23.38 27.29

18.30 18.32

32.36 53.49

aspect are larger than that of the west-aspect (Fig. 10). Areas of the west-aspect with dense vegetation are mostly located at the territory of the park; however sparse vegetation could be spotted at the areas of the east-aspect outside the park (Fig. 10c). Secondary succession of Taiwan red pine could easily occur and develop at the park due to the protection of artificial interference prohibited under national park law. Pine forest with tc less than 15 years could form fire-prone environment and cause the risk of periodic fire. The phenomena coupled with tc index could be employed as an effective countermeasure of forest fire prevention. As a result of erosion control and artificial rehabilitation, the eastaspect of Sheng-Guang is prone to prolong the time of complete recovery. Vegetation engineering method of staking and wattling which was applied at the east-aspect slopeland could interfere natural succes-

Fig. 9. Time series of exponential decay model of Huan-Shan site (left) and Sheng-Guang site (right), bold lines are regression lines, dot lines are 95% of confidential interval (C.I.) and long dash lines are 95% of y(t) threshold value. ∗∗∗ p < 0.001.

203

Ecological Indicators 79 (2017) 196–206

C. Chompuchan, C.-Y. Lin

Fig. 10. Spatial distribution of time of complete recovery (tc) in Huan-Shan site (left) and Sheng-Guang (right). The specific cases of high tc value are highlighted with dash quadrates. (a)–(d) are images from Google Earth® on 8th Feb. 2016.

conditions. Therefore, BRR and tc may not adequately clarify the other properties (e.g. forest height, biomass or species diversity) which required the various approaches/instruments to obtain (e.g. airborne hyperspectral remote sensing, LIDAR data, ground-based investigation) (Frolking et al., 2009). In addition, previous studies concluded that NBR has limitations for assessing regeneration of grasslands and shrublands (Chen et al., 2011; Meng et al., 2014). Another limitation of NBR is that it requires the short wave infrared band at wavelengths 2.08–2.35 μm, which is present only in several satellite platforms e.g., TM/ETM+, OLI, MODIS, ASTER.

sion and result in prolong tc. Results of this study are consistent to the idea that the post-fire rehabilitation treatments could be effective for runoff and sediment yield reduction (Fernández et al., 2011; GimenoGarcía et al., 2007; Robichaud, 2005) but prolong the vegetation recovery (Dodson et al., 2010). In addition, the Rehabilitation Projects do disturb the succession process during the beginning of recovery stage and result in prolonged tc, but if without artificial rehabilitation, the succession of fire-adapted species could probably dominate the burned site and then be at risk of fire recurrence. From the ecological perspective, the shorter the recovery time is, the higher the resilience (Di-Mauro et al., 2014). In this study, high resilience areas could imply that thespectral index value approaches the steady state which refers to undisturbed forest nearby the fire scars. Low resilience areas involve the human perturbation and deterioration sites. Criticism of engineering resilience to measure in ecosystems has pointed out the fact that the natural succession including human influence could recover toward multiple stable states (Folke et al., 2004). Multiple disturbances can occur during the recovery period which increase the risk of degrading stable state (Magnuszewski et al., 2015). On the contrary, artificial rehabilitation might promote the stable state. Thus, other mathematical response functions should be challenged to accurately identify various equilibriums of ecosystem responses.

4. Conclusions Spectral indices derived from remotely sensed imagery were a common approach to measure the magnitude of ecosystem change as a result of fire and to monitor ecosystem response. This study demonstrated the ability of time series Landsat data for dynamic monitoring the post-fire of Wu-Ling fire scars in central Taiwan. Burn boundary with respect to delay mortality effect was extracted by dNBR, and the burn severity could be then quantitatively described. However, for a long-term assessment, the ordinal severity level categorized by dNBR is inadequate to interpret forest recovery context. BRR, a newly developed index, was purposed and introduced to categorize the forest recovery condition. Results show that BRR coupled with dNBR could enhance the post-fire recovery interpretation. On-site investigation, aerial photos and the images extracted from Google Earth® revealed that the heterogeneity of forest landscapes are formed as a result of different levels of recovery condition. Since the trend of trajectory of forest recovery shows a good fit to the exponential decay model, the model could be employed to predict the time of complete recovery (tc). Mapping of tc could display the spatial distributions of resilience across

3.3. Limitation of indices There are several limitations of BRR and tc applications that should be remarked. Since the indices estimation were based on NBR, the context of recovery could refer to the regeneration of forest following a fire disturbance in terms of the amount of chlorophyll content in vegetation, forest canopy and soil moisture content, and some soil 204

Ecological Indicators 79 (2017) 196–206

C. Chompuchan, C.-Y. Lin

the fire scars. Fire regime affects tc obviously. Although areas with lower tc could recover within a short duration, the problem of fire recurrence should be concerned. Areas with higher tc are strongly recommended to restore forest ecosystem with artificial rehabilitation.

60, 153–162. http://dx.doi.org/10.1016/S0034-4257(96)00178-2. Fernández, C., Vega, J.A., Jiménez, E., Fonturbel, T., 2011. Effectiveness of three post-fire treatments at reducing soil erosion in Galicia (NW Spain). Int. J. Wildl. Fire 20, 104–114. http://dx.doi.org/10.1071/WF09010. Flood, N., 2014. Continuity of reflectance data between Landsat-7 ETM+ and Landsat-8 OLI, for both top-of-atmosphere and surface reflectance: a study in the Australian landscape. Remote Sens. 6, 7952–7970. http://dx.doi.org/10.3390/rs6097952. Folke, C., Carpenter, S., Walker, B., Scheffer, M., Elmqvist, T., Gunderson, L., Holling, C.S., 2004. Regime shifts, resilience, and biodiversity in ecosystem management. Annu. Rev. Ecol. Syst. 35, 557–581. http://dx.doi.org/10.1146/annurev.ecolsys.35. 021103.105711. Franklin, J., Spears-Lebrun, L.A., Deutschman, D.H., Marsden, K., 2006. Impact of a highintensity fire on mixed evergreen and mixed conifer forests in the Peninsular Ranges of southern California, USA. For. Ecol. Manage. 235, 18–29. http://dx.doi.org/10. 1016/j.foreco.2006.07.023. Frolking, S., Palace, M.W., Clark, D.B., Chambers, J.Q., Shugart, H.H., Hurtt, G.C., 2009. Forest disturbance and recovery: a general review in the context of spaceborne remote sensing of impacts on aboveground biomass and canopy structure. J. Geophys. Res. Biogeosci. 114. http://dx.doi.org/10.1029/2008jg000911. Gómez, C., White, J.C., Wulder, M.A., 2011. Characterizing the state and processes of change in a dynamic forest environment using hierarchical spatio-temporal segmentation. Remote Sens. Environ. 115, 1665–1679. http://dx.doi.org/10.1016/j. rse.2011.02.025. García, M.J.L., Caselles, V., 1991. Mapping burns and natural reforestation using Thematic Mapper data. Geocarto Int. 6, 31–37. http://dx.doi.org/10.1080/ 10106049109354290. Gimeno-García, E., Andreu, V., Rubio, J.L., 2007. Influence of vegetation recovery on water erosion at short and medium-term after experimental fires in a Mediterranean shrubland. Catena 69, 150–160. http://dx.doi.org/10.1016/j.catena.2006.05.003. Gouveia, C., DaCamara, C.C., Trigo, R.M., 2010. Post-fire vegetation recovery in Portugal based on SPOT/VEGETATION data. Nat. Hazards Earth Syst. Sci. 10, 673–684. http://dx.doi.org/10.5194/nhess-10-673-2010. He, L., Chen, J.M., Zhang, S., Gomez, G., Pan, Y., McCullough, K., Birdsey, R., Masek, J.G., 2011. Normalized algorithm for mapping and dating forest disturbances and regrowth for the United States. Int. J. Appl. Earth Obs. Geoinf. 13, 236–245. http:// dx.doi.org/10.1016/j.jag.2010.12.003. Holling, C.S., 1973. Resilience and stability of ecological systems. Annu. Rev. Ecol. Syst. 4, 1–23. http://dx.doi.org/10.1146/annurev.es.04.110173.000245. Ireland, G., Petropoulos, G.P., 2015. Exploring the relationships between post-fire vegetation regeneration dynamics, topography and burn severity: a case study from the Montane Cordillera Ecozones of Western Canada. Appl. Geogr. 56, 232–248. http://dx.doi.org/10.1016/j.apgeog.2014.11.016. Kane, V.R., Lutz, J.A., Roberts, S.L., Smith, D.F., McGaughey, R.J., Povak, N.A., Brooks, M.L., 2013. Landscape-scale effects of fire severity on mixed-conifer and red fir forest structure in Yosemite National Park. For. Ecol. Manage. 287, 17–31. http://dx.doi. org/10.1016/j.foreco.2012.08.044. Keeley, J.E., 2009. Fire intensity, fire severity and burn severity: a brief review and suggested usage. Int. J. Wildl. Fire 18, 116–126. http://dx.doi.org/10.1071/ WF07049. Keeling, E.G., Sala, A., 2012. Changing growth response to wildfire in old-growth ponderosa pine trees in montane forests of north central Idaho. Glob. Change Biol. 18, 1117–1126. http://dx.doi.org/10.1111/j.1365-2486.2011.02574.x. Kennedy, R.E., Andréfouët, S., Cohen, W.B., Gómez, C., Griffiths, P., Hais, M., Healey, S.P., Helmer, E.H., Hostert, P., Lyons, M.B., Meigs, G.W., Pflugmacher, D., Phinn, S.R., Powell, S.L., Scarth, P., Sen, S., Schroeder, T.A., Schneider, A., Sonnenschein, R., Vogelmann, J.E., Wulder, M.A., Zhu, Z., 2014. Bringing an ecological view of change to landsat-based remote sensing. Front. Ecol. Environ. 12, 339–346. http://dx.doi. org/10.1890/130066. Key, C.H., Benson, N.C., 2006. Landscape assessment (LA): sampling and analysis methods. In: Lutes, D.C., Keane, R.E., Caratti, J.F., Key, C.H., Benson, N.C., Sutherland, S., Gangi, L.G. (Eds.), FIREMON: Fire Effects Monitoring and Inventory System (General Technical Report RMRS-GTR-164-CD). Rocky Mountains Research Station, USDA Forest Service, pp. LA1–LA55. Key, C.H., 2006. Ecological and sampling constraints on defining landscape fire severity. Fire Ecol. 2, 34–59. http://dx.doi.org/10.4996/fireecology.0202034. Kokaly, R.F., Rockwell, B.W., Haire, S.L., King, T.V.V., 2007. Characterization of post-fire surface cover, soils, and burn severity at the Cerro Grande Fire, New Mexico, using hyperspectral and multispectral remote sensing. Remote Sens. Environ. 106, 305–325. http://dx.doi.org/10.1016/j.rse.2006.08.006. Lee, B., Kim, S.Y., Chung, J., Park, P.S., 2008. Estimation of fire severity by use of Landsat TM images and its relevance to vegetation and topography in the 2000 Samcheok forest fire. J. For. Res. 13, 197–204. http://dx.doi.org/10.1007/s10310-008-0072-x. Lentile, L.B., Morgan, P., Hudak, A.T., Bobbitt, M.J., Lewis, S.a., Smith, A.M.S., Robichaud, P.R., 2007. Post-fire burn severity and vegetation response following eight large wildfires across the Western United States. Fire Ecol. 3, 91–108. http://dx. doi.org/10.4996/fireecology.0301091. Lhermitte, S., Verbesselt, J., Verstraeten, W.W., Veraverbeke, S., Coppin, P., 2011. Assessing intra-annual vegetation regrowth after fire using the pixel based regeneration index. ISPRS J. Photogramm. Remote Sens. 66, 17–27. http://dx.doi. org/10.1016/j.isprsjprs.2010.08.004. Lin, Y.F., Qiu, Q.A., 2002. Change of the Vegetation After Fire on Fansan and Mt. Shei East Peak. Shei-Pa National Park, Republic of China (Taiwan). Lin, C.Y., Lo, H.M., Chou, W.C., Lin, W.T., 2004. Vegetation recovery assessment at the Jou–Jou Mountain landslide area caused by the 921 earthquake in Central Taiwan. Ecol. Modell. 176, 75–81. http://dx.doi.org/10.1016/j.ecolmodel.2003.12.037. Lin, C.-C., Chiou, C.-R., Chou, C.-Y., 2005. Identifying and evaluating fire severity: a case

Acknowledgments This research was supported by a grant from the Ministry of Science and Technology of Taiwan R.O.C. (105-2621-M-005-002-MY3). Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.ecolind.2017.04.038. References Baret, F., Guyot, G., 1991. Potentials and limits of vegetation indices for LAI and APAR assessment. Remote Sens. Environ. 35, 161–173. http://dx.doi.org/10.1016/00344257(91)90009-U. Bartels, S.F., Chen, H.Y.H., Wulder, M.A., White, J.C., 2016. Trends in post-disturbance recovery rates of Canada’s forests following wildfire and harvest. For. Ecol. Manage. 361, 194–207. http://dx.doi.org/10.1016/j.foreco.2015.11.015. Brewer, C.K., Winne, J.C., Redmond, R.L., Opitz, D.W., Mangrich, M.V., 2005. Classifying and mapping wildfire severity: a comparison of methods. Photogramm. Eng. Remote Sens. 71, 1311–1320. Cansler, C.A., McKenzie, D., 2012. How robust are burn severity indices when applied in a new region? Evaluation of alternate field-based and remote-sensing methods. Remote Sens. 4, 456–483. http://dx.doi.org/10.3390/rs4020456. Chafer, C.J., 2008. A comparison of fire severity measures: an Australian example and implications for predicting major areas of soil erosion. Catena 74, 235–245. http:// dx.doi.org/10.1016/j.catena.2007.12.005. Chen, J.C., Chen, C.T., 2015. Discourse on Taiwanese forest fires. In: Douglas, P., Petra, T.B., Sarah, M., Fantin, T., John, F.S. (Eds.), Wildfire Hazards, Risks and Disasters. Elsevier, pp. 145–166. http://dx.doi.org/10.1016/B978-0-12-410434-1.00008-7. Chen, X., Vogelmann, J.E., Rollins, M., Ohlen, D., Key, C.H., Yang, L., Huang, C., Shi, H., 2011. Detecting post-fire burn severity and vegetation recovery using multitemporal remote sensing spectral indices and field-collected composite burn index data in a ponderosa pine forest. Int. J. Remote Sens. 32, 7905–7927. http://dx.doi.org/10. 1080/01431161.2010.524678. Chou, W.-C., Lin, W.-T., Lin, C.-Y., 2009. Vegetation recovery patterns assessment at landslides caused by catastrophic earthquake: a case study in central Taiwan. Environ. Monit. Assess. 152, 245–257. http://dx.doi.org/10.1007/s10661-0080312-8. Chu, T., Guo, X., Takeda, K., 2016. Remote sensing approach to detect post-fire vegetation regrowth in Siberian boreal larch forest. Ecol. Indic. 62, 32–46. http://dx.doi.org/10. 1016/j.ecolind.2015.11.026. Chuvieco, E., Martín, M.P., Palacios, A., 2002. Assessment of different spectral indices in the red-near-infrared spectral domain for burned land discrimination. Int. J. Remote Sens. 23, 5103–5110. http://dx.doi.org/10.1080/01431160210153129. Cohen, W.B., Goward, S.N., 2004. Landsat’s role in ecological applications of remote sensing. Bioscience 54, 535–545. Díaz-Delgado, R., Lloret, F., Pons, X., Terradas, J., 2002. Satellite evidence of decreasing resilience in Mediterranean plant communities after recurrent wildfires. Ecology 83, 2293–2303. http://dx.doi.org/10.1890/0012-9658(2002)083[2293:SEODRI]2.0. CO;2. Díaz-Delgado, R., Lloret, F., Pons, X., 2003. Influence of fire severity on plant regeneration by means of remote sensing imagery. Int. J. Remote Sens. 24, 1751–1763. http://dx.doi.org/10.1080/01431160210144732. De Santis, A., Chuvieco, E., 2007. Burn severity estimation from remotely sensed data: performance of simulation versus empirical models. Remote Sens. Environ. 108, 422–435. http://dx.doi.org/10.1016/j.rse.2006.11.022. Di-Mauro, B., Fava, F., Busetto, L., Crosta, G.F., Colombo, R., 2014. Post-fire resilience in the Alpine region estimated from MODIS satellite multispectral data. Int. J. Appl. Earth Obs. Geoinf. 32, 163–172. http://dx.doi.org/10.1016/j.jag.2014.04.010. Dodson, E.K., Peterson, D.W., Harrod, R.J., 2010. Impacts of erosion control treatments on native vegetation recovery after severe wildfire in the Eastern Cascades, USA. Int. J. Wildl. Fire 19, 490–499. http://dx.doi.org/10.1071/WF08194. Epting, J., Verbyla, D., 2005. Landscape-level interactions of prefire vegetation, burn severity, and postfire vegetation over a 16-year period in interior Alaska. Can. J. For. Res. 35, 1367–1377. http://dx.doi.org/10.1139/x05-060. Epting, J., Verbyla, D., Sorbel, B., 2005. Evaluation of remotely sensed indices for assessing burn severity in interior Alaska using Landsat TM and ETM+. Remote Sens. Environ. 96, 328–339. http://dx.doi.org/10.1016/j.rse.2005.03.002. Escuin, S., Navarro, R., Fernández, P., 2008. Fire severity assessment by using NBR (Normalized Burn Ratio) and NDVI (Normalized Difference Vegetation Index) derived from LANDSAT TM/ETM images. Int. J. Remote Sens. 29, 1053–1073. http:// dx.doi.org/10.1080/01431160701281072. Fernández, A., Illera, P., Casanova, J.L., 1997. Automatic mapping of surfaces affected by forest fires in Spain using AVHRR NDVI composite image data. Remote Sens. Environ.

205

Ecological Indicators 79 (2017) 196–206

C. Chompuchan, C.-Y. Lin study of the Wulin fire. Taiwan J. For. Sci. 20, 203–213. Lin, W.-T., Lin, C.-Y., Chou, W.-C., 2006. Assessment of vegetation recovery and soil erosion at landslides caused by a catastrophic earthquake: a case study in Central Taiwan. Ecol. Eng. 28, 79–89. http://dx.doi.org/10.1016/j.ecoleng.2006.04.005. Lippitt, C.L., Stow, D.A., O’Leary, J.F., Franklin, J., 2013. Influence of short-interval fire occurrence on post-fire recovery of fire-prone shrublands in California, USA. Int. J. Wildl. Fire 22 (2), 184–193. http://dx.doi.org/10.1071/wf10099. Liu, W., Wang, L., Zhou, Y., Wang, S., Zhu, J., Wang, F., 2015. A comparison of forest fire burned area indices based on HJ satellite data. Nat. Hazards. http://dx.doi.org/10. 1007/s11069-015-2115-x. Lo, S., Feng, F., 1987. An evaluation of the stand conversion project in Taiwan forestry. Bull. Exp. For. NCHU 8, 89–100. Magnuszewski, P., Ostasiewicz, K., Chazdon, R., Salk, C., Pajak, M., Sendzimir, J., Andersson, K., 2015. Resilience and alternative stable states of tropical forest landscapes under shifting cultivation regimes. PLoS One 10, 1–20. http://dx.doi.org/ 10.1371/journal.pone.0137497. Meng, R., Dennison, P.E., D’Antonio, C.M., Moritz, M.A., 2014. Remote sensing analysis of vegetation recovery following short-interval fires in Southern California Shrublands. PLoS One 9, 14–17. http://dx.doi.org/10.1371/journal.pone.0110637. Meng, R., Dennison, P.E., Huang, C., Moritz, M.A., D’Antonio, C., 2015. Effects of fire severity and post-fire climate on short-term vegetation recovery of mixed-conifer and red fir forests in the Sierra Nevada Mountains of California. Remote Sens. Environ. 171, 311–325. http://dx.doi.org/10.1016/j.rse.2015.10.024. Miller, J.D., Thode, A.E., 2007. Quantifying burn severity in a heterogeneous landscape with a relative version of the delta Normalized Burn Ratio (dNBR). Remote Sens. Environ. 109, 66–80. http://dx.doi.org/10.1016/j.rse.2006.12.006. Milodowski, D.T., Mudd, S.M., Mitchard, E.T.A., 2015. Erosion rates as a potential bottom-up control of forest structural characteristics in the Sierra Nevada Mountains. Ecology 96, 31–38. http://dx.doi.org/10.1890/14-0649.1. Murphy, K.A., Reynolds, J.H., Koltun, J.M., 2008. Evaluating the ability of the differenced Normalized Burn Ratio (dNBR) to predict ecologically significant burn severity in Alaskan boreal forests. Int. J. Wildl. Fire 17, 490–499. http://dx.doi.org/10.1071/ WF08050. Newton, A.C., Cantarello, E., 2015. Restoration of forest resilience: an achievable goal? N. For. 46, 645–668. http://dx.doi.org/10.1007/s11056-015-9489-1. Nolan, R.H., Lane, P.N.J., Benyon, R.G., Bradstock, R.A., Mitchell, P.J., 2015. Trends in evapotranspiration and streamflow following wildfire in resprouting eucalypt forests. J. Hydrol. 524, 614–624. http://dx.doi.org/10.1016/j.jhydrol.2015.02.045. Olson, J.S., 1963. Energy storage and the balance of producers and decomposers in ecological systems. Ecology 44, 322. http://dx.doi.org/10.2307/1932179. Pasquarella, V.J., Holden, C.E., Kaufman, L., Woodcock, C.E., 2016. From imagery to ecology: leveraging time series of all available Landsat observations to map and monitor ecosystem state and dynamics. Remote Sens. Ecol. Conserv. 1–19. http://dx. doi.org/10.1002/rse2.24. Pereira, J.M.C., 1999. A comparative evaluation of NOAA/AVHRR vegetation indexes for burned surface detection and mapping. IEEE Trans. Geosci. Remote Sens. 37, 217–226. http://dx.doi.org/10.1109/36.739156. Picotte, J.J., Robertson, K., 2011. Timing constraints on remote sensing of wildland fire burned area in the southeastern US. Remote Sens. 3, 1680–1690. http://dx.doi.org/ 10.3390/rs3081680. Pimm, S.L., 1984. The complexity and stability of ecosystems. Nature 307, 321–326. http://dx.doi.org/10.1038/315635c0. Riaño, D., Chuvieco, E., Ustin, S., Zomer, R., Dennison, P., Roberts, D., Salas, J., 2002. Assessment of vegetation regeneration after fire through multitemporal analysis of AVIRIS images in the Santa Monica Mountains. Remote Sens. Environ. 79, 60–71. http://dx.doi.org/10.1016/S0034-4257(01)00239-5. Robichaud, P.R., 2005. Measurement of post-fire hillslope erosion to evaluate and model rehabilitation treatment effectiveness and recovery. Int. J. Wildl. Fire 14, 475–485. http://dx.doi.org/10.1071/WF05031. Schepers, L., Haest, B., Veraverbeke, S., Spanhove, T., Vanden Borre, J., Goossens, R., 2014. Burned area detection and burn severity assessment of a heathland fire in

belgium using airborne imaging spectroscopy (APEX). Remote Sens. 6, 1803–1826. http://dx.doi.org/10.3390/rs6031803. Shakesby, R.A., Doerr, S.H., 2006. Wildfire as a hydrological and geomorphological agent. Earth-Sci. Rev. 74, 269–307. http://dx.doi.org/10.1016/j.earscirev.2005.10. 006. Shakesby, R.A., 2011. Post-wildfire soil erosion in the Mediterranean: review and future research directions. Earth-Sci. Rev. 105, 71–100. http://dx.doi.org/10.1016/j. earscirev.2011.01.001. Smith, A.M.S., Drake, N.A., Wooster, M.J., Hudak, A.T., Holden, Z.A., Gibbons, C.J., 2007. Production of Landsat ETM+ reference imagery of burned areas within Southern African savannahs: comparison of methods and application to MODIS. Int. J. Remote Sens. 28, 2753–2775. http://dx.doi.org/10.1080/01431160600954704. Soil and Water Conservation Bureau, 2007. Ecological Engineering [WWW Document]. URL http://webarchive.swcb.gov.tw/web/eco.swcb.gov.tw/po.html (Accessed 28 January 2016). Souza, C.M., Roberts, D.A., Cochrane, M.A., 2005. Combining spectral and spatial information to map canopy damage from selective logging and forest fires. Remote Sens. Environ. 98, 329–343. http://dx.doi.org/10.1016/j.rse.2005.07.013. Spasojevic, M.J., Bahlai, C.A., Bradley, B.A., Butterfield, B.J., Tuanmu, M.N., Sistla, S., Wiederholt, R., Suding, K.N., 2016. Scaling up the diversity-resilience relationship with trait databases and remote sensing data: the recovery of productivity after wildfire. Glob. Change Biol. 22, 1421–1432. http://dx.doi.org/10.1111/gcb.13174. Todman, L.C., Fraser, F.C., Corstanje, R., Deeks, L.K., Harris, J.A., Pawlett, M., Ritz, K., Whitmore, A.P., 2016. Defining and quantifying the resilience of responses to disturbance: a conceptual and modelling approach from soil science. Sci. Rep. 6, 28426. http://dx.doi.org/10.1038/srep28426. U.S. Geological Survey, 2015a. Production Guide: Provisional Landsat 8 Surface Reflectance Product. Version 1.8, December 2015. [WWW Document]. URL http:// landsat.usgs.gov/documents/provisional_l8sr_product_guide.pdf (Accessed 20 January 2016). U.S. Geological Survey, 2015b. Product Guide: Landsat 4–7 Climate Data Record (CDR) Surface Reflectance. Version 6.1, December 2015. [WWW Document]. URL http:// landsat.usgs.gov/documents/cdr_sr_product_guide.pdf (Accessed 20 January 2016). van Wagtendonk, J.W., Root, R.R., Key, C.H., 2004. Comparison of AVIRIS and Landsat ETM+ detection capabilities for burn severity. Remote Sens. Environ. 92, 397–408. http://dx.doi.org/10.1016/j.rse.2003.12.015. Veraverbeke, S., Lhermitte, S., Verstraeten, W.W., Goossens, R., 2010. The temporal dimension of differenced Normalized Burn Ratio (dNBR) fire/burn severity studies: the case of the large 2007 Peloponnese wildfires in Greece. Remote Sens. Environ. 114, 2548–2563. http://dx.doi.org/10.1016/j.rse.2010.05.029. Veraverbeke, S., Lhermitte, S., Verstraeten, W.W., Goossens, R., 2011. A time-integrated MODIS burn severity assessment using the multi-temporal differenced normalized burn ratio (dNBRMT). Int. J. Appl. Earth Obs. Geoinf. 13, 52–58. http://dx.doi.org/ 10.1016/j.jag.2010.06.006. Veterans Affairs Commission R.O.C, 2011. Wuling Story [WWW Document]. URL http:// www2. wuling-farm.com.tw/en/profile/index.php (Accessed 26 January 2016). Viedma, O., Garcia-Haro, J., Melia, J., Segarra, D., 1997. Modeling rates of ecosystem recovery fires by using Landsat TM data. Remote Sens. Environ. 61, 383–398. Vlassova, L., Pérez-Cabello, F., Mimbrero, M.R., Llovería, R.M., García-Martín, A., 2014. Analysis of the relationship between land surface temperature and wildfire severity in a series of Landsat images. Remote Sens. 6, 6136–6162. http://dx.doi.org/10.3390/ rs6076136. Walker, L.R., Wardle, D.A., Bardgett, R.D., Clarkson, B.D., 2010. The use of chronosequences in studies of ecological succession and soil development. J. Ecol. 98, 725–736. http://dx.doi.org/10.1111/j.1365-2745.2010.01664.x. Whittier, T.R., Gray, A.N., 2016. Tree mortality based fire severity classification for forest inventories: a Pacific Northwest national forests example. For. Ecol. Manage. 359, 199–209. http://dx.doi.org/10.1016/j.foreco.2015.10.015. Wu, H.L., Feng, Z.Y., 2006. Ecological engineering methods for soil and water conservation in Taiwan. Ecol. Eng. 28, 333–344. http://dx.doi.org/10.1016/j. ecoleng.2006.09.005.

206