Supplemental Table 2: Summary of Studies using Spectral Data
Paper Author Year Location Forest Type(s) Satellite Imagery Source Additional Processing Composite Type Measure of Fire Severity Method to Assess Recovery Spectral Metrics Spectral Baseline Years to ID Trajectories Field Validation Field Metric Years of Field Measures Measurement Spectral Field Findings Modeling of Drivers Application of Driver Model Relationship of Y to Coniferous Growth Estimate of Model Variance Provide a 3 sentence summary of the paper
4037 Aspinall 2025 Southern Alberta Lodgepole Pine, Engelman Spruce RPA Orthoimages to calculate a measure of greenness from three band images Green Chromatic Coordinate Creen Chromatic Coordinate NULL NULL X Biomass 4.5 Correlation between plot estimates of biomass and biomass estimatesd from RPAs lidar data Plot level biomass and biomass calculated as a transformation of the green chromatic coordinate had a strong correlation. Correlations were stronger in the dry site, where there confusion between tree cover and herbaceous material was lower X Biomass Investigates the relationship between recovery of biomass and height in distince plot types (wet and dry) in Waterton Lakes National Park. Develops a model to test the relationship between plot lovel biomass and the biomass estimated from repeart RPA flights. Uses the RPA flights to assess the temporal change in biomass and height among the moist and dry sites
3 Bright 2019 Western US forests Ponderosa Pine Google Earth Engine Landsat 4,5,7, 8 Surface Reflectance Collections Cloud Masking (CFMASK), transformation of Landsat 8 bands to correspond with Landsat 7, medioid compositing of 'summer-season' imagery, medioids used in LandTrendR algorithm Median dNBR Magnitude of recovery NBR Landtrendr requires 6 9 NULL X Pure Spectral 0.5 For 12 different forests in different forest types investigate what explains variability in NBR recovery 9 years post-fire. Investigates the variance explained by three different models (1) pre-fire NBR and dNBR (2) pre-fire NBR, dNBR, climate anomalies (3) pre-fire NBR, dNBR, climate anomalies and topography. Models the included post-fire climate anomalies, namely maximum temperature had the highest explanatory power (variance for model with the highest predictive power noted)
Mixed conifer RdNBR 0.4
522 Buma 2012 Northwest Colorado Subalpine forest MODIS 16-day TERRA NDVI (MOD13Q1) Masked pixels with reliability scores > 2 Conversion of NDVI with elevation, slope, and aspect | |dNBR |NDVI Threshold |NDVI |NULL |8 (year of highest correlation) |X |Stem Density |8 |90 paired plots measured of conifer seedlings nad percent cover of forbs |Peak NDVI values > 0.6396 had a 62% accuracy for identifying sites with 382 trees per ha |X |Spectral Response Indicative of Conifer Growth |Stem Density |0.18 |Investigates the relationship between MODIS measures of recovery and forb as well as stem densities. Finds that conifer relationships with MODIS NDVI peak in early June, whereas the correlation with forb cover peaks in late summer. Generally, peak NDVI better explained for cover but was also associated with regeneration density. Develops a peak NDVI cutoff value for plots that successfully regenerated in year 8
439 Celebrezze 2024 Central Oregon, Central Washington Mixed Conifer Google Earth Engine Landsat Collection 2 Google Earth Engine implementation of LandTrendr for growing season (June to August) Median MTBS Slope of Recovery, Magnitude of recovery NBR 10 years pre-fire average 4 X Stem Density, Composition 16-21 99 field plots with estimates of percent cover, and variable width transects for tree seedlings Spectral clustering of "fast-recovery" NBR slopes > 0.02 had more shrubs,Spectral clustering of "slow-recovery" NBR slopes < 0.02 had more grass ~ Models field based recovery with environmental drivers + field plot data Pure Structural Investigates the spectral patterns of recovery and compares to different spectral responses in relation to structural differences. Spectral patterns partition into two different responses at year four, the "fast" spectral recovery generally has a higher proportion of shrub and a lower proportion of grasses. However, the fast spectral recovery is also higher in areas that are not in the at the trailing edge of forests. In this case, authors posit the faster spectral responses are associated with lodgepole pine serotinous regeneration. Also consider how commonalities in field plot measurements relate to spectral recovery. Supports that plots that are shrub or tree dominant have similar rates of spectral recovery, whereas grass is slower.
Slope of Recovery, Magnitude of recovery, Data clustering (unsupervised classification) 13 clustering of field base metrics showed that shurb and tree dominated had similar spectral recovery rates post fire, but that grass was slower
NDVI 2 Findings similar to NBR, but fewers years to identify trend and generally less seperability among calsses
Years to a certain value (i.e. Y2R), Slope of Recovery, Magnitude of recovery, Data clustering (unsupervised classification) 11
129 Chen 2011 South Dakota Black Hills Predominantly Ponderosa Pine + Grassland USGS EROS Landsat 4,5 and 7 Top of Atmosphere Reflectance Calculated for indices CBI Slope of Recovery NDVI 0.3 years pre-fire 3 Field visits and post-fire imager Cover 3 - 7 Describes trends of recovery in grasslands vs dense and open conifer forests. NDVI recovery achieved for most sitees 3 years post-fire Measuring the spectral recovery of NDVI, EVI, NBR, and IFI (inhanced forest index) across a time-series of post fire images. Also determines which metrics are best aligned to field based measures of burn severity. Finds that NBR is the most sensitive for both fire severity mapping, but mostly of the canopy. Also supports the NBR is the best at differentiating between coniferous and grassland areas, but does not help distinguish between different grasses.
USGS EROS Landsat 4,5 and 8 EVI EVI was generally less variable and sensity to changes in vegetation conditions
USGS EROS Landsat 4,5 and 9 NBR NBR was the most variable post-fire and dense and open forest had consistently lower NBR compared to grassland. NBR poorly differentiated differences in grass composition
USGS EROS Landsat 4,5 and 10 Integrated Forest Index (Huang et al., 2008) Consistently high IFI aligned with areas that converted to grass or sparse canopy forest
120 Crockett 2024 Arizona, New Mexico Ponderosa Pine, Mixed conifer, Sub-Alpine Google Earth Engine Landsat 5, 7 , and 8 Surface Reflectance collections Masking clouds (USGS Quality Band) Google Earth Engine implementation of LandTrendr for growing seson (June 20th to Sept. 20th) |Maximum |MTBS |Modeling (e.g. random forests) |NBR |Landtrendr requires 6 |5 | | |NULL | | |X |Pure Spectral |Forest State |0.743 |Uses random forests to model to post-fire NBR values (magnitude only) from a collection of seasonal variables as well as soil factors. Finds that the largest impacts on post-fire NBR values are the climate factors. While measures of slope erodibility are important, they are less important than climate. Additionally, support that the greatest differences among forest types and in modelling success are the year post-fire and year five post-fire. They do not include years past year 5, so their power to say year five is the "best" is limited.
916 Epstein 2017 Northwest Montana Mixed Conifer Google Earth Engine Landsat Tier 1 Images Landsat 8 transformed to use same reflectance as 5 and 7 Cloud, shadow, snow masking Median and maximum composite images May - Ottober |Median, Maximum |dNBR, MTBS, Lidar Imputed Canopy Cover Loss |Modeling of Lidar Imputed Canopy Cover for both slope of recovery and recovery time |NBR, NDVI, TCW, TCG, TCB, and 5 spectral reflectance bands |3 |5 |~ (Lidar imputed canopy cover) |Cover |5-15 |Computes annual estimates of canopy cover using spectral + landscape level variables |Median canopy cover recovery ~ 40 years, and the time period used to estimate this trend (5,10,15) did not alter findings. | |Pure Spectral | | |Creates lidar imputed canopy cover metrics and then uses these to calculate canopy recovery times. Finds that the overall canopy recovery times do not vary in the last 40 years, and that most areas take ~ 40 years to recover the canopy, but this depended on the time to their lowest canopy cover. Also finds that the time to recovery does not change based on the observation window (5-15 years of spectral measurements). Notes there are two patterns of recovery, one is trending towards recovery and one is not trending to canopy recovery in the next 200 years - and that these trends are binary. Finds that many sites continued to lose their canopy cover after the fire event. About 50% of sites lost more canopy cover in the years that preceded the fire compared to the fire itself. Notes that since 2013 there is an increased in the number of sites that were 'not recovering' which could be related to droughts
93 Franks 2013 Greater Yellowstone Ecosystem Mixed Conifer USGS EROS Landsat 5- 7 Top of Atmosphere Surface Reflectance derived from LEDAPS and resampling 30 to 60 m MTBS, Other - Field assessed tree canopy Slope of Recovery NDVI 1 5 X Lodgepole Leaf Area Index, Annual Net Productivity, Sapling Density 11-17 Compare rate of change in lodgepole LAI, ANPP, and stem density to rate of change of NDVI Polynomial regression with NDVI and LAI R² = 0.62, ANPP R² = 0.62, Collects NDVI and SWIR (band 5) to compare to density and derived LAI and above ground biomass 11-17 years post-fire. Conducts a first level regression comparison between lodgepole leaf area index (LAI), annual net primary productivity (ANPP), and sapling density. Correlations between NDVI and different metrics was highest for ANPP and varied by burn severity. Conducts a second level of analysis compraing rate of change of structural measures with rate of spectral change. Changes were correlated across metrics, but SWIR was slighlty better correlated.
SWIR (band 5) Polynomial regression with SWIR (band 5) and LAI R² = 0.62, ANPP R² = 0.70,
61 Guz 2022 Northern New Mexico Mixed Conifer Google Earth Engine Landsat 5, 7, 8 Tier 1 Collection Masking clouds (CFMask) Google Earth Engine implementation of LandTrendr for growing season (June to September) |Median |MTBS |Slope of Recovery |NBR |Landtrendr requires 6 |15 | | |NULL | | |NBR recovery ~ topography (high-severity fire) |Pure Spectral | |0.1812 |Creates three different models to explain the NBR recovery rates for high, medium and low severity fire for three different fires in Northern Colorado. Creates a model for climate and then a second model for topographic variables. Summer precipitation emerged as strong and consistent driver of NBR recovery. Accuracy of models vary among individual fire events
|NBR recovery ~ topography (all fires) |Pure Spectral (all fires) |0.49
|NBR recovery 2002 high severity fires ~ winter climate |Pure Spectral |0.87
|NBR recovery 2002 high severity fires ~ summer climate |0.65
547 Hermosilla 2024 Canada Cordilleran Landsat 4,5 , 7, and 9 surface reflectance collection Best Available Pixed Composites (BAP) for peak growing season (± 30 days August 1st) Maximum dNBR Supervised modeling (fitting a classifier or model to set categories) NBR, NDVI, TCA, TCW, TCG, TCB, EVI 2 years pre-disturbance NULL X Composition Variable National Forest Inventory dataset used to create tree species dataset Produce a annual tree composition map for Canadian ecozones including the mountain cordillera. Identifies key differences between harvested forests and fire forests. Support that fire forests recover slower than harvested forests. Also support that lodgepole pine, black-spruce, and balsam fir type forests are the most likely to shift in composition post fire. Comparatively, Engelmann spruce and subalpine fir are more likely to maintain coniferous species, but shift in the type of coniferous species. Uses spectral data to model species composition, but does not investigate trends in spectral data
10501 Ireland 2015 Southern BC Ponderosa Pine, Interior Douglas Fir Landsat TM surface reflectance Top of Atmosphere Reflectance Calculated and atmospheric correction dNBR Recovery Indicator NDVI 0.16 4 NULL Areas identified as conifer had a lower rate of spectral recovery compared to the grassy agricultural regions Assesses the spectral recovery using NDVI and six Landsat TM images. They computed the recovery indicator by comparing burned locations to unburned locations (which theoretically have similar spectral responses). They found that coniferous areas had a lower spectral recovery compared to grassy areas. Additionally, they also found that the highest correlation between pre-fire and post-fire NDVI was year 3 - and that it then decreased after this.
445 Jakubauskas 1996 Greater Yellowstone Ecosystem Mixed Conifer Landsat 5 TM DN values corrected to surface using Markam and Barker (1986) and haz using Chavez (1988) Data clustering (unsupervised classification) All Landsat Bands NULL NULL X Basal area, Composition, Height, Biomass, LAI, 0-500 69 sample sites in 10 distinct spectral classes Tree speces, age, and height to understand disturbance history |Lodgepole pine age classes and stand composition based on Landsat 5 bands Early succession (<20 years) is brightest in NIR Bare ground immediately following fire (LP0) is brightest in all other bands | | | | |Paper investigates the relationship between different stages of succession in lodgepole pine forests using a Landsat image. Includes data from 68 forest plots, representing 10 distinct 'spectral classes' and uses these to associated spectral responses with 6 different stages of succession. Early succession, or LP0, is related to the prior successional stages identified by Romme et al. in 1982 and is generally noted as the "herbaceous-stage". The overall ability of Landsat to discriminate amongst the different stand ages was 84% but there is not temporal analysis of efficacy. The LP0 was brightest overall, but had the lowest reflectance in the SWIR2 .
117 Kiel 2022 Greater Yellowstone Ecosystem Mixed Conifer Landsat 5-8 Top of Atmosphere Landsat 8 Harmonized to Landsat 5 Masking NDVI values < 0.01 and > 0.9 Maximum composites of spring (January - Jun 5th ) and fall (September 9th to December 21) |Maximum |RdNBR, MTBS |Magnitude of recovery |NDVI |1 year pre-fire |24 |X |Stem Density |24 |Stem density for 71 plots sampled in 24 years post-fire |NDVI threshold >0.4 for both spring + fall maximum values had fewer than 1000 saplings |X |Spectral Response Indicative of Conifer Growth |Stem Density |0.92 |Investigates the patterns of sparse and dense recovery using NDVI after the yellowstone fires. Supports that after 30 years areas that are recovered (greater than 1000 stems) have a higher NDVI threshold. Also supports that the most important variables in this modeling are distance and elevation
1 Menick 2024 Northwest Montana Douglas-Fir Google Earth Engine Landsat 4,5,7, 8 Surface Reflectance Collections Image masking using NDSI or NDFSI (normalized difference forested snow index) > 0.4, median composites (December - April) Median RdNBR Piecewise Linear regression NDVI (Snow cover) 1 year pre-fire 11.5 ~ (spectral metrics input into a binary classifier to assess accuracy, but regression independent validation) Seedling / Sapling Precense 29-34 Validated conifer presence from NAIP imagery + GEE 29-34 years post-fire for 4700 locations Winter NDVI values decreased until year 11.5. Estimated Y2R 36.9 0.98 Models NDVI recovery for different forest types in northwest Montana and calculates the recovery time based on a piecewise regression. Winter NDVI has a ~ 10 year wherein it decreases. Develops a classifier with NDVI, NBR, NDVI, EVI, NDSI, NDWI, NDFSI to identify areas with conifers. Random forest developed with data from current imagery ( 2018-2021) and found NDVI, NDWI, NBR2, NBR metrics were most important and successfully detected conifers 98% of the time.
Fir-Spruce 19.4 Winter NDVI values decreased until year 19.4. Estimated Y2R 48.7 Spectral Response Indicative of Conifer Growth Stem Density
Lodgepole pine 14.6 Winter NDVI values decreased until year 14.6. Estimated Y2R 36.9
196 Mitchell 2010 South Dakota Ponderosa Pine USGS EROS Landsat 5 Top of Atmosphere ATCOR3 module applied to convert TOA to surface reflectance dNBR, dNDVI Slope of Recovery NDVI 1 year pre-fire 5 X Composition 5 17 Field samples to identify predominant cover types areas with greater change in NDVI were mostly associated with grass recovery. Heavily burned areas (which had high rates of NDVI recovery) did not have any seedlings X Pure Spectral 0.171 Compares change in NDVI after wildfire as a funciton of burn severity, slope, aspect, elevation and slope erodability. All of landscape factors explained varability in change in NDVI, but had poor overall accuracy. Geographically weighted regression results were more accurate, suggesting the relationship between spectral recovery and landscape drivers is highly location specific Paper discusses the rate of recovery and accuracy of fire severity mapping in the South Dakota region. This area is dominated by Ponderosa Pine, and the post-fire compositions generally shifted to grasslands. In general, there was a mix between forb and grass cover.
1041 Mumma 2024 Northern BC Lodgepole pine dNBR Slope of Recovery NBR 1 pre-fire value NULL NULL Spectral Response Indicative of Conifer Growth Investigates the pattern of moose in addition to the regrowth rates in interior BC.
57 Shrestha 2024 Western US Mixed Conifer USGS Archive MODIS LAI Product, USGS Archive Collection-6 Broadband white-skiy Albedo Snow Cover Masking based on MODIS10A1 Snow Cover Dataset MTBS Magnitude of recovery MODIS Summer LAI 3 years pre-fire 10 - 20 NULL X Pure Spectral 0.73 (10 -year MODIS LAI) Investigates that spectral patterns of post-fire recovery among seven main forest types in the western US. Uses MODIS data to measure the average rate of LAI recovery and Albedo Recovery at year 10 and at year 20. Models the relationship between trends of spectral recovery and climate factors. Spectral recovery for summer LAI was most impacted by average temperature, annunal precipition, and fire severity class. Forest types showed divergent responses to elevation and temperature. pinyon pin increased with average temperature and elevation. Models for albedo were more sensitive to unique forest types.
Lodgepole pine
Douglas Fir
Ponderosa Pine
Pinyon Juniper
Oak
Spruce/Fir/Hemlock
140 Smith-Tripp 2024 Central British Columbia SBS, SBPS Landsat 4,5 , 7, and 9 surface reflectance collection Best Available Pixed Composites (BAP) for peak growing season (± 30 days August 1st) Maximum dNBR Data clustering (unsupervised classification) NBR, NDVI, NDMI TCA, TCW, TCG, TCB 10 years pre-fire 13 X Stem Density, Basal area, Composition 14 Lidar imputation of structural metrics from 26 field plots Most important spectral metric to differentiate among classes TCA slope, Magnitude of NBR growth utilize a clustering on 7 different spectral indices and their trajectories (NBR, NDVI, NDMI, TCA, TCB, TCG, TCW) to differentiate spectral trajectories. then use field data and LiDAR data to relate different clusters with measures of forest structure: Basal Area, Stem Density, and Coniferous to deciduous ratio.
88 Smith-Tripp (b) 2024 Interior BC SBS, SBPS Landsat 4,5 , 7, and 9 surface reflectance collection Best Available Pixed Composites (BAP) for peak growing season (± 30 days August 1st) Maximum dNBR Data clustering (unsupervised classification) NBR, NDVI, NDMI TCA, TCW, TCG, TCB 3 years pre-fire 5 or less X Stem Density, Basal area, Composition 5-16 Lidar imputation of structural metrics from 26 field plots Most important spectral metric to differentiate among classes NBR regrowth and NBR slope Not applicable first, use unsupervised learning to spectrally cluster (using various measures of trajectory of NBR, NDVI, NDMI, TCA, TCB, TCW, and TCG) an enormous area covering the SBS and SBPS ecozones of interior BC using only data from up to 5 years post-fire. Then, draw conclusions about structural patterns within those unsupervised clusters up to 16 years post-fire using a space-for-time substitution of structural attributes obtained from LiDAR flights.
999 Vanderhoof 2018 US Rocky Mountains Mixed conifer Google Earth Engine Landsat 5 & 7TM 32-DAY NDVI Composite Collection Mean composite of summer (June, July, August Masked all NDVI values < 0.1 or > 0.9. Mean composite of January, Feburary, March. Winter: masked pixels where MODIS 10CM snow-cover likelihood < 85% Mean MTBS Absolute change NDVI (Growing Season), NDVI (Snow cover) 2-4 years 20 ~ Quickbird and Worldview imagery to map vegetation class and density Composition / Cover 20 Defined cover types for high-density, moderate density, low-density conifer, mixed forest, decduous dominated forest, grassland. Conifer density was interpreted from imagery Compare spectral differences between snow and growing season. Identify four patterns that are consistent in trend: (1) Winter + Growing season less green than pre-fire: pre-fire high-density conifer -> post-fire minimal coniferous regowth (2) Winter + Growing season more green than pre-fire: pre-fire sparse conifer, herbaceous -> post-fire possible increase in coniferous or shrub (3) More green in growing season: pre-fire moderate density conifer -> post-fire possible decidous OR herbaceous vegeation (4) More green in winter season: Pre fire mixed forest -> strong coniferous recovery but sparse decidous recovery | | | | |Classify forest recovery 20 years post-fire into distinct coniferous, deciduous and herbaceous classes using snow and growing season NDVI values. Compare the absolute change in NDVI pre-fire and post-fire to identify four broad categories. At 20 year time-point differences in seasonal patterns of NDVI were accurate (43%) indentifying areas of high to sparse coniferous density. Patterns are identified as a comparison between winter and summer NDVI values, not temporal trends.
154 Vanderhoof 2018 Southern Colorado Mixed Conifer Worldview 2 imagery Aerial imagery, SAVI, ground-validation, soil burn severity from Burned Area Rapid Response Supervised modeling (fitting a classifier or model to set categories) mean red edge band, the NIR2 band, standard deviation of green band, red edge band, NIR1 band, NIR2 band, SAVI, pixel ratios coastal band/all bands, green band/all bands, red edge band/all bands, shape, brightness with a near-infrared focus (red edge þ NIR1 þ NIR2), GRVI index 1 year (Image Classification) 1 (defined thresholds more effectively than four year time-series) ~ training objects from Wordview 2-m imagery Composition / Cover 1735948800 Defined cover types for conifer, riparian, bare soil, low and high density shrub Object identified recovered areas (trained using classifier) had higher accuracies than pixel-level estimates. Establishes NDVI and SAVI thresholds as well as object identification structures to understand different patterns of post-fire forest recovery. Explores the accuracy using thresholds for pixels and supervised classification of recovery objects. Supervised classification had higher overall accuracies. Pixel based approaches were highly variable in accuracy particularly for differentiating between aspen, dense shrub, and coniferous areas. Additionally, thresholds for 'spectral' recovery established 1-year post-fire were more accurate than pixel thresholds defined in each year.
Spectral Cutoff SAVI Image classification 4 1 - 4 Pixel SAVI recovery threshold for coniferous = 0.56 ± 0.14 Spectral Response Indicative of Conifer Growth Notes that there is generally a rapid increase in areas classified as aspen/herbaceous, and shrub. Minimal recovery, where there is not bare ground and the area that is conifer is highest the year. Aspen peaked in year 3.
NDVI Pixel NDVI recovery thresholds > 0.3 Not reported Establishes NDVI and SAVI thresholds as well as object identification structures to understand different patterns of post-fire forest recovery. Notes that there is generally a rapid increase in areas classified as aspen/herbaceous, and shrub. Minimal recovery, where there is not bare ground and the area that is conifer is highest the year. Aspen peaked in year 3.
138 Vanderhoof 2021 Western US Coniferous (Snow Cover) Google Earth Engine Landsat 4,5,7, 8 Surface Reflectance Collections Cloud Masking (CFMASK), transformation of Landsat 8 bands to correspond with Landsat 7, mean compositing of ''snow cover" remove NDVI <-0.05 AND NDFSI >0.4 (suggesting snow free) Mean dNBR Years to a certain value (i.e. Y2R), Slope of Recovery, NDVI Threshold NDVI (Snow cover) 2-4 years pre-fire (exclude year before fire) 10 X Stem Density 10 230 field plots with measures of sapling density and percent cover NDVI slopes > 0.00497 had > 200 saplings / ha X Spectral Response Indicative of Conifer Growth Stem Density Oregon data set (SE=.0017); California data set (SE=.0070); Colorado data set (SE=.0037) Investigates the differences in NDVI slope of recovery as related to field site trends. Finds NDVI recovery rates varied by (a) season (b) forest type and (c) field-plot differences. NDVI Growing season slope did not distinguish between areas with with or without saplings, NDVI snow cover slope did distinguish between areas with stems and areas without stems. NDVI growing season successfully seperated deciduous from non-deciduous areas. The correlations between NDVI snow cover and NDVI growing season were low Found that GS recovery rates were mostly impacted by RdNBR (so an estimate of burn severity), followed by maximum summer temperature, where recovery was slower with higher temperatures. Comparatively, NDVI Snow cover slope was most impacted by the maximum aridity in the five years post-fire, where areas that had higher VPD had faster recovery values. This was followed by normal summer temperatures, areas that were warmer had faster recovery.
Coniferous Forest (Growing Season) Cloud Masking (CFMASK), transformation of Landsat 8 bands to correspond with Landsat 7, medioid compositing of ''snow cover" remove NDVI <0.05 and > 0.5 NDVI (Growing Season) 5 Composition 5 231 field plots with measures of sapling density and percent cover NDVI slopes > 0.00541 associated with > 50% grass/foor/shrubs Composition Oregon data set (SE=.0048); California data set (SE=.019)
104 Walker 2019 New Mexico, Arizona Ponderosa Pine Google Earth Engine Landsat 4,5,7, 8 Surface Reflectance Collections Cloud, shadow, and snow masking Savitzy-Golay filter to derive NDVI temporal values at 8 and 16 day timesteps | |dNBR, MTBS |Phenological Metrics (change amplitude, base value, peak value, timing of peak value) |NDVI |NULL |5 - 8 |~ Google Earth High Resolution Reference imagery |Composition |Variable |Visually identify aeas of decidous, forest, grass, and shrub after fire |Base NDVI values strongly differentiated coniferous forest compared to ther forest types. Amplitude of seasonal change was highest in deciduous forests. Shrubs had an greater peak NDVI compared to grass, but lower than other forest types. The difference between 8 and 16 day compositing did not impact findinsg | | | | |Investigatees variability between four different post-fire trajectories of NDVI. Compares changes in amplitude, peak NDVI, day of year for peak NDVI values, and the baseline NDVI values. Finds that amplitude for NDVI change is greatest in deciduous and shrub classes, and lowest in these grass classes. Peak NDVI is greatest, but also the most variable for coniferous forest classes. Differences in trends of recovery begin around year 8
10240 White 2017 Canada Cordilleran Landsat 4,5 , 7, and 9 surface reflectance collection Best Available Pixed Composites (BAP) for peak growing season (± 30 days August 1st) Maximum dNBR Years to recovery Magnitude of Recovery NBR 2 year pre-fire values 5 NULL Investigates the trends of spectral recovery across all of Canada using three different metrics spectral metrics (years to recovery and the absolute and normalized change 5 years post-fire). These different attributes are related to different classes of spectral recovery - years to recovery is the long-term recovery rates, where as the change metrics.
6669 White 2022 Canada Cordilleran Landsat 4,5 , 7, and 9 surface reflectance collection Best Available Pixed Composites (BAP) for peak growing season (± 30 days August 1st) Maximum dNBR Years to recovery, Magnitude of Recovery NBR 2 year pre-fire values 5 X Cover, Height Variable Lidar imputation of canopy cover and height validated with various plot level data Areas identified as conifer had slower rates of spectral recovery compared to mixed and broadleaf stands. Also, 97% of wildfire pixels achieved 80% of the pre-fire NBR value, but this value of the recovery decreased when the number of years to spectrally recovery was low (20% less had structurally recovered when years to recovery was < 5). The accuracy of these metrics varied by region - the mountain cordillera most commonly achieved both height and cover when spectrally recovery (and took 8 years to do so). The second most common was cover only recovery. Updates the spectral recovery assessment of White et al. 2017 and compares achievement of spectral recovery to successful recovery as forest (as per FAO standards). Spectral recovery generally aligned with FAO standards of forests. Confirms that the spectral recovery rate is 8 years for the montane cordillera, and that this rate is both less variable than other areas. Also finds the cordillera has clustered regions of slow spectral recovery.
144 Wilson 2021 Columbia River Basin Douglas Fir Google Earth Engine MODIS 16 Day Composite Imagery Maximum EVI values June 1 - October 1 Maximum RdNBR Slope of Recovery EVI 1 year pre-fire values 4 NULL X Pure Spectral 0.42 Assesses relationship between post-fire greenup (via EVI) and drivers of recovery by regions within the Columbia River Basin. Summer precipitation was the most important variable for spectral recovery. Snow dynamics (timing and date of snowmelt) were more important in the Cascades compared to farther east. Notes that absolute climate values (peak snow for any given post-fire year) best explained variabiltiy in that year of spectral recovery, but differences in trends were more sensistive to anomalies of cliamte conditions.
Sub Alpine Fir / Engelman Spruce Maximum EVI values June 1 - October 2 0.69
Oregon Cascades Grand fir Maximum EVI values June 1 - October 3 0.7
Mountain hemlock Maximum EVI values June 1 - October 4 0.57
Central Idaho Douglas fir Maximum EVI values June 1 - October 5 0.43
Englemann spruce/Subalpine fir Maximum EVI values June 1 - October 6 0.39
Montana Rockies Douglas fir Maximum EVI values June 1 - October 7 0.33
Englemann spruce/Subalpine fir Maximum EVI values June 1 - October 8 0.39
Columbia River Basin Douglas Fir Maximum EVI values June 1 - October 9 0.42