Remote Sensing for Fire Management

Transcription

1 Remote Sensing for Fire Management FOR 435/535: Remote Sensing for Fire Management 5. Severity A Source of Confusion Field Measures Remote Sensing Measures FOR 435: A Source of Confusion The terms fire severity and burn severity are used inconsistently in the fire effects and remote sensing literature. Source: Jain T, Pilliod D, Graham R (2004) 1

2 FOR 435: A Source of Confusion Severity is by nature a value laden term, with negative perceptions often applied. Negative Connotations: severity = bad This can lead to public and policy miscommunications The problem is that although some fires may appear severe, they might not be ecologically bad for the ecosystem. FOR 435: A Source of Confusion There are lots of different definitions of severity, depending greatly on if the objective of a given study has been to assess vegetation or soil impacts of the fire. * Fire duration and heat transfer * Vegetation consumption *Whit White ash production * Change in surface reflectance * Alteration in soil properties * Changes in the litter and duff layers * Long-term vegetation mortality and recovery FOR 435: A Source of Confusion: Ecosystem Differences 2

3 FOR 435: A Source of Confusion Most studies use the term fire severity to describe immediate (minutes to days) post-fire effects, while burn severity is often reserved for longer duration impacts (weeks to years). FOR 435: A Source of Confusion Lentile et al (2006) proposed simplifying the terms: Limit use of the Terms Fire Severity & Burn Severity Describe and Quantify the Actual Processes Being Assessed Make sure that satellites CAN also measure these processes Other recent studies (e.g., Keeley, 2009) have argued keeping all the terms as they are what managers know. Pre-Fire Condition Active Fire Characteristics Post Fire Effects During Combustion Following Combustion FOR 435: A Source of Confusion: What we need Need immediately active or post-fire measures that relate to active fire characteristics (i.e. intensity) and predict both short and long term post-fire effects (i.e. severity). We have added challenges: 1. We need field measures that are easy both to understand and to implement. 2. We need remote sensing measures that directly relate to those easy to implement field measures of severity. 3. We need robust measures that work across scales and environments. 4. If possible we need measures that do not require pre-fire conditions data as most wildfires are assessed in an opportunistic manner. 5. If possible we need measures that can predict long term effects from pre-fire, active, or immediately postfire data. 3

4 Land managers are tasked with assessing the impact wildfires and Rx fires have on multiple resources including: vegetation, soils, water, recreation, habitats, etc. Researchers also often seek to understand basic ecological effects of fires on soils, erosion, invasive species and vegetation mortality. Many field based methods were developed to assess severity. The most common have included: Ryan and Noste (1985) USFS % Brown, % Black, % Green CBI The Ryan and Noste method described fire severity in terms of the heat received by both the overstorey and the soil. Fire severity is defined by combining the flame length classes and ground char classes. Preceded the widespread use of satellite imagery for mapping burn severity Does not provide an explicit sampling methodology, only a framework Widely cited, and still applied in modified versions in the field today 5 5-U Flame Length class L 3-M 2-D Unburn Light Moderate Deep Char Depth Class 4

5 The CBI field measure of severity was developed to calibrate the dnbr severity spectral index. Measures scaled m radius plots Used 5 Strata: Soil Understorey Shrubs / saplings Sub-canopy trees Overstorey The CBI is now a part of FIREMON, a national protocol developed by the US forest service for measuring and monitoring vegetation and forests. CBI Measures: Soil Measures (0-3) - change in color - change in cover - post-fire erosion - fuel consumed (1,10,100 hr) Understory (0-3) - Amount of veg. Consumed - Changes in species composition - Colonizing species? CBI Measures: Overstorey (0-3) % Torch (black) % Scorch (brown) % Green (unburned) Height of bole char 5

6 The Good: Very fast and easy to understand method Does not require any knowledge of the pre-fire condition. It assumes you can guess the conditions based on what you see around you. The Bad: Ocular assessments User dependent Some measures are very hard to estimate The Ugly: These estimates are done after the fire (often in a new area unseen before the fire) Many of these field measures are not measurable by satellites (which is bad given CBI was developed as ground validation for the dnbr spectral index ) The Ugliest: Given there is often no pre-fire data, how do you know whether effects are caused by the fire; and if they are what magnitude of those effects are due to the fire? The NPS developed their own field severity protocol called the Fire Effects Monitoring (FEM) program. Within each plot, dead fuels are measured using browns transects. A series of 1 meter quadrats are used to estimate herbaceous cover by species. Shrubs, overstorey tree diameter, height, and health are measured. Post-fire everything is re-measured, providing detailed data on the consumption and plant mortality. Further re-measurement enables managers to assess how vegetation is changing at these sites over time. These plots require pre-fire data and so are only appropriate for use in Rx fires. 6

7 Direct Measure via Remote Sensing Comparable Measure via Field Methods Similar to traditional USFS %Green, %Brown, % Black Source: Hudak, Morgan, Hardy et al Inherently Scalable Use Existing 30m Immediate Post-fire Landsat Data Physically Related to Carbon and Water Processes Source: Lentile et al (2009) The aim of any method we use to infer the severity of a fire is to physically relate surface process to remotely derived measures Remote sensing approaches that have been used to infer severity of fires include: NDVI The NBR series of Spectral Indices The Burn Severity Index (BSI) Fractional cover via linear spectral unmxing Regression-based methods Disney s Radiative Transfer Method 7

8 The NDVI takes advantage of the differential absorption of red wavelength radiation and reflectance of near-infrared wavelength radiation exhibited by healthy green vegetation. NDVI = In the case of both green and senesced vegetation, replacement of the vegetation by charred surfaced results in a significant drop in NIR reflectance and slight drop in red reflectance. Using change detection of NDVI allows us to highlight a rapid fall from values near 1 (green/senesced vegetation) to values near 0 (soil/char). The Normalized Burn Ratio (NBR) takes advantage of how the NIR and longer infrared wavelength change following a fire. NBR = As with NDVI, in the case of both green and senesced vegetation, replacement of the vegetation by charred surfaced results in a significant drop in NIR reflectance. The difference is the case of the longer infrared (band 7). Charcoal and soil often has a higher reflectance than green vegetation. The result is values near 1 represent both loss of green vegetation + increase in soil/char cover. The Differenced Normalized Burn Ratio (dnbr) is a change detection method that calculates the difference between post- and pre-fire NBR values as a measure of severity. dnbr = NBR pre -NBR post Most people use a postfire image from 1 year post-fire. This mainly represents the canopy condition after tree mortality has occurred. 8

9 The Differenced Normalized Burn Ratio (dnbr) is known to exhibit some problems. Problems: The calibration method, CBI has its own laundry list of problems Non-linear asymptotic relationship with CBI these relationships vary with differences in spatial scale (sensor resolution on same fires!) and ecosystem type (forest, range). NBR was originally developed in 1990 to detect burned area NBR does not do well in some ecosystems - grasslands, shrub land, woodlands, or ecosystems with underground consumption. Non-constant variance (greater spread with higher severity levels) The Differenced Normalized Burn Ratio (dnbr) is known to exhibit some problems. The Relative Differenced Normalized Burn Ratio (RdNBR) was developed to try and overcome some of the problems of dnbr. Pre-fire overstorey condition will influence the resulting dnbr measure of severity Therefore, the same dnbr value is possible independent of actual post-fire effects The relative dnbr corrects for this effect by dividing the overall dnbr value by the initial pre-fire NBR. Thus the resulting post-fire severity value takes into account the pre-fire stand conditions. 9

10 The RdNBR corrects the variance problems that were seen with dnbr. Another standard and simple method is to use linear spectral unmixing to estimate fraction cover (and changes) in the cover. Pixel: Green (Leaf) + Brown (dry grass) + Black (char) We can break a pixel down into component fractions 10

11 Smith et al. RSE (2005) Smith et al. IJRS (2007) Landsat >50% ETM Charcoal (4:3:2) Jasper Fire, South Dakota started 24 th Aug 2000 ~33,800 ha Burned in 9 days Ponderosa Pine Forest Lentile et al (2005,2006b): 1-Yr post-fire Measures in 80 sites Landsat Image: 14 th Sept 2000 Compare both NBR and Char Fraction Cover to 1-yr Post-Fire Field Canopy and Sub-Canopy Measures Landsat 7:5:4 NBR Char Fraction 11

12 Canopy Variables: 1-Yr Post Fire Immediate Char Fraction NBR R 2 = R 2 = Percentage Live Tree Smith, Lentile et al. IJRS (2007) Canopy Variables: 1-Yr Post Fire Immediate Char Fraction NBR Lentile, Smith et al. (2009) RR 2 = =0.52 Total Crown Effects (Crown Scorch + Crown Consumption) Char Fraction R 2 =0.69 Bole Scorch (surrogate of flame length Intensity) 12

13 What s Potentially Happening with Bole Scorch? Char Fraction = 50% 75% 100% Prediction of Sub-Canopy Ecosystem Condition Char Fraction R 2 =0.5 5 Litter Organic Weight (g/m2) Sub-Canopy Variables Immediate Char Fraction NBR Lentile, Smith et al. (2009) R 2 = =0.47 Average Bark Thickness 13