Philip E. Dennison

Endmember selection for multiple endmember spectral mixture analysis using endmember average RMSE

Abstract Multiple endmember spectral mixture analysis (MESMA) models mixed spectra as a linear combination of endmembers that are allowed to vary in number and type on a per pixel basis. For modeling an image using MESMA, a parsimonious set of endmembers is desirable for computational efficiency and operational simplicity. This paper presents a method of selecting endmembers from a spectral library for use in MESMA. Endmember average root mean square error (EAR) uses MESMA to determine the average error of an endmember modeling spectra within its land cover class. The minimum EAR endmember is the most representative endmember for a land cover class within the spectral library and can be used to model the larger image. These techniques were used to map land cover, including four dominant vegetation species, soil, and senesced grass, in the Santa Ynez Mountains above Santa Barbara, CA, USA. Image spectra were extracted from a 20-m resolution airborne visible infrared imaging spectrometer (AVIRIS) reflectance image using reference polygons and combined into a library of 915 spectra. Possible confusion between land cover classes was determined using the class average RMSE (CAR). EAR was used to select the single most representative endmember within each land cover class. The six minimum EAR endmembers were used to map the AVIRIS image. Land cover class accuracy was assessed at 88.6%. Using a fractional accuracy assessment, undermodeling of dominant land cover classes and overmodeling of absent land cover classes was found at the pixel scale. Land cover mapped using the minimum EAR endmembers represents a substantial improvement in accuracy over previous efforts.

Evaluation of the potential of Hyperion for fire danger assessment by comparison to the Airborne Visible/Infrared Imaging Spectrometer

Parameters derived from remote sensing that can be used to assess fire danger include surface reflectance, live and dead biomass, canopy water content, species composition, and fuel state. Spectral bands and wavelength locations of traditional multispectral data make assessment of fire danger in Mediterranean shrublands difficult, although fire danger parameters have been derived from Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) data. We compare nearly simultaneous acquisition of Hyperion and AVIRIS to evaluate spaceborne monitoring potential of fire danger in Southern California chaparral. Field spectra were acquired to support reflectance retrieval and construct a spectral library for vegetation mapping. Reflectance spectra retrieved from Hyperion and AVIRIS had similar shape and albedo, but SNR was five times higher in AVIRIS. Fuel condition was assessed using the endmember fractions from spectral mixture analysis, with both Hyperion and AVIRIS imaging spectrometer data providing similar fractions and spatial distributions. Hyperion demonstrated good capability for separating spectral signals from bare soil and dry plant litter. Canopy water content was compared using the 980- and 1200-nm liquid water bands, the water index, and the normalized difference water index. Results showed that Hyperion is capable of retrieving canopy water at 1200 nm, but demonstrates poor performance at 980 nm. Sensor noise and instrumental artifacts account for poor performance in this spectral region. Overall, full-spectrum measures outperformed band ratios because of a lower sensitivity to sensor noise in individual bands. Species and community mapping showed similar patterns with better accuracy for AVIRIS relative to Hyperion, but with both instruments achieving only 79% and 50% overall accuracy, respectively.

Assessment of vegetation regeneration after fire through multitemporal analysis of AVIRIS images in the Santa Monica Mountains

Abstract Spectral mixture analysis (SMA) from Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) was used to understand regeneration patterns after fire in two semiarid shrub communities of the Santa Monica Mountains, California: northern mixed chaparral and coastal sage scrub. Two fires were analyzed: the Malibu Topanga fire (3 November 1993) and the Calabasas fire (21 October 1996). SMA was compared to the results of the Normalized Difference Vegetation Index (NDVI) to assess vegetation recovery. An unburned control plot (within the past 20 years), having similar environmental features, was used to generate two relative fire regeneration indices, Regeneration Index (RI) and Normalized Regeneration Index (NRI). Indices were calculated using the Green Vegetation (GV) endmember and the NDVI. These indices were determined to be largely independent of AVIRIS radiometric calibration uncertainty, minor errors in the atmospheric correction, topographic distortions, and differences in the phenological state of the vegetation because of interannual or seasonal differences. The temporal evolution of the two fires were combined to produce a longer observation period and used to fit a logarithmic regression model for each Mediterranean shrub community. The NRI developed from the GV endmember (NRIGV) produced the closest estimate for the time of recovery in both communities based on recovery times in the literature. The use of NDVI worked very well for recovery in the northern mixed chaparral, but was less successful in the coastal sage scrub, mainly because of extensive herbaceous cover during the first years of the regeneration process. Endmembers generated from hyperspectral images were more accurate because they are tuned to capture the greenness of the shrub type of vegetation. Use of matching plots having similar environmental features, but which were burned in different years were demonstrated to improve estimates of the recovery within each community.

Use of Normalized Difference Water Index for monitoring live fuel moisture

Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI) were compared for monitoring live fuel moisture in a shrubland ecosystem. Both indices were calculated from 500 m spatial resolution Moderate Resolution Imaging Spectroradiometer (MODIS) reflectance data covering a 33‐month period from 2000 to 2002. Both NDVI and NDWI were positively correlated with live fuel moisture measured by the Los Angeles County Fire Department (LACFD). NDVI had R 2 values ranging between 0.25 to 0.60, while NDWI had significantly higher R 2 values, varying between 0.39 and 0.80. Water absorption measures, such as NDWI, may prove more appropriate for monitoring live fuel moisture than measures of chlorophyll absorption such as NDVI.

Setting Wildfire Evacuation Trigger Points Using Fire Spread Modeling and GIS

Warning communities in the path of an advancing wildfire is a challenging problem. Decision makers need the most current information available to determine who should evacuate, when they should leave and what type of order to issue (e.g. mandatory, recommended, voluntary). This paper presents a new method for delimiting wildfire evacuation trigger points using fire spread modeling and GIS. Using data on wind, topography, and fuel in conjunction with estimated evacuation time, a trigger buffer can be computed for a community whereby an evacuation is recommended if a fire crosses the edge of the buffer. A case study is presented for the Corral Canyon section of the 1996 Calabasas Fire near Malibu, California, USA. The paper concludes with a discussion of the strengths and weaknesses of this approach.

Evaluating predictive models of critical live fuel moisture in the Santa Monica Mountains, California

Large wildfires in the Santa Monica Mountains of southern California occur when low levels of live and dead fuel moisture coincide with Santa Ana wind events. Declining live fuel moisture may reach a threshold that increases susceptibility to large wildfires. Live fuel moisture and fire history data for the Santa Monica Mountains from 1984 to 2005 were used to determine a potential critical live fuel moisture threshold, below which large fires become much more likely. The ability of live fuel moisture, remote sensing, and precipitation variables to predict the annual timing of 71 and 77% live fuel moisture thresholds was assessed. Spring precipitation, measured through the months of March, April, and May, was found to be strongly correlated with the annual timing of both live fuel moisture thresholds. Large fires in the Santa Monica Mountains only occurred after the 77% threshold was surpassed, although most large fires occurred after the less conservative 71% threshold. Spring precipitation has fluctuated widely over the past 70 years but does not show evidence of long-term trends. Predictive models of live fuel moisture threshold timing may improve planning for large fires in chaparral ecosystems.

Adapt to more wildfire in western North American forests as climate changes.

Wildfires across western North America have increased in number and size over the past three decades, and this trend will continue in response to further warming. As a consequence, the wildland–urban interface is projected to experience substantially higher risk of climate-driven fires in the coming decades. Although many plants, animals, and ecosystem services benefit from fire, it is unknown how ecosystems will respond to increased burning and warming. Policy and management have focused primarily on specified resilience approaches aimed at resistance to wildfire and restoration of areas burned by wildfire through fire suppression and fuels management. These strategies are inadequate to address a new era of western wildfires. In contrast, policies that promote adaptive resilience to wildfire, by which people and ecosystems adjust and reorganize in response to changing fire regimes to reduce future vulnerability, are needed. Key aspects of an adaptive resilience approach are (i) recognizing that fuels reduction cannot alter regional wildfire trends; (ii) targeting fuels reduction to increase adaptation by some ecosystems and residential communities to more frequent fire; (iii) actively managing more wild and prescribed fires with a range of severities; and (iv) incentivizing and planning residential development to withstand inevitable wildfire. These strategies represent a shift in policy and management from restoring ecosystems based on historical baselines to adapting to changing fire regimes and from unsustainable defense of the wildland–urban interface to developing fire-adapted communities. We propose an approach that accepts wildfire as an inevitable catalyst of change and that promotes adaptive responses by ecosystems and residential communities to more warming and wildfire.

Critical live fuel moisture in chaparral ecosystems: a threshold for fire activity and its relationship to antecedent precipitation

Large wildfires in southern California typically occur during periods of reduced live fuel moisture (LFM) and high winds. Previous work has found evidence that a LFM threshold may determine when large fires can occur. Using a LFM time series and a fire history for Los Angeles County, California, we found strong evidence for a LFM threshold near 79%. Monthly and 3-month total precipitation data were used to show that the timing of this threshold during the fire season is strongly correlated with antecedent rainfall. Spring precipitation, particularly in the month of March, was found to be the primary driver of the timing of LFM decline, although regression tree analysis revealed that high winter precipitation may delay the timing of the threshold in some years. This work further establishes relationships between precipitation and fire potential that may prove important for anticipating shifts in fire regimes under climate-change scenarios.

Mapping Plant Functional Types at Multiple Spatial Resolutions Using Imaging Spectrometer Data

Imaging spectrometer data have been used to map plant functional types (PFTs—plant species grouped by similarities in their resource use, ecosystem function, and responses to environmental conditions) at spatial resolutions of 30 m and finer, but not at coarser spatial resolutions that may be necessary for global PFT mapping. This study uses spatially resampled Airborne Visible InfraRed Imaging Spectrometer (AVIRIS) data acquired over the Wasatch Mountains of northern Utah, USA to examine changes in PFT classification accuracy as spatial resolution is degraded from 20 to 60 m. Accuracy was dependent on the spatial resolution of the classified data and the spatial resolution of endmembers used in the multiple endmember spectral mixture analysis classifier.

Modelling long-term fire regimes of southern California shrublands

This paper explores the environmental factors that drive the southern California chaparral fire regime. Specifically, we examined the response of three fire regime metrics (fire size distributions, fire return interval maps, cumulative total area burned) to variations in the number of ignitions, the spatial pattern of ignitions, the number of Santa Ana wind events, and live fuel moisture, using the HFire fire spread model. HFire is computationally efficient and capable of simulating the spatiotemporal progression of individual fires on a landscape and aggregating results for fully resolved individual fires over hundreds or thousands of years to predict long-term fire regimes. A quantitative understanding of the long-term drivers of a fire regime is of use in fire management and policy.