Broadband, red-edge information from satellites improves early stress detection in a New Mexico conifer woodland
Highlights
► Multiple plant stresses can affect the health of forests. ► The utility of red-edge satellite data for detecting tree stress was examined. ► A red-edge employing index detected stress 13 days after it was induced. ► Traditional red and green employing indices detected stress 12 to 16 days later. ► Red-edge satellite data may improve forest health monitoring from space.
Introduction
Stressors affecting forests are highly dynamic in space and time and therefore can affect the health, esthetic condition, and ecosystem services provided by forests in multiple ways. Due to tight coupling among temperature, precipitation, and various types of plant stressors, climate change might intensify the effects of stress on forests (Logan et al., 2003, Niinemets, 2010). For instance, predicted warming surface temperatures and droughts in North American forests are expected to result in more intensified insect outbreaks (Logan et al., 2003). This development is of great socio-economic importance given that forest insects and disease are already the main agents of natural disturbance in North American forests. Remarkably, relative to fire, insects and pathogens impact 45 times more forest area in the US and have a fivefold greater economic impact (Logan et al., 2003). Timely, accurate, and cost-effective information is therefore needed via remote sensing to monitor spatio-temporal dynamics of forest stress conditions, test ecological hypotheses related to forest stress, and guide the allocation of resources for stress mitigation and control (e.g., sanitation logging) (Pontius et al., 2005, Wulder et al., 2005). Further, maps of remotely sensed stress and mortality at multiple scales could provide insight into the dynamics of stress and mortality patterns caused by different agents of stress in a changing climate (Hatala et al., 2010).
Plants experience stress if suboptimal growth conditions cause their plant physiological functions (e.g., light and dark reactions of photosynthesis) to decline from their physiological standard (Niinemets, 2010). To be of value for forest health monitoring, stress detection should be early to allow timely intervention and minimize the spread of stress agents. Numerous studies have explored the utility of remote sensing for early stress detection. Spectral vegetation indices (SVIs) have been investigated that employ short-wave infrared wavelengths sensitive to plant water content (e.g. Ceccato et al., 2001, Eitel et al., 2006, Stimson et al., 2005, Toomey and Vierling, 2005). The use of SVIs using visible bands such as the Normalized Difference Vegetation Index (NDVI; Tucker, 1979) can also be useful because a wide variety of different stress agents such as temperature, light, disease, ozone, or plant nutrition induce the loss of chlorophyll a + b (Chlab) which strongly affects absorption of photosynthetically active radiation (Carter, 1993, Carter and Knapp, 2001, Hendry et al., 1987). As a result, increasing visible reflectance has shown to be one of the most universal responses of leaf spectral reflectance to stress (Carter, 1993, Carter and Knapp, 2001). Within the visible spectrum, reflectance bands centered within the red (670 nm) and the green (550 nm) spectral regions have generally been used to remotely detect stress from satellite platforms. However, Chlab strongly absorbs in the red spectral region, leading to red reflectance saturation at low Chlab levels and making the red band often unresponsive to an initial loss in Chlab at earlier stress stages (Carter and Knapp, 2001, Jacquemoud and Baret, 1990). This explains the finding by Carter (1993) who showed a sensitivity minimum in the red (centered at 670 nm) to eight different stress agents (competition, herbicide, pathogen, ozone, insufficient mycorrhizae, barrier island environment, senescence, and dehydration). In contrast to red reflectance, green (centered at 550 nm) and red-edge (centered at 700 nm) reflectances have been found to be sensitive to a wide range of Chlab levels. Between the green and the red-edge reflectances there is evidence that the wavelength in the red-edge region is superior to the green in regards to its responsiveness to stress induced changes in Chlab (Carter, 1993, Carter, 1998, Carter and Knapp, 2001, Carter and Miller, 1994, Eitel et al., 2007, Eitel et al., 2008, Eitel et al., 2009, Eitel et al., 2010). This might be partly due to the red-edge band picking up some stress induced increase in fluorescence (Carter and Miller, 1994, Lichtenthaler and Rinderle, 1988).
Findings by Carter and Miller (1994) indicated that the ratio of red-edge (690–700) to NIR (760 nm) reflectance could improve early stress detection in soybean. This ratio corresponded more closely to plant physiological measures of stress (fluorescence and plant water status) than wavebands and waveband combinations in the visible-NIR region of the electromagnetic spectrum. Carter and Knapp (2001) studied the effect of a wide variety of stressors (dehydration, flooding, freezing, ozone, herbicides, competition, diseases, insects, N fertilization) on the spectral response between 400 and 850 nm of different plant species including conifers and deciduous trees. They found that, within that range of wavelengths, an increase in reflectance at 700 nm was the most universal and most sensitive spectral response of plants to stress. In Balsam fir (Abies balsamea (L.) Mill), Luther and Carroll (1999) showed that the red-edge reflectance at 711 nm was most sensitive to stress induced by root pruning, light, and nutrient availability. Eitel et al. (2010) showed for Scots pine (Pinus sylvestris) that red-edge reflectance information improved active ground optical remote sensing estimates of stress induced changes in Chlab (r2 > 0.73) over those based on red wavelength (590–670 nm) information (r2 = 0.57). Narrow band reflectance imagery acquired from a ground-based platform that provided red-edge reflectance information (695 ± 5 nm) allowed detecting herbicide-induced stress in loblolly pine (Pinus taeda L.) and slash pine (Pinus elliottii Engelm.) 16 days prior to visual signs of stress (Cater et al., 1996).
These findings suggest that the use of red-edge information could help to improve the early detection of plant stress. However, most of the aforementioned studies were not based on satellite data. An explanation for the latter is that until recently, only a limited number of hyperspectral satellite platforms provided radiance data in the red-edge portion of the spectrum, while multispectral satellites such as Landsat did not provide red-edge information. This recently changed with the launches of the RapidEye (Brandenburg, Germany) and DigitalGlobe WorldView-2 (Longmont, CO, USA) satellites now providing commercially available red-edge band information. As many of the studies employing red-edge information were based on narrow band (=<10 nm FWHM spectral sampling) ground spectral data, relatively little is known if broadband, red-edge satellite data (> 10 nm) can respond to and assist with tracking stress induced changes that have been reported for narrow band red-edge reflectance.
The objective of this study was to determine whether broadband, red-edge information from the RapidEye satellites improves early stress detection in conifer forests relative to information provided by combinations of other non red-edge spectral bands. We addressed this objective by examining the utility of red-edge and non red-edge indices for early stress detection. Spectra used for this analysis were simulated with the PROSPECT + SAIL radiative transfer model and acquired with the RapidEye satellite constellation. The use of a physically based canopy reflectance model allowed us to simulate satellite spectra for varying levels of Chlab while holding other variables (e.g., leaf area index (LAI), viewing and illumination geometry) constant that may complicate the interpretation of these data when actual satellite data are used. By evaluating the simulated satellite spectra combined with a temporally dense time series of satellite data for Chlab-related stress detection, we aim to shed light on improving satellite-based early warning systems and to guide future earth observing satellite designs for future monitoring of forest stress.
Section snippets
PROSPECT + SAIL simulations
The PROSPECT radiative transfer model (Jacquemoud & Baret, 1990) was used to simulate leaf reflectance and transmittance spectra between 400 and 900 nm at a spectral resolution of 1 nm. Model input parameters are leaf structural parameter N, leaf water content Cw, leaf dry matter content Cdm (g cm− 2), and leaf Chlab (μg cm− 2). Thirteen simulated leaf reflectance and transmittance spectra were obtained with typical evergreen needle values (Ollinger, 2011) for N = 1.0, Cw = 0.02 cm and Cdm = 0.02 g cm− 2, and
Results and discussion
The PROSPECT + SAIL model allowed us to simulate satellite spectra by varying Chlab while holding constant other variables such as LAI, leaf angle distribution, and viewing and illumination geometry. This modeling approach is a useful diagnostic technique that can be used to complement real satellite data, where it is more difficult to separate and thus interpret the effects of the variable of interest from the effects of other variables. The PROSPECT + SAIL simulation results thus helped in
Conclusion
A broadband SVI incorporating the red-edge Chlab detection band detected plant foliar stress as related to Chlab changes earlier than other broadband SVIs incorporating the green and red Chlab detection bands. This finding is mainly explained by the sensitivity of the red-edge band to stress induced changes in Chlab and possibly due to the sensitivity of the red-edge band to stressed induced increase in fluorescence. However, during later stages of stress, characterized by low Chlab levels (< 30
Acknowledgments
We greatly appreciate John McCallum for processing and analyzing leaf samples in the laboratory. We also thank RapidEye for making the dense timeseries of imagery available for this study. This work was supported by a cooperative agreement between the University of Idaho and USDA, Forest Service Pacific Northwest Research Station agreement 10JV11261900065, and by the University of Idaho Harold Heady Professorship. We thank two anonymous reviewers for their helpful comments on earlier versions
References (45)
- et al.
Biochemical responses of Pinus pinaster trees to fire-induced trunk girdling and crown scorch: secondary metabolites and pigments as needle chemical indicators
Journal of Chemical Ecology
(2002) - Barnes, E. M., Clarke, T. R., Richards, S. E., Colaizzi, P. D., Haberland, J., Kostrzewski, M., Waller, P., Choi, C.,...
- et al.
Forest health monitoring in the United States: a program overview
Australian Forestry
(2008) - et al.
Tree die-off in response to global change-type drought: mortality insights from a decade of plant water potential measurements
Frontiers in Ecology and the Environment
(2009) - et al.
Narrow-band reflectance imagery compared with thermal imagery for early detection of plant stress
Journal of Plant Physiology
(1996) Response of leaf spectral reflectance to plant stress
American Journal of Botany
(1993)Reflectance wavebands and indices for remote estimation of photosynthesis and stomatal conductance in pine canopies
Remote Sensing of Environment
(1998)- et al.
Leaf optical properties in higher plants: linking spectral characteristics to stress and chlorophyll concentration
American Journal of Botany
(2001) - et al.
Early detection of plant stress by digital imaging within narrow stress-sensitive wavebands
Remote Sensing of Environment
(1994) - et al.
Detecting vegetation leaf water content using reflectance in the optical domain
Remote Sensing of Environment
(2001)