In this study, several vegetation indices were examined in order to determine the most sensitive vegetation index for monitoring southern Appalachian wetlands. Three levels of platforms ( in situ, airborne, and satellite) for sensors were also examined in conjunction with vegetation indices. Net primary production (NPP) data were gathered to use as a measure of wetland function. Along with the in situ radiometers, National Agricultural Imagery Program (NAIP) data and Landsat 8 Operational Land Imager (OLI) data were gathered in order to calculate vegetation indices at three platforms. At the in situ level, VARI700 was the most sensitive vegetation index in terms of NPP (r 2 = 0.65, p < 0.05). At the airborne level, the NDVI was the most sensitive vegetation index to NPP (r 2 = 0.35, p = 0.11). At the satellite level, the DVI appeared to have a positive relationship with NPP. For most indices there was a drop in the coefficient of determination with NPP when the platform altitude increased, with the exception of NDVI when increasing altitude from in situ to airborne. This study provides a novel methodology comparing reflectance and vegetation indices at three platform levels.
The use of remote sensing in vegetation-related studies is a rising trend as it allows for the estimation of vegetation characteristics (chlorophyll content and net primary production, NPP) over large areas [
In all of the aforementioned in-situ remote sensing studies, two inter-calibrated hyperspectral radiometers were used. This methodology was employed because the use of two radiometers allows for one to capture incoming radiation while the other captures the ground reflectance simultaneously [
Remote sensing allows for the estimation of vegetation characteristics over large areas in a short amount of time, though error is introduced because of atmospheric interference (e.g. Rayleigh and Mie scattering). Differences in platform (airborne vs. satellite) might determine the amount of the aforementioned error due to the amount of atmosphere with which light must interact. Thus, using data from an airborne platform for comparison to proximally sensed data might serve as an intermediate step between comparing proximally sensed and satellite data.
The National Agricultural Imagery Program (NAIP) generates plane platform data that served the purpose of this comparison. NAIP was selected among the varieties of remotely sensed data for a few reasons. The spatial resolution of NAIP data is 1 m of ground distance with ground control points within 6 m [
While Landsat 8 OLI has limitations associated with spatial resolution (30 m) for visible and NIR bands, imagery from Landsat 8 is available at 16 day intervals [
Attempts to estimate biomass from remotely sensed data have been numerous [
where ρNIR is the reflectance of NIR, ρRed is the red reflectance, and α = 0.96916 [
where a serves as an estimation of the vegetation fraction (VF) and can range from 0.05 to 0.2. Other vegetation indices that have been found to have linear relationships with aboveground vegetation characteristics are the modified simple ratio (MSR) [
and the Renormalized Difference Vegetation Index (RDVI) [
The Soil-Adjusted Vegetation Index (SAVI), which was developed by [
where L is a correction factor based on LAI [
Two indices were developed by [
and
where Rgreen is green reflectance, Rred is red reflectance, Rblue is blue reflectance, and R700 is the reflectance at 700 nm (red edge).
All vegetation indices have benefits and drawbacks (
NPP is an important indicator of ecosystem function. Gross primary productivity (GPP) is the amount of carbon (C) converted from inorganic forms in the atmosphere to part of the plant body through the process of photosynthesis. NPP is the GPP minus the amount of C lost to the environment through cellular respiration in the root system and herbivory. NPP is important because it serves as a marker of C storage as well as an indication of ecosystem function, where more productive systems will result in higher NPP values. Thus, NPP is important as a measure of both C sequestration and ecological function. The importance of this study is, in part, tied to the importance of NPP.
The significance of this research lies in the study of NPP in montane wetlands, an ecosystem that is little studied, as well as the methods used in this study. As already discussed, NPP is an important factor to study and the development of a model that can estimate NPP in southern Appalachian montane wetlands using vegetation indices would provide a useful tool in management of montane wetlands with less disturbance. In order to create this model, we employed a methodology using proximally sensed data to help improve the accuracy of remotely sensed data for the model. The model developed in this study could be employed in other settings to also improve management, while limiting disturbance.
Vegetation Index | Bands Needed | Ancillary Data Needed? | Can Be Used with Multispectral Data? |
---|---|---|---|
NDVI | Red, NIR | No | Yes |
DVI | Red, NIR | No | Yes |
WDRVI | Red, NIR | Yes | Yes |
RDVI | Red, NIR | No | Yes |
SAVI | Red, NIR | Yes | Yes |
MSR | Red, NIR | No | Yes |
VARI700 | Red Edge (700 nm), NIR | No | No |
VARIGreen | Blue, Green, Red, NIR | No | Yes |
The objective of this study was to create a model using vegetation indices generated from remotely sensed data in order to estimate NPP to best serve in the management of southern Appalachian wetlands. Among the myriad of vegetation indices in existence, NDVI, DVI, RDVI, WDRVI, SAVI, MSR, and VARI will be examined. The model is designed to inform wetland managers to which type of data acquired and used (platform, spatial resolution, spectral resolution, and temporal resolution) to the appropriate needs of and resources allocated to the wetland managers. Many vegetation indices were tested to provide the best estimate for NPP throughout southern Appalachian wetlands.
Three wetland sites in the southern Appalachians, one in Alabama and two in Tennessee (
Leaf area index (LAI) data were collected along with peak standing aboveground biomass, which was assumed to be NPP, for each quadrat [
were oven-dried for 12 hours at 50˚C [
Generating a spectral signature involved collecting top of canopy reflectance using OceanOptics JAZ hyperspectral radiometers (OceanOptic Inc., Dunedin, FL, USA) following [
After field collection of radiometer data, a reflectance value was generated for each proximally sensed data sample. Radiance and irradiance data were downloaded from the radiometers and a reflectance value was generated by dividing the radiance by the irradiance. The reflectance value generated was a measure of the percentage of irradiance that was radiated back by the vegetation (radiance) for each band, unless vegetation indices dictated otherwise (e.g. VARIGreen). Reflectance values generated from the radiometers were averaged over the bandwidths of NAIP radiometers. For example, the red band for NAIP captured by an ADS radiometer is 607 - 662 nm, thus the values for the reflectance of radiometers were averaged over the same range in order to generate a red reflectance for the radiometers for comparison to NAIP data.
NAIP data were acquired by downloading data from the USDA Geospatial Data Gateway (http://datagateway.nrcs.usda.gov/). The year of the field work was used for the data collection. If the year of the field work was not available, the closest year was used. Local governments proximal to each site were contacted to inquire about data collected closer to field work dates. GPS coordinates were used to find the pixel associated with the NPP sample collection point and reflectance values for red, green, and NIR were recorded for vegetation index calculations.
NAIP imagery was calibrated using ERDAS Imagine (Hexagon Geospatial, Norcross, GA). NAIP images were calibrated using the empirical line calibration (ELC). Images were calibrated by band by using a spectral library (ASTER). Ground cover types were selected (e.g. grass, asphalt) that occurred near the site and five points were taken of each ground cover within the image. The spectral signatures for the ground points were run against the spectral library signatures for that band. A new image was generated from the ELC. Brightness values were taken from that image for vegetation indices.
Landsat 8 OLI images were downloaded from the USGS Earth Explorer website (http://earthexplorer.usgs.gov/). Imagery was downloaded for the date closest to the date of field work before field work was conducted. 20% cloud cover was the maximum cloud cover; if an image exceeded 20% cloud cover for the collection date, the image from the previous collection period was used. GPS coordinates were used to locate the pixel for each quadrat. Brightness values were recorded for blue, green, red, and NIR for each pixel in order to calculate vegetation indices.
After reflectance values were generated from radiometer and brightness values were recorded from NAIP data, vegetation indices were calculated for all NPP sites using combinations of green, red, and NIR reflectance data, depending on the vegetation index equation. NDVI was employed because it is the most common vegetation index and generally accepted as the best vegetation index for estimating biomass. The MSR (Equation (3)), RDVI (Equation (4)), DVI (Equation (1)), and WDRVI (Equation (2)) were also used because success has been found in estimating biomass from these indices [
The objectives of this study were to determine which vegetation index best identifies variance in NPP and, thus, best informs wetland managers on wetland health. Linear regressions of the vegetation indices calculated for each sample point for in situ and NAIP data and compared to NPP will determine which of the vegetation indices was best able to explain variance in NPP. Because Landsat has coarser spatial resolution than in situ and NAIP imagery, NPP values that fell within one Landsat pixel were averaged [
There was a large amount of variability in NPP among sites in this chapter as well as a smaller amount of variability in NPP within each site (
Vegetation indices calculated from in situ radiometer data were able to explain a moderate amount of variance in NPP in southern Appalachian wetlands (
able to explain the most amount of variance in NPP (r2 = 0.65, p < 0.05), though all of the index values were negative and there appeared to be an outlier. After removing the outlier, it appears as though VARI700 is not as sensitive to NPP (
Most of the vegetation indices calculated from NAIP data were unable to explain more than 10% of variance in NPP in southern Appalachian wetlands (
The amount of variance in NPP by vegetation indices from Landsat 8 OLI were the lowest among all vegetation indices tested in this study (
In examining NPP by vegetation indices at three platform levels, a trend was noted for most of the vegetation indices along a change in height of the platform (
There was a disparity in the ability of vegetation indices to explain variance in NPP at all levels, though no ve-
getation index was best able to explain variance consistently at all levels. The difference in the ability of vegetation indices to explain variance in NPP is due in part to differences in the indices themselves, but also is related to the type of vegetation and ecosystems in which the indices were employed. No vegetation indices have been developed to specifically study montane wetland NPP. Another cause for the lack of a vegetation index that was able to explain NPP variance at all three levels is that the data is different for all three levels. At the in situ level, hyperspectral radiometers were employed and thousands of bands were used at a high spatial resolution. At the airborne level, NAIP data are limited in the context of spectral resolution, though spatial resolution is still high. At the satellite level, spatial resolution is low, though spectral resolution is higher than NAIP.
At the in situ level, the VARI700 was best able to explain variance in NPP and the SAVI was the second most sensitive vegetation index with respect to NPP. The VARI700 was only applicable to the in situ radiometers because the VARI700 requires hyperspectral data. The VARI700 is a useful vegetation index that was developed by [
examining reeds [
The NDVI, which is one of the main vegetation indices used for examining or monitoring vegetation, was the most sensitive vegetation index for NAIP data. While the NDVI may be criticized, it is still very useful in monitoring vegetation. The NDVI was one of the original vegetation indices developed by [
The DVI was the most sensitive vegetation index when using Landsat 8 OLI data. Similarly, [
There were some limitations associated with this research. More specifically, there was a lack of sample size to examine the associations between field NPP estimates and the estimates gathered from vegetation indices. There were limitations associated with the congruence of dates between the field work and the collection of remotely sensed data. Also, there might be some differences in the season in which the biomass was collected and when the imagery was collected. For imagery with lower spatial resolution (i.e. Landsat 8 OLI), there may exist a difference in the NPP from field samples and the ecosystem NPP. There might also exists differences between scales concerning NPP and imagery.
This study provided valuable tools for wetland managers in the southern Appalachians and contributes to the scientific community with intellectual merit. By conducting an examination on vegetation indices and NPP, the people in charge of southern Appalachian wetlands can now make better informed decisions on how to monitor wetlands with remote sensing data. The results of this study also contribute to the scientific community by further testing vegetation indices in an environment where few, if any, similar studies have been conducted.
The choice in vegetation index depends on the platform that is being used to examine southern Appalachian wetlands. At the in situ level, the VARI700 (r2 = 0.65) was most sensitive to changes in NPP, with SAVI (r2 = 0.39) and VARIGreen (r2 = 0.33) also showing sensitivity toward NPP. At the airborne level, most vegetation indices saw a drop in coefficient of determination with NPP. The exception was the NDVI, which was more sensitive to NPP at the airborne level (r2 = 0.35) than the in situ level (r2 = 0.28). All other vegetation indices were insensitive to changes in NPP. At the satellite platform, the DVI (r2 = 0.37) was the most sensitive vegetation index.
Within the broader context of vegetation index usage across ecosystems and in different areas of the world, it would appear that a variety of indices should be tested and performance may depend on the platform level. For future studies concerning the use of vegetation indices, it is important to start with the NDVI as that is still the main index and was effective at the airborne level with high spatial resolution. It is also important to consider the ecosystem in which the study takes place. As few studies have been focused on remote sensing use in wetlands, it is important to expand efforts to study wetlands with remote sensing so as not to disturb these sensitive ecosystems, yet to understand them because of their sensitivity, especially as it relates to climate change.
First and foremost, we would like to thank Chanda Maguigan for all of her hard work in this study. We would like to thank the Department of Geosciences for partial funding through an assistantship and use of the department truck. We would also like to thank the permitting agencies including The Nature Conservancy of Shady Valley, TN and Lake Guntersville State Park. This material is based upon work supported by the National Science Foundation under Grant No. DGE-0947419 at Mississippi State University. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author and do not necessarily reflect the views of the National Science Foundation.
Michael Maguigan,John Rodgers,Padmanava Dash,Qingmin Meng, (2016) Assessing Net Primary Production in Montane Wetlands from Proximal, Airborne, and Satellite Remote Sensing. Advances in Remote Sensing,05,118-130. doi: 10.4236/ars.2016.52010