Wetland Monitoring and Assessment Using Hyperspectral Remote Sensing

Participants: MERI, Ball State University and Rutgers University-Newark


The current wetland monitoring techniques used in the Meadowlands are labor intensive, time consuming and subjective, hence there is a need for developing cost effective methodologies for accurately determining the health and extent of the remaining open areas, specifically the spatial distribution of Phragmites and its associated mixtures types at the landscape level.


To design and implement a sustainable wetland monitoring and assessment program based on high resolution remote sensing that can be used in the Meadowlands and adopted by other agencies responsible for coastal wetland integrity and reporting.


Use both systems (Hyperspectral and LIDAR) to build a model that will link image values from vegetation types and texture and height to range values of sulfide concentration, salinity and redox-potential in the field. After the model is built and tested, only hyperspectral images will be required for the long term monitoring and assessment.


Image processing

· Hyperspectral imagery was taken in June, 2007. Spectral resolution: 5 nm, spatial resolution: 2.5m

· LIDAR imagery was taken in May, 2007. Spatial resolution: vertical: 0.6m; horizontal: 0.6m

· As ground reference, aerial photography was taken using a tethered balloon platform. Spatial resolution: 0.1 m

· Image Classification: 2004 and 2007 imagery were classified using rectified balloon imagery as training areas.

· Image data: vegetation indices (Normalized Difference Vegetation Index (NDVI); Sum Green Index; Photochemical Reflectance Index (PRI); Structure Insensitive Pigment Index (SIPI); Plant Senescence Reflectance Index (PSRI)) were calculated from pixel data in order to link plant productivity to sediment chemistry.

Field campaign

· Four sites were considered representative for the District (see image). At each site three transects were established with three sampling points along each. Lysimeters were installed at each sampling point.

· Coordinates were recorded at each sampling points to spatially relate results of field measurements to hyperspectral image information.

· During the field campaign there were three seasonal samplings during June, August and October in 2007. Biometric measurements were carried out in January, 2008.

Lab work

· Sample Analysis – Analysis concerning sediment chemistry and water quality were completed after each sampling session at the MERI laboratory.

· Data Analysis – descriptive statistical tests were calculated after all the field samples were processed. Regression models were tested to link field results to image data.


Sediment chemistry vs. plant community

· Oxidation reduction potential was always lower in high marsh sediments

· Sulfide was always higher in high marsh sediments

· Salinity was usually higher under high marsh

· Stunted Phragmites occupied slightly lower elevations than high marsh and vigorous Phragmites

Image data vs. sediment chemistry

· Vegetation indices derived from October images explained 84% of the variation in sediment sulfide concentration

· Vegetation indices explain 86% of the variation in salinity for June images

· October images explain 78% of the variation in ORP

Vegetation classification of hyperspectral imagery

· The overall accuracy in determining vegetation cover using hyperspectral and LIDAR was 68%

· The most reliable classification classes were Phragmites (93%) and low marsh (92%)

· Stunted Phragmites and mixtures of Spartina and Phragmites were ~50% accurate

· Vegetation indices explain 85% of the variation of plant height (October image)


Remote Sensing Summary Report.doc – 7.73 MB (8,107,520 bytes)

Final Presentation.ppt – 33.6 MB (35,271,680 bytes)

Final Data.xls – 48.0 KB (49,152 bytes)