Vegetation Trend Analysis with Remote Sensing of a Lacustrine Environment in the New Jersey Pine Barrens
Final Project Report by Thomas A. Woolman

ES771 - Remote Sensing, Emporia State University, Fall 2014 Semester


·ES771 Course Syllabus

Web page last updated:
3 DEC 2014

Remote sensing involves obtaining information about a location without directly physically observing that site on the ground. Generally, the techniques utilized in remote sensing make use of aerial-based sensor technologies (primarily satellites) to receive electromagnetic emissions and then convert those signals into useful data for classification and identification purposes. Remote sensing may be further categorized as either active remote sensing, such as when signals (generally either laser light or microwave radar signals) are emitted from aircraft or satellites (Jenson, J.R., 2007) or passive reception of selected frequencies of naturally transmitted electromagnetic energy (generally frequencies from sunlight) in which this data is merely recorded.

Remote sensing systems allow data to be collected over vast areas, typically hundreds of square kilometers, both quickly and homogenously (from the same time period), without regard to human safety concerns. Access issues, either due to political concerns or logistical demands are also not a factor with remote sensing, provided the satellite or aerial observation platforms provide coverage for that location. The usefulness of these systems for environmentally sensitive areas can be significant because those areas can remain undisturbed by people and trails, encampments and other environmental impacts can be avoided. Furthermore, data can be obtained almost immediately compared to traditional ground-based geomorphology survey techniques.

Common applications of remote sensing involve the use of orbital systems to enhance natural resource management, national security and military operations investigations, agricultural and land use planning and development inquires. Applications continue to grow as the number and capabilities of remote sensing satellites and aerial systems increases and data processing capabilities expand (NASA, 2014).

Study Area

The area chosen for this investigation was a portion of the Pine Barrens of New Jersey known as Success Lake. Success Lake is a reservoir located just 5.2 miles from Lakehurst, in Ocean County, in the state of New Jersey, United States, near Woodair Estates, NJ. It is located within the New Jersey Wildlife and Game Refuge. The area is a portion of a larger region of the Atlantic coastal plain known locally as the New Jersey Pine Barrens.

Location map for the Atlantic Coastal Pine Barrens ecoregion. The underlying land cover is from the 1992
National Land Cover Database (Vogelmann and others, 2001). The 20 km X 20 km sample blocks are shown in black.
Map compliments of USGS Land Cover Trends Project - Atlantic Coastal Pine Barrens

The New Jersey Pine Barrens region is a heavily forested coastal plain encompassing more than seven counties of southern and central New Jersey. The term "pine barrens" refer to the acidic, nutrient-poor sandy soil which is unfavorable to agricultural cultivation. However, the unique ecology of the Pine Barrens supports a diverse spectrum of native plant life such as dwarf pines, oak forests and rare orchids and carnivorous plants. The area is also known for the rare pygmy Pitch Pines as well as other dwarf coniferous trees and shrubs. Naturally occurring forest fires are a primary component of the reproduction cycle of many of these species. The fine, acidic sand that composes much of the area's topsoil is porous, poor in nutrients and retains little moisture. "Due to generally acidic soils, agricultural activity is often limited to acid-loving crops such as blueberries and cranberries, although small areas of the ecoregion support fruits, vegetables, and other crops." (USGS, 2014).

View north from a fire tower on Apple Pie Hill in Wharton State Forest, the highest point in the New Jersey Pine Barrens at approximately 250 feet elevation
Image compliments of wikipedia

Despite its proximity to major east coast population centers of the United States, including Philadelphia and New York City, and the fact that the state's largest highway systems, the Garden State Parkway and Atlantic City Expressway run through it, the Pine Barrens remains largely rural and undisturbed.

In 1978 Congress passed legislation to designate 1.1 million acres (4,500 km²), or one-fifth of the state of New Jersey, designating this area as the Pinelands National Reserve (the nation's first National Reserve) to preserve its unique ecological diversity (Pearce, J., 2002)

The scene selected in the New Jersey Pine Barrens was Success Lake, an area near the east-central portion of the Pine Barrens nearing the Atlantic coast. Two datasets for comparative analysis were used, the first was LC80140322014219LGN00 , from 7 AUG 2014 using Landsat 8 data and the second was LT50140322011243GNC01 , from 31 AUG 2011 using Landsat 5 data. Both scenes were from the approximate same time period of the season and both gave extraordinarily good nearly cloud-free images on those days, lending themselves to a stronger quantitative scrutiny.

New Jersey Pinelands Management Area, compliments of the State of New Jersey Pinelands Commission. Black star in the north-center image represents the approximate location of Success Lake.


The objective of this study was to ascertain the level of change in the area of Success Lake across a span of several years, during the same seasonal period. Late summer (August) was chosen as the constant time period, and a span of 3 years between 2011 and 2014 was chosen. The variance investigation between the two datasets primarily focused on ground moisture and vegetation biomass change.

I began this overall process by querying the US Geological Survey EarthExplorer site to locate appropriate datasets from various remote sensing satellites for the proposed study area described previously. After selecting and downloading the resulting Landsat 5 and Landsat 8 datasets, I converted .TAR file collections to .tif files, and then converted each of the two sets of .tif band files into Idrisi .RST raster images files. Bands 3 and 4 for both the Landsat 5 and Landsat 8 image sets were then enhanced with the SCALAR function to perform a haze correction, adding a whole number value of 1 to each of these 4 images to provide a minimum value of 1 in each image (the previous minimum value was 0.0 in each image).

I then created a "natural color" composite image based on Landsat 8 OLI bands 2-3-4 to get an overview of this complex central New Jersey scene. I then used Google Earth to locate the exact coordinates of Success Lake, NJ and its nearby major landforms. I used those major landforms to locate Success Lake, NJ in the natural color 1-2-3 band Landsat 8 composite image. Lastly, upon zooming in to an appropriate level of detail around Success Lake in this Landsat 8 composite image, I saved the composite raster image as a new .RST file, keeping only the zoomed in level of detail selected.

Landsat 8 OLI bands 2-3-4 "natural color" composite of Success Lake, Ocean County, New Jersey, U.S.A. from August 7, 2014

I then used the Idrisi WINDOW function to zoom into the same level of detail and coordinates selected in my last Landsat 8 2-3-4 band composite image, so all from both Landsat 5 and Landsat 8 would have the same scene coverage area so we could begin comparative NDVI review, comparing August 2014 to August 2011.

All new "windowed" scenes from each of the two datasets from Landsat 5 and 8 were now successfully created. The VEGINDEX function in Idrisi was then used to begin a comparative quantitative analysis between the two image scenes. The haze corrected band images from the previously used SCALAR function were of course used in the NDVI investigation. Band 3 was used for "red" and Band 4 was used for "infrared" in this NDVI analysis for the Landsat 5 TM image datasets. For the Landsat 8 OLI dataset, band 4 was used for red and band 5 was used for infrared. These bands were selected based on near-equivalent wavelengths (USGS 2014) .

Landsat 5, August 2011 NDVI Analysis

Next, the resulting NDVI output files from the VEGINDEX function were rescaled and converted to byte-binary format. SCALAR function was used to add 1.0 to each of the two NDVI output images, then multiply them by 125, and then converted using the CONVERT function to byte-binary format with rounding. The final resulting two NDVI processed images were then displayed with the NDVI palette.

The results were conclusive in that they showed a measurable increase in vegetation for the 2011 (Landsat 5) scene compared to the lesser vegetation shown in the 2014 (Landsat 8) scene. A follow-up cluster analysis was next conducted to further quantify the variance between the two image sets.

Landsat 8, August 2014 NDVI Analysis

According to the IDRISI help system, "CLUSTER provides an unsupervised classification of input images using a histogram peak technique." Six bands were selected for both datasets. To perform the cluster analysis, all of the windowed and haze-corrected bands for both image sets was utilized. TM imagery was selected for the Landsat 5 images and OLI was selected for the Landsat 8 images. Broad generalization was selected, with dropping the least 1.0 percent clustering rule option.

Landsat 5 2011 Dataset CLUSTER Analysis

The resulting cluster investigation again supported the NDVI review and the two sets of composite form results. The cluster results for the Landsat 8 August 2014 were 16 clusters but the vast majority of the surface area was assigned to cluster 1 (shown in red), indicating a homogenous area of vegetation that correlated to the low NDVI values. The Landsat 5 August 2011 cluster analysis was a much more dynamic area made up of only 13 clusters. Those clusters were varied in their location assignments and the most complex swaths of territory covered by the cluster analysis correlated to the NDVI values showing high vegetation readings.

Landsat 8 2014 Dataset CLUSTER Analysis

The use of NDVI, bands 345 (TM) / 456 (OLI) composite imagery and blue-NDVI-red composite imagery for both datasets all support the analytical conclusion that August 2011 was significantly more moist and vegetation-dense than August 2014.

Landsat 5 TM bands 3-4-5 composite image from the 2011 dataset
Landsat 8 OLI bands 4-5-6 composite image from the 2014 dataset

Landsat 5 band 2-NDVI-5 composite
Landsat 8 band 2-NDVI-4 composite

Measurements of the maximum horizontal width of Success Lake were taken from the NDVI images (to maximize contrast) for both datasets. For the 2011 Landsat 5 NDVI scene, the maximum width of Success Lake (measured horizontally, east to west) was 911 meters. In the 2014 Landsat 8 NDVI scene, the maximum width was 844 meters. This represented a maximum width variability of 7.4% year over year in the August period.

Geostatistical Analysis

To begin the process of creating a geostatistical trend analysis, the two CLUSTER raster files were subjected to a GROUP function. Idrisi Selva defines the GROUP function as determining the "contiguous groupings of identically valued integer cells in an image. Cells belonging to the same contiguous grouping are given a unique identifier, numbered consecutively in the order found" (Idrisi online help, 2014). The results of the GROUP function on both cluster datasets is shown below:

Landsat 5 2011 dataset GROUP image showing contiguously related data points
Landsat 8 2014 dataset GROUP image showing contiguously related data points

The results of this GROUP function dramatically show the level of variance between the two timeframes in terms of relative vegetation biomass and ground moisture. Quantitative values supplied by the GROUP function further indicate a significantly higher moisture/biomass reading in the southeastern quadrant of the scene in 2011 compared to 2014. The reservoir and its immediate borders in 2011 are also much stronger relative to 2014, as is the area north and northeast of the reservoir. The 2014 was weaker overall with the majority of cells shown fall into the green (weak) zone outside the perimeter of the reservoir. Fewer red (high) cells are shown relative to 2011, and the red pixels are quantitatively smaller values than 2011.

The final process utilized was a TREND function in IDRISI, using the previous raster datasets resulting from the above GROUP function. IDRISI defines the TREND function as calculating "1 to 9 order trend surface polynomial equations for spatial data sets, and interpolates surfaces based on those equations. In addition to the trend surface image, TREND also reports some information about the polynomial equation calculated and the goodness of fit. The number of cells used to determine the surface is reported, followed by the surface coefficients of the polynomial fitted. Coefficient "b0" is the intercept and coefficients "b1" and greater are the slope coefficients associated with the order of equation chosen. The goodness of fit is expressed as a percent, and is accompanied by the F ratio and associated degrees of freedom required to test whether the goodness of fit is significantly greater than zero."

This geostatistical modeling tool provided interesting results. For the Landsat 5 (2011) group dataset, nine orders of surface were fitted and the number of cells for surface determination was 12995. The Goodness of Fit as measured by the coefficient R² was a high value of 71.05%, along with an F ratio of 588.08. For the Landsat 8 (2014) group dataset, nine orders of surface were also fitted which provided the highest R² value for all attempts, and the number of cells for surface determination was also 12995. The Goodness of Fit R² value was a moderately high value of 55.97%, which supports the statistical validity of the model as well. The Landsat 8 TREND model provided an F ratio of 304.60. The result of the two trending datasets is shown below.

Landsat 5 TREND function showing a strong biomass and soil moisture progression moving from the reservoir at center to the southeast
Landat 8 TREND function showing a much weaker biomass and soil moisture progression in both the reservoir and to the southeast quadrant of the image. Furthermore, a nearly "dead zone" is apparent in the Landsat 8 northeast and north central areas.


Ground truth evidence was provided by USGS monitoring stations for hydrological conditions for both time periods in question, which served to corroborate the remote sensing results. The various USGS monitoring stations are shown in the map below:

U.S. Geological Survey Hydrological Monitoring Stations in New Jersey. Note that the Morrell 1 ground water survey station
is the closest monitor to Success Lake (denoted as a red star). Map compliments of the U.S. Geological Survey, 2014 (modification by the author)

USGS New Jersey Hydrological Conditions, August 2014 and August 2011:

8/2014: Summary of August 2014 Monthly Hydrologic Conditions (USGS, 2011) stated that "Precipitation was below normal at the Newark and Trenton index stations, and above normal at the Atlantic City index station for the reference period 1981-2010. " Note: Success Lake is closest to the Trenton index station and furthest from the Atlantix City index station.

Groundwater levels, as measured in water-table observation wells for the month of August, were below normal at the Readington School 11 and Morrell 1 index wells." Note: Morrell 1 groundwater index well is the closest USGS monitoring station to Success Lake (see map above).

8/2011: Summary of August 2011 Monthly Hydrologic Conditions stated that "August 2011 was the wettest month on record in New Jersey. On average there was 16.64 inches statewide which is 4.66 inches more then the second wettest month on record in October 2005. This is 12.43 inches above the 1981-2010 monthly average of 4.21 inches. Despite below average precipitation in June and July, the precipitation in August made the summer of 2011 the wettest season on record in New Jersey. There was 23.90 inches of precipitation, which is 4.23 inches above the previous season record of 19.67 inches in 1928."

"Groundwater levels, as measured in water-table observation wells for the month of August, were above normal at all three index wells. Levels increased from last month, and were higher than one year ago. Readington School 11 index well was 10.12 ft higher than last month and 12.88 ft higher than last year. Morrell 1 index well was 3.44 ft higher than last month and 4.96 ft higher than last year. Vocational School 2 index well was 2.87 ft higher than last month and 2.21 ft higher than last year."

My analysis of the remote sensing data coupled with the hydrological groundwater and precipitation monitoring stations is that while drought conditions were moderate in August 2014, the tremendously strong August 2011 NDVI results were actually the result of record precipitation levels throughout the state during that period. It provided a much more vegetation-lush, "false positive" baseline image from which to compare the August 2014 NDVI image. The above exercise appears to indicate that the various remote sensing techniques employed all validated the ground truth hydrological condition monitoring and confirmed the dynamic variability of ground moisture and vegetative biomass within the same seasonality.


(1) Jensen, J. R., 2007, Remote sensing of the environment: an Earth resource perspective (2nd ed.). Prentice Hall. ISBN 0-13-188950-8.

(2) National Aeronautics and Space Administration, 2014, Earth Science: NASA's Mission to Our Home Planet Accessed 1 December, 2014

(3) United States Geological Survey, 2014, Frequently Asked Questions about the Landsat Missions Accessed 2 December, 2014

(4) United States Geological Survey, 2014, New Jersey Hydrological Conditions, August 2014, Accessed 30 November, 2014

(5) United States Geological Suvey, 2011, New Jersey Hydrological Conditions, August 2011, Accessed 30 November, 2014

(6) Pearce, J., 2002, "Trouble in Paradise", The New York Times Accessed 29 November, 2014

(7) Clark Labs, 2014, IDRISI Selva online help IDRISI Selva online helpAccessed 4 December 2014

(8) United States Geological Survey, 2014, Land Cover Trends Accessed 4 December, 2014

This webpage is copyrighted by Thomas A. Woolman.
All original content is © copyrighted by Thomas A. Woolman, with permission granted to Emporia State University for academic use with attribution to the author.