Marilyn G. Hughes
Rutgers Cooperative Extension & Center for Remote Sensing
Phone (732-932-1582)
E-mail (mghughes@crssa.rutgers.edu)

Peter V. Oudemans
Rutgers Blueberry & Cranberry Research Center,Chatsworth, NJ
Phone (609)-726-1590 x 20
E-mail (oudemans@aesop.rutgers.edu)

Joan R. Davenport, Soil Sciences, Washington State University
Keri Ayres, Rutgers Blueberry & Cranberry Research Center
Teuvo M. Airola, Rutgers Center for Remote Sensing
Abbott Lee, Lee Brothers, Inc., Chatsworth, New Jersey


This study uses GPS/GIS/RS techniques to analyze cranberry (Vaccinium macrocarpon Ait.) crop health and yield. Extensive field sampling has been used in the past as a means of estimating potential bed yields. The major problem for predicting yield appears to be to high intra-bed spatial variability. For this study, color-IR photography from commercial cranberry beds (May 1996) was rectified to earth coordinates using GPS technology. An unsupervised multi-spectral classification and an NDVI were done to statistically group pixels in the image. Results indicate that a number of features within cranberry beds can be identified, including variations of vegetative cover, irrigation and drainage systems, and areas of beds damaged by insects and fungal disease (Phytophthora cinnamomi). In the future, remotely sensed imagery will be linked to ground based data to gain further insight into the spatial variation of factors affecting crop yield and health.


Cranberries are a low growing perennial crop indigenous to the sandy wetland soils found in the Pine Barrens region of New Jersey. The berries develop on uprights along a network of vegetative runners that grow as a ground cover over the bed (Eck, 1990) and are harvested from late September to early November. Currently, New Jersey is home to over 3,300 acres of cranberry beds valued between $6,000 and $30,000 per acre annually (Roper and Vorsa, 1997). Due to stringent wetland laws, expanding the cultivated acreage to meet an increasing demand for the crop is difficult. Therefore, cranberry growers are turning to new technologies such as precision agriculture to spatially map, predict, and ultimately increase current crop yield in existing cranberry beds.

Cranberry yield is influenced by the physical characteristics of the soil environment (i.e. soil profile, percent sand, percent organic matter) and management practices (i.e. irrigation, fertilizer, pesticide, and fungicide applications). These factors affect the fruit set, berry size, and number of flowers per upright (Eck, 1990, Baumann and Eaton, 1986). Currently, estimates of cranberry yield are made using a combination of two techniques: 1) determination of bed areas in production using aerial photography to define the boundaries, 2) through intensive field sampling during the growing season. In the future prediction of yield and crop health using remote sensing (RS) techniques will provide a non-intrusive means of acquiring this information from individual sites as well as on a regional scale.

The detection and monitoring of crop health and soil drainage properties have been successfully undertaken using RS techniques in field crops (see Frazier et al., 1997 for review). Remote sensors are made up of detectors that record specific wavelengths of the electromagnetic spectrum (ERDAS, 1997). All types of land cover absorb a portion of the electromagnetic spectrum giving it a "signature" for identification. For example, areas of healthy green vegetation reflect strongly in the near-IR band and absorb in the red band, and areas covered by water absorb strongly in the near-IR part of the spectrum. Indicators of plant stress can be developed using a variety of techniques based upon the spectral reflectance properties of the crop of interest and the radiometric information available from the remote sensing instrument in use. To date, multi-spectral information available from satellites by virtue of their spatial resolution (5-80 meters ground resolution), have little utility in monitoring cranberries that are grown in beds on the order of less than 10 acres. Recent advances in technology are leading to the development of new instruments that will allow access to a wide range of digital imagery from both aircraft and space borne platforms in the conversion of conventional imagery into digital format. The pending launch of a number of new commercial satellite remote sensing devices will provide significantly better spatial resolution data and a more frequent data acquisition in the future. In particular, the launch of a pointable instrument having spatial resolutions of 1m panchromatic and 4m multi-spectral is planned for this year by the SpaceImaging Corp. (Thornton, CO). At these higher resolutions satellite imagery will become a viable way to monitor cranberry production at regional and global scales.


For this study, we used the available 1:12000 color-IR aerial photography obtained and archived each May from Ocean Spray, Inc. (Lakeville-Middleboro, MA). The CIR photography is currently used to catalogue production acreage. Data were collected in three bands, the green (.5-.6), red (.6-.7), and near-IR (.7-2.0). The overall goal of this study was to determine the feasibility of using remotely sensed data in conjunction with ground-truthed field data to map spatial variations in cranberry yield. To that end the following approach was taken. 1) image processing techniques were used on the color-IR photography to map bed features and variability, 2) spectral information was used to derive statistical relationships to predict bed yield, and 3) the information was used to develop a protocol for future studies incorporating both traditional field sampling and remote sensing techniques for mapping, understanding and predicting cranberry yield.


Color infrared (CIR) photography from May 20, 1996 providing coverage of the Lee Brother's cranberry farm was scanned in using a high-resolution color scanner at 600 dots per inch (dpi). This resulted in a ground resolution of approximately 2 feet and provides information about reflectance in the green, red, and near infrared portions of the electromagnetic spectrum. Using a series of ground control points obtained in the field with a GPS unit, the image was rectified and re-projected into the New Jersey State Plane Coordinate System. In addition, this imagery serves as a backdrop for information surveyed in the field, and as the source for multi-spectral data for remote sensing classification and yield estimation. The location of irrigation sprinkler heads within two of the beds provided the initial data sample locations in 1996 (Figure 1). These locations were digitized on screen by scanning in a picture of their precise positions, and later refined using GPS data from within the beds. Based on a preliminary study using this data set it was found that the distance between the irrigation sprinkler heads as sampling points (50-80 feet) did not reflect the actual spatial variability within the beds. Large areas of low yield occurred between sampling points, leading to the consensus that a new sampling method was needed. In 1997, ground sampling was modified to better reflect the heterogeneous conditions of the cranberry beds. This procedure, called "smart sampling" was laid out using flags and the coordinates for each point were determined using a GPS unit (Figure 1). Field data collected at each point include parameters such as number, weight and condition of fruit, upright density, canopy height, soil characteristics, and pH.

Figure 1. Locations of sampling points on two cranberry beds in Speedwell, NJ. Crosses identify sprinkler heads, and filled circles identify smart sampling points. The bed on the lower left is planted with the cultivar "Stevens" and the bed in the upper center is planted with the cultivar "Early Black".

An unsupervised multi-spectral classification was performed on the color-IR imagery to cluster the digital reflectance numbers into 10 statistically based classes. The percent of each class (%CL) within each bed for the whole farm was then computed. A correlation analysis between %CL and total bed yield at harvest was done to determine significant relationships between the spectral data and yield. A stepwise multiple regression was done using the spectral classes with the highest correlation coefficients to predict bed yield. In addition, a normal difference vegetative index (NDVI) was calculated for each pixel based on the formula NDVI=(NIR-red)/(NIR+red) to determine variations in vegetative health over the beds. In-situ ground data for two test beds were overlain onto the NDVI surfaces to visually assess variations in observed data within the beds.


Initial studies utilizing the color-IR photography show that remote sensing techniques have great potential for mapping bed boundaries, estimating yield and providing valuable information to growers regarding crop health. The results of the unsupervised classification on all beds on this farm for 1996 are shown in Figure 2. Major yield components based on this technique are identifiable on the imagery as verified by field observations. For example, a close-up of the Stevens bed from 1996 (Figure 3) reveals many significant features. The dark gray pixels shown in the lower left-hand part of the bed are lower lying areas with poor drainage that are wet. Inspection of this bed in the field revealed the presence of acute symptoms of Phytophthora infestations, specifically, P. cinnamomi that is introduced into the bed via irrigation (Oudemans, 1998). The areas surrounding the dark, wet areas are also poorly drained and appear to be affected by, but not killed by P. cinnamomi. Based on these results, the grower placed several under drains in the same bed prior to the 1997-growing season. The white pixels located above the water in Figure 2 represent areas of low vine coverage and lower yield possibly due to insect damage.

The percent of area in the 18 beds occupied by pixels for each of the ten statistical clusters was computed (%CL). Results of the correlation analysis between %CL and yield at harvest are shown in Table 1. Four of the %CL classes gave correlation coefficients over 0.4 and were selected for use in a multiple regression analysis (eq. 1). The yield values predicted from (eq.1) compare favorably to the actual yields determined at harvest (Table 2). These four classes explain approximately 85% of the variation in yield. Residuals range from a low of -7100 lb./acre to a high of +4600 lb./acre. Average deviation is 1900 lb./acre. These results support development of a method for performing supervised classifications whereby the operator delineates the ground features on the imagery.

Figure 2. Results of an unsupervised classification on a CIR image taken on May 20, 1996.

Table 1. Correlation coefficients between yield in individual beds and 10 spectral classes resulting from unsupervised classification

Spectral class defined in the unsupervised classification
r0.050.02-0.18-0.25-0.4- 0.48-0.58

Yield=263.022-0.21*%CL5+1.87*%CL8+1.54*%CL9-45.0*%CL10 (r2=0.85) equ. 1

Figure 3. Results of an unsupervised classification on a CIR image taken on May 20, 1996.

The above analysis shows that RS information may be used to predict yield. Ground observations based on this map were made in an attempt to identify these spectral classes in the field. Although significant work remains to be done, it appears that the classes contributing positively to yield reflect areas in the bed that have high vine density. Areas that contribute negatively to yield appear to be those high in weeds, and damaged by root disease and/or insects. The results of the NDVI (data not shown) indicate that areas within beds of low or no vegetation are aligned with areas of high water/poor drainage, root damage and disease. Of the two test beds, the "Early Black" bed had the higher overall NDVI, less disease and insect damage and produced approximately 50% more fruit than the "Stevens" bed. The associations between the yield in the bed and the other sampled variables indicate that yield at harvest is strongly associated with upright density similar to results found by Baumann and Eaton (1986).

Table 2. Raw data used to calculate multiple regression from unsupervised multi-spectral classification of image. %CL represents area for each statistical class defined in the unsupervised classification. Actual is the measured yield at harvest. Pred. is the yield computed using the regression equation (eq. 1). Resid. is the residual difference between the actual and computed yields. Raw data are measured in bbl/acre, where 1 bbl/acre=100 lbs./acre.



The above study shows that spectral analysis of color-IR imagery provides valuable insights into the composition of, and variability within cranberry beds. Results suggest that there are at least four significant factors affecting cranberry yield that can be mapped and identified using CIR imagery. For the upcoming season the following objectives are planned.

  • Digital color-IR imagery of the cranberry acreage will be obtained in early May 1998.
  • For selected beds, ground data will be collected at smart sampling sites and irrigation sites.
  • An unsupervised classification will be performed on the images to identify the spectral clusters in the beds.
  • Ground based data will be collected within the clusters to identify the contributing factors for each spectral class.
  • The in-situ data will be linked to the remotely sensed imagery via geostatistical techniques within a GIS.


    Current agricultural methodology is aimed at maximizing productivity while minimizing the area of cultivated land. This is extremely important in cranberry production because strict federal guidelines prevent new cranberry acreage from being developed in the wetland environment. Thus, with little opportunity to expand the area of production, cranberry growers must utilize new technologies to decrease the impact of farming on the environment and maximize yields. These preliminary results indicate that color infrared photography provides useful information regarding crop health and yield. Using an unsupervised classification, a number of features within the beds are identifiable, including variations in vegetative cover within and between beds, irrigation system features, and areas of the beds impacted by both root disease and insect damage. Results from the correlation analysis suggest that aerial photography flown early in spring may be used to predict yield at harvest. In addition, it may be possible to identify areas within beds vulnerable to fungal disease, insect damage, and drainage problems early enough in the growing season that mitigating actions could be taken. Further research linking the remotely sensed data with the site-specific data will aid in understanding the complex and interrelated factors that contribute to crop yield and should help to improve current agricultural practices for cranberry.


    Baumann, T. E., and G. W. Eaton. 1986. Competition among berries on the cranberry upright. J. of the Amer. Soc. for Hort. Sci. 111(6): 869-872.

    Eck, P. 1990. The American Cranberry. Rutgers University Press. New Brunswick, NJ.

    ERDAS, Inc. 1997. Remote Sensing. 5-10. In ERDAS Imagine Field Guide. Fourth Ed. Atlanta, GA.

    Frazier, B. E., C. S. Walters, and E. M. Perry. 1997. Role of Remote Sensing in Site-Specific Management. 149-160. In: The Site-Specific Management for Agricultural Systems. ASA-CSSA-SSSA, Madison, WI.

    Oudemans, P. V. 1998. Detection and monitoring of Phytophthora species in cranberry irrigation water by lupine baiting. Plant Disease 82: (accepted).

    Roper, T. R., and N. Vorsa. 1997. Cranberry: botany and horticulture. Hort. Rev. 21:215-249.