Table Of ContentsNext Page

The Application of Remote Sensing and GIS for Improving Vineyard Management

Sarah Pitcher-Campbell, Mike Tuohy and Ian Yule

Massey University
Private Bag 11222
Palmerston North
New Zealand
0064 63569099
mtuohy@massey.ac.nz

Abstract

A case study was completed on the Stoneleigh vineyard in Marlborough as the initial phase of a project to compare three methods that might be used to produce surrogate maps of soil texture on a field scale.

The objectives of the study were to:
- To map soil variability within the Stoneleigh Vineyards by image analysis of colour infrared aerial photographs.
- To complete a detailed contour and electromagnetic survey of certain blocks of the Stoneleigh Vineyards.
- To make a visual comparison of the maps produced in the two above objectives.

Colour infrared aerial photography was digitised and then interpreted using digital image analysis to delineate stony (drier) soils from silty (wetter) soils. The pattern of ridges and meander channels was easily identified visually on the aerial photographs and the classified image could be used in lieu of a traditional soil map. A high resolution digital elevation model was produced using a real-time kinematic GPS. An electro-magnetic (EM) survey was also carried out to provide additional data that could be used to map the pattern of soils.

Using the digital map, management decisions can be made with more accuracy and confidence. For instance, a more efficient irrigation system could be designed or a soil sampling regime for fertility assessment could be determined.

Introduction

Variations in grape yield and quality and the consequent variation in wine production can present viticulturists with management challenges. Identification of low producing areas may result in replanting while regions producing high quality fruit may be selectively harvested for vintage wine. (Bramley et al., 1999) sited that grape yield can vary from almost zero to sixteen tonnes ha-1. This occurs in blocks that receive uniform management in the form of irrigation, fertiliser, pesticide inputs and weed control measures.

One of the main causes of this variation is soil type. Traditionally pedologists delineate soils based on field observations and sampling of profile pits and augurings carried out in a regular grid pattern with spacing of twenty metres or more and more usually on a grid of 75m sq. Such techniques can be expensive and often do not provide the degree of detail required in situations like that present in many vineyards.

The Stoneleigh Vineyards in Marlborough, New Zealand is typical of many recent plantings on young alluvial terraces where the stoniness of the soil is often a reason for this land use being preferred. By following the normal practice of aligning the rows north-south the natural topographic features, especially the stony ridges and lower-lying meander channels, are crossed almost at right angles. This has resulted in marked variation along the rows, within relatively short distances (~10m), and with subsequent dichotomous soil patterns.

Little literature is cited about the classification of remotely sensed images within a GIS, for soil mapping. A literature review revealed only one relevant research article on the use of colour infrared aerial photography for soil mapping, however this study was aimed at locating tile drains rather than classifying soil texture, (Verma et al., 1997). The drain mapping procedure was based in the fact that soil over efficiently draining tile lines dries faster than soil at other locations in the field and has a higher reflectance in the infrared region of the radiation spectrum. In the study colour infrared aerial photographs were digitised and converted to a format for image processing in the GIS software- IDRISI; where they were then geometrically corrected. The images were then separated into three separate bands (red, green, and blue). Images representing each band and various combinations of the three bands were evaluated to see how well differences in soil moisture content could be displayed. The product of the green and blue bands yielded satisfactory results for tile line mapping, where three distinct moisture regions were observed. On-screen digitising option in IDRISI was then used to digitise the tile drains and the results stored in vector format.

Several research studies (Kachanoski et al., 1988, Doolittle et al., 1994, Doolittle et al., 1995, and James et al., 2000) have shown that mapping soil EC can be a good surrogate measurement for spatial variations in soil properties. (Kachanoski et al., 1988) determined the relationships among field scale spatial variations of soil EC, soil water content and soil texture in a study site in Ontario, Canada, where soluble salts were low. The site was chosen because of the wide textural differences and moisture regimes present. Soil EC measurements were made at 52 positions at various depths using the EM38 sensor. At each of the 52 positions, measurements of volumetric soil water content were taken to a depth of 0.5m using a Textronic 1502 cable tester, hand probe, and Time Domain Reflectrometry. Measurements of soil texture were obtained from core samples from 37 of the sampling positions. Soil EC, determined with the EM38, explained 96% of the spatial variation of soil water content independent of a wide range of soil texture.

(Doolittle et al., 1995) assessed the application of an EM sensor and GPS, to map values of soil electrical conductivity, combined with computer graphics to prepare interpretation maps in a test area of 3725 ha in Virginia, USA. It was concluded that the application of EM and GPS were compatible and provided a rapid means to conduct a preliminary assessment of a soil survey area. The patterns obtained reflected differences in mineralogy, parent materials and depths to water table.

(James et al., 2000) compared traditional methods and EM scanning with the EM38 for determining soil textural boundaries. A soil texture survey was created using a tractor-mounted soil-coring device. A total of 168 cores, one metre deep were extracted and analysed to determine soil texture and horizon depth the soil map was produced from the core information by defining boundaries between soil sore classes. Scanning with the EM38, when the soils were at field capacity, produced the soil EC map. Confusion matrices were used to determine whether electrical conductivity data derived from the EM38 accurately describes soil textural boundaries. These matrices confirmed there were a highly significant correlation between the EM survey and the soil texture survey.

Methodology

There were two separate phases of the research: (1) the creation of a soil variability map of the Stoneleigh Vineyards from image analysis of colour infrared aerial photographs and (2) a contour and electromagnetic (EM) survey of certain blocks of the Stoneleigh Vineyards.

Soil Variability Map

The data sources used in this study were seven colour infrared aerial photographs, each covering an area of the Stoneleigh Vineyards, taken at a scale of 1:5000 and vector data files containing information on the blocks within the Stoneleigh Vineyards. Each colour infrared aerial photograph was scanned (digitised) using a desktop flatbed scanner at 0.5 metre resolution and stored in a Tiff format. Each of the seven digitised images was converted from the Tiff format to an ER Mapper Raster Dataset (.ers) for use and processing in the image processing software - ER Mapper, version 6.1. The seven converted raster images (.ers) were then geometrically aligned (registered) to correspond to the New Zealand Map Grid (NZMG) projection.

The seven colour infrared aerial photographs of the vineyard were taken in such a way that adjacent photographs overlap. Using ER Mapper’s Mosaic Wizard the seven rectified raster images where assembled to create a continuous representation (seamless mosaic) of the entire Stoneleigh Vineyards. The image mosaic contained significant regions of shadows from the grapevines. To compensate for the lack of data in the shadow regions the image mosaic was digitally enhanced (filtered) using two successive low-pass average filters (avg3.ker) on each layer (red, blue and green) of the image. The ‘avg3.ker’ filter averages each pixel based upon surrounding pixels. The ‘3’ in the filter name is the dimension of the square filter e.g a 3 x 3 matrix. In addition to giving the shadowed areas values similar to the adjacent areas these filters also removed any ‘noise’ – small atypical values within the blocks.

The filtered image mosaic of the Stoneleigh Vineyards was then visually interpreted to identify areas that would be classified as fine-textured soils (wetter) and areas that would be classified as stony soils (drier) based on the colours within the image. Figure 1 shows the filtered image mosaic. The pattern of ridges and meandering channels was easily identified visually on this image. The decision was made to classify the colour infrared image into just two soil textural classes as stony and fine textured soils require very different management regimes, such as irrigation requirements, for successful wine grape production. It is therefore important to know the location of these differing soil textures.

The image mosaic was classified on a block-by-block basis and was based on the cell (pixel) values for the green layer only – this corresponds to the red portion of the radiation spectrum. The classification was based on the green layer only as this layer depicted the greatest difference within the image of the potential wetter fine-textured soils and the drier stony soils by removing the influence of the vegetation.

Figure 1: Filtered Image Mosaic of Stoneleigh Vineyards

The type of classification utilised was a basic ‘density slice’ where using the “INREGION()” formula contained in the algorithms of ER Mapper relevant raster data for each block was selected. The INREGION() function returns a ‘1’ if the current cell being processed is within the specified region otherwise it returns ‘0’ (null).

The two INREGION() formulas used to classify each block were:

If ((inregion(r1) and i1 >= v1) then 1 else null

and

If (inregion(r1) and i1 < v1) then 1 else null

Where r1 was the block number, i1 was the green layer, and v1 was the cell ‘threshold’ value. The threshold cell value for each block was chosen based on the visual interpretation of the image mosaic. In the first formula, if the cell values were greater than or equal to (>=) the threshold value (v1) the cells selected corresponded to fine-textured soils. In the second formula, if the cell values were less (<) the threshold value (v1) the cells corresponded to stony soils. It is important to note here that the resulting map created from this classification, based on stony and fine-textured soils, is called a ‘soil variability map’ in this paper.

Several blocks of the Stoneleigh Vineyards were not included in the classification as the rows were either aligned east west, instead of north south, which seemed to mask out the soil features or, grapevines had not yet been planted and the presence of weeds in these areas again mask out the soil features so the soil variability within these blocks could not be determined with any confidence.

In December 2000 the Stoneleigh Vineyards were ground-truthed to evaluate the accuracy of the soil variability map created using the colour infrared aerial photographs. To ground-truth the vineyard 40 data points from around the vineyard were entered into a Trimble PRO - XRS GPS®. For each of these points a description of the soil surface was noted based on the presence of stones. These GPS points were then entered into ER Mapper and overlaid on the soil variability map to confirm if the areas mapped as stony and fine-textured soils corresponded to the presence of stony and fine-textured soils within the vineyard.

Contour and Soil Electrical Conductivity Survey

The EM survey covered approximately 48ha of the Stoneleigh Vineyards and was carried out in early February 2001 using the Geonics EM38 sensor. The sensor was mounted in the vertical dipole mode in polystyrene boxes and attached to a rubber sled. Rubber and polystyrene materials were used as the EM38 responds strongly to the presence of metallic objects within approximately 1m of the sensor (Sudduth et al., 2001). The sled was attached to the rear of a four-wheel all-terrain vehicle (ATV). In this configuration the EM38 was suspended slightly above the ground surface during data collection and was towed approximately 2.4m behind the ATV. This distance was necessary for eliminating the effects of ATV engine noise on instrument readings.

The EM survey was carried out by driving the ATV down every fourth row of a block, which corresponded to approximately 10m intervals. Analog soil EC data from the EM38 was read and stored in the Trimble AgGPS® 170 Field Computer which was mounted in front of the ATV. Data was transferred to the AgGPS® 170 Field Computer via a DL 720 Polycorder® data logger. Soil EC data was recorded at one-second intervals corresponding to a measurement approximately every 2-3m. Differentially corrected RTKGPS data was integrated with the EM38 data to provide the coordinates of each measurement point and to give surface ground height. A Trimble AgGPS® 214 RTKDGPS (real-time kinematic differential global positioning system) receiver was used.

All map production and spatial analysis was conducted using SSToolbox®, which is a GIS specifically written for precision agricultural applications. To create a contour map for each region low accuracy GPS points were filtered out from the database. Using the hybrid vector-raster data model in SSToolbox a grid, with a cell size of 10m was created from the GPS point data. Using this model values for ‘ground height’ were assigned to the cells using the geo-statistical technique – kriging, for interpolating between the sampling points. The contour grid surface was divided into five classes each with an equal interval so the land surface could be observed. The grid surface was then converted to polygons to produce contour maps.

To create soil EC maps for each region, the hybrid vector-raster data model was used in the same way as in the creation of the contour maps. However, two soil EC maps were created for each region. For one of these the surface grid data was divided into two classes before converted to polygons, and the second map was created using five classes each with an equal interval. The soil EC maps divided into two classes were created so they could be compared with the soil variability map, and the soil EC maps in five classes were created so they could be compared with the contour maps.

Results

The results of the trial are in three separate sections: (1) the soil variability map of the Stoneleigh Vineyards derived from image analysis of colour infrared aerial photographs; (2) the contour and EM survey of certain blocks of the Stoneleigh Vineyards; and (3) the visual comparison of the three maps produced for each region to observe if any relationships exist.

Soil Variability Map

Figure 2 presents the soil variability map of the Stoneleigh Vineyards. The blue areas represent fine-textured soils and the yellow areas represent stony soils. This map illustrates the extent of the soil spatial variability that exists within individual blocks of the Stoneleigh Vineyards; down the rows of a block soil texture can change numerous times. For example, along the rows of three particular blocks, going from north to south, the soil texture changes from one texture to another 11 times.

The accuracy of the soil variability map was tested by seeing how well it corresponded with the surface soil observed at 40 locations throughout the vineyard. The result of the testing showed the soil variability map corresponded well to the positions of stony and fine-textured soils within the vineyard. However, the results of the digital analysis showed that some blocks maybe more accurately mapped than others. For example in four particular blocks the soil variability map visually do not look that accurate, and no observation points were taken in these blocks to confirm their accuracy. Many more ground observation points are obviously needed to be confident in the accuracy of the map.

Figure 2: Soil Texture Variability Map

One possible reason for discrepancies in the classification may be the stage of vine growth, where the filtering of the image may not have been successful in masking out the effect of shadowing from older more mature vines. Also the aerial photograph was only separated into two classes, however within the vineyard there are areas of excessively stony soils and areas of fine-textured soil, but there are also soils which could have been classified as slightly stony or slightly fine-textured soil, and these textural classes were not delineated.

Electromagnetic and Contour Survey

Figure 3 to 4 show the results of the EM and contour survey of a small region within the vineyard. Ground height ranged from 15.4 to 17.8m (difference of 1.6m) above sea level and soil EC values ranged from 0.2 to 14.9mS/m across the blocks.

Figure 3: Ground Height and Soil Electrical Conductivity

Figure 4 illustrates how the data collected in the EM survey can be manipulated within a GIS. The maps of soil EC divided into two classes is expected to relate to the extreme soil textural differences i.e. the higher ranged of EC values is expected to relate to fine-textured soils and the lower range to the stony soils. Where the maps of soil EC in five classes are presumed to resemble the more subtle changes in soil texture that exist within a block.

Figure 4: Soil Electrical Conductivity

Visual Comparison of Maps

The higher soil EC values were presumed to relate to fine-textured soils and the lower soil EC values relate to stony soils as sandy soils have a low conductivity, silts have a medium conductivity and clays have a high conductivity (Lund et al., 1999). Moreover, (James et al., 2000) found lower soil EC values (less than 12) correlated to clay loam over sandy clay, and higher soil EC values (greater than 17mS/m) correlated to sandy loam over loamy sand.

As well as this, ground height was expected to have a relationship with soil textural properties. Stony soils might reasonably be expected to be present on the higher areas of the blocks surveyed and the fine-textured soils in the low-lying areas. As levees (occurring adjacent to former river beds) form the highest parts of the landscape and, consequently, are free drained (Gerrard, 1992). Behind the levees lower-lying areas occur formed of heavier-textured clays and silts.

Figure 5 present the maps of soil variability and soil EC, from their visual interpretation, a weak relationship between soil texture and soil EC on the lower half of the maps is revealed. In this respect the fine-textured soils in blue on the soil variability map correspond to the higher soil EC values (5.5 – 14.9mS/m) collected by the EM38, and the stony soils in yellow on the soil variability map correspond to the lower soil EC values (0.2 – 5.5mS/m). However, in other regions there was no visual relationship between soil variability and soil EC. The visual comparison of the soil variability and contour maps suggests there is no simple relationship between soil variability and ground height in these areas.

Figure 5: Soil Electrical Conductivity and Soil Variability

As higher soil EC values are presumed to represent fine-textured soils (Lund et al., 1999) it was expected the high soil EC values would be present in the lower lying areas in the landscape on the contour map. There appeared to be no obvious relationship between soil EC and contour. However, the visual comparison of the same maps in other regions of the vineyard show a possible relationship between soil EC and ground height.

Conclusions

There are a number of reasons why there were no significant relationships between the soil variability maps, the contour maps and the soil EC maps. Firstly, a possible cause for the low correlation between soil variability and soil EC is the fact that aerial photographs look at the reflectance of the soil surface where the sensitivity of the EM38 in the vertical mode is highest at about 0.4m below the instrument, making that approximately 0.3 - 0.35m below the ground surface; the instrument was raised slightly above the ground in the survey. Secondly, the results of the EM survey of soil EC by the EM38 may not be totally reliable because the ground was extremely dry at the Stoneleigh Vineyards at the time the survey was undertaken. Soil EC is partly a function of water content, with the soil being so dry the readings from the EM38 may not be a good indication of changes in soil texture1 . For this reason it would be useful to redo the soil EC survey when the soil was at field capacity, such as James et al, (2000) who measured soil EC with the EM38 when the soil were at field capacity to compared traditional methods and EM scanning with the EM38 for determining soil textural boundaries.

Further work is required in order to provide further evaluation of these techniques

  • Undertake soil EC survey when the soils are at or close to field capacity rather than excessively dry, as occurred within this study.
  • If areas of water ponding are present, which is perceived to be from a man made event, such as an irrigation leak the EM38 sensor data needs to be filtered out.
  • Soil EC and topography surveys should be undertaken in other blocks of the vineyard.
  • Further investigation into the accuracy of digital image processing of colour infrared aerial photographs for creating soil variability maps. Digging soil pits and auguring could achieve this.
  • Obtaining colour infrared aerial photographs taken during the winter months when there is no vegetation on the vines to confuse with the reflectance of the soil surface would also be useful.
  • Obtain true colour aerial photographs, again taken during the winter months, to see if this form of photography can more accurately differentiate soil textural variations within the vineyard. Smith and Culter (1982) concluded that for bare ground true colour air photos gave more information for soil mapping than did colour infrared, where as in vegetated areas under moisture stress both kinds of photography were equally useful.

Measuring yield at the Stoneleigh Vineyards, using a yield monitor attached to a harvester and a DGPS, and using a GIS create maps of yield for various blocks of the vineyards would be the next logical set in the management of the vineyard. These yield maps could then be compared to the soil EC and contour maps and to the soil variability map to observe if any relationships exist.

Information on soil textural boundaries within blocks is critical for viticultural managers where variation in yield is expected to relate to soil texture. The maps produced illustrate how the use of remote sensing, GPS and GIS technology can provide more information about in field soil spatial variation than is available through the use of more traditional soil mapping methods, such as regional soil surveys. Therefore allowing grape growers to gain greater control over their production process by managing their land in a way that takes into account variable performance and allows them to maximise the efficiency of resources.

References

Doolittle, J. A.; Sudduth, K. A. Kitchen, N. R.; Indorante, S. J. (1994). Estimating depths to claypans using electromagnetic induction methods. Journal of Soil and Water Conservation 49 (6): 572-575.

Doolittle, J. A.; Ealy, E.; Secrist, G.; Rector, D.; Crouch, M. (1995). Reconnaissance soil mapping of a small watershed using electromagnetic induction and global positioning system techniques. Soil Survey Horizons: 36 (3) 86-94.

Gerrard, J. (1992). Soil geomorphology: an integration of pedology and geomorphology. London, Chapman & Hall.

James, I. T.; Waine, T. W.; Bradley, R. I.; Godwin, R. J.; Taylor, J. C. (2000). A comparison between traditional methods and EMI scanning for determining soil textural boundaries. In Proceedings of EurAugEng 2000, University of Warwick.

Kachanoski, R. G. Gregorich, E. G. Wesenbeeck, I. J. (1988). Estimating spatial variations of soil water content using non-contacting electromagnetic inductive methods. Canadian Journal of Soil Science 68: 715-722.

Lung. E. D.; Christy, C. D.; Drummond, P. E. (1999). Practical application of soil electrical conductivity mapping. In J. V. Stafford. (Ed). Proceedings of the 2nd European Conference on Precision Agriculture, 771-779.

Smith, S. M.; Cutler, E. J. B. (1982). A comparative analysis of true colour and colour infrared aerial photography as aids in the mapping of soil. New Zealand Journal of Science 25: 325-334.

Sudduth, K. A.; Kitchen, N.R.; Drummond, S.T. (1999). Soil conductivity sensing on claypan soils: comparison of electromagnetic induction and direct methods. In: Proceedings of the Fourth International Conference on Precision Agriculture, American Society of Agronomy, Madison, WI, pp. 979¯990.

Sudduth, K. A.; Kitchen, N.R.; Drummond, S.T. (2001). Accuracy issues in electromagnetic induction sensing of soil electrical conductivity for precision agriculture. Computer and Electronics in Agriculture [Online] 31 239-264. Available: http://www.elsevier.com/locate/compag.html [Retrieved 20 April 2001]

Verma, A.; Cooke, R.; Wendte, L. (1997). Mapping subsurface drainage systems with colour infrared aerial photographs. Applications in Remote Sensing [Online] Available: http://umbc7.umbc.edu/~tbenja1/verma/mod6graphic.html [Retrieved 23 April 2001].

1 R. Bramley, 12 February 2001, personal communication.

Top Of PageNext Page