Home

Using Remote Sensing To Manage North Dakota's Rangelands

Paul E. Nyren and Bob D. Patton


Article Contents

Summary
Introduction
Applications

-- GIS uses
-- Mapping Vegetation types
-- Technology becoming more user friendly
CGREC Research
-- Determining Forage production
-- Sward Stick
-- Radiometer
-- Neural Network


Summary

The advent of powerful home computers and optical devices such as digital scanners is allowing sophisticated technologies to be used by farmers and ranchers as another management tool. These techniques hold the promise of providing more range monitoring with less time and labor. One of these tools is what is commonly called GIS or geographical information systems. There are many functions in a full GIS system. They vary from simply looking at aerial images and drawing out field boundaries for determining acreages to developing complex maps with numerous overlays.

 Introduction

With the use of today's site-specific crop monitoring capable of estimating crop yields down to a fraction of an acre, it is difficult to imagine a crop whose production remains largely unmeasured. Rangeland has always been the prime example of low input sustainable agriculture. Today, however, with more pressure to manage rangelands for an ever-increasing number of interest groups there is a need for better monitoring of this resource.

Rangeland monitoring is complicated by the vast areas needing study and the labor-intensive techniques required. The new technologies for site-specific crop monitoring are also being used to develop new techniques for rangeland monitoring.

Applications

GIS Uses

Recently you may have read articles or heard discussions about precision farming, which is the ability to map soil fertility or productive potential from soil surveys or soil maps and monitor crop inputs and yields accordingly. While farmers or agricultural consultants may not use sophisticated computer software to produce these field maps, they are, in essence, working with GIS systems. While these processes seem impressive, some may feel that this type of management is only available to those who can afford to pay for the information. To some extent this is true but the field of GIS, like that of computers, is changing rapidly and the information and availability of information will become more easily accessible and less expensive in the future.

The uses of GIS are limited only by the skill of the user, the availability of inputs such as digitized aerial photos or satellite imagery, and the user's imagination. Applications that have been documented by users in North Dakota and other states include sophisticated soil and vegetation maps, noxious or undesirable vegetation maps, and land use planning, to name a few.

The use of GIS in land management requires some type of image. Generally when large areas are being managed satellite imagery is used. Somewhat more useful imagery for individual ranch operations can be acquired with a light single engine aircraft and a 35mm camera. The higher the altitude the more area will be covered in each image but the resolution will be lower.

This principle is similar to the photographs taken on the family vacation. A photo taken a few feet from a family member will show details of the person's face and clothing but may not show the panoramic view of the mountain range in the background. A photo taken of a hiker from a long distance may show the grandeur of the mountains being traversed but lack the detail to recognize who the hiker is or what he or she is wearing.

Mapping Vegetation Types

Mapping vegetation types is one function of GIS systems. This allows users to identify distinct vegetation types such as troublesome or noxious weeds or woody plants. Images are scanned into the computer and the plant species or plant communities identified. The unique color pattern of the vegetation is mapped by the computer and the areas of infestation measured.

An example of this was described in an article in the June issue of Beef Magazine entitled "Aid from Above." Texas rancher Bert Wallace used GIS to map his 66,000 acre ranch to determine the highest infestations of mesquite, a woody plant which, if left unchecked, competes with forage grasses for scarce soil water. Mesquite grows no higher than 4 feet in dry areas but can reach 60 feet in areas near water. Aerial spraying is used to control the mesquite but spraying the entire ranch is too expensive. Wallace contracts with an aerial survey company to fly the ranch and take infrared photos of the entire ranch at a scale of 1:40,000. Wallace then scans the photos into his computer and runs a classification routine called Adaptive Resonance, which automatically clusters the pixels into groups of similar reflectance values. It also compensates for differences in reflectance numbers resulting from shadows or other distortions on the photo. Each cluster contains pixels representing similar ground features. Grasses would be in one cluster, and mesquite would be in a different one while bare soil might be in yet another. This first round of classification is somewhat arbitrary and two ground features with similar reflectance might end up in the same cluster such as cottonwoods and mesquite which look similar on an infrared aerial photo.

The next phase of processing, called Feature Mapping, refines this classification and uses operator input to actually identify a cluster which is something the computer cannot do. "The beauty of Feature Mapping is that I can keep inputting ground truth information to further subdivide the classes," says Wallace. For instance, once the mesquite stands are identified Wallace measures their canopy density on screen. He then enters this as additional training information and reclassifies the already identified mesquite classes as high, medium or low density. This step is very important because grass is inhibited from growing under mesquite trees once canopy closure exceeds 45%, which is when spraying becomes cost effective. Once the densest stands of mesquite have been identified, soil maps are digitized and placed over the vegetation maps to identify the areas of best soils. These areas are where chemical application will be the most cost effective.

 Technology Becoming More User Friendly

While this technology is becoming more affordable for individual producers, the time it takes to learn the GIS programs and the expense of obtaining the images is still a large deterrent to more individual use of this tool. New companies are beginning to emerge that will collect the images, usually satellite, and sell just the small portion of the image an individual producer needs. Currently software companies are developing application software that will automatically perform the type of analysis that has been described in this article. Researchers at the Central Grasslands Research Extension Center (CGREC) have contributed to a national grant proposal to NASA for funding to assist a software company to develop this type of application for individual producers.

CGREC Research

Determining Forage Production

Another function of GIS systems is measuring current growth or above ground biomass. This is done by using color infrared film to capture the images of the area to be studied. Chlorophyll, the green pigment in plants, reflects nearly all near infrared light (NIR) (Figure 1 a) and absorbs nearly all visible red light (VR) (Figure 1 b). By using the computer to compare the amount of NIR to VR the amount of green plant material can be calculated (Figure 1 c). This technique will not separate the species of plants but gives a total biomass for the area.

The CGREC has concentrated its efforts in GIS on developing techniques to measure the above ground biomass on a grazing intensity study. The aerial photos used at the CGREC for this study are obtained using a standard 35mm camera and color infrared transparency film. The photos are taken from an elevation of 3000 feet above ground level. Once the film is developed, it is scanned or digitized into a desk top computer. This is accomplished by a device which projects a very narrow light beam through the film and measures the density of each of the color bands (NIR, VR and green) in the film. Each point measured on the film produces a number that the computer records. These numbers create a dot on the computer monitor referred to as a pixel and this number represents the density of the color on the film. The ratio of reflectance of NIR and VR is then calculated by the computer and a third raster created from the results (Figure 1 c). The results of these ratio calculations produce an image that can be viewed on the computer monitor or printed as a picture. The lighter the color in the image the more green biomass on the ground. The area on the ground represented by a pixel on the screen is referred to as the resolution and is approximately 1.7 feet square for the images in this study.

The calibration of the transverse vegetation index (TVI) values on the aerial photos to above ground biomass is accomplished by locating areas on the image that were sampled by conventional clipping methods. Five rectangular areas covering 20 X 16 pixels are located inside these sites and the values for each pixel are then downloaded into a spreadsheet program and averaged. The correlation between the average brightness value from the image and the forage production obtained by conventional clipping methods is calculated. Once the image is calibrated the production represented by each pixel can then be calculated allowing for a very rapid estimate of the production on the entire image.

Sward Stick

Other methods of remote monitoring at the CGREC include the use of a sward stick and radiometer. Like GIS these devices use methods other than actual clipping and weighing of the forage to determine the current year's production. The sward stick (figure 2) measures the height and density of the forage underneath it, so it is essential that the plate and hollow tube weigh 0.98 lb/sq ft. if our calibration tables are used. The sward stick was originally constructed from a sheet of plastic, plastic water pipe, and an aluminum tube (all available at a local hardware store). We found after using the device on several windy days that the thin sheet of plastic cracked and it was subsequently replaced by a thin sheet of fiberglass.

The sward stick is designed to not only measure the height of the vegetation under the square plate but also the density. This is done by slowly lowering the plate down on the area to be measured and reading the height at which the plastic tube and plate are supported by the vegetation. This reading can be taken in a few seconds and the operator can walk to a new location, repeat the process and collect another reading. Fifty readings can be collected in the time it would take a person to collect one sample by the traditional clipping method.

Before the sward stick can be useful it must be calibrated. This is done by placing the device over an area and obtaining a reading. The vegetation under the plate is then clipped, dried and weighed in the traditional manner. A number of these calibration readings must be taken on a variety of vegetation densities to accurately calibrate the device. The sample weight and the sward stick reading are then entered into a computer and a regression formula calculated to determine how closely the sward stick readings are related to the actual sample weights. From this formula, a table of values can be constructed that will allow the user to obtain a number of readings with the sward stick, average these readings and look up the corresponding production in the table. Areas with different production potentials or species such as range sites and seeded pastures should be calibrated separately.

The advantage of the sward stick is that once calibrated it can be used relatively rapidly and the more samples taken the more accurate will be the results. One disadvantage is that a new calibration table may have to be developed each year for each site. Another is that the sward stick cannot be used to measure the forage production under woody species such as buckbrush and, in the fall, on areas where stiff forbs are growing, without first removing the woody species which reduces its practicality.

 

 

 

Radiometer

The radiometer is a device developed by Dr. Vernyl Pederson at North Dakota State University to monitor crop health. It consists of light sensors mounted on an aluminum pole and connected to a portable computer (figure 3). It works by measuring the incoming light striking the top of each sensor and comparing that value to the light reflected from the crop canopy. The size of the area sampled is varied by raising or lowering the height of the sensor above the ground. The sensor head contains eight individual sensors, each sensitive to a narrow band of light. By separating the incoming light and reflected light from the crop canopy into eight narrow bands, the reflectivity of the vegetation in these separate bands can be measured. The actual measurement of forage production is based on theradiomtr.gif (22423 bytes)principle that green growing vegetation reflects almost all near infrared (NIR) light and absorbs all visible red (VR) light. By calculating the ratio of incoming and reflected VR and NIR light, the amount of green vegetation can be calculated.

The calibration process for the radiometer is similar to the sward stick. A radiometer reading is taken from an area which is then sampled by clipping and weighing the current year's production. The radiometer readings collected and stored in the portable computer in the field are later transferred to a desk top computer and the ratio of VR to NIR is calculated. A regression formula is then calculated using this ratio and corresponding actual yield value to get a calibration formula similar to the sward stick. Once this is done, yield tables can be constructed by the computer which will convert the vegetation ratio from the radiometer into yield data.

The radiometer is less sensitive than the sward stick to being calibrated for each range site or species but less accurate in measuring vegetation that has begun to lose its green color. The time of day may affect the accuracy of the radiometer making it more sensitive to when it is used. The radiometer, like the sward stick, must be hand operated and the operator must walk over the area being sampled.

 

Neural Network

As part of this study the CGREC is looking at the possibility of using artificial intelligence to assist in the prediction of above ground biomass using data obtained from the aerial photographs. Artificial intelligence or neural networks, are computer programs comprised of highly interconnected processing elements or nodes. These nodes are analogous to the neurons in the human brain. Neural networks differ from traditional predictive technologies by being able to learn from their own errors like human beings. The basic structure of a neural network consists of an input layer, an output layer and hidden layer(s) between the input and output layers. Each layer in the network contains a number of processing elements or nodes depending on the type of application. The input information to the network are represented in the input layer. The desired results of the network are given by the output layer. For example in this study the treatment, range site, and measurements will be input to the network while the above ground biomass will be predicted through the output layer. There are various algorithms within neural networks and they continue to be developed and improved by software manufacturers. The advantage of a neural network is that it can examine several inputs at one time to determine the relationship to the output (in this case the above ground biomass).

The generalization capability of neural networks makes it possible for this technology to learn and make decisions from inconsistent or incomplete data in much the same manner as a human being. For biological and agricultural applications, where the data are often inconsistent and ill-defined, this technology has a higher potential of success than other traditional technologies.

Table 1 shows the results of the general regression neural network (GRNN) applied to the data sets obtained from the sward stick, radiometer and aerial photos. The average actual and predicted forage yields for the entire study area are shown. The R2 was highest for the radiometer (0.98) and lowest for the sward stick (0.87). The next four columns of the table show the percentage of observations which fall within a given percentage of the actual production values. The aerial photographs were highest with 97% of the predicted values within 10% of the actual value. The sward stick and the radiometer were within 67% and 79%, respectively. The final two columns in the table give the average percent difference between the value predicted by the network and the actual production value. The first of the two columns gives the average value predicted for the entire study area. This value was calculated by averaging all the predicted values into one number for the entire study area and comparing that number with the average of all the actual values. These values range from a high of 1.3% to a low of 0.8%. The final column gives the average difference between the network predicted value and the actual value by pasture and study site. Here the differences are greater since the network is trying to predict the outcome of a smaller area. These values range from a high of 8.9% for the sward stick to a low of 3.4% for the aerial photographs. In this case the neural network is no different than any other model. The larger the area of prediction, the easier it is to predict accurately. While the use of remote sensing and high speed computers may seem futuristic, the technology is fast becoming more available and user friendly. However, this type of technology must be evaluated as any other ranch management tool is evaluated based on a cost benefit ratio. Where the need exists, remote sensing technology may prove to be very useful to North Dakota producers.

 


Table 1. Results of applying a General Regression Neural Network (GRNN) to the aerial infrared photos, radiometer, and sward stick data sets in 1998.






Treatment


Average for entire study area




R2

Percent of predicted forage yield on sample sites
Average percent difference
Actual
production
(lbs/acre)
Predicted
production
(lbs/acre)

within 5%
of actual
within 10%
of
actual

within 20%
of actual
within 30%
of
actual

Entire
area

By sample site
Aerial infrared photos

1,972


1,987
0.95 70 97 97 97 0.8% 3.4%
Radiometer 1,989 1,986 0.98 64 79 96 100 0.2% 4.8%
Sward stick 1,795 1,772 0.87 40 67 83 97 1.3% 8.9%


Paul E. Nyren, Director
North Dakota State University
Central Grasslands Research Center
4824 48th Ave. SE
Streeter, ND 58483
Phone: 701-424-3606
E-mail: p.nyren@ndsu.edu

Bob Patton
Assistant Range Scientist
North Dakota State University
Central Grasslands Research Center
4824 48th Ave. SE
Streeter, ND 58483
Phone: 701-424-3606
E-mail: bob.patton@ndsu.edu


Home

Back to Range Contents