© 2004 Plant Management Network.
A Guide to Predicting Spatial Distribution of Weed Emergence Using Geographic Information Systems (GIS)
Christopher L. Main, Darren K. Robinson, J. Scott McElroy, Thomas C. Mueller, Department of Plant Sciences, University of Tennessee, 252 Ellington Building, Knoxville 37996; and John B. Wilkerson, Biosystems Engineering and Environmental Science, University of Tennessee, 101 Agricultural Engineering Building, Knoxville 37996
Main, C. L., Robinson, D. K., McElroy, J. S., Mueller, T. C., and Wilkerson, J. B. 2004. A guide to predicting spatial distribution of weed emergence using geographic information systems (GIS). Online. Applied Turfgrass Science doi:10.1094/ATS-2004-1025-01-DG.
A Geographic Information System (GIS) combines layers of spatially related information to better understand relationships that vary geographically. Scientists could benefit from the use of GIS to better explain new research, expand on prior research, and potentially develop integrated pest management (IPM) programs. Objectives of this manuscript are to demonstrate GIS applications in a posteriori weed management decision-making. The model system utilized empirically derived temperatures for large crabgrass (Digitaria sanguinalis L.) germination. Numerous studies have been conducted on the germination parameters of large crabgrass; however, there have not been attempts to apply the data to a spatial context. This report compares four techniques for creation of prediction maps to determine which method would be most useful in the creation of maps that predict time of weed emergence. Inverse Distance Weighted (IDW) technique formed distinct rings of high and low temperature values around weather station locations (producing an inaccurate prediction map). The two Spline techniques investigated had a tendency to over- and under-predict temperature along the borders of Tennessee due to the procedures’ fitting of a minimum curve to the data. Interpolation by Kriging produced the most accurate prediction map due to the spatial autocorrelation introduced by this geostatistical model.
Geographic Information Systems (GIS) help manage, analyze, and present spatially related information combining multiple layers of environment and biological information related to a spatial location to gain a better understanding of a specific location. Researchers can benefit from the use of a GIS to more fully investigate data and develop spatially accurate graphical data displays. Data presented in a table of rows and columns is much different than data presented spatially. While this difference is conceptual, the way one comprehends data may have a profound effect on the conclusions drawn from that data (1). GIS provides the layout and drawing tools that present research results in visual documents. Spatial application of empirical data could possibly aid researchers in decision-making that could lead to a better understanding of biological systems.
The example used in this demonstration is large crabgrass (Digitaria sanguinalis L., herein referred to as crabgrass for brevity) germination as affected by interaction of environmental and biological factors. This paper describes a method for taking climatic data (air temperature), biological data (temperature for crabgrass germination), and GIS coordinate data to develop a model for crabgrass emergence to make improved control recommendations.
Crabgrass is an ideal weed for demonstration of GIS modeling in the field of weed science. Both empirically derived crabgrass germination (5,9,13) and historical weather data from the National Oceanographic and Atmospheric Administration (NOAA) are available (14). Spatial analysis of these parameters using ArcGIS 8.2 (ESRI, Redlands, CA) software can be utilized with existing crabgrass management recommendations to develop an IPM program for this weed.
Crabgrass is a common summer annual grass weed in both cultivated and non-cultivated fields as well as in turfgrass (17). Crabgrass annually invades turf areas (10). This weed is an undesirable, expensive-to-control pest in most of the United States (6,16).
Annual grass weeds that can be anticipated in established turfgrasses are best controlled by preventative herbicides. Preemergence crabgrass control is performed annually with a variety of herbicides at various application timings (11). Preemergence herbicides prevent germinating seedlings from developing. To be effective, these herbicides must be applied prior to seed germination (11). Preemergence herbicides generally have residual control, extending over a period of a few weeks or months. Ideally, a preemergence herbicide should be applied immediately prior to weed germination, thus allowing maximum resdiual control into the growing season.
In turfgrass situations, preemergence herbicides are applied to existing vegetation, thatch, and the soil surface. After application, rainfall or irrigation is needed to evenly distribute and move the herbicide into the root zone where it is effective in controlling emerging weeds. Some products such as pendimethalin rapidly degrade (~7 days) when exposed to sunlight, so movement into the soil immediately after application with rainfall or irrigation is crucial (18). However, excessive rainfall or irrigation can lead to herbicide leaching through the soil profile with water or to herbicide runoff from the soil surface with both circumstances reducing weed control (15). More precise timing of preemergence herbicide application just prior to weed emergence could possibly reduce herbicide losses and lessen environmental impacts from herbicide use.
Timing of weed seed germination is life cycle dependent. Crabgrass germination begins when environmental conditions reach certain measurable parameters. Several studies have been conducted on the germination parameters of crabgrass. Crabgrass begins to germinate at temperatures between 10°C to 15°C (9) with 40% germination reported at 15°C (13). Degree-day analysis demonstrated that a base temperature of 12°C was useful as a beginning point for determining crabgrass emergence (5).
The objective of this project was to determine how a GIS could allow for integration of historical weather data and biological data to create maps that predict crabgrass emergence. This manuscript investigates and contrasts three different interpolation procedures for creating a temperature distribution map, and a method described to create a map for predicting crabgrass emergence using the state of Tennessee as an example. Development of a method for predicting weed germination could be applied with current management strategies to aid in development of IPM programs for weed management.
Data Sets, Interpolation, and Map Creation
The first step in utilizing GIS is to determine the availability of both relational and spatial data sets relevant to weed emergence. Examples of these types of data sets are yield maps from previous years, soil series maps, topography maps, or environmental measurements (rainfall, temperature, degree days, etc.). For our demonstration, maximum daily air temperature was selected. ArcGIS is a family of software products that form a complete GIS with integrated systems for geographic data creation, management, integration, and analysis.
Data set creation. Historical weather data were obtained from NOAA Daily Station Normals (1971-2000) for 87 weather stations across Tennessee (Fig. 1) (14). Weather data were reported in two files. The first file (denoted as the weather station location file) indicated station identification number, location name, latitude and longitude (in degrees and minutes), and elevation of the weather station. The second file (denoted as weather data file) contained station identification number and average high, low, and mean temperature, as well as warming and cooling degree days for each day of the year (totaling 365 data points for each station location). Maximum daily temperature was used to simplify this demonstration of the GIS. This simplification indicates that the researcher must use judgment to select appropriate data from what is available to utilize in a GIS. For example, these NOAA data sets did not include soil temperature, which would be a more precise measure of the temperature at the soil to seed interface. Therefore, in this method, a surrogate measurement (maximum air temperature) was used as an indirect measure of soil surface temperature. GIS software is robust in the sense that any type of modeling (hydrothermal, degree-day, and others) parameters can be utilized to create prediction maps.
The data sets were converted into database files with a spreadsheet program by opening the NOAA file as a space-delimited text file and then saving this file in a database file format (.dbf). The station location file was added to the ArcMap (ESRI, Redlands, CA) program as X-Y data (since it contains the latitude and longitude coordinates of each weather station) and projected with a map of Tennessee counties. The temperature data file was relationally joined to the coordinate data by a common attribute (station identification number) for each file. After joining, the combined file with temperature data and weather station location was exported to a shape file for interpolation to a raster using Spatial Analyst (ESRI, Redlands, CA). This step-by-step building of data is different from other types of data examination techniques because all the observational data is geo-referenced to each collection location. Thus, one can build multiple layers of data describing the same geographic location.
Interpolation techniques. The temperature data consists of individual points across Tennessee rather than a continuous map surface. To graphically display the data as a continuous surface, a technique must be used to determine what happens between the actual data points. Usually the different data points are not uniformly distributed, or if they are, they do not fully describe the heterogeneity of the examined system. In our example, not all counties have weather stations, some counties have multiple stations, and counties are of different sizes. This inconsistency illustrates the need to correctly examine each data point as it relates to other data. Another factor to consider is that some data points do not have other points surrounding them (such as a weather station located near the state boundary) creating an "edge effect." Edge effects are typical in fields, since various measurements are often made on a field-specific basis when utilizing GIS, such as in soil fertility or pH measurements. The process of using multiple data points to ascertain data values between those points is called interpolation. All interpolation procedures utilized ArcMap default settings for simplification of prediction map creation.
Three interpolation techniques, Inverse Distance Weighted (IDW), Spline (Minimum Surface Curvature), and Kriging (Ordinary Point Kriging) were investigated. Air temperature values for a four-day period beginning March 28 and ending March 31 were selected for interpolation comparison. These dates were chosen to coincide when crabgrass germination should occur in Tennessee based on preliminary analysis (analysis not shown). Inverse Distance Weighted and Spline techniques are deterministic interpolation methods that are based on the surrounding measured values or on specific mathematical formulas that determine the smoothness of the resulting map (commonly called a surface) (8).
Inverse Distance Weighted technique estimates temperature for cells averaging known values from sample points close to the cell of interest. IDW assumes that variables being mapped decrease in influence with the increase in distance from the sample location (3,7,12).
An interpolation using the Spline method fits a minimized curved surface to known sample points (2,12). This fitted data results in a smoothing effect passing through the known data points. Two types of Spline interpolations are possible: regularized and tension. Use of the regularized method creates a smooth, gradually changing surface, but has a tendency to over- and/or under-predict for a given area (12). The tension method uses the character of the modeled data to create a surface with values that are more closely constrained by the range of the data set (12).
Interpolation by Kriging is a geostatistical method based on statistical models that predict spatial autocorrelation of sampled data points (3). Autocorrelation, or variography, is the statistical relationship among measured points in a data set. Kriging can also provide some measure of the certainty or accuracy of the prediction models based on autocorrelation. Kriging interpolation is similar to IDW in that it uses surrounding data points to predict an unknown value for an unmeasured location. The difference with Kriging is that the predicted point depends on a fitted model to the measured points, the distance from the unknown point to measured points, and the spatial relationship among the measured points around the predicted point. Kriging is considered the best predictor of non-sampled locations, because mean residual error is minimized by its calculation (8). Kriging models use a semivariogram to depict the spatial autocorrelation between measured sample points. Semivariogram modeling is what separates spatial modeling from simple spatial description. The model assumes that measurements that are geographically close together are more similar than ones that are farther apart (4).
Semivariograms are described by the parameters of range, sill, and nugget, which are needed to interpolate data with a Kriging method (Fig. 2). The range is the distance from a measurement (known sample) point to the point where the semivariance stops increasing with distance from the sample point. The value at which the semivariogram model attains the range is known as the sill, meaning the change in semivariance is no longer increasing with increasing distance from the sample point. The nugget or nugget effect is created by measurement errors or spatial sources of variation at distances smaller than the sampling interval and is the value of semivariance when the distance from the sample point equals zero. The partial sill is the sill minus the nugget and is the value needed for Kriging interpolation.
Prediction map creation procedure. The method utilized to create this crabgrass prediction map is illustrated as a flow chart in Fig. 3. After determination that the Kriging method was the most useful model to use, each day temperature report was analyzed by the Geostatistical Wizard in Geostatistical Analyst to determine range, partial sill, and nugget. The temperature for each day was subsequently interpolated by the Kriging method and a temperature layer for each day from March 1 to April 15 was created. During the Kriging process, the range, partial sill, and nugget were entered as advanced parameters for the semivariogram.
For this demonstration, the environmental requirement for crabgrass germination is set as four consecutive days of 18°C air temperature. Data layers were analyzed in combinations of four (i.e., March 1, 2, 3 and 4; then March 2, 3, 4 and 5; etc.) using Cell Statistics in Spatial Analyst averaging the temperature for each consecutive four-day period. Air temperatures were averaged over a four-day period since the warming of soil temperature is more dynamic than air temperature. This analysis resulted in creation of 43 separate four-day periods predicting temperature from March 1 to April 15. Since the biological germination trigger was set at 18°C, the layers were searched for the first four-day period that had a temperature of 18°C or greater somewhere in Tennessee. The first day of consecutive four-day periods with average high temperatures of 18°C or greater was March 23, and was set as the starting date for further analysis. At the same time, the layers were searched for the last layer that contained a region of Tennessee with temperatures less than 18°C (April 4), and this was set as the ending date for analysis.
Beginning with the March 23-26 layer, the data were reclassified using Spatial Analyst to indicate areas with temperature less than 18°C and areas where the temperature was 18°C or greater. Each layer then represented discrete variables, 0 indicating areas where temperatures were < 18°C (no crabgrass germination), and 1 indicating areas where temperatures were > 18°C (crabgrass germination may occur). The next step utilized the Raster Calculator in Spatial Analyst to add the discrete layers (13 layers in this simulation) together to create a map with areas that had 18°C temperatures for four days beginning on March 23 (a cell value of 13), and the areas that would not reach 18°C for four days until April 4 (a cell value of 0). Each day after March 23 was indicated by a cell value equal to the value for March 23 (cell value = 13) - (1 × the number of days after March 23). Therefore March 24 and 25 would have cell values of 12 and 11, respectively. This procedure allows for graphical display of what day crabgrass emergence is predicted.
After the final calculation, some cells existed that were artifacts of the interpolation procedure and misrepresented small areas of the state. These cells are misrepresented data from the interpolation method found in a large geographic area that is homogeneous for a specific temperature. To correct for these artifacts, Neighborhood Statistics tool in Spatial Analyst was utilized to smooth the data. Neighborhood statistics is a function that computes a cell value based on neighboring cells to correct for small pockets of cells that exist as artifacts of the interpolation procedure. The data are smoothed as a result of this procedure. After smoothing, a map was created that depicts the predicted pattern of crabgrass germination across Tennessee.
GIS Techniques and Making Weed Emergence Prediction Maps
The term germination will be used in this discussion to designate predicted germination based on historical temperature data.
Comparison of interpolation techniques. The four-day period used for interpolation model comparison had an actual average high temperature range of 15.5 to 21.1°C for 29-year average high temperature (14). Inverse Distance Weighted technique formed distinct rings of high and low temperature values around weather station locations since this method assumes that the relevance of each data point diminishes in importance with distance from a known point (Fig. 4a). These rings around sample points are common with the use of IDW technique. However, one advantage with IDW is that interpolated values will fall between the maximum and minimum temperatures for the state. The interpolated temperature range using IDW was 15.5 to 21.1°C, which is identical to the actual range of values of the input data set.
The Spline technique had a tendency to over- and under-predict some data points since it was fitting a mathematical minimum curvature technique to the data. This results in temperatures below and above the measured values for station locations (Fig. 4b). The predicted temperature range for this four-day period using Spline was 12.8 to 21.7°C, which is 3.3°C larger than the range for the NOAA data set. To help minimize this under/over-prediction the tension option was utilized to reduce the peaks and valleys around known data points, but some predictions outside the measured temperature range still occurred (Fig. 4c). The tension option reduced the under/over-prediction that was associated with fitting a curve to the modeled data. This option predicted a temperature range of 15 to 21.7°C, which is only 1.1°C larger than the reported data set range. Analyzing data points from outside the state boundary could possibly minimize over- and under-prediction along with reducing the edge effect produced by Spline interpolation.
On the Kriging map, no circles around measurement points were apparent when compared to the IDW map, which was due to spatial autocorrelation introduced by geostatistical model used by the Kriging method (Fig. 4d). The temperature range for the Kriging method was 16.7 to 20.6°C, and was 1.7°C smaller than the actual range of the data set. In this case, using temperature as a trigger for weed germination, predicting a higher temperature than the actual minimum temperature would be acceptable. Higher predicted temperatures would allow the temperature for germination to be reported before the weed seeds would actually germinate, allowing preemergence herbicides to be applied before germination occurs for maximum herbicide efficacy.
Dille and others (3) determined that no single interpolation technique was more precise than another in creation of weed maps. In our research the "default settings" for each interpolation procedure are utilized to facilitate the uses of the ArcMap program by novice users. To determine the most correct interpolation procedure, weather station data was selectively removed, and each interpolation procedure was allowed to predict what temperature should occur for the data that was removed. Interpolation by Kriging utilizing the default settings produced the most conservative and correct temperature prediction map based on the previous discussion of each interpolation methods’ tendencies and weaknesses. For these reasons, Interpolation by Kriging was selected for creation of a map that predicts the time of crabgrass germination.
Creation of a crabgrass germination map. The Kriging method was determined to be the most useful for predicting the date of crabgrass germination since it did not display the inadequacies of IDW or Spline techniques. Crabgrass germination is predicted to begin on March 24 (Fig. 5). Crabgrass germination progresses north and east until April 4, except for an area in the eastern, north-central part of the state known as the northern highlands of the Cumberland Plateau where germination more closely resembles northeastern Tennessee.
A potential use of this technique is to more optimally time preemergence crabgrass control strategies. Development of an IPM program for crabgrass control will rely on selected preemergence herbicides being applied prior to crabgrass germination. According to our GIS methods, in the western part of Tennessee, preemergence herbicides should be applied beginning March 20 in the southern part of this region and by March 25 in the extreme northwest corner of Tennessee (Fig. 5). In the central part of the state, preemergence herbicides should be applied beginning March 23 and progress northward until March 28. The eastern third of Tennessee should make applications starting March 25 and progress north and east through the Tennessee River valley until March 30. Extreme northeastern Tennessee and the Cumberland Plateau should make preemergence applications beginning March 30 until April 3.
This project has demonstrated that GIS can be used to make maps that theoretically predict weed emergence from weather data. Confirmation of the predicted map can be accomplished with field research to determine the actual date of crabgrass germination across a region. Hopefully, this unique use of GIS technology can be used to develop an IPM program for crabgrass management in Tennessee, which can then be transferred to other states based on their historical weather data.
GIS presents information graphically, which may allow researchers to more fully investigate data resources and develop spatially accurate graphical data displays. GIS provides the layout and drawing tools that present research results with visual documents. Graphically displayed data can have a profound effect on the conclusions drawn from a data set. Utilization of GIS will allow researchers to better understand how research results apply to our natural environment.
Acknowledgments and Disclaimer
Mention of any trademark or product name does not constitute endorsement by the University of Tennessee to the exclusion of similar products. Names used are for clarity and brevity of the discussion.
1. Brown, T. L. 2003. Making Truth: Metaphor in Science. University of Illinois Press, Urbana, IL.
2. Cooke, R. A., Mostaghimi, S., and Campbell, J. B. 1993. Assessment of methods for interpolating steady-state infiltrability. T. Am. Soc. Agric. Eng. 36:1333-1341.
3. Dille, J. A., Milner, M., Groeteke, J. J., Mortensen, D. A., and Williams, M. M. 2002. How good is your weed map? A comparison of spatial interpolators. Weed Sci. 51:44-55.
4. Donald, W. W. 1994. Geostatistics for mapping weeds, with a Canada thistle (Cirsium arvense) patch as a case study. Weed Sci. 42:648-657.
5. Fidanza, M. A., Dernoeden, P. H., and Zhang, M. 1996. Degree-days for predicting smooth crabgrass emergence in cool-season turfgrasses. Crop Sci. 36:990-996.
6. Gregg, M. F., Higingbottom, J. K., Lowe, D. B., McCarty, L. B. 2000. Preemergence crabgrass and goosegrass control in bermudagrass. Proc. South. Weed Sci. Soc. 53:56-57.
7. Gotway, C. A., Ferguson, R. B., Hergert, G. W., and Peterson, T. A. 1996. Comparison of kriging and inverse-distance weighted methods for mapping soil parameters. Soil Sci. Soc. Am. J. 60:1237-1247.
8. Isaaks, E. H., and Srivastava, R. M. 1989. An Introduction to Applied Geostatistics. Oxford University Press, Oxford, UK.
9. King, C. A., and Oliver, L. R. 1994. A model for predicting large crabgrass (Digitaria sanguinalis) emergence as influenced by temperature and water potential. Weed Sci. 42:561-567.
10. McCarty, L. B., Everest, J. W., Hall, D. W., Murphy, T. R., and Yelverton, F. 2001. Color Atlas of Turfgrass Weeds. Ann Arbor Press, Chelsa, MI.
11. McCarty, L. B., and G. Miller. 2002. Managing Bermudagrass Turf: Selection, Construction, Cultural Practices, and Pest Management Strategies. Ann Arbor Press, Chelsa, MI.
12. McCoy, J., and Johnston, K. 2001. Using ArcGIS Spatial Analyst. ESRI, Redlands, CA.
13. Moreno, J. E., and McCarty, L. B. 1994. Factors affecting crabgrass (Digitaria Spp.) germination. Proc. South. Weed Sci. Soc. 47:71.
14. NOAA. 2001. National Oceanic and Atmospheric Administration. Daily Station Normals 1971-2000. National Climatic Data Center, Asheville, NC.
15. Ross, M. A., and Lembi, C. A. 1999. Applied Weed Science, 2nd ed. Prentice-Hall Inc., Upper Saddle River, NJ.
16. Toubakaris, M. R., Higingbottom, J. K., Bunnell, B. T., McCarty, L. B. 1999. Preemergence crabgrass control in bermudagrass. Proc. South. Weed Sci. Soc. 52:74.
17. Webster, T. M. 2000. Weed survey: Southern states. Proc. South Weed Sci. Soc. 53:267-271.
18. Vencill, W. K. ed. 2002. Pendimethalin. Pages 340-341 in: Herbicide Handbook, 8th edition. Weed Sci. Soc. Am., Lawrence, KS.