Predicting Invasions Using the Garp System

GARP for Dummies-Predicting invasions using the GARP System

Darren Houniet and Paul Emms

Abstract:

The prediction of the geographical extent of the invasive potential of alien plants has important implications for the conservation of South African fynbos ecosystems. The GARP system is a programme which can be used to predict the extent of these potential invasions. We surveyed fifteen of the most invasive species in the South African fynbos biome. Due to the difficulties of using the present GARP manual we attempted to provide a more simplistic step-by-step approach to using the GARP system.

Introduction:

The GARP System is a modelling program used to predict the extent of distribution range of wild plant species (User Manual). GARP stands for Genetic Algorithm for Rule-set Production, meaning that it creates ecological envelopes for species based on presence, in the form of point localities and fixed environmental variables, which determine species distribution limits (User Manual). These envelopes are graphically represented as maps, which depict these distribution limits.

Although the GARP system is used for wild species, we were interested in determining the invasive potential of current invaders within the South African fynbos biome. As the GARP System does not discriminate between genetic and evolutionary traits it can determine distribution limits of both wild and introduced weed species. This report highlights the main methodologies used in the compilation of these species prediction envelopes.

Methods:

Jack-knifing:

To run the GARP System we downloaded the ‘JAVA J2SE Runtime Environment 5.0’ software which we acquired online from .com. The ‘DesktopGARPSetup 1 1 3.exec’ was downloaded from the ‘course documents’ window of the NISL website . The data was in the form of a ‘zip.’ File and so had to be extracted and unzipped, thus making it accessible to us.

For each species all 42 variables were run once through the GARP System with the following ‘Optimization Parameters’:

10 runs per experiment,
A convergence limit of .01,
And 1000 maximum iterations.

This data was then arranged into an excel spreadsheet for each species (raw data). This process is known as jack-knifing.

The relevant variable could then be selected via looking at the Omission and Commission values. The raw data for a relevant species was then selected. In the excel spreadsheet a new column was inserted to the right of the ‘Omission (ext)’ column, which was named the ‘Average’ column (Table 1). An average was then calculated for the first ten Omissions and Comissions. This was then repeated for every ten consecutive Omission and Comission values (Table 1).

A new worksheet was then created (Table 2) within this excel document and all the environmental data from the original worksheet was copied and transposed into the first column of this new worksheet. A reference column was then also created (Table 2) in this new worksheet in which the following values were inserted: 2, 12 ,22,32,42….using increment of 10 since this corresponds to the averages of the ten Comission and Omission values. Each value was then concatenated following which an addition column was created so that we could apply the indirect function to these values. They were then sorted in ascending order using the sort function. Ten of the most significant variables (with the least amount of skewed data Type I or Type ii errors, i.e. the lowest values) were used to draw up the envelope maps for the five species.

Table1 Illustration of format for calculating averages.

Commission / Omission (int) / Omission (ext) / Average
2.087145869 / 4 / 96.07843137 / 33.89300873
24.55697375 / 22.4 / 60.37735849 / 0
1.36672149 / 38.56 / 96.15384615 / 0
39.71831638 / 31.44 / 33.96226415 / 0
47.30709537 / 5.28 / 30.76923077 / 0
36.91074614 / 20.16 / 43.39622642 / 0
46.5959369 / 9.12 / 34.61538462 / 0
1.155922074 / 2.08 / 98.18181818 / 0
37.95779379 / 10.4 / 42.10526316 / 0
26.95452755 / 13.84 / 59.25925926 / 0
31.22842781 / 0 / 56.36363636 / 31.37545031
7.996478955 / 0 / 89.09090909 / 0
30.83231022 / 8.56 / 48.07692308 / 0
32.10405615 / 27.76 / 48.14814815 / 0
36.0744979 / 7.92 / 60 / 0
9.882091316 / 0 / 88.46153846 / 0
39.51678288 / 22.96 / 43.1372549 / 0
40.51286803 / 11.28 / 38.88888889 / 0
42.99149853 / 0 / 35.18518519 / 0
28.52278255 / 0 / 55.76923077 / 0
1.58215386 / 1.44 / 98.11320755 / 31.73116582

Table 2 The ten most significant environmental variables selected via jacknifing.

Environmental variable / Indirect / Reference / Concatenate
Altitude / 27.93458 / 122 / Results_Ac_cycl!AE122
Potential annual evaporation / 28.09638 / 162 / Results_Ac_cycl!AE162
CV of annual precipitation / 28.19046 / 172 / Results_Ac_cycl!AE172
Daily mean temp (Jan) / 28.36092 / 132 / Results_Ac_cycl!AE132
Daily mean temp (July) / 28.62257 / 262 / Results_Ac_cycl!AE262
Frost variability (STDEV # frost days) / 29.16688 / 272 / Results_Ac_cycl!AE272
Soil water stressJan) / 29.1833 / 282 / Results_Ac_cycl!AE282
Soil water stress (July) / 29.32602 / 222 / Results_Ac_cycl!AE222
Average daily heat units (Jan) / 29.58618 / 292 / Results_Ac_cycl!AE292
Average daily heat units (July) / 29.58956 / 72 / Results_Ac_cycl!AE72

Example of formulae used:

Concatenate column: =concatenate(“Results_Ac_cycl!Average value from Comission and Omission!”, Reference value), Thus a completed formulae would be: =CONCATENATE("Results_Ac_cycl!AE",C2).

Indirect column: =Indirect(concatenate value).

GARP run:

The GARP System was opened under the program ‘Desktop GARP’ (see Diagram 1). The data for the desired species was then uploaded using the ‘Upload Data Points’ function. This allowed us to browse the C – drive and open up the Program Files folder, in which can be found the subdirectories Desktop GARP – Sample Dataset –GARP Layers - Data – species information (i.e. Microsoft excel spreadsheet for Acacia Cyclops). The ‘Datasets’ function from the toolbar was selected followed by the selection of the ‘Scan Directory’ function. We then selected the SAAAC worksheet using the same former directory site. This then allowed us to select the SAAAC option in the Dataset dialog box. From here the previously selected ten most significant environmental layers (Table 3) could be entered for the GARP run.

The Convergent limit (Diagram 1) was set to .005 with 100 runs and 1000 iterations. The ‘All combinations of the selected layers’ option was selected. The Output directory selected was extracted from the Results folder of the above-mentioned Desktop GARP directory. The ‘model’ option from the toolbar initiated the run. The above running procedure was then repeated using the ‘All selected layers’ option in place of the ‘All combinations of the selected layers’ option.

Diagram 1 Desktop GARP window.

Table 3 The environmental variables selected for each species.

SPECIES
VARIABLE / 1 / 2 / 3 / 4 / 5 / 6 / 7 / 8 / 9 / 10 / 11 / 12 / 13
Altitude / x / x / x / x / x
Potential annual evaporation / x / x / x / x / x / x / x / x / x / x / x / x
CV of annual precipitation / x / x / x / x / x / x / x / x / x / x
Daily mean temp (Jan) / x / x / x / x / x / x / x
Daily mean temp (July) / x / x / x / x / x
Frost variability (STDEV # frost days) / x / x / x / x / x / x
Soil water stressJan) / x / x / x / x / x
Soil water stress (July) / x / x / x / x / x / x / x / x / x / x
Average daily heat units (Jan) / x / x / x / x / x / x / x
Average daily heat units (July) / x / x / x / x / x
Mean annual precipitation / x / x / x / x / x / x
Mean annual temperature
Mean relative humidity (Jan) / x / x / x / x
Mean relative humidity (July) / x / x / x / x / x
Median rainfall (July) / x
Median rainfall (Jan) / x / x
Minimum humidity (Jan) / x / x / x
Minimum humidity (July) / x / x / x / x / x / x
Mean of daily maximum temperature (Jan) / x
Mean of daily maximum temperature (July)
Mean of daily minimum temperature (Jan)
Mean of daily minimum temperature (July)
Potential Evapotranspiration (Jan) / x / x / x / x / x / x
Potential Evapotranspiration (July) / x
Positive chill units (May to September)
Rain seasonality / x / x
Solar radiation units (Jan) / x / x / x / x
Solar radiation units (July) / x / x / x / x / x / x / x
Temperature range (Jan) / x / x / x
Temperature range (July) / x / x
Fertility 1 (very low)
Fertility 2 (low)
Fertility 3 (medium)
Fertility 4 (high)
pH1 (acidic)
pH2 (neutral)
pH3 (alkali)
Texture 1 (fine) / x
Texture 4 (very coarse) / x
Texture 3 (coarse)
Texture 2 (medium)
Terrain morphology / x / x

Key

1 = Acacia cyclops
2 = Acacia longifolia
3 = Acacia melanoxylon
4 = Atriplex
nummularia subsp. nummularia
5 = Eucalyptus camaldulensis
6 = Hakea sericea
7 = Metrosideros excelsa
8 = Nicotiana glauca
9 = Prosopis glandulosa
Tarr. Var. torreyana
10 = Pinus pinaster
11 = Populus X canescens
12 = Schinus molle
13 = Acacia mearnsii

Idrisi32:

Initially, the ‘Compiler.jar’ programme had to be downloaded from the NISL website, and saved within the ‘Desktop GARP’ folder (in the C-drive). The compiler program was then opened and the ‘Add’ function selected. The desired species could then be selected by browsing through the results folder. The ‘Start Run’ function was then selected within the ‘GARP Utility’ window. We then re-opened the Results folder as the previous results had to be placed into the correct folder (i.e. correct species folder).

Then using the ‘Start’ function, followed by the “Programs’ function, the IDRISI-32 program was initiated. In the File menu the ‘Data Paths’ function was selected. The ‘Browse’ function was then used to locate the correct species folder again. We then again opened the ‘File’ menu and selected the import function followed by the following sequence of prompts:

Software-Specific Format;
ESRI Formats;
ARCRASTER.

This opened the ‘ArcInfo Raster Exchange Format’ window. The ‘ArcInfo raster ASCII format to Idrisi’ option was selected. The input file was the ‘asci’ file for that species (i.e. Ac_cycl.asc). The output file consisted of only the species name (i.e. Ac_cycl). An output reference was also selected. The reference system used was ‘latlong’. The 'O.K.' button was then pressed. The 'O.K.' function in the ‘ArcInfo Raster Exchange Format’ window was then selected to generate a final species prediction map.

Boolean mapping:

There were two Idrisi files per species:

Idrisi Raster Documentation File,
RST File.

These were then each placed into separate folders, with the title of the plant species. This was done for convenience sake. The RST file was opened, from which a Boolean map could be generated. To generate the Boolean map in the Idrisi programme we started by selecting the ‘Analyses’ function, from the tool bar. This was then followed by selecting the 'Database Queries' function, followed by the 'Recclass' function.

Within this working dialog box the following procedure was carried out:

Within the input file selection: The browse option led us to search for the Idrisi file of the species in question.
Within the output file selection: The output file can be assigned any name, and must be directed to the original species folder.
We used the Recclass parameters specified in table 2 below:

Table 2 Recclass parameters and format assigned to Boolean maps.

0 / -9999 / 66
2 / 66 / 100

References:

1) Desktop GARP manual

2) Henderson, L. 2001. Alien weeds and invasive plants. A complete guide to declared weeds and invaders in South Africa. PPRI Handbook No, 12.

Acknowledgements:

Thanks to Richard, James, and Russel for all the help.