Water hyacinth occurrence and satellite derived covariates for South Africa
Authors/Creators
-
1.
Stellenbosch University
-
2.
University of the Witwatersrand
Description
Table S1. Feature descriptions, associated units, temporal coverages, and data sources of the features considered to investigate the likely drivers of water hyacinth occurrence. All explanatory variables were downloaded at a 30 m spatial resolution. Those variables that were only available at a coarser spatial resolution (> 30 m) were automatically resampled to 30 m using bilinear interpolation within Google Earth Engine.
|
Variable |
Units |
Temporal coverage |
Source |
|
Number of consecutive nights with less than 10 degrees Celsius |
days |
2013 |
MODIS LST- night |
|
Area of waterbody |
m2 |
2013 |
(Singh et al., 2020) |
|
Median river width at mean discharge |
m |
1984 - 2018 |
Global River Width from Landsat (GRWL) (Allen & Pavelsky, 2018) |
|
*Area of aquatic vegetation |
m2 |
2013 |
(Singh et al., 2020) |
|
*Minimum Temperature in the coldest month between1970 and1990 |
°C |
1970 - 2000 |
WorldClim |
|
Connectivity- upstream and downstream river count |
river count |
2000 |
WWF hydrosheds (Grill et al., 2019) |
|
*Water seasonality- number of months water is present |
month count |
**1984 - 2019 |
JRC GSW (Pekel et al., 2016) |
|
*Total precipitation |
mm |
**1958 - 2021 |
TerraClimate |
|
*Distance to nearest coastline |
m |
2009 |
NOAA (https://oceancolor.gsfc.nasa.gov/docs/distfromcoast/) |
|
*Distance to roads |
m |
1979 - 2010 |
Global Road Infrastructure Project (GRIP) (Meijer et al., 2018) |
|
Elevation |
m |
2000 |
NASADEM |
|
*Continuous Heat Insolation Load Index (0=cool, 255=warm). Surrogate for effects of shading and topographic insolation |
0-255 |
2006 - 2011 |
SRTM, (Theobald et al., 2015) |
|
*Global Human Modification (1= high modification) |
0 - 1 |
2016 |
(Kennedy et al., 2019) |
|
*Topographic diversity- Surrogate for the variety of temperature and moisture conditions available to species as local habitats |
-1323 - 8.81 |
2006 - 2011 |
SRTM, (Theobald et al., 2015) |
|
Frost duration - Median duration of frost |
day count |
2007 |
(Schulze & Maharaj, 2007) |
|
Number of days below 0°C (0-365) |
day count |
**2000 - present |
MODIS- LST, night temperature |
|
Number of days below 10°C (0-365) |
day count |
**2000 - present |
MODIS- LST, night temperature |
|
*South African National Land Cover (73 class) |
km2 |
2013 - 2014 |
GeoTerraimage |
|
*Mean Annual Runoff |
mm/year |
2005 |
Strategic Water Source Areas (SWSA) |
|
*Broad (10) class Landcover |
% (cover fraction) |
2015 |
Copernicus Global Land Landcover Layers (CGLS-100)(Buchhorn et al., 2020) |
|
*Flood hazard with a 10-year return period |
water depth (m) |
2016 |
Joint Research Commission (Dottori et al., 2016) |
|
Riparian soil nitrogen (5km buffer) |
g/kg |
1905 - 2016 |
Soil grids |
|
Riparian soil ph (5km buffer) |
pH |
1905 - 2016 |
Soil grids |
|
*Riparian soil carbon (5km buffer) |
g/kg |
1905 - 2016 |
Soil grids |
|
*Stream Power Index (SPI) |
-1 - 1 |
1987 - 2017 |
GeoMorpho90 GeoMorphometric layers (Amatulli et al., 2020) |
|
*Relative wealth index |
-1 - 1 |
2001 - 2019 |
Facebook (https://dataforgood.fb.com/tools/relative-wealth-index/) |
|
13 iSDA Soil layers (includes *total nitrogen, etc.) |
various |
2013 - 2019 |
(Hengl et al., 2021) |
|
*Mean wind speed for a 10-year period |
m/s |
2008 - 2017 |
Global wind atlas |
|
* Included in final model, ** Multi-temporal data - 2013 or temporally closest data was selected
|
Abstract (En)
Despite a long history of management efforts targeting invasive water hyacinth, Pontederia crassipes, their success has been inconsistent. Identifying the most appropriate management strategy remains challenging and is shown here to be contingent upon the ecological and socio-economic context in which the weed proliferates. Motivated by the advent of numerous Earth Observation (EO) derived map products, and recent advances in explainable Artificial Intelligence (xAI), we discerned the environmental and socio-economic contexts impacting water hyacinth prevalence across multiple spatial scales in South Africa. We used SHaply Additive exPlanations (SHAP), a model-based approach, to interpret the response of water hyacinth to an array of biotic and abiotic factors captured by a Species Distribution Model (SDM). Consistent patterns emerge between modelled plant responses and known physiological constraints limiting water hyacinth growth, such as minimum temperature. Additionally, our findings revealed novel spatial patterns with climatic factors in coastal regions and with vegetation type in the country's interior. These findings have the potential to facilitate future laboratory and field-based investigations aimed at corroborating and elucidating the fundamental mechanisms underpinning novel non-linear trends and invader-covariate interaction effects. The spatially explicit outcomes presented here, covering all South African waterbodies, serve to facilitate risk-prioritisation and the selection of context-specific management strategies based upon a data-driven approach. This study demonstrates the power of combining EO data, SDMs, and xAI techniques to enhance our understanding of invasive species dynamics and inform adaptive management practices.
Methods (En)
The data has been extracted from Google Earth Engine for all South African waterbodies detectable in the South African subset of the Global Surface Water dataset for 2013. Refer to code and research paper for more details.
Files
merged_31082021.csv
Additional details
Funding
- FitzPatrick Institute of African Ornithology
- 40470
Software
- Repository URL
- https://github.com/Geethen/PhD--EO_and_ML_for_water_hyacinth_management/tree/main/Chapter_2
- Programming language
- Python
- Development Status
- Inactive