Remote sensing

Open data for Metaldehyde risk

This is a proof of concept and has not yet been worked into an operational product.  It’s a first pass to demonstrate the utility of open data. If you’d like to work with Geoger to develop this further then please get in touch.

The idea

Metaldehyde is a pesticide found primarily in slug pellets and associated products. I have heard a lot about metaldehyde as a pollutant of water courses in recent months and I wanted to know which open datasets could be used to create workable maps of potential risk areas. This blog post reports on the successful implementation of an initial method.

The data

Landsat-5 satellite imagery was obtained for an area covering Hampshire, UK. The crop of interest for this study is oil seed rape and the simplest way to recognise this is to use imagery captured when the crop is in bright yellow flower in late April / early May.  The image used was captured on 30 April 2011 and a subset was created (see Figure 1) that related to four Ordnance Survey OpenData OS Terrain 50 tiles. OS Terrain 50 is an elevation dataset provided at 50m resolution (Figure 3 – black relates to low elevations and white to higher values). Further OpenData layers were also downloaded to provide water body information and wider basemap context (Figure 2 – roads, watercourses, woodland and buildings are plotted on top of the satellite imagery). Open aerial imagery and elevation data were obtained for the corresponding area from the Hampshire Hub. The aerial data were collected during the summer of 2013.

The method

Open source GIS software was used  for all the data processing in this trial. The .asc elevation files were stitched together into a mosaic as were the true colour aerial photographs. Oil seed rape fields were identified in the Landsat imagery using a simple classification scheme to keep the processing quick and easy (Figure 4). The waterbody layer was buffered to 100m and areas of slope greater than 4 degrees were calculated from the 50m OS Terrain data (Figure 5). These were then overlain and areas of overlap used to highlight potential risk areas (Figure 6). The images in the gallery show some of these processing steps – click on an image to see a larger version.

Once the areas of potential risk had been identified for an area on the Landsat image, the aerial photography was loaded for one of the risk hotspots (Figure 7). Orange areas show moderate potential risk and red areas show highest potential risk. The field boundaries for the oil seed rape fields identified in the satellite imagery were digitised from the aerial imagery and the analysis was re-run using the finer resolution elevation information. This then shows within field risk areas, allowing pesticide application to be better monitored and controlled (Figure 8).

In summary

This work takes open datasets as the source of the potential risk model parameters. The aerial imagery is used to create detailed field boundaries and higher resolution  slope information. Ordnance Survey data are used to provide the location of watercourses (which is the impacted environment) and a lower resolution elevation dataset for regional analysis. As the method stands, the satellite imagery is the only dataset that would require updating on an annual basis. This would be to obtain crop details and update maps of risk for a specific growing season, based on which fields are growing high risk crops. The results are different from many existing studies as they show within field risk areas, as opposed to just classifying a field as ‘at risk’ or not.

Taking it further

This is currently a broad brush approach, successfully demonstrating the utility of open data and open source software for investigating real world issues. Information that would improve the methodology include soil type, additional crop types (known for metaldehyde risk), time of year (for metaldehyde application risk) and rainfall (to improve real-time risk potential). The results would also need to be validated prior to the method being operationalised. Automating some of the steps will also help provide a more efficient model.

Other information

Many of the parameter values used in this work were obtained from other studies. The following presentations are worth reading if you are interested in this subject:

  • Yorkshire Water catchment management document.
  • Anglian Water, Affinity Water and Essex and Suffolk Water farming document.

Also have a look at these websites: