Ground cover monitoring for Australia: Sampling strategy and selection of ground cover control sites: appendicies and references

Acknowledgments

The authors thank all participants at the Expert Workshop (17–18 August 2010) for their involvement and their ongoing input to achieving a validated remotely sensed fractional cover product for Australia. The workshop and this report were funded by the Australian Government Department of Agriculture, Fisheries and Forestry.

Executive summary

High frequency, spatially explicit ground cover information is required to improve wind and water erosion modelling and monitoring, and our understanding of the impact of climate variability and changing management practice on pasture condition, particularly in the rangelands. Ultimately, a national network of sensor independent reference sites is required to improve validation of land cover products derived from remotely sensed data and which are monitored using agreed standardised sampling approaches.

To meet these needs the Australian Government Department of Agriculture, Fisheries and Forestry has funded delivery of a validated remotely sensed time–series of fractional cover for Australia using MODIS satellite data based on the method of Guerschman et al. (2009). The fractional cover product will provide monthly ground cover data to better parameterise the CEMSYS (Shao et al. 2007) wind erosion model and the SedNet (Wilkinson et al. 2004) water erosion model.

The fractional cover dataset is derived from two indices which provide information on vegetation amount and colour. Plotting these indices against each other, the amounts of three cover components – for each satellite pixel – can be estimated, namely:

• photosynthetic (green) vegetation cover
• non–photosynthetic (brown) vegetation cover
• bare soil.

Before it can be routinely used the fractional cover product needs to be validated. Validation will establish confidence in the approach and encourage operational use of the data products. The successful operational use of medium resolution data will help build a case to fund work based on sensors operating at higher spatial resolution. To date, only a qualitative analysis of algorithm retrieval and accuracy of the fractional cover product has been undertaken.

Defining the uncertainty of the cover fraction estimates requires quantitative validation of the fractional cover product. Users need to know whether the product (and version) is suitable for their purpose and/or region. Defining uncertainty enables: assessment of its impact as a data input into other models (e.g. CEMSYS and SedNet); informed decisions on the basis of the data itself and; ultimately improvement in the predictions through modifying the fractional cover algorithm by understanding the possible sources of the error.

A limited qualitative evaluation showed that the fractional cover algorithm works well in areas with high greenness, such as the intensive land use zone. This includes land covers such as cropping, woodlands and aquatic vegetation. It performs poorly for some soil types – gibber and to a certain extent red, black and bright soils – and low vegetation covers, particularly those typical of the rangelands. This lack of performance for these surfaces may reflect their absence from the area upon which the end–member dataspace was developed – the Northern Territory savanna. It could also be that differences in soil spectral reflectance may be influencing fractional cover estimation using the algorithm.

An Expert Workshop on Sampling Strategy and Selection of Ground Cover Control Sites was held at CSIRO Land and Water, Canberra in August 2010. The workshop brought together experts from federal and state agencies with the aim to:

• develop a statistically robust sampling strategy for ground cover validation sites
• prioritise target areas for field measurement of fractional cover
• validate the fractional cover product.

To implement the sampling strategy a national network of field sites will be established through collaboration with state agencies. This report elaborates on the main recommendations made at the workshop.

The required precision of the satellite fractional cover data was established as ± 15% of the bare ground component. This is the level of precision required by the erosion models.

Currently it is not possible to quantitatively estimate the number of validation sites required to achieve the desired precision. This requires field knowledge to establish actual variability in the fractional cover data. Previous experience of validating other national satellite datasets support a minimum of 1500 sites to achieve credible validation for the fractional cover product.

A systematically distributed sampling approach is recommended. This is guided by 8 principles to ensure sampling effort is representatively spread across priority areas, is geographically distributed, encompasses ground cover variability and meets requirements for homogeneity at the MODIS scale. These principles are:

• sample all non–woody vegetation types used for grazing and broadacre cropping
• target field validation effort at 90% in rangeland areas and 10% in broadacre cropping areas
• field sites to have less than 12% foliage projected cover or 20% tree canopy cover
• sample the full range of ground cover from 0 to 100%
• field sites to be spatially homogeneous at the MODIS scale
• target key soil colours: gibber, red soils, black soils, bright soils, and others
• consider other issues such as soil moisture, timing of sampling and the need for repeat visits to sites
• ensure an adequate number of sites in each priority environment.

The sampling strategy should be iteratively implemented with annual reviews. Initially the focus should be on wide spatial coverage across the priority cover ranges and soil colours and on the analysis of existing ground cover data from Queensland and New South Wales. Sampling effort should be reviewed annually to assess both the validity of the estimated 1500 sample size and the impact of current ground cover sampling effort on the overall uncertainty of the product. Sampling should be adapted to allow visits to areas poorly represented in field data already obtained, to assess temporal variability and the inclusion of future sampling sites.

As field validation data becomes available, the number of validation sites required to establish confidence in the data will become increasingly apparent. A data simulation approach based on the variance between field observations and the satellite data is recommended to achieve this.

Errors due to 'scaling', caused by differences between the scale of ground–based validation measurement and the 500m MODIS pixel may be important source of uncertainty in the validation process. To understand these errors a method is proposed using higher resolution Landsat satellite data as an intermediate step. It is suggested this validation approach will reduce scaling errors.

The sampling strategy addresses the need to achieve confidence in the data while addressing the practical challenges of achieving rigour in a purely statistical sense. Rigour is addressed by consistency, encapsulating:

• a standard methodology for the identification of priority areas for site visits
• guidance to field teams on the selection of specific sites to sample
• a standard methodology for acquiring field data by the state–based validation teams, including consistent training in the method and common guidelines
• refinement of the sampling effort after annual review as new validation information becomes available
• a statistical methodology to better estimate the number of samples potentially required as field validation data is acquired.