Sample Design Tools
Sampling Design for Rangeland Monitoring and Assessment
One of the biggest challenges that rangeland researchers and managers face is how to gather the necessary data for decision-making as efficiently and accurately as possible. Often we know how we want to collect the data – meaning what method to use – based on objectives, but where to sample is an altogether different problem that has a large impact on not only efficiency, but in terms of how (or even if) the data can be generalized to other areas.
Tools for Calculating Sample Size and Power
Multi-scale Sample Requirements Evaluation Tool (MSSRET) - web version: A web-based tool for determining the appropriate number of samples needed to meet monitoring or assessment objectives. MSSRET uses pilot sample data or data collected from similar areas to calculate sample sizes needed to detect a specified minimum detectable difference. MSSRET can also be used to estimate the power of actual monitoring data to detect a specified amount of change.
- Microsoft Excel version of the MSSRET tool.
- The Rangeland Assessment and Monitoring Methods Guide has descriptions of and links to additional sample size and power calculation tools.
Tools for Sample Design
- Jornada Sampling Tools for ArcGIS - An ArcGIS Toolbox developed by the Jornada Experimental Range with a number of tools in it for selecting or generating random samples within ArcGIS. These tools were initially developed to do unequal probability sampling (both simple and stratified), but more basic sampling tools have been included. More sampling tools will be added to this toolbox over time.
- The Rangeland Assessment and Monitoring Methods Guide has descriptions of and links to additional sample design tools (e.g., Hawth's Tools, Spatially-balanced sample design tools).
- Web-based Sample Design Tools: Coming soon...
Tools for Analyzing Sample Data
- Unequal-Probability Sample Analysis Tool - An Excel-based tool (using the Horwitz-Thompson estimator) for calculating descriptive statistics from monitoring and assessment data that was collected using an unequal probability sample design.
- The Rangeland Assessment and Monitoring Methods Guide has descriptions of and links to additional data analysis tools.
- A web-based Assessment and Monitoring Analysis Tool: Coming soon...
Sample Design Discussions
The Rangeland Assessment and Monitoring Methods Guide has a growing collection of wiki-page reviews of sample design concepts and tools.
- Review of sample designs commonly used in rangeland monitoring and assessment.
- Inference space - what it is and why it's important to you.
- Why stratify? When and why stratification is helpful for designing monitoring and assessment programs.
- When key areas aren't anymore - the case for randomized sample site selection.
- Additional references and links for practical rangeland monitoring and assessment sample design.
Sample Design using Object-based Image Analysis (OBIA)
The basic idea behind using OBIA for sample selection is that an image segmented into polygons, or objects, at the right scale captures the pattern and distribution of vegetation community patches on the landscape. If we used an appropriate scale in segmenting the image into objects, we may assume that the objects are homogeneous. The graph below shows that sample plots that stradle image objects have higher variability than those that are mostly or entirely within a single object. This allows us to treat them as sampling units, and in order to measure that polygon, we need only to be able to navigate to it and stay away from the boundaries when sampling.
Graph of average shrub cover variance within field plots by the proportion of the plot that is within a single image object. Bars on the left represent plots that overlap multiple image objects. Bars on the right are for plots mostly (or entirely) within a single image object.
Another powerful aspect of using image objects for sampling comes when sampling is used to estimate rangeland condition for a large area. Because every place in a landscape is part of an image object whose area is known, we can determine the probability that any object is selected for sampling. This is important because if we know the selection probabilities, we can adjust them based on factors like study objectives or logistics (e.g., distance to roads) to make sampling more efficient. The altered selection probabilities are then used in calculating the estimate to “unbias” the result.
Stratification is another common approach that is used in selecting sampling locations in order to reduce variation between samples and divide a landscape into more homogeneous units. The best strata are based on characteristics of the landscape, and this is an area where coarse-scale image objects derived through OBIA can help greatly.
In order to use OBIA for sample design, it is necessary to have a set of image objects created using an OBIA software application. Presently, however, access to OBIA software is very limited, largely due to cost. We are working on developing web-based tools that implement this approach, and in the mean time have created sets of image objects and used existing sample selection tools (e.g., the Jornada sample design tools above, R, Hawth's Tools) to design sample schemes for specific projects.