While the overall strategy often takes the form of the generic workflow, as illustrated in Figure 2, the data analysis and sampling design depend strongly on the situation, the available data and the specific goals and constraints. We however attempt to bring some structure to the typical workflow here, by dividing the process in a pre-processing step, the exploratory data analysis, the actual data analysis, potentially a post-processing step and the sampling design. Checking if the objective is achieved, and whether the constraints are violated, is of course also necessary at some point. The pre-processing step is still relatively straightforward, and compasses checking for errors, outliers, making possible corrections and removing parts of the data irrelevant to the problem at hand. The exploratory data analysis can also still be structured in a way that it is applicable to most problems in D&D for constrained environments, by looking into four aspects of the data:

  1. Is this a univariate or multivariate problem?
  2. Is this a problem involving spatial trends?
  3. Is this a problem involving spatial structure?
  4. Is this data requiring robust methods?

We use the outcome of the exploratory data analysis, i.e. the answers to the above questions, to bring some structure in the range of methods for the actual data analysis, possibly applicable to the problem at hand. As the range of situations and methods is vast, we cannot discuss every possible road to selecting a specific approach. This only provides some guidance on the type of methods to use, and the expertise of the user of this strategy comes in at this point to make a final decision, based on the individual method descriptions, suggestions and remarks contained in this document. A post-processing step can be required to translate the obtained results into the required information for checking if the objective is achieved. If it is not, a sampling design approach has to be selected. Similar to the data analysis, we can also only provide general recommendations here, and use a classification in terms of:

  1. Is this sampling design probabilistic or not?
  2. Are the selection probabilities equal or not?

We do try to relate the different approaches to the four, above defined, aspects of the data as well, in the list of approaches in Section 5. When a sampling design is proposed, a final check on violation of the defined constraints should be made, before moving to the corresponding characterization campaign. The different steps outlined here are further discussed in detail one by one below.