Visualizing and interpreting the results

Visualizing the results is an important step in any Zonation analysis. You will want to look at your results immediately after the analysis has finished to verify that the analysis seems to work technically. Visualization is also needed to interpret the results. The basic interpretation includes things such as analyzing the spatial patterns of priorities and the performance of different biodiversity features. At some point, you may also want to take a closer look at the results and do further quantitative analysis either by using the facilities in Zonation or some external tools. This section introduces the basics of visualizing Zonation results as well as important points related to interpreting what you are seeing. If you are working with a real-life conservation planning project, visualizations alone may not be enough to provide the support that is needed. Instead, the results alongside with informative visualizations can be developed into planning products that are relevant and easy enough to understand by the intended users. The exact form and function of such planning products will vary case-by-case, but some considerations are given in Section 7.2.

Standard Zonation outputs and their interpretation

There are two main outputs produced by Zonation: 1) Priority rank maps and 2) performance curves. Almost all other results are derived from these two outputs and consequently the interpretation of the results is based on them. Here we give a short description of each of the outputs, for a more thorough explanation please see the Zonation manual (Moilanen et al., 2014) and the “Quick introduction to Zonation” document (Di Minin et al., 2014).

The priority rank map is the first main Zonation output visualization. By default, Zonation outputs a raster file with the value in each cell indicating the relative conservation priority of that cell. These values are derived from the order of iterative cell ranking (removal). Each raster cell has a value between 0 and 1, meaning that values close to 0 were removed first (low relative conservation value and priority), while high values close to 1 were retained towards the end (high relative conservation value and priority). Assigning a suitable colour scale to these values gives rise to a rank priority map in which the colours indicate the conservation priority over the area of interest (Figure 3). This type of map is quite easy and intuitive to interpret for most people, but it is good to bear in mind that the map – as all Zonation results – shows conservation priorities relative to the data and the model of spatial prioritization that you used. The priority rank map does not necessarily tell you anything about the absolute conservation value over your area of interest, although usually it does: it makes little sense to run a Zonation analysis using data that has no relevance for conservation.

Figure 3 An example of a priority rank map (from Pouzols et al. (2014)). This priority map shows which areas are the globally the most important (and least important) for protected area expansion, based on analysis taking into account the distributions of c. 24 000 terrestrial vertebrates. The red color indicates higher priority for protection. The original pixel values (0-1) have been converted to percentages (0-100%) corresponding to the priority fraction of landscape to be protected. In terms of visualization, the colors have been selected to print out well both on screen and on paper. The projection has been set to equal area (Eckert IV) to retain the comparability at different latitudes. The latitudinal bar chart describes how large a fraction of the current protected area network (grey) and the top priority areas (red) is located in that latitudinal bin.

Performance curves, the second main output of Zonation, are automatically produced and exported as text files for each feature (or group of features) during a Zonation analysis. The content of performance curves files can be visualised as a curve that quantifies the proportion of the original occurrences retained for each biodiversity feature, at each top fraction of the landscape chosen for conservation (Figure 4). Performance curves start from 1.0 because the full landscape includes the full distribution of the biodiversity features. At the other end, no areas are chosen, and correspondingly, the protection level for the feature is zero. Because the number of these curves can be high, it is common to average and visualize curves across feature groups, such as taxonomic groups. Average curves usually are concave because the initial aggregate losses for biodiversity are low (low-priority areas, such as densely populated urban areas contain relatively little biodiversity), but aggregate losses unavoidably accelerate when moving to high-priority areas with high feature richness, rarity and local occurrence levels.

The priority rank map has direct correspondence with the performance curves - top-priority areas selected from the priority rank map include feature representation summarized by the respective performance curves (Figure 4). For example, if we select the top 10% of the priority rank map, we can evaluate the corresponding representation for each biodiversity feature or for broader groups through the performance curves.

Figure 4. An example of performance curve visualization (from Pouzols et al. (2014)). The figure shows how the biodiversity features perform on average in the priority ranking presented on a map in Figure 3. In this visualization, the background colouring corresponds to those used on the map for different proportions of the terrestrial area. The average performance across the 24 000 input features is shown with a turquoise line (Global priorities, future 2040). To compare the performance of this Zonation prioritization result with three other variants, mean curves for all have been plotted to the same figure. The figure reveals that the current protected areas (11 % of the surface area) protect on average 19 % of the input biodiversity feature distributions. The next 6 % of land surface area (up to 17 %) in the prioritization order would cover on average 54 % of the species distributions. The figure also shows that land use change expected by 2040 reduces the species distributions with 12 % from their current extent (impact of the use of a so-called condition layer in the analysis).

Creating visualizations

You can create visualizations of Zonation results in two ways: Using the Zonation GUI or some external tool. The Zonation GUI provides multiple tools for visualizing and inspecting the results as priority maps and performance curves (Di Minin et al., 2014). If you are using the GUI (Zonation can also be run from the command line), examining both the rank priority map and the performance curves in the GUI should be your first stop in visualizing your results. You can also compare the performance of different features or groups in your analysis by using the interactive plots functionality in the GUI. The merged map tool allows you to make simple visual and quantitative comparisons between the distributions of priorities in different runs. In case you are running Zonation from the command line only, Zonation will produce an image file of the priority rank map using a pre-set colour scale, but the performance curves you will have to plot yourself. Consult Chapter 4 in the manual and Chapter 10 in the “Quick introduction to Zonation” document for an in-depth description of Zonation’s built-in visualization capabilities.

For a more detailed analysis of the results, you need to be able the dig deeper into the results themselves. You can do this outside the Zonation environment by working with the numerical outputs produced by Zonation (see Section 3.4 in the Zonation manual). The priority rank rasters (files ending in .rank.tif) and weighted species range size rarity rasters (files ending in .wrscr.tif) are usually visualised using a GIS tool such as QGIS or ArcGIS. The performance curve files are numerical tables that you can processes and visualize using some type of statistical software such as R or Excel. The textual outputs created by Zonation, such as the performance curves, are primarily designed to be human- rather than machine-readable. This means that getting the data into some other software takes a little care, but can be done. If you use R, it might be worthwhile to get familiar with package zonator introduced in Section 6.1.2. Zonator includes functionality for automatically parsing all the common Zonation outputs into useful data structures in R. It can also easily visualize many of the outputs and it provides some analytical tools for the outputs.

As mentioned, the priority rank map and the performance curves always go hand in hand and almost every representation of Zonation results should include both. If you can, it may also be a good idea to link the rank priority maps with the performance curves to emphasize that they are closely connected (Figures 5 and 6).

Figure 5. Example of a visualization showing how the rank priority map and the performance curves produced by Zonation can be visually linked. For the explanation of the priority rank map and the performance curves in this figure, see Figure 3 and Figure 4.

It is good to remember that while the initial visualizations of the Zonation results can be done quickly, fine-tuning them to meet specific communication needs usually takes more time than you would expect. In addition, normal rules of cartographic and statistical communication apply here. Therefore choosing the presentation style for the intended audience and media is important (see Figure 6 for an example). Equally important is to aim at consistency between different representations of the same results (e.g. maps and performance curves).

Figure 6. An example of a visualization showing how different area reservations in a plan (protected areas and recreational areas) perform in biodiversity protection and where additional green areas should be planned. The visualization combines performance curves and the map, and tries to make some relevant settings in the Zonation process visible (connectivity, masking with waters and the source data). Adapted and translated from Kuusterä et al. (2015).

6.2.3 Visualising differences between Zonation variants

Assuming that you have followed the suggestions of developing different analysis variants, you may be wondering which variants should go into the final visualisations. In case you have arrived at one or a few production runs (i.e. the “final” versions), then those are typically the ones that you base your visualizations on. Note however, that sometimes it is useful to highlight particular features of the analysis using development variants. Say you were developing your Zonation analysis using a sequence (as in Section 5.1 starting from page 35). The first type of connectivity (distribution smoothing) is introduced in variant 4 and the second type of connectivity (interaction connectivity) in variant 6. Variant 7 has all the analysis components in it and can be considered a production run. While your final visualizations would probably be based on variant 7, you might still want to create a separate map from variant 4 to demonstrate the effect of connectivity. Which analysis components you should want to demonstrate is up to you. However, try to keep it simple. Including everything will make the product large and complicated, and might confuse the main message.

There are many ways you can visualize the differences between Zonation variants. You can compare, for example, the performance curves of different variants (e.g. Figure 4). Another common option is to make a map series showing the spatial variation in the priority rankings (Figure 7). If you are interested in a particular fraction of the priority ranking (e.g. the 17 % of the landscape in Figure 8) you can examine the spatial overlap and/or similarity of different variants in terms of the fraction of interest. For example, the Jaccard index (Warman et al., 2004) and fuzzy similarity index (Mas et al., 2012) can be useful in inspecting the similarity between priority patterns of different variants.

Figure 7. An example of visualisation intended to allow the user to evaluate outputs of two different Zonation variants spatially. Panel A shows a variant without connectivity and panel B with connectivity between different forest types in the landscape. The visualization and the close ups demonstrate how the overall pattern stays the same between the runs, but the more detailed pattern of priorities changes. The marginal plots on top and on the left side of each panel can facilitate the comparisons. They show the count of cells in the top 10% of the landscape along both latitudinal and longitudinal gradients. Adapted from Lehtomäki et al. (2015).

Figure 8. Example of visualisation to compare the top 17% of two different Zonation runs, one made globally and the other made nationally using the administrative units feature of Zonation. In practice, the continuous priority maps have been converted into a binary 0/1 surface before the visualisation. The most distinctive colour (red) has been chosen for overlapping priorities (value 1 in both Zonation variants). Figure from Pouzols et al. (2014).

While group-based performance curves can be very informative, you need to think a little about what exactly you want to visualize. If grouping functionality is enabled, Zonation gives you the following statistics on the representation levels for each group: minimum, mean, weighted mean and maximum. In many cases, you are probably interested in the mean representation levels, but in some cases looking at, for example, the minimum representation level (i.e. the biodiversity feature with the worst performance in that group) would make sense. Summarising information into a statistic such as the mean will also hide other types of potentially useful information. For example, it may be useful to visualize the distribution of individual performance levels within the group from the feature-based performance data (Figure 9).

Figure 9 Boxplots are a useful way to visualize and compare the distributions of feature-specific performance levels within groups. This example presents the distributions of species ranges covered by expanding the global protected area network to 17% of the global terrestrial area based on Zonation prioritization (Pouzols et al., 2014). Results are grouped by species in different taxons on the left and by IUCN Red-list categories on the right. The colour indicates one of two scenarios corresponding to those in Figure 4.

6.2.4 Providing context for interpretation

Providing sufficient context for the results is crucial for the usability of the results. It is important to bear in mind that rank priority maps and performance curves tell very little about how they were produced and it is unfortunately easy to interpret them out of context. In our experience, maps produced with a specific model of spatial prioritization easily start to represent “generic” conservation value of the areas covered by the analysis. In other words, a prioritization based on the occurrence of valuable forest habitats can be implicitly and mistakenly expanded to represent a broader set of biodiversity features (e.g. peatlands, freshwater habitats etc.). It is therefore very useful to combine additional information to your visualizations providing information on what is the context of the analysis and its results.

Often the intended users, such as land-use planners or other implementers, also need more detailed information on individual planning units. Combining the information produced by the post-processing (Section 6.1.1) analyses to selected planning units (cadastral units, forest stands, individual protected areas or whatever other spatial delineation) will give direct and accessible information on a given location’s priority rank. This information is especially useful for illustrating the range-size normalization that is built-in into Zonation. For any given planning unit it is easy to check what features occur there. What is less obvious, however, is how large a fraction of the overall occurrences of that feature is contained in the planning unit. Both these types of information, what features and what fractions of their overall occurrences are within the planning unit, are given by suitably tuned post-processing analysis. The information can then be combined to other spatial data in a GIS or turned into other planning products such as unit-specific description cards (Figure 10). This type of a planning product also reminds whoever is using it about data and analysis features that were used, thereby summarizing the all-important planning context.

Figure 10. An example of a site information card created for top each priority area in a real-life planning case in Finland. The card proves information on why the site has been selected as a top priority area, and what biodiversity features it contains at high densities compared to the study area in general. In addition to the broad scale priority maps and performance curves, such visualizations are useful when decisions over individual land parcels will be implemented on the ground. This example card was adapted from Kuusterä et al. (2015) and has been simplified for illustrative purposes.

Including information on the input data and some key parameters to the visualisations will also remind the user that the results cannot be generalised beyond their context. An example of such a visualization is given in Figure 6, where the data sources and some settings (use of connectivity) are presented as an integral part of the visualisation. Whenever possible, it is good to try to visualize also the analysis process together with the results to make the user of the results aware of choices made. This can be done, for example, by using flowcharts that reveal the input data sets, different steps of the process, and the differences between Zonation analysis variants (Figure 11).

Figure 11. An example of a flowchart visualising a Zonation analysis presented in a scientific article (adapted from Lehtomäki et al. 2015). The flowchart works with the text and describes in detail the different input data sets, Zonation variants and the outputs.