Introduction
It may come as no surprise that identifying ideal locations for the construction of a new wind farm is not as trivial as finding the nearest open field without a towering wind turbine. In fact, identification of the most suitable locations requires knowledge of a whole plethora of characteristics pertaining to the site in question. Some of these are regulatory, “Are we far enough away from the nearest town?”, “What degree of natural protection is prescribed for this area?”. Some of these are practical “Is the ground suitable for construction?”, “Does this area get enough wind to make a new wind turbine profitable?”. Answering this set of questions accurately, as prescribed by key stakeholders, is a prerequisite for the success of any large construction project. In this project, we turned to leveraging climate data, earth observation imagery and regulations, in order to efficiently identify areas in Salzburg that fulfilled a set of predefined criteria. This is a model example of how big data can be leveraged for a real-life use case and gives insight into the number of considerations that have to be made at each step of the process.
Sourcing the Data
Naturally the first step for any task that wants to leverage geospatial datasets to answer a question of practical/commercial concern is sourcing the data. As any Data Scientist would know this is the most critical step: every problem that follows ultimately flows from the decisions/choices made here. Of particular concern when developing a solution for a commercial application is that the sourced data is not only accurate, but is ideally certified by the relevant authorities. A detailed map of zoning information, for example, is of zero value if it cannot be certified to have come directly from the relevant zoning authority. For this reason official data sources are of extremely high value as they provide not only the necessary ground truth information but also legitimacy to any conclusions or predictions that are to be drawn from it. In this project the Bundesamt für Eich- und Vermessungswesen (Federal Office of Metrology and Surveying) of Austria as well as the Open Government Data resource (OGD) provided by the state of Salzburg served as an excellent foundation for the development of our actionable wind farm land survey.
Leveraging Insights
As already mentioned, the identification of ideal sites for the construction of new wind farms is broken down into a number of quantifiable criteria that provide a binary characterization of a given area of interest (“Fit for the construction of a wind farm”, “Not fit for the construction of a wind farm”). As such, failure to fulfill any one of these criteria would result in the area being disregarded in any further analysis. What this methodology allows is a “quick and easy” classification scheme for identifying the fitness of a region for the construction of a wind farm.
The implementation of the above described process can be broken down as follows:
Firstly, each criteria was associated with a data source that provided the sufficient ground truth information for answering this question in the positive or in the negative. Then for each criteria the associated ground truth information was processed to classify the suitability of each region in Salzburg with regards to the criteria in question. In this way a vector layer was generated for each and every criteria classifying areas of Salzburg into “Suitable” or “Unsuitable” regions. In the following we will take a quick look at the generation of two of these layers.
Identifying Residential Areas
One of the key criteria to follow for the construction of a new wind farm is that it is an appropriate distance away from major residential areas. Fulfilling this criteria is a necessity in most areas due to government regulation in place to address concerns with regards to noise pollution, urban planning and the visual impact of these installations. Utilizing ground truth data sourced from the Bundesamt für Eich- und Vermessungswesen gives us a good starting point for attempting to generate an appropriate layer that segments Salzburg into areas as defined by their suitability with regards to this criteria.
As can be seen in this case the ground truth data consists of a number of polygons corresponding to existing structures present in the state of Salzburg. This offers a promising starting point, but a number of questions still need to be addressed in order to transform this input into the appropriate binary classification layer. Namely, how does one make a meaningful distinction between a town, appearing here as a tight clustering of polygons, and two lone houses in a field? In this case we turned to the Density-Based Spatial Clusterring algorithm which provides an intuitive density based classification of an area into “residential” or “not-residential” based on the density of these structure polygons and their number of neighbors. In this way we can achieve for example the following classification.
Identifying Level Regions
Another important criteria for any construction project is that the ground is of the appropriate grade. Generally speaking construction projects must be built level in order to provide the appropriate foundation for the planned structure. Above a certain grade of incline the costs that will have to be incurred to level the area become far too prohibitive for the project to be practical. Luckily open source raster data captured from surveys conducted with laser measurements are available at 5m resolution.
In this case since this is raster, and not vector, data the translation of this layer into a binary classification mask is slightly more involved. In this case this was accomplished by defining a threshold angle above which a point of interest was considered “too steep” querying the raster data for all of these regions, extracting these regions to vectors with OpenCV and thresholding the extracted areas gives us the desired classification.
Conclusion
After having successfully extracted a binary classification mask for each of the criteria that define a region’s suitability for the construction of a wind farm, putting this altogether is then straightforward: simply combine all of the classification masks into one. The resulting regions are the optimal regions as determined by official ground truth data provided by the state government of Salzburg and the federal government of Austria, quite a find!