Two of the fundamental questions natural resource managers should pose to themselves from time to time are: “What is it that is being managed and over what time frame?”. In this article I will delve into these questions using the boreal forest as an example and consider the consequences of the answers for natural resource data management.
Forest from above
Let’s first step inside the forest (Fig. 1); it seems that the forest could be roughly described as a collection of trees. Imagine then that we could get a lift from a crane with the range of 1,000 kilometers. The first stop would be at the altitude of 1 kilometer (Fig. 2). Now a second pattern begins to emerge; the collections of trees seem to be themselves somewhat grouped.
{sidebar id=157 align=right} Ascending to the next stop at 10 km (Fig. 3) this impression gets even stronger; the forest indeed is a collection of rather homogeneous collections of trees; i.e., stands. However, ascending to the end of the range of the crane at 1,000 km (Fig. 4), the first two impressions of forest dissolve and are replaced with a new one; forests are almost uniform land cover that change gradually over large distances. Naturally this description of forests only applies to a given location at the globe at a given time, in this case 21st century Finland.
No matter what the actual definition of forest used is, it is clear that it changes with the scale of observation and therefore the data management system used should accommodate different representations of the natural objects.
{sidebar id=158 align=left}
Action through time
While the first question was about what is being managed, the second question of the time frame is linked to the “how” part of management. Natural resource management can be operationalised as a form of adaptive management that consists of a cycle (Fig. 5) of planning, action, monitoring and evaluation. This allows for the integration of new goals, knowledge and technology into the management at the stage where the results of monitoring the previous round of planning and action is evaluated.
The practical problem is that the full management round can take quite a long time; in managed boreal forests the full circle for an individual tree from a seed to a harvested tree can take a century and even the period between successive management actions for an individual stand is measured in decades.
{sidebar id=159 align=right} The need for the ability to travel backwards (evaluation) and forwards (planning) in time introduces a serious problem to the data management system : changes. Because trees grow and die and human interventions either fuel or slow down these processes, a forest inevitably changes as time passes by, even if one would forget about the changing scales and would fi the viewpoint to the forest.
In addition to the changes in the nature, the things that one would want to record into the data system, the reflection of the forest, will change; e.g., is the forest just a collection of trees or does it include other components as well, or does one store for a stand that “there’s a lot of timber” or “the amount of pine timber is 150 cubic metres and the amount of pine pulp wood is 57 cubic metres” or “the amount of A-grade pine timber is 20 cubic metres, B-grade 100 and C-grade 30. The amount of pulp fibre is 30 cubic metres, biocomposite fibre 20 and biofuel fibre 15 cubic metres”.
{sidebar id=160 align=left}
Describing forests by scales
The question from the data management point of view hence becomes “How would it be possible to store different descriptions of a forest in different scales, and to store the past, the current, and the predicted future state of the forest, all in different scales simultaneously with varying attribute content?”
The first part of the answer tackles the different views of forest. The way to do this is to define that the forest is described with a single object type, no matter whether it’s seen as a collection of trees, stands or regions. This object has a collection of generic properties and links to other objects. These links enable us to travel between the different views of the forest: a tree knows which stand and region it belongs to and similarly a stand knows which are its trees and which regions it belongs to.
The natural argument against this solution is “But a tree can’t be of the same object type as a region, they have completely different attributes!”. While intuitively true, this argument doesn’t hold if we define the object attributes at a suitably abstract level.
{sidebar id=161 align=right} Trees and regions have exactly the same set of attributes if we say that an object has a location and some other attributes. “Some other attributes” sounds too vague to be useful, but once we define that this equals to a collection of attributes, each attribute being defined as a pair of value interpretation and the actual attribute value, things start to sound feasible.
The multi-scale data model for forest resource management can be represented as a connected graph in which the nodes represent the objects at different data levels and the edges between the nodes represent the connections between the data levels (Fig. 6).
Forest changes
The second part of the answer deals with the passage of time; i.e., the changes. To accommodate the temporal dimension, all objects and their attributes have a life span. The attribute life span combined with the value interpretation-value pairs caters for the possibility of changes in what kind of data is being stored for each object at any given time (Fig. 7). Also the location of the object has similar life span, thus allowing the object to change its geometry several times through its entire life history.
In addition, the objects have a parent-child relationship with other objects. The temporal relations between objects are handled using an identical method to that applied to the spatial relations between observation scales. The relations are regarded as edges in a time graph of object nodes forming links to other objects.
The objects are either predecessors or successors to each other (Fig. 8). The temporal edges between the object nodes are directional and labeled to indicate the time passes forwards or backwards when the edge is traversed.
{sidebar id=162 align=right} Both the spatial and temporal relationships between objects may be subject to uncertainty should the spatial delineation of objects contain inaccuracies or if the delineation principles are changed. Uncertainty in the relations between objects may be expressed using edge labels identifying for example the certainty of the existence of the link as a number between 0 and 1 (Fig. 8).
To facilitate the analysis of change over time, the data model contains a class for events that drive the changes in spatio-temporal objects. An event occurs at a given point in time, it has a location and information about the event type. Events and objects are again connected through edges at a graph having objects and events as nodes. Each object is connected to the events that have affected it, and each event is connected to the objects that it has affected. A more detailed relationship between events and features can be computed from the stored locations, i.e., the exact spatial intersection of an event and a features can be derived from the geometries between the event and the features it has been linked to.
Should the evolution of an event over time be of interest, e.g., the development of a forest fire or insect outbreak, events can also be modelled as a sub-class of spatio-temporal objects, thus allowing them to occur over a period of time within which they can have multiple geometries and attribute values.
{sidebar id=163 align=left}
Data model evolution
The main benefit of the presented conceptual data model is its generic nature, which absorbs the changes in data content that are bound to happen over time. On the flip side of this generic nature are the complications that this causes for using the mainstay of data management tools; relational databases and SQL; for the implementation of the conceptual model.
Although possible, it’s not advisable to implement heavily network-based, hierarchical data models in pure relational form. One option is to implement the generic data model in an object-relational database, in which case data access is a combination of SQL-operations and procedural program code. Moreover, the circle that led from the hierarchical databases of the late 1960s to relational databases has closed with the arrival of markup languages as part of the data management mainstream.
This has returned hierarchical data access as one of the focal areas again. Extensible Markup Language (XML) and its data access tool XQuery are one of the viable options for managing spatio-temporal data over time. Yet another option is to use a pure network database.
{sidebar id=164 align=right} The emphasis in data analysis is not always on tracking changes and their possible causes over time, however, or on the relationships between objects at different scales over time. Quite often the focus is on the various objects and their properties in a particular area at one point in time.
This is where the databases in use today excel, as they are able to store and give access to data that has a constant structural content and unambiguous location. Therefore, it may be sometimes feasible to derive more staticly structured content from the generic model implementation to be stored and analysed in a relational database management system.
Summary
Let’s return to the question at the beginning of the article: “What is it that is being managed and over what time frame?”. It’s clear by now that the answer is “I’m not too sure”. We may have quite a good grasp of what it is that is being managed at the moment, but what is almost guaranteed is that it won’t be the same as the time passes. Should one embrace this uncertainty and constant change when designing support systems for natural resource management, one should also be more successful in implementing systems that are relevant for adaptive management.
Dr. Jussi Rasinmäki – Simosol Oy, Riihimäki, Finland. Email: jussi.rasinmaki [at] simosol.fi
For more information:
www.simosol.fi
www.metla.fi/dissertationes/df49.htm
Reference
Walters, C.J. 1986. Adaptive Management of Renewable Resources. MacMillan Publishing Company, New York.
