Category Archives: Socially informed use of remote sensing

What is telecoupling?

Show new policy regimesand regulations in one country have direct consequences for land use in others, forexample, in relation to forest protection policies resulting in leakages of deforestation abroad (Meyfroidt & Lambin,2009; Meyfroidt et al.,2013; Meyfroidt, Rudel, & Lambin,2010) .

A telecoupling ariseswhen an action produces flows between two or more place-based human–environment systems, which create a change and/or response in one or both of the systems–regardless of whether or not these effects are intended. Within each system, a varietyof agents can create or hinder the flows, and hence set in motion different causes andeffects, including feedbacks.

Systems are classified as sending, receiving or spill-over systems. Sending systemsrefer to places where the flow originates, whereas receiving systems are the recipients ofthe flow. Spill-over systems are understood as places that affect or are affected by the flowof interaction between sending and receiving systems, but without direct influence on thenature or direction of the flow. The complexity of the simple schematics increases asmultiple sending, receiving and spill-over systems interact over distances. Depending onthe particular flow being analysed, any system can act as a sending, receiving and/or spill-over system. Although the spatial extent of telecouplings is not explicitly addressed byLiu et al. (2013), telecouplings are implicitly characterised as interactions over (large)geographical distances, for example, the soybean trade between the US and China.
(2) (PDF) From teleconnection to telecoupling: taking stock of an emerging framework in land system science. Available from: [accessed Feb 14 2019].

Eakin et al.(2014) stress that the outcomes or results of telecoupled interactions are often indirect,emergent or of a second or third order because different land use systems are governedindependently of each other. This approach suggests that telecoupling can be analysedas the outcome of five key features: the trigger that sets the telecoupling in motion, thedirect impacts in the system with the initial change, the indirect/unexpected impacts inthe distantly coupled system, the feedback processes that influence the existinggovernance structures, and finally, the potential institutional change in both systems.

A further distinction of this approach is the explicit emphasis on the networkedinteractions across scales in the creation of telecouplings, which substitute the spatialhierarchy and nested scales of analysis featuring prominently in the structuredapproach. For example, Eakin et al. (2014) note that the rising influence of informa-tion technology and social networks have made it possible for actors toskip scaleand interact, influence and create outcomes in telecoupled systems (p. 159). Finally,the question of analytical entry pointis left open in the heuristic approach totelecoupling analysis, where the analysis, for example, could start from an observedland use change, a policy expected to trigger change or in adverse social or environ-mental impacts.

Whereas Liuet al. (2013) and Liu et al. (2014) frame telecouplings in a structured spatial hierarchy,Eakin et al. (2014) define them as the outcomes of networked interactions across scales.Furthermore, the structured approach in essence presents a type of‘checklist’of compo-nents to include in an exhaustive analysis that encourages, though does not require, theanalysis to begin from the flow of interest, while the heuristic approach focuses onnetworks, actors and processes with a more open analytical entry point (Friis &Nielsen,2014). Both approaches highlight the need for continued engagement withdifferent theoretical tools and methodologies in order to capture the full complexity ofthe dynamics and processes involved in telecoupling.

POLITICAL ECOLOGY These insights from political ecology can provide telecoupling research with the meansto address the challenge related to power asymmetries and asymmetrical relations betweensystems. By analysing interactions between distantly linked systems as (potential) distribu-tion conflicts, actors at both‘ends’of the interaction become active agents with (potential)power to influence the outcome of the interaction. Instead of analysing‘effects’of telecou-plings on (passive) receiving or spill-over systems, telecoupling research could ask whichactors, regardless of their‘location’in the interaction, have the power to decide on land useoutcomes and to shape the interconnectedness of (telecoupled) human–environment systems.The contested nature of the processes of production of (unequal) telecouplings could thus beexplored, with particular attention to dynamics of resistance and struggle for alternativetelecouplings and political ecological orders across the world.
(2) (PDF) From teleconnection to telecoupling: taking stock of an emerging framework in land system science. Available from: [accessed Feb 14 2019].

(2) (PDF) From teleconnection to telecoupling: taking stock of an emerging framework in land system science. Available from: [accessed Feb 14 2019].

(2) (PDF) From teleconnection to telecoupling: taking stock of an emerging framework in land system science. Available from: [accessed Feb 14 2019].

(2) (PDF) From teleconnection to telecoupling: taking stock of an emerging framework in land system science. Available from: [accessed Feb 14 2019].

(2) (PDF) From teleconnection to telecoupling: taking stock of an emerging framework in land system science. Available from: [accessed Feb 14 2019].

(PDF) From teleconnection to telecoupling: taking stock of an emerging framework in land system science. Available from: [accessed Feb 14 2019].

What is Bagging in statistics?: “two heads are better than one”

“Bagging” or bootstrap aggregation is a specific type of machine learning process that uses ensemble learning to evolve machine learning models. Pioneered in the 1990s, this technique uses specific groups of training sets where some observations may be repeated between different training sets.

The idea of bagging has been used extensively in machine learning to create better fitting for models. The idea is that if you take several independent machine learning units, they can function collectively better than one unit that would have more resources.

To really illustrate how this works, think of each part of the bagging process as an individual brain. Without bagging, machine learning would consist of one really smart brain working on a problem. With bagging, the process consists of many “weak brains” or less strong brains collaborating on a project. They each have their domain of thinking, and some of those domains overlap. When you put the final result together, it is a lot more evolved than it would be with just one “brain.”

In a very real sense, the philosophy of bagging can be described by a very old axiom that predates technology by quite a few years: “two heads are better than one.” In bagging, 10 or 20 or 50 heads are better than one, because the results are taken altogether and aggregated into a better result. Bagging is a technique that can help engineers to battle the phenomenon of “overfitting” in machine learning where the system does not fit the data or the purpose.

What is training data? Basically data

Training data (or training set) refers to that portion of data used to fit a model. Unsupervised learning refers to analysis in which one attempts to learn something about the data. other than predicting an output value of interest (whether it falls into clusters, for example).

Overfitting in Statistics

Figure 1.  The green line represents an overfitted model and the black line represents a regularized model. While the green line best follows the training data, it is too dependent on that data and it is likely to have a higher error rate on new unseen data, compared to the black line.
Figure 2.  Noisy (roughly linear) data is fitted to a linear function and a polynomial function. Although the polynomial function is a perfect fit, the linear function can be expected to generalize better: if the two functions were used to extrapolate beyond the fit data, the linear function would make better predictions.

In statistics, overfitting is “the production of an analysis that corresponds too closely or exactly to a particular set of data, and may therefore fail to fit additional data or predict future observations reliably”.[1] An overfitted model is a statistical model that contains more parameters than can be justified by the data.[2] The essence of overfitting is to have unknowingly extracted some of the residual variation (i.e. the noise) as if that variation represented underlying model structure.[3]:45

Underfitting occurs when a statistical model cannot adequately capture the underlying structure of the data. An underfitted model is a model where some parameters or terms that would appear in a correctly specified model are missing.[2] Underfitting would occur, for example, when fitting a linear model to non-linear data. Such a model will tend to have poor predictive performance.

Overfitting and underfitting can occur in machine learning, in particular. In machine learning, the phenomena are sometimes called “overtraining” and “undertraining”.

The possibility of overfitting exists because the criterion used for selecting the model is not the same as the criterion used to judge the suitability of a model. For example, a model might be selected by maximizing its performance on some set of training data, and yet its suitability might be determined by its ability to perform well on unseen data; then overfitting occurs when a model begins to “memorize” training data rather than “learning” to generalize from a trend.

As an extreme example, if the number of parameters is the same as or greater than the number of observations, then a model can perfectly predict the training data simply by memorizing the data in its entirety. (For an illustration, see Figure 2.) Such a model, though, will typically fail severely when making predictions.

The potential for overfitting depends not only on the number of parameters and data but also the conformability of the model structure with the data shape, and the magnitude of model error compared to the expected level of noise or error in the data.[citation needed] Even when the fitted model does not have an excessive number of parameters, it is to be expected that the fitted relationship will appear to perform less well on a new data set than on the data set used for fitting (a phenomenon sometimes known as shrinkage).[2] In particular, the value of the coefficient of determination will shrink relative to the original data.

To lessen the chance of, or amount of, overfitting, several techniques are available (e.g. model comparisoncross-validationregularizationearly stoppingpruningBayesian priors, or dropout). The basis of some techniques is either (1) to explicitly penalize overly complex models or (2) to test the model’s ability to generalize by evaluating its performance on a set of data not used for training, which is assumed to approximate the typical unseen data that a model will encounter.

%d bloggers like this: