Research
 

Research at the Centre of Biostochastics, Umeå


Within forestry/agricultural research, with its studies of biological, economic, technical and related questions, one meets a large variety of problems, which have to be tackled with the aid of different mathematical-statistical theories and methods.

From the earliest times, foresters have been keenly interested in the measurement of trees and the determination of volume, yield and increment. However, at the time of the breakthrough for mathematical statistics within forestry research - that is, during the 1920's - statistical methods were widely met with distrust. This was partly due to the fact that the methods used were inadequate for the difficult problems to which they were applied. Bertil Matérn, who became the first professor in Forest Biometry in Sweden, was employed by the Swedish Forest Research Institute in 1945 (which was later incorporated into the Royal College of Forestry and the Swedish University of Agricultural Sciences), for working with problems related to the Swedish National Forest Inventory. These problems had a spatial (and a temporal) dimension, for which there were no theory at the time. Matérn initiated new theory and became one of the pioneers within spatial statistics.

Today, foresters not only have ground measurements as a basis for the assessment of the state of and changes in forest resources, but also data from orbital platforms generating a myriad of remote sensing data streams across a range of spatial, temporal, and radiometric resolutions. Further, when studying biological systems one must confront an enormous range of scales, in space, time, and organizational complexity, e.g., one deals with phenomena that range from molecular processes to evolutionary, ecological and population processes. Thus, it is becoming clear that forest biometrics, environmental statistics, mathematical biology, etc., demand more and more development of statistical approaches, and the veritable data deluge we have in many disciplines of forest research and practice today necessitates new ways of statistical thinking that will enable prediction of the usual and the identification, quantification, and assessment of the unusual.

Statistical methods utilizing quantification of variations in time and space have had an energetic development during the last decades, together with the development of computer capacity, needed for complex models, algorithms, and huge amounts of data. Furthermore, techniques for mathematical and statistical modeling and analysis have become vastly more powerful than they were previously. The exponential growth of computer power, both in terms of speed and storage, and the appearance of computationally intense techniques such as the Markov Chain Monte Carlo (MCMC) and various resampling techniques mean that what was impossible a decade ago is now possible.

Current research activities include:

Classification methods

Resampling methods

Wavelets

Estimation of parameters under nonstandard conditions

Gibbs sampler and boundary detection

Time series analysis

Image analysis