Potential of Bayesian formalism for the fusion and assimilation of sequential forestry data in time and space

Cheikh Mohamedou,Alireza Hamedianfar,Annika Kangas,Jari Vauhkonen

doi:10.1139/cjfr-2021-0145

Cheikh Mohamedou, Alireza Hamedianfar + Show 2 more

Open Access

https://doi.org/10.1139/cjfr-2021-0145

Copy DOI

Abstract

Forest resource assessments based on multi-source and multi-temporal data have become more common. Therefore, enhancing the prediction capabilities of forestry dynamics by efficiently pooling and analyzing time-series and spatial sequential data is now more pivotal. Bayesian filtering and smoothing provide a well-defined formalism for the fusion or assimilation of various data. We ascertained how often the generic, standardized Bayesian framework is used in the scientific literature and whether such an approach is beneficial for forestry applications. A review of the literature showed that the use of Bayesian methods appears to be less common in forestry than in other disciplines, particularly remote sensing. Specifically, time-series analyses were found to favor ad hoc methods. Our review did not reveal strong numeric evidence for better performance by the various Bayesian approaches, but this result may be partly due to the challenge in comparing a variety of methods for different prediction tasks. We identified methodological challenges related to assimilating predictions of forest development; in particular, combining modelled growth with disturbances due to both forest operations and natural phenomena. Nevertheless, the Bayesian frameworks provide possibilities to efficiently combine and update prior and posterior predictive distributions and derive related uncertainty measures that appear under-utilized in forestry.

Highlights

Even though the adoption of a Bayesian approach resulted in unexpectedly small numeric benefits, it may rationalize the analyses by accounting for the features of the Bayesian approaches that we identified in the introduction section
Bayesian approaches have been increasingly utilized in many applications during the last decade, less so in forestry than in the other reviewed scientific disciplines
State-of-the-art data processing frameworks and cross-disciplinary technologies allow for the integration of remotely sensed and forest inventory data, which are available at more frequent intervals and produce a longer time-series

Summary

Introduction

Systematic forest inventories have been carried out for over a century (Kangas et al, 2018a) and have utilized remote sensing and other digital map data already for decades to estimate forest variables (Katila and Heikkinen, 2020). While historical and new data have considerable potential to improve forestry-related predictions, this is not self-evident and may not be realized unless sampling, modelling and estimation methods are used appropriately with respect to the different properties of the data sources (Kangas et al, 2018b, 2019). Many of the observations made by Särkkä (2013) on generic measurement systems can be extended to forestry data: even with the most carefully measured field plots, much of the signal may remain hidden (i.e., the forestry dynamics that we attempt to model). We must deal with “noise” (Särkkä, 2013) in the form of measurement, model and sampling errors (cf., Kangas et al, 2019)

Methods

Findings

Discussion

Conclusion