Open AccessConsistent forest biomass stock and change estimation across stand, property, and landscape levels
Author(s) -
Victor Strîmbu,
Hans Ole Ørka,
Erik Næsset
Publication year - 2021
Publication title -
canadian journal of forest research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.677
H-Index - 121
eISSN - 1208-6037
pISSN - 0045-5067
DOI - 10.1139/cjfr-2020-0203
Subject(s) - estimator , mean squared error , stock (firearms) , statistics , environmental science , forest inventory , sampling (signal processing) , estimation , mathematics , climate change , biomass (ecology) , econometrics , physical geography , geography , forest management , ecology , computer science , agroforestry , economics , management , archaeology , filter (signal processing) , computer vision , biology
Stimulating climate change mitigation actions in the forest sector requires methods to quantify the biomass stocks and changes at different geographical levels. Often, differences in data and estimation methods that are available at each level cause inconsistencies in forest parameters estimated at different levels. We propose a method to align model-based and model-assisted estimators to ensure cross-sectional and time series consistency of stock and change estimates of aboveground biomass (AGB). The method adjusts estimates within their confidence intervals using heuristic optimization to minimize the estimation errors. The method is evaluated under simulated sampling in a case study representing a forested area of approximately 50 km 2 in southeastern Norway. The area is divided into 93 forest properties encompassing 3324 forest stands. The artificial forest population is generated for two time points using wall-to-wall airborne laser scanning data acquired in 2001 and 2016, as well as field surveys conducted within a similar timeframe. The adjusted AGB stock and change estimators at different levels of aggregation are compared with the original unadjusted estimators in terms of bias and root mean squared error (RMSE). The results show that the adjusted estimators do not introduce bias, and the increase in RMSE is small for the forest stand-level estimators, and even decreasing for the forest property-level estimators. The method can easily be adapted to complex systems of estimators that need to be consistent.