Long-term strategic forest planning based on biased remote sensing predictions

Accurate forest data is essential for informed decisions regarding forest policy and management. Traditionally collected through field surveys, this type of data has increasingly been produced with remote sensing (RS). RS provides comprehensive resource maps produced with data from sensors, including airborne laser scanning (ALS) and satellite imagery. However, RS predictions can include large uncertainties, including both random and systematic errors. The systematic errors often stem from the problem of regression towards the mean, whereby small true values are overestimated while large true values are underestimated. These errors pose challenges for effective forest management planning since they can lead to wrong assumptions about forest conditions, for example, that a forest conforms to average conditions due to reduced variability. In this study, we quantified the differences between expected and realised outcomes in forest planning informed by RS predictions, specifically evaluating inventories based on ALS and optical satellite imagery. The evaluation was made according to a business-as-usual scenario with additional concerns about biodiversity and carbon sink targets. The satellite-based forest inventory, more impacted by both general uncertainty and regression towards the mean, performed worse than ALS. Our results indicate that reliance on RS predictions led to 10% to 12% overestimated harvest levels, with notable fluctuations over time, alongside a decrease in net present value of -6% to -9%. Furthermore, carbon stocks were unintentionally reduced in the satellite-based plans, with overestimations ranging from 8% to 24%. Across both RS methods, achieving stable development for biologically valuable forests proved difficult. Our findings underscore the relevance of these issues for forestry and are important to ongoing policy development related to forest monitoring and planning.