Leveraging incomplete remote sensing for forest inventory

Researchers have harnessed a new statistical technique that allows forest managers to use incomplete satellite imagery for precise forest inventories, bypassing the need for complex and often uncertain data repair processes. The method, known as a "hybrid estimator," is particularly valuable for leveraging decades of archived data from the aging Landsat 7 satellite, which has been collecting images with systematic gaps since 2003.

The study, published in Forest Ecosystems, addresses a critical challenge in forestry and climate science. Remote sensing is vital for large-scale forest monitoring, but missing data, such as the wedge-shaped gaps in Landsat 7 imagery, compromised the credibility of estimates for forest volume, biomass, and carbon stocks. Traditionally, scientists have used "gap-filling" algorithms to reconstruct this missing data, but these methods propagate their own errors that are not always quantifiable, and thereby misleading decision-makings relying on inventory estimates.

In contrast, instead of trying to create a perfect, wall-to-wall image, the hybrid estimator uses probability-based sampling to work directly with the available, non-wall-to-wall data.

"The core idea is to use a statistically robust sample of the existing good pixels, rather than relying on a guess-filled complete map," the researchers explained. "This allows us to generate reliable population-level estimates for forest attributes without the uncertainty that comes from gap-filling."

The research team put their method to the test in the forests of Inner Mongolia, China. They compared their hybrid estimator, which used the Landsat 7 data, against a conventional model-based method that required pristine, wall-to-wall imagery from the Sentinel-2 satellite.

The results indicated that hybrid estimator achieved a sampling precision of over 90%, meeting China's national standard for forest inventory. Most importantly, its efficiency was comparable to the conventional model-based method using the superior Sentinel-2 data.

"Our findings show that we don't have to discard the archived Landsat 7 data," the authors stated. "By using this hybrid approach, we can extract valuable, reliable information from it directly. This activates a huge historical dataset for long-term trend analysis and provides a cost-effective tool for large-scale forest surveys."

The study also provides practical guidance for forest managers, suggesting that using a larger number of smaller sample clusters further optimizes the precision of the estimates. This new method holds the potential to make forest monitoring more flexible, cost-effective, and reliable, especially in regions where access to the latest satellite imagery is limited.

END