British Columbia Mine Reclamation Symposium

Monitoring success : using remote sensing and machine learning to get more out of revegetation trials Anderson, Jeff; Pearse, Ben

Abstract

Reclamation trials are critical for understanding the impact of different variables on revegetation outcomes. However, drawing meaningful conclusions from these trials is difficult due to the uncontrolled nature of the environment and traditional plot-based sampling methods that suffer from limited spatial coverage, observer bias, and the inability to capture subtle spatial trends. In July 2023, a two-hectare revegetation trial was established on a mined waste rock dump. The relative revegetation performance of two cover materials, three handling methods and three cover depths on the health of tree seedlings 14 months post-planting was evaluated using remote sensing methods, calibrated and validated against plot-based data. Aspen (Populus tremuloides), lodgepole pine (Pinus contorta), and white spruce (Picea glauca) were planted across all treatment blocks. Seedlings were classified as dead or alive and assigned health indicators using high-resolution multispectral imagery and object-oriented machine learning classifications. Confounding factors were isolated, impacted seedlings removed, and the health of the remaining seedlings was evaluated in relation to cover material, bulk density, and cover depth. Seedlings planted within 10 m of the slope crest or planted on the tops of mounds were identified as confounding factors, showing reduced seedling size. After excluding those factors, cover material and bulk density emerged as the primary influences on seedling health. Seedling mortalities differed by 5% and average size by 0.29 m2 between the two borrow source materials. Within the cover material associated with poorer seedling performance, a statistically significant impact of bulk density on the growth of aspen (R2 = 0.71) and white spruce (R2 = 0.35) was observed. This study highlights the importance of accounting for confounding factors when evaluating treatment variables, which is only possible using high-resolution spatially explicit data. The results emphasize that low-quality and compacted cover material can impede seedling establishment and growth.

Item Citations and Data

Rights

Attribution-NonCommercialNoDerivatives 4.0 International