Researchers from the Institute of Applied Ecology (IAE), Chinese Academy of Sciences, mapped and estimated species-level biomass (species composition and aboveground biomass) in Chinese boreal forests (the Great Xing’an Mountains) by integrating MODIS imagery with forest inventory data using k nearest neighbor (kNN) method.
To design effective forest management plans and to sustain forest structure and function in response to various disturbances, forest managers need to get timely and accurate information about forest stand characteristics such as species-level biomass.
To map forest attributes over large areas, researchers often integrate field inventory data with remote sensing data by using imputation models. The k nearest neighbor (kNN) imputation is one of such models. The accuracy of kNN model is determined by the k value (i.e., the number of nearest neighbors), the distance metrics and the temporal coverage of remote sensing data.
Dr. ZHANG Qinglong and Dr. LIANG Yu from the IAE, in collaboration with researchers from the Geological Survey of the United States, compared the performance of kNN models based on different distance metrics, k values and temporal MODIS data (May – October).
They found that Random Forest (RF) showed the best performance among the six distance metrics (i.e., RF, Euclidean distance, Mahalanobis distance, most similar neighbor in canonical correlation space, most similar neighbor computed using projection pursuit, and gradient nearest neighbor).
They concluded that k = 6 was an appropriate parameter for the kNN model. They also found that MODIS data obtained in June produced better accuracy for the estimate of species-level biomass than other single-month MODIS data.
“Using MODIS data to impute species-level attributes can produce accuracy comparable with using lidar data for young and middle-aged boreal forests,” they said.
The maps of species-level biomass also captured the effects of disturbances including fire and harvest and the accuracy of most tree species was improved up to the ecoregion scale.