Estimating rangeland fine fuel biomass in western Texas using high-resolution imagery and machine learning

Authors: LI, ZHENG - Texas A&M University; Angerer, Jay; JAIME, XAVIER - Texas A&M University; Yang, Chenghai; WU, X.BEN - Texas A&M University

Submitted to: Remote Sensing
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/29/2022
Publication Date: 9/2/2022
Citation: Li, Z., Angerer, J.P., Jaime, X., Yang, C., Wu, X. 2022. Estimating rangeland fine fuel biomass in western Texas using high-resolution imagery and machine learning. Remote Sensing. 14(17). Article 4360. https://doi.org/10.3390/rs14174360.
DOI: https://doi.org/10.3390/rs14174360

Interpretive Summary: Fire spread and intensity on rangelands are greatly influenced by fine fuel biomass. However, the ability to accurately estimate fine fuel biomass is a challenge because of the heterogeneity in plant communities across rangeland landscapes. Machine learning offers opportunities to use remote sensing imagery and limited field data to train models to estimate fine fuel biomass across these heterogenous landscapes. In this study, high spatial resolution (0.23m) images were used to classify the landscape into different fuel types (e.g., grass, grass/shrub mix, shrub, etc.) and to predict rangeland fine fuel biomass using a machine learning model. The model performed well for predicting fuel types, having an overall accuracy of 95%. For biomass estimation, the model that considered image texture information (i.e., spatial variation in pixel brightness) performed better than the model that did not include texture (81% vs. 77% of variation explained). The machine learning model was then used to predict biomass at the scale of moderate resolution satellites (Landsat, 30 m resolution) using fuel type cover information derived from the high-resolution imagery and Landsat spectral information. Results of this analysis indicated that the fine fuel biomass accuracy was slightly lower than for the high-resolution biomass estimation (56% vs. 77% of the variability explained). These findings indicate that high spatial resolution images have the potential to effectively estimate rangeland fine fuel biomass and can be helpful for rangeland monitoring and management.

Technical Abstract: Rangeland fine fuel biomass is a key factor in determining fire spread and intensity, while the accuracy of biomass estimation is limited due to inherent heterogeneity in rangeland ecosystems. In this study, high spatial resolution (0.23m) images were used to classify fuel types and predict rangeland fine fuel biomass in west Texas based on the random forest algorithm. Two biomass models, including one with the fuel type, spectral curves from original bands and vegetation indices, and another one that contained a combination of the fuel type, spectral curves from original bands, vegetation and texture indices as explanatory variables were assessed. Furthermore, the biomass models were also examined by upscaling the remote sensing images from high to medium (30-m) spatial resolution with the spectral curves derived from Landsat images. The fuel type map had an accuracy of more than 95%, and non-woody types were kept for estimating fine fuel biomass. The results showed that around 77% and 81% of biomass variances were explained by models without texture indices and with texture indices, respectively. The fuel type and NDVI were two significant input variables influencing fine fuel biomass for both models and adding texture indices can contribute to the improvement of model accuracy. The upscaling analysis for biomass estimation using medium spatial resolution images showed that approximately 56% of the variance in biomass was explained by the model. The addition of fractional cover improved the model performance with another 6% of variance explained in biomass estimation. These findings indicate that high spatial resolution images have the potential to effectively estimate rangeland fine fuel biomass and can be helpful for rangeland monitoring and management.