研究目的
To describe the uncertainty characterisation framework and the related uncertainty validation exercise of the Fire-CCI project, focusing on the generation of Essential Climate Variables (ECVs) with associated per pixel uncertainty characterisation.
研究成果
The presented framework and methodology provide a basis for producing uncertainty-quantified burnt area products at both pixel and Climate Model Grid scales. Results highlight the need for algorithms to more closely align uncertainty estimates with true product uncertainties, suggesting further refinement and consideration of uncertainty within the retrieval algorithm itself.
研究不足
The framework assumes no uncertainty introduced by the retrieval algorithm itself, focusing only on the uncertainty in remote sensing observations. The sensitivity of uncertainty estimates to the level of noise in the input datasets varies between algorithms, indicating a need for refinement.
1：Experimental Design and Method Selection:
The study employs a Monte-Carlo framework to approximate the integral of the probability distribution of the observation, focusing on the uncertainty in remote sensing measurements.
2：Sample Selection and Data Sources:
Three test sites representing significant pyromes for burnt area (savanna, boreal forest, and tropical forest) were selected. Data from MODIS Collection 6 surface reflectance products were used.
3：List of Experimental Equipment and Materials:
MODIS surface reflectance products (MOD/MYD09) with their uncertainty characterisation.
4：Experimental Procedures and Operational Workflow:
Realisations of reflectance were generated by sampling from the data distribution, considering different levels of observational noise. Algorithms were run on these realisations to estimate the true uncertainty.
5：Data Analysis Methods:
The study compares algorithm estimates of uncertainty with true uncertainty derived from the sampling framework, assessing the accuracy and sensitivity of the uncertainty characterisation.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
