Validation of the MODIS Clumping Index: A Case Study in Saihanba National Forest Park

Highlights: What are the main findings? A novel multi-scale validation (field, UAV, Landsat) demonstrates that MODIS CI products show good agreement with reference data (R = 0.75, RMSE = 0.05) in a temperate forest. Direct “point-to-pixel” comparisons are highly susceptible to subpixel heterogeneity. Semivariogram analysis of the high-resolution CI map reveals that a ~209 m observational footprint is required for a spatially representative sample, critically informing future validation design for coarse-resolution products. What is the implication of the main finding? The study provides a robust framework that enables diagnosis of error sources, distinguishing between uncertainties from satellite retrieval (e.g., land cover misclassification causing errors up to 0.33) and those introduced by the validation process itself (e.g., upscaling method choice). Findings confirm the operational utility of MODIS CI while underscoring the necessity for international cooperative campaigns to obtain representative field data and further research on scaling methods for extensive global validation. The clumping index (CI) describes the level of foliage grouping relative to the random distribution within the canopy. It plays a vital role in the derivation of other important parameters (e.g., the leaf area index, (LAI)) that are usually employed in hydrological, ecological and climatological modeling. In recent years, several satellite-based CI products have been developed using multi-angle reflectance data. However, these products have been validated through the use of a “point-to-point” comparison, which rarely involves a quantitative analysis of spatial representativeness for field-measured CIs in most cases. In this study, we developed a methodological framework to validate the MODIS CI at three different data scales on the basis of intense field measurements, high-resolution unmanned aerial vehicle (UAV) observations and Landsat 8 data. This framework was used to understand the impacts of the scale issue and subpixel variance of the CI in the validation of the MODIS CI for a case study of 12 gridded 500 m pixels in Saihanba National Forest Park, Hebei, China. The results revealed that the MODIS CIs in the study area were in good agreement with the upscaled field CIs (R = 0.75, RMSE = 0.05, bias = 0.02) and UAV CIs. Through a comparison of the observed CIs along the 30 m transects with the 500 m MODIS CIs, we gained insight into the uncertainty caused by the direct “point-to-pixel” evaluation method, which ranged from −0.21~+0.27 for the 10th and 90th percentiles of the observed-MODIS CI error distribution for the twelve pixels. Moreover, semivariogram analysis revealed that the representativeness assessments based on high-resolution albedo and CI maps could reflect the spatial heterogeneity within pixels, whereas the CI map provided more information on the variation in vegetation structures. The average observational footprint needed for a spatially representative sample is approximately 209 m according to an analysis of the high-resolution CI map. The uncertainty of mismatched MODIS land cover types can lead to a deviation of 0.33 in CI estimates, and compared with the CLX method, the scaled-up CI method based on simple arithmetic averages tends to overestimate CIs. In summary, various validation efforts in this case study reveal that the accuracy of the MODIS CIs is generally reliable and in good agreement with that of the upscaled field CIs and UAV CIs; however, with the development of surface process modeling and remote sensing technology, substantial measurements of field CIs in conjunction with high-resolution remotely sensed CI maps derived from single-angle advanced methods are urgently needed for further validation and potential applications. Certainly, such a validation effort will help to improve the understanding of MODIS CI products, which, in turn, will further support the methods and applications of global geospatial information.