LAUSR

Abstract Leaf area index (LAI) products are now routinely generated from remote sensing and widely used in most land surface process models. Assessing the uncertainties associated with these LAI products is essential for their proper application. In validation activities, LAI reference maps serve as the bridge to upscale the field measurements to coarse resolution. Currently, the temporally continuous validation is attracting increasing attention. Therefore, there is an urgent requirement for the temporally continuous LAI reference maps. However, two main problems hinder the generation of temporally continuous LAI reference maps: 1) How to obtain temporally continuous field measurements, effectively and inexpensively? 2) How to obtain temporally continuous and fine spatial resolution satellite images which are synchronous with the field measurements? This paper proposed a method to address the above two problems based on the combination of the wireless sensor network technology and a data blending approach. Firstly, the temporally continuous effective LAI was obtained through the analysis of multi-angle gap fraction measured by LAINet observation system (developed with wireless sensor network technology), and the temporally continuous NDVI was reconstructed through CACAO (Consistent Adjustment of the Climatology to Actual Observations, a data blending approach). Then, a transfer function relating reconstructed NDVI to field measured LAI was calibrated through exponential function fitting. Finally, the temporally continuous LAI reference maps were generated by applying the calibrated transfer function to the reconstructed NDVI. Performances of the proposed method were evaluated over a crop site. Results show that the reconstructed LAI reference maps agree well with the original LAI reference maps derived from the Landsat-8 OLI NDVI (R 2  = 0.90, RMSE = 0.27 at 30 m resolution, R 2  = 0.97, RMSE = 0.09 at 1 km resolution). The derived temporally continuous LAI reference maps were then used as benchmark to validate the MOD15A2 LAI product. Generally, the MOD15A2 LAI has a relatively high accuracy (R 2  = 0.53, RMSE = 0.31), and captures the overall vegetation phenology well. But its value shows an obvious underestimation with a bias of −0.30. The results of this study contribute to the assessment of temporal dynamics of uncertainty in LAI products, which will benefit the long-term vegetation monitoring and data assimilation.