A Multi-Scale Validation Strategy for Albedo Products over Rugged Terrain and Preliminary Application in Heihe River Basin, China

Xingwen Lin, Jianguang Wen, Qinhuo Liu, Qing Xiao, Dongqin You, Shengbiao Wu, Dalei Hao, Xiaodan Wu
2018 Remote Sensing  
The issue for the validation of land surface remote sensing albedo products over rugged terrain is the scale effects between the reference albedo measurements and coarse scale albedo products, which is caused by the complex topography. This paper illustrates a multi-scale validation strategy specified for coarse scale albedo validation over rugged terrain. A Mountain-Radiation-Transfer-based (MRT-based) albedo upscaling model was proposed in the process of multi-scale validation strategy for
more » ... regating fine scale albedo to coarse scale. The simulated data of both the reference coarse scale albedo and fine scale albedo were used to assess the performance and uncertainties of the MRT-based albedo upscaling model. The results showed that the MRT-based model could reflect the albedo scale effects over rugged terrain and provided a robust solution for albedo upscaling from fine scale to coarse scale with different mean slopes and different solar zenith angles. The upscaled coarse scale albedos had the great agreements with the simulated coarse scale albedo with a Root-Mean-Square-Error (RMSE) of 0.0029 and 0.0017 for black sky albedo (BSA) and white sky albedo (WSA), respectively. Then the MRT-based model was preliminarily applied for the assessment of daily MODerate Resolution Imaging Spectroradiometer (MODIS) Albedo Collection V006 products (MCD43A3 C6) over rugged terrain. Results showed that the MRT-based model was effective and suitable for conducting the validation of MODIS albedo products over rugged terrain. In this research area, it was shown that the MCD43A3 C6 products with full inversion algorithm, were generally in agreement with the aggregated coarse scale reference albedos over rugged terrain in the Heihe River Basin, with the BSA RMSE of 0.0305 and WSA RMSE of 0.0321, respectively, which were slightly higher than those over flat terrain. Remote Sens. 2018, 10, 156 2 of 23 to estimate land surface albedos because of their large spatial scale coverage and a high revisit frequency [4] . However, the retrieved albedos suffer from large uncertainties due to the inherent complexity of the physical processes and their parameterization of retrieval algorithms [5] . Thus, it is critical to evaluate the performance of the retrieved albedos prior to their wide application. Many scientists have focused on assessing the accuracy of albedo products in recent decades over flat and homogeneous land surfaces. Taking the MODIS albedo products validation as an example, the validation results showed that the MODIS albedo products displayed high accuracy with an uncertainty (Root-Mean-Square-Error, RMSE) below 0.03 at the snow-free land covers and 0.07 at the snow-covered land surface, respectively, when the validation activities occurred in the homogeneous land surface or in sites with high spatial representativeness [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] . Generally, the albedo product can be assessed by direct comparison with in situ albedos at the sites where the land surface is sufficiently homogeneous [6, 9, 10, 20] . However, in the heterogeneous land surface, the in situ albedo cannot be directly compared with albedo products because of the scale mismatching between the in situ albedo and the albedo products, unless that in situ albedo can be considered with high spatial representativeness over the sampled area [10, 16, 20, 21] . The scale mismatching will result in about 15% disagreement between the MODIS albedo and in situ albedo [6, 9, 22] . As the sample sites with limited spatial representativeness, multi-points albedo observing is generally adapted to capture the spatial distribution characteristics of the albedo over the sampled area. The simplest and most efficient method is to average these albedos within the area and as the reference truth to compare with the albedo products [23] . Alternatively, the multi-scale validation strategy provides a solution to deal with the scale mismatching over a heterogeneous land surface by introducing fine scale albedo products (e.g., the Enhanced Thematic Mapper plus (ETM+) or China HJCCD (HJ) albedo) as an upscaling bridge between in situ albedo and coarse scale satellite albedo products (e.g., MODIS and Global LAnd Surface Satellite (GLASS) albedo) [8, 17, 24, 25] . In situ albedos are used to calibrate the fine scale albedo. Then, the calibrated fine scale albedos are aggregated to the coarse scale and for albedo validation. Previous studies have indicated that the upscaling of albedo from fine scale to coarse scale is highly linear over flat terrains [8, 16, 17, [24] [25] [26] . Therefore, in the case of flat terrain, the linear weighted average model was considered as a good performance model for upscaling the fine scale albedo to a coarse scale. As a special heterogeneous land surface, the topography has vast effects on the land surface albedo [24, 26] . Topographic slope, aspect, shadow, and solar location influence albedo values and their spatial distribution when compared with that over flat terrain [13, [27] [28] [29] [30] . The coarse scale albedo decreased with the increase of the slope facing away from the sun and increased when facing toward the sun [30, 31] , and generally showed larger values over the slope facing toward the sun than that facing away from the sun, especially, in the shadowing case [27, 30, 32] . The complex topography leads to the intensive scale effects on albedo products among different spatial resolutions over the rugged terrain [17, 33] . However, neglecting the scale effect caused by the complex topography in albedo products results in unreliable validation results [33] [34] [35] . Peng et al., (2014) assessed the MODIS products by using the multi-scale validation strategy with the HJ albedo as the bridge to aggregate the in situ albedo linearly to the MODIS pixel scale. The uncertainty distribution analysis showed that the largest scaling uncertainty was at the pixels over rugged terrain and its uncertainty of MODIS was as high as 0.07, when neglecting the scale effects at the upscaling progress over rugged terrain [17] . Therefore, neither the direct comparison nor the linear upscaling model in the multi-scale validation strategy were suitable for the coarse scale albedo products validation over rugged terrain [33] . The albedo spatial scale issue caused by topography should be emphasized in the multi-scale validation strategy over rugged terrain. The objective of this paper was to develop an upscaling method in the procedure of the multi-scale validation strategy for albedo products validation over rugged terrain. Simulated data with different mean slopes and solar zenith angle over nine Digital Elevation Models (DEM) were used to validate the albedo upscaling method. Based on the proposed upscaling method, the aggregated HJ albedo, Remote Sens. 2018, 10, 156 3 of 23 which had been validated by the in situ albedo over the Heihe River Basin, was used as the reference truth for the MODIS albedo products preliminary validation. The paper is organized as follows: Section 2 describes the multi-scale validation strategy, including the upscaling method and the fine scale albedo products retrieval algorithm; Section 3 describes the experimental area and validation dataset; Section 4 shows the performance of the albedo upscaling model and the preliminary validation results for MODIS albedo products. The discussions are summarized in Section 5. Finally, a brief conclusion is drawn in Section 6. Multi-Scale Validation Methodology Multi-Scale Validation Procedure over Rugged Terrain The general multi-scale validation strategy includes three key procedures over rugged terrain. The first one is the retrieval and accurate evaluation of fine scale albedo. The second is the calibration of fine scale albedo products. Finally, the third process is to scale up the fine scale albedo products to the coarse scale [17, 24] . The multi-scale validation strategy over rugged terrain has similar procedures than those over flat terrain, which is shown in Figure 1 . The DEM was used here to calculate the topographic factors (e.g., slope, aspect), and to couple with fine scale reflectance to retrieve fine scale albedo. The fine scale albedo was assessed by direct comparison with in situ albedos and was calibrated for reducing its uncertainties. An albedo upscaling model, which was based on the mountain radiation transfer theory (MRT-based albedo upscaling model), was proposed to aggregate the fine scale albedo to a coarse spatial scale. Consequently, the aggregated albedos could be directly compared with the coarse scale albedo products. To implement a successful multi-scale validation strategy over rugged terrain, two key issues should be solved including the albedo upscaling method and the fine scale albedo-generated algorithm on sloping surfaces.
doi:10.3390/rs10020156 fatcat:g37463plxbbhflxdf25lrsmnma