Variation among in situ soil moisture upscaling approaches for SMAP calibration/validation: implications for estimations of retrieval bias
Abstract
An important component of the SMAP mission is a comprehensive validation strategy that assesses the soil moisture retrievals using a number of monitoring networks located around the world. Among the features necessary for the networks to be selected as part of the core validation sites for the SMAP mission was a description of an appropriate upscaling procedure. Upscaling procedures are necessary to ensure that the measurements from the individual network locations are representative of the soil moisture within the larger SMAP pixel. Several different upscaling strategies have been described including arithmetic average (AA), invese distance weighting (IDW), Kriging (K), Voronoi diagrams (VD), temporal stability (TS) and soil weighted averages (SWA). This study evaluates differences from the use of six upscaling techniques to generate a representative footprint scale soil moisture average for the validation of the SMAP products. Further we evaluate the sensitivity of these strategies to station loss. The in situ soil moisture data was obtained from 30 stations within a network located near Kenaston, Saskatchewan during the 2015 growing season. An estimate of the upscaled mean was determined using each approach. Station drop out was simulated using a Monte Carlo procedure where the upscaled average was calculated using each approach 10000 times for VD, IDW, AA and SWA with the number of available stations reduced from 30 to 5 stations. Method agreement was highest during dry down periods but reduced following precipitation events. Comparing among the upscaled estimates to the SMAP soil moisture retrievals found average differences (bias) of 0.017, 0.01, 0.04, 0.032, 0.042 and 0.039 m3 m-3 for K, VD, SWA, IDW, TS and AA, respectively. A statistically significant association was found between method variance with the SMAP retrieval bias (and similarly between the network CV and retrieval bias). Based on our results it is suggested that when the network CV is beyond 0.4 calibration/validation results should be treated with some caution. With respect to station drop out, the VD, IDW and AA showed the least sensitivity, with minimal change in the pixel-scale soil moisture regardless of the number of stations used. However the SWA was most sensitive to station drop out.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2016
- Bibcode:
- 2016AGUFM.H31G1481B
- Keywords:
-
- 1833 Hydroclimatology;
- HYDROLOGYDE: 1843 Land/atmosphere interactions;
- HYDROLOGYDE: 1855 Remote sensing;
- HYDROLOGYDE: 1866 Soil moisture;
- HYDROLOGY