Premium
Effect of sampling density on regional soil organic carbon estimation for cultivated soils
Author(s) -
Sun Weixia,
Zhao Yongcun,
Huang Biao,
Shi Xuezheng,
Landon Darilek Jeremy,
Yang Jinsong,
Wang Zhigang,
Zhang Beier
Publication year - 2012
Publication title -
journal of plant nutrition and soil science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.644
H-Index - 87
eISSN - 1522-2624
pISSN - 1436-8730
DOI - 10.1002/jpln.201100181
Subject(s) - kriging , sampling (signal processing) , mean squared error , environmental science , soil water , spatial variability , soil science , soil carbon , standard error , mathematics , statistics , filter (signal processing) , computer science , computer vision
An extensive knowledge of how sampling density affects soil organic C (SOC) estimation at regional scale is imperative to reduce uncertainty to a meaningful confidence level and aid in the development of sampling schemes that are both rational and economical. Using kriging prediction, this paper examined the effect of sampling density on regional SOC‐concentration estimations in cultivated topsoils at six scales in a 990 km 2 area of Yucheng County, a typical region in the N China Plain. Except the original data set ( n = 394), five other sampling densities were recalculated using grids of 8 km × 8 km ( n = 28), 8 km × 4 km ( n = 44), 4 km × 4 km ( n = 82), 4 km × 2 km ( n = 142), and 2 km × 2 km ( n = 257), respectively. Experimental SOC semivariances and kriging interpolations at six sampling density scales were calculated and modeled to estimate regional SOC variability. Accuracy of the effects of the five sampling densities on regional SOC estimations was assessed using the indices of mean error (ME) and root mean square error (RMSE) with 100 independent validation samples. By comparison with the kriged grid map derived from the 394 samples data set, the relative error (RE,%) was spatially calculated to highlight the spatial variability of prediction errors at five sampling‐density scales due to the intrinsic limitations of ME and RMSE in accuracy assessment. The results indicated that sampling density significantly affected the estimation of regional SOC concentration. Particularly when the sampling density was < 4 km × 4 km, the large spatial variation of SOC was concealed. Semivariance analysis indicated that different sampling density had significant effect on reasonable detection of the dominant factors which influenced SOC spatial variation. Greater sampling density could more exactly reveal regional SOC variation caused by human management. The prediction accuracy for regional SOC estimation increased with the increasing of sampling density. The critical areas with larger RE values should be intensified in the future sampling scheme, and the areas of lower RE values should be decreased relatively. A specific sampling scheme should be considered in accordance with the demand to the estimation accuracy of regional SOC stock at a certain confidence level. Our results will facilitate a better understanding of the effect of sampling density on regional SOC estimation for future sampling schemes by providing meaningful confidence levels.