Sie sind auf Seite 1von 6

Digital Soil Assessments and Beyond Minasny, Malone & McBratney (eds) 2012 Taylor & Francis Group,

p, London, ISBN 978-0-415-62155-7

Spatial agricultural soil quality evaluation based on digital soil maps and uncertainty analysis
X.L. Sun
State Key Laboratory of Soil and Sustainable Agriculture, Institute of Soil Science, Chinese Academy of Sciences, Nanjing, China College of Agriculture, Guangxi University, Nanning, China

Y.G. Zhao, M.S. Zhao & G.L. Zhang


State Key Laboratory of Soil and Sustainable Agriculture, Institute of Soil Science, Chinese Academy of Sciences, Nanjing, China

ABSTRACT: Soil quality information is used for managing soil resources. This study evaluated the soil quality for agricultural lands of Hong Kong and analyzed its associated uncertainty. First, spatially correlated soil quality indicators were mapped using geostatistical tools, while spatially uncorrelated indicators were mapped by combining soil series means and inverse distance-weighted residuals from the means. Then, the correlated indicators were simulated using stochastic simulation and values of the uncorrelated indicators were randomly generated based on the statistical characteristics of a soil series. Results showed that the agricultural soil quality of this study area was moderate, with an average index value of 0.68. The produced soil quality information contained moderate uncertainty and was credible with a confidence level above 0.90 on most of the area, given the derived uncertainty. However, based on probability samples, the soil quality information was moderately accurate and the derived uncertainty actually was not so good. Efforts are needed to improve uncertainty characterization for digital soil quality assessment. 1 INTRODUCTION (Carr et al., 2007). Thus, recent studies on digital soil assessment generally focused on this issue, e.g., Nol et al. (2010), Heuvelink et al. (2010) and de Gruijter et al. (2011). One reason for this is that uncertainty associated with digital soil information indicates how to use the information appropriately. The main objective of this study was to provide soil quality information for agricultural lands of Hong Kong. We first evaluated soil quality based on digital soil mapping of soil quality indicators and then analyzed the associated uncertainty.

Soil quality describes soil function in the most comprehensive way and therefore it is the most desired information for managing soil resources for various uses (Allen et al., 2011), particularly for long-term uses on a large scale. When agriculture, environment and ecology are increasingly being managed more precisely, soil quality information are required in greater detail and at high accuracy. Digital soil assessment was proposed by Carr et al. (2007) to infer difficult-to-measure soil information, such as soil quality, based on digital maps of individual soil attributes and/or types. By inheriting the biggest advantage of digital soil maps, i.e. deriving uncertainties for outcomes, the digital soil assessment approach provides soil information and the associated uncertainty. Since digital soil maps for digital soil assessment usually depict every grid of an interested area, the outcomes would be detailed. Therefore, the digital soil assessment approach could be the best way to generate soil quality information. The crucial issue with the digital soil assessment approach is uncertainty propagation

2 2.1

MATERIALS AND METHODS Study area and sampling

More than 95% of the agricultural areas of Hong Kong (located on the southern tip of China) are distributed throughout the northern central region (Fig. 1). The total area is 110.6 km2. The climate is subtropical, with an average annual air temperature of 23C and precipitation of 2214 mm. The topography of the study area is quite complex. Steep hills and rugged mountains separate the agricultural

37

lands into flat patches (Fig. 1). Less than half of the area is cultivated with vegetables (major), flowers and scenic trees (minor), and fruits (very few), whereas the remainder is left to fallow. The soils are similar to Ultisols, Oxisols, Alfisols and Inceptisols of the US Soil Taxonomy and

Figure 1.

Study area and sampling locations.

Anthrosols of the World Reference Base for Soil Resources. Forty-one soil profiles were firstly collected to obtain relevant information on the local 25 soil series (Sun, 2011), where the number of profiles for each soil series was determined in proportion to its area. Sampling sites were generally located evenly within the patch/patches of the soil series. Ninetytwo topsoil samples were secondly collected, where the locations of the sampling points were selected using SPCOSA package (Walvoort et al., 2010) for R, with the above 41 soil profile locations as prior points. At each of the sampling sites (including the prior points), soil samples at a depth of 020 cm were collected. In addition to these samples, another 35 topsoil samples were randomly selected, in order to validate evaluated soil quality and derived uncertainty of the evaluation. Soil properties listed in Table 1 were measured, except saturated hydraulic conductivity (KS).

Table 1. PCs

Results of principal component analysis of soil quality indicators in agricultural areas of Hong Kong. PC1 2.33 24.77 PC2 1.63 12.03 PC3 1.58 11.32 PC4 1.34 8.17 PC5 1.23 6.85 PC6 1.16 6.12 PC7 1.06 5.08 PC8 1.02 4.73

Eigenvalue Proportion Eigenvectors A-deptha (cm) Obs-depth (cm) CA (m2 m1) KS (cm d1) pH Bulk density (g cm3) Sand (%) Silt (%) Clay (%) SOM (g kg1) CEC (cmol kg1) TN (g kg1) TP (g kg1) TK (g kg1) NH4+-N (mg kg1) A-P (mg kg1) A-K (mg kg1) A-Cu (mg kg1) A-Fe (mg kg1) A-Mn (mg kg1) A-Zn (mg kg1) A-B (mg kg1)

0.01 0.09 0.10 0.21 0.22 0.16 0.38b 0.37c 0.31 0.12 0.02 0.01 0.32 0.16 0.05 0.33 0.12 0.13 0.24 0.15 0.28 0.08

0.10 0.02 0.10 0.28 0.05 0.38 0.06 0.04 0.08 0.39 0.25 0.24 0.05 0.06 0.31 0.04 0.13 0.12 0.11 0.25 0.27 0.43

0.44 0.28 0.05 0.18 0.13 0.22 0.01 0.02 0.03 0.12 0.33 0.34 0.10 0.28 0.21 0.17 0.01 0.06 0.20 0.34 0.17 0.20

0.30 0.37 0.11 0.01 0.31 0.07 0.22 0.21 0.09 0.21 0.29 0.29 0.20 0.10 0.34 0.17 0.16 0.01 0.11 0.27 0.17 0.04

0.03 0.14 0.01 0.13 0.31 0.13 0.18 0.20 0.04 0.05 0.24 0.32 0.05 0.50 0.12 0.18 0.23 0.34 0.36 0.04 0.13 0.07

0.16 0.14 0.30 0.43 0.25 0.41 0.04 0.05 0.06 0.37 0.15 0.12 0.08 0.07 0.13 0.14 0.30 0.22 0.01 0.09 0.01 0.30

0.05 0.14 0.48 0.19 0.01 0.12 0.01 0.06 0.24 0.12 0.20 0.07 0.08 0.28 0.06 0.24 0.42 0.44 0.11 0.23 0.06 0.06

0.16 0.30 0.41 0.22 0.18 0.15 0.10 0.04 0.36 0.32 0.04 0.02 0.10 0.15 0.24 0.10 0.39 0.22 0.02 0.05 0.16 0.17

a A-depth: layer A depth; obs-depth: obstacle horizon depth; CA: catchment area; KS: Saturated hydraulic conductivity; SOM: soil organic matter; CEC, cation exchange capacity; TN, total nitrogen; TP, total phosphorous; TK, total potassium; NH4+-N, ammonium nitrogen; A-P, available phosphorous; A-K, available potassium; A-Cu, available copper; A-Fe, available iron; A-Mn, available manganese; A-Zn, available zinc; A-B, available boron. b Loadings in bold correspond to selected indicators. c Underlined loadings are considered highly weighted.

38

KS was determined using a pedotransfer function based on clay, silt, soil organic matter (SOM) and bulk density (Sun, 2011). Catchment area (CA) was derived from the local 25 meter digital elevation model. 2.2 Soil quality evaluation system

Table 2. Selected soil quality indicators and their scoring functions, communalities and weights for agricultural soil quality assessment of Hong Kong. Indicator A-depth Obs-depth CA KS Sand TK NH4+-N A-K A-Cu A-B Scoring function Upper limit Upper limit Peak limit Upper limit Peak limit Upper limit Upper limit Upper limit Upper limit Upper limit Lower limit 12 0 5,000 30,000 5 20 40 8 10 25 3 0.15 Upper limit 25 100 80,000 700,000 15 70 85 12 14 45 6.5 0.3 Weight 0.165 0.100 0.012 0.171 0.232 0.099 0.066 0.022 0.109 0.023

Minimum data set and integrated quality index were used in this study. First, principal component analysis was conducted to select the most appropriate soil quality indicators for the study area (Qi et al., 2009). The results of this analysis was presented in Table 1. Only principal components (PC) with eigenvalue larger than or equal to 1 were considered for soil quality evaluation. Within each PC, indicators receiving weighted loading values within 10% of the highest weighted loading were selected as the most appropriate, which are highlighted with loadings in bold in Table 1. Then, the selected soil quality indicators were scored using standard scoring functions (Sun et al., 2003; Qi et al., 2009). According to the indicators functions on soil quality, the following two types of standard scoring functions were used: (1) upper limit: 0.1, x L f x ) = 0.9 + 0.1, U L 1, (2) peak limit:
0.1, x L1 + 0.1, 0.9 L 2 L1 f x) = 1, U2 x 0.9 U U + 0.1, 2 1

where Wi is the assigned weight, Ni is the indicator score, and n is the number of indicators. The weights shown in Table 2 for the indicators were obtained according to the indicators communalities, which were derived from the standardized factor analysis on all indicator values using the factor procedure of SAS. 2.3 Soil mapping and uncertainty analysis

x L

L x U (1) Multivariate geostatistical analysis was conducted to analyze spatial structures of the selected soil quality indicators and the soil properties for calculating KS, i.e., SOM, bulk density, silt and clay (clay was computed by 100 minus sand and silt). However, layer A-depth and obstacle horizon depth were not considered in this analysis since only 41 profiles were collected in this study. The Matherons and MVE the robust estimators in Lark (2003), i.e., u,vV and u,v , were used to construct auto- and crossM variogram models, in order to tackle outliers in the sample data, because outliers would lead to poor estimates of mapping uncertainty. For soil properties that were not correlated with others, variogram models were then constructed using four estimates for univariate geostatistics analysis in Lark (2000), i.e., Matheron, Cressie-Hawking, Dowd and Gentons estimators. All constructed variogram models were cross-validated and appropriate models were selected using the method of Lark (2000). Based on the selected models, ordinary kriging and cokriging were performed to make maps for spatially correlated and cross-correlated soil properties, respectively. Unfortunately, some soil properties, even their residuals from soil series means, didn't have a spatial structure indicated by the sample data of this study. Therefore, these soil properties were mapped by combining soil series means

x U

x L1 or x U 2 L1 x L2 L2 x U1 (2)

U1 < x U 2

where f is the score, x is the soil property value, L and U are the lower and upper threshold values, respectively. The values of L and U for each indicator were determined according to many other studies, local standards and farming experiences of local farmers (Sun, 2011), which are shown in Table 2. Finally, soil quality index was calculated based on the scores obtained from the above and weights, using the method of integrated quality index: SQI

Wi Ni
i =1

(3)

39

and extrapolated residuals from the means using inverse distance-weighting (Sun, 2011). All the maps were then combined into the above soil quality evaluation system, to generate a soil quality map of this study area. For indicators with a valid variogram model, their uncertainties were derived using sequential Gaussian simulation, where co-simulation was used for cross-correlated indicators, to generate 500 realizations at each site. For an indicator without a valid variogram, 500 values at each site were randomly drawn from normal distributions with parameters of locations and scales of the corresponding soil series mean and standard deviation. The 500 realizations and values were then inputted into the soil quality evaluation system, in order to analyze uncertainty of the mapped soil quality. The uncertainty was calculated as the standard deviation of the 500 soil quality realizations. Cumulative probability of the mapped soil quality in the 500 soil quality realizations was also computed. Further, the soil quality map and the derived uncertainty were validated based on the 35 randomly selected samples. The mean error (ME), mean absolute error (MAE) and standard deviation of error (sE) of the soil quality map were calculated: ME = ME = 1 N ( Pi N i =1 1 N Pi N i =1 Oi ) Oi Oi ME )2 N 1 (4) (5)

where Qi,(1p)/2 and Qi,(1+p)/2 are the (1p)/2 and (1+p)/2 quantiles of the 500 simulated soil quality indice at site i, Oi is the observed soil quality index at site i.

3 3.1

RESULTS AND DISCUSSION Geostatistical analysis

Figure 2 shows the constructed vairogram models for cross-correlated soil quality indicators, i.e., sand, silt and TK, based on Matherons estimator, while Figure 3 for auto-correlated indicators, i.e., SOM, box-cox transformed NH4+N, A-K and A-B. Variograms of bulk density and A-Cu had only nuggets, so they were not presented. Results of cross-validation on all the models are presented in Table 3, indicating all the models based on Matherons estimator gave squared standardized prediction errors (Lark, 2000) in their confidence levels of 95%. Therefore, mapping uncertainty would not be impacted by outliers in this study and these models were used to map the indicators. 3.2 Soil quality and uncertainty

sE =

( Pi
i =1

(6)

where Pi and Qi are mapped and observed soil quality indices at site i, respectively. The accuracy of a p-probability interval ( ( p )) , goodness of probability intervals for a series of p (G), and average wideness for a probability p(W ( p )) (Goovaerts, 2001) were calculated for the derived uncertainty:

1, if Qi ,( p ) / 1 N ( p) i , i = 0, otherwise N i =1 otherw Oi Qi ,(1+ p ) / 2

(7) G 1 1, if ( p ) p a( p ) = 0, otherwise W ( p) =
N 1 (Q i (Qi N ( p ) i =1 p) /

0 (3a( p)

2 )( ( p ) p )dp, p (8)

Qi

p) / 2 )

(9)

Figure 4(a) shows the evaluated soil quality index based on maps of soil quality indicators made using ordinary cokriging (for sand, TK, silt), ordinary kriging (for SOM, NH4+N, A-K and A-B), soil series mean and inverse distance-weighted residuals from soil series mean (for A-depth, Obsdepth, bulk density and A-Cu). The average soil quality index was 0.68, with a standard deviation of 0.10 and a range of 0.320.93. The low spatial variability of soil quality index reflects that most of the study area is of medium quality. Visually, spatial distribution of the index was very complex. This can be explained by the complex distribution of topography and soil series in this study area (Sun, 2011). Based on the 35 validation samples, the ME, MAE and sE of the soil quality map were 0.09, 0.12 and 0.11, respectively. The ME and MAE accounted for 13% and 18% of the average soil quality index (i.e., 0.68), respectively, reflecting that the soil quality map was moderately accurate. Figure 4(b) shows the uncertainty of the evaluated soil quality, characterized using the standard deviation of the 500 soil quality index realizations at each site. The uncertainty was generally normally distributed in the area, with a mean of 0.10 and a standard deviation of 0.09. The uncertainty accounts, on average, for 14% of the soil quality index in Figure 4(a), with a range of 831%.

40

Figure 2. Variogram models of cross-correlated soil quality indicators, i.e., sand, silt and TK. Sph represents a spherical model and numbers in parentheses indicate ranges of the models in unit of meter.

Figure 3. Variograms of auto-correlated soil quality indicators: SOM (a), NH4+-N (b), A-K (c) and A-B (d), Except NH4+-N was transformed using box-cox, the others were transformed using natural logarithm. Sph and lin represents spherical and linear models, respectively, and numbers in parentheses indicates ranges of the models in unit of meter (except no range in (d)).

Table 3. models.

Squared standardized prediction error of the Matheron MVEa Mb C-Hc Dowd Genton

Sand Silt TK SOM NH4+-N A-K A-B


a b c

0.37 0.40 0.47 0.54 0.43 0.49 0.35

0.66 0.44 0.52

0.42 0.42 0.45 0.55 0.44 0.47 0.43 0.51 0.43 0.47 0.51 0.55 0.46 0.49 0.42

MVE u,vV M u,v Cressie-Hawking.

This suggests that the uncertainty of the soil quality information was moderate. Figure 4(c) shows the cumulative probability of the soil quality index in Figure 4(a) among

the 500 soil quality index realizations. Most of the study area, i.e., 82%, had a probability between 0.2 and 0.7, while only 0.3% and 1.1% had a probability less than 0.1 and larger than 0.9, respectively. This indicates that on most of the study area, the obtained soil quality index in Figure 4(a) was credible, given the above uncertainty. Figure 5 plots the accuracy and average wideness of the uncertainty against a series of probabilities. Only for p > 0.93, more than p percent of p-probability intervals included true values. Therefore, p-probability intervals based on the uncertainty were correct only for p > 0.93. However, the average wideness for p > 0.93 was larger than 0.36, a relatively high value for soil quality index. Particularly, the goodness of the uncertainty was 0.68. Thus, the uncertainty was not so good (Goovaerts, 2001). This is due to that many uncertainty sources, i.e., the maps of soil quality indicators, collectively contributed uncertainty to the soil quality information.

41

area with a confidence level above 0.90, given the derived uncertainty. However, the soil quality information was moderately accurate and the derived uncertainty actually was not so good. Therefore, more efforts are needed to improve uncertainty characterization for digital soil quality assessment. ACKNOWLEDGEMENT This study was supported by the Basic Research Program of Jiangsu (BK2008058), the International Science & Technology Cooperation Project of MOST, China (2010DFB24140) and the Public Policy Research (2002-PPR-3) of Research Grants Council of Hong Kong. REFERENCES
Allen, D.E., Singh, B.P. & Dalal, R.C. 2011. Soil health indicators under climate change: a review of current knowledge. In B.P. Singh, A.L. Cowie & K.Y. Chan (eds), Soil Health and Climate Change: 2545. Heidelberg: Springer. Carr, F., McBratney, A.B., Mayr, T. & Montanarella, L. 2007. Digital soil assessments: Beyond DSM. Geoderma 142: 6979. de Gruijter, J.J., Walvoort, D.J.J. & Bragato, G. 2011. Application of fuzzy logic to boolean models for digital soil assessment. Geoderma 166: 1533. Goovaerts, P. 2001. Geostatistical modelling of uncetainty in soil science. Geoderma 103: 326. Heuvelink, G.B.M., Burgers, S.L.G.E., Tiktak, A. & Van Den Berg, F. 2010. Uncertainty and stochastic sensitivity of the GeoPEARL pesticide leaching model. Geoderma 155: 186192. Lark, R.M. 2000. A comparison of some robust estimators of the variogram for use in soil survey. European Journal of Soil Science 51: 137157. Lark, R.M. 2003. Two robust estimators of the crossvariogram for multivariate geostatistical analysis of soil properties. Geoderma 54: 187201. Nol, L., Heuvelink, G.B.M., de Vries, W. & Kros, J. 2010. Uncertainty propagation analysis of an N2O emission model at the plot and landscape scale. Geoderma 159: 923. Qi, Y., Darilek, J.L., Huang, B., Zhao, Y., Sun, W. & Gu, Z. 2009. Evaluating soil quality indices in an agricultural region of Jiangsu Province, China. Geoderma 149: 325334. Sun, B., Zhou, S.L. & Zhao, Q.G. 2003. Evaluation of spatial and temporal changes of soil quality based on geostatistical analysis in the hill region of subtropical China. Geoderma 115: 8599. Sun, X.L. 2011. Digital Soil mapping and its application for assessing the effects of urbanization on soil properties and agricultural soil quality in Hong Kong. Hong Kong: Hong Kong Baptist University, PhD thesis. Walvoort, D.J.J., Brus, D.J. & de Gruijter, J.J. 2010. An R package for spatial coverage sampling and random sampling from compact geographical strata by k-means. Computer & Geosciences 36: 12611267.

Figure 4. The evaluated soil quality index (a), associated uncertainty (b) and cumulative probability (c).

Figure 5. The plots of accuracy (a) and average wideness (b) against a series of probabilities. The line is a 1:1 relationship. Average wideness is not available for p < 0.2 because accuracies for p < 0.2 were all 0 in (a).

CONCLUSIONS

Only three soil properties were cross-correlated in the agricultural areas of Hong Kong. Spatial outliers in the collected samples could not impact seriously the soil mapping of this study. Soil quality of this study area was moderate, with an average index value of 0.68. Associated uncertainty of the derived soil quality information was moderate, according to the standard deviation of the soil quality index, and the derived soil quality information was credible on most of the study

42

Das könnte Ihnen auch gefallen