Yu-Guo Zhao

Modeling spatio-temporal distribution of soil moisture by deep learning-based cellular automata model

Soil moisture content (SMC) is a key hydrological parameter in agriculture, meteorology and climate change, and understanding of spatio-temporal distributions of SMC in farmlands is important to address the precise irrigation scheduling. However, the hybrid interaction of static and dynamic environmental parameters makes it particularly difficult to accurately and reliably model the distribution of SMC. At present, deep learning wins numerous contests in machine learning and hence deep belief network (DBN), a breakthrough in deep learning is trained to extract the transition functions for the simulation of the cell state changes. In this study, we used a novel macroscopic cellular automata (MCA) model by combining DBN to predict the SMC over an irrigated corn field (an area of 22 km2) in the Zhangye oasis, Northwest China. Static and dynamic environmental variables were prepared with regard to the complex hydrological processes. The widely used neural network, multi-layer perceptron (MLP), was utilized for comparison to DBN. The hybrid models (MLP-MCA and DBN-MCA) were calibrated and validated on SMC data within four months, i.e. June to September 2012, which were automatically observed by a wireless sensor network (WSN). Compared with MLP-MCA, the DBN-MCA model led to a decrease in root mean squared error (RMSE) by 18%. Thus, the differences of prediction errors increased due to the propagating errors of variables, difficulties of knowing soil properties and recording irrigation amount in practice. The sequential Gaussian simulation (sGs) was performed to assess the uncertainty of soil moisture estimations. Calculated with a threshold of SMC for each grid cell, the local uncertainty of simulated results in the post processing suggested that the probability of SMC less than 25% will be difference in different areas at different time periods. The current results showed that the DBN-MCA model performs better than the MLP-MCA model, and the DBN-MCA model provides a powerful tool for predicting SMC in highly non-linear forms. Moreover, because modeling soil moisture by using environmental variables is gaining increasing popularity, DBN techniques could contribute a lot to enhancing the calibration of MCA-based SMC estimations and hence provide an alternative approach for SMC monitoring in irrigation systems on the basis of canals.

Vertical distribution and storage of soil organic and inorganic carbon in a typical inland river basin, Northwest China

Knowledge of soil carbon (C) distribution and its relationship with the environment can improve our understanding of its biogeochemical cycling and help to establish sound regional models of C cycling. However, such knowledge is limited in environments with complex landscape configurations. In this study, we investigated the vertical distribution and storage of soil organic carbon (SOC) and soil inorganic carbon (SIC) in the 10 representative landscapes (alpine meadow, subalpine shrub and meadow, mountain grassland, mountain forest, typical steppe, desert steppe, Hexi Corridor oases cropland, Ruoshui River delta desert, Alxa Gobi desert, and sandy desert) with contrasting bioclimatic regimes in the Heihe River Basin, Northwest China. We also measured the 87Sr/86Sr ratio in soil carbonate to understand the sources of SIC because the ratio can be used as a proxy in calculating the contribution of pedogenic inorganic carbon (PIC) to total SIC. Our results showed that SOC contents generally decreased with increasing soil depth in all landscapes, while SIC contents exhibited more complicated variations along soil profiles in relation to pedogenic processes and parent materials at the various landscapes. There were significant differences of C stocks in the top meter among different landscapes, with SOC storage ranging from 0.82 kg C/m2 in sandy desert to 50.48 kg C/m2 in mountain forest and SIC storage ranging from 0.19 kg C/m2 in alpine meadow to 21.91 kg C/m2 in desert steppe. SIC contributed more than 75% of total C pool when SOC storage was lower than 10 kg C/m2, and the proportion of PIC to SIC was greater than 70% as calculated from Sr isotopic ratio, suggesting the critical role of PIC in the C budget of this region. The considerable variations of SOC and SIC in different landscapes were attributed to different pedogenic environments resulted from contrasting climatic regimes, parent materials and vegetation types. This study provides an evidence for a general trade-off pattern between SOC and SIC, showing the compensatory effects of environmental conditions (especially climate) on SOC and SIC formation in these landscapes. This is largely attributed to the fact that the overall decrease in temperature and increase in precipitation from arid deserts to alpine mountains simultaneously facilitate the accumulation of SOC and depletion of SIC.

An Insight into Machine Learning AlgorithmstoMapthe Occurrence of Soil MatticHorizon in the Northeastern Qinghai-Tibetan Plateau

Abstract Soil diagnostic horizons, which each have a set of quantified properties, play a key role in soil classification. However, they are difficult to predict, and few attempts have been made to map their spatial occurrence. We evaluated and compared four machine learning algorithms, namely, the classification and regression tree (CART), random forest (RF), boosted regression trees (BRT), and support vector machine (SVM), to map the occurrence of the soil mattic horizon in the northeastern Qinghai-Tibetan Plateau using readily available ancillary data. The mechanisms of resampling and ensemble techniques significantly improved prediction accuracies (measured based on area under the receiver operator characteristic curve score (AUC)) and produced more stable results for the BRT (AUC of 0.921 ± 0.012, mean ± standard deviation) and RF (0.908 ± 0.013) algorithms compared to the CART algorithm (0.784 ± 0.012), which is the most commonly used machine learning method. Although the SVM algorithm yielded a comparable AUC value (0.906 ± 0.006) to the RF and BRT algorithms, it is sensitive to parameter settings, which are extremely time-consuming. Therefore, we consider it inadequate for occurrence-distribution modeling. Considering the obvious advantages of high prediction accuracy, robustness to parameter settings, the ability to estimate uncertainty in prediction, and easy interpretation of predictor variables, BRT seems to be the most desirable method. These results provide an insight into the use of machine learning algorithms to map the mattic horizon and potentially other soil diagnostic horizons.

Mapping Soil Organic Carbon Using Local Terrain Attributes: A Comparison of Different Polynomial Models

Abstract Local terrain attributes, which are derived directly from the digital elevation model, have been widely applied in digital soil mapping. This study aimed to evaluate the mapping accuracy of soil organic carbon (SOC) concentration in 2 zones of the Heihe River in China, by combining prediction methods with local terrain attributes derived from different polynomial models. The prediction accuracy was used as a benchmark for those who may be more concerned with how accurately the variability of soil properties is modeled in practice, rather than how morphometric variables and their geomorphologic interpretations are understood and calculated. In this study, 2 neighborhood types (square and circular) and 6 representative algorithms (Evans-Young, Horn, Zevenbergen-Thorne, Shary, Shi, and Florinsky algorithms) were applied. In general, 35 combinations of first- and second-order derivatives were produced as candidate predictors for soil mapping using two mapping methods (i.e., kriging with an external drift and geographically weighted regression). The results showed that appropriate local terrain attribute algorithms could better capture the spatial variation of SOC concentration in a region where soil properties are strongly influenced by the topography. Among the different combinations of first- and second-order derivatives used, there was a best combination with a more accurate estimate. For different prediction methods, the relative improvement in the two zones varied between 0.30% and 9.68%. The SOC maps resulting from the higher-order algorithms (Zevenbergen-Thorne and Florinsky) yielded less interpolation errors. Therefore, it was concluded that the performance of predictive methods, which incorporated auxiliary variables, could be improved by attempting different terrain analysis algorithms.

Mapping Soil Thickness by Integrating Fuzzy C-Means with Decision Tree Approaches in a Complex Landscape Environment

Predictive soil mapping depends on understanding the relationships between soil properties and environmental factors. However, in a complex soil landscapes, there is a shortage of suitable approaches to establish these relationships. The main objective is to predict soil thickness in an alpine watershed relating to soil environmental factors based on an unsupervised fuzzy clustering method (fuzzy c-means, FCM) and decision tree (DT) method. In this study, FCM method was used for stratifying the landscape, and then, a representative soil thickness was assigned to each class. For each class, a number of points were randomly chosen in proportion to representative areas, and then, the environmental factors at these point locations were extracted as a training data set (3626 points). For the training data set, DT method was used to obtain the critical threshold of soil–environment relationships. Finally, soil thickness map was created by applying the results of the DT across the region. An independently collected field sampling set (31 points) was used to evaluate the effectiveness of the proposed approach. For training set, 95.48 % of the total training data were correctly predicted. For validation set, the overall accuracy and Kappa coefficient could reach 74.2 % and 0.659, respectively. Evaluation accuracy of soil map was up to 74.2 %. In conclusion, it is suggested that soil–landscape modeling using FCM and DT methods can be efficiently used as a valuable research technique for spatial soil thickness prediction in a complex soil landscape where soil characteristics and properties are not available.

Digital Soil Mapping Across Paradigms, Scales, and Boundaries: A Review

Accurate spatial soil information is urgently needed for dealing with the global issues such as agricultural production, environmental pollution, food security, water security, and human health. This need has been motivating the development of digital soil mapping. We reviewed recent advances in digital soil mapping with respect to paradigms, scales, and boundaries, with the intent to improve our understanding on current status of soil mapping. Some important challenges thus research opportunities emerged recently were then outlined, such as 3D digital mapping of the soil properties beyond soil organic matter, soil mapping in areas with intensive human activities, and multi-source soil data integration for soil mapping.

Mapping Soil Organic Matter in Low-Relief Areas Based on Time Series Land Surface Diurnal Temperature Difference

Accurate estimates of the spatial variability of soil organic matter (SOM) are necessary to properly evaluate soil fertility and soil carbon sequestration potential. In plains and gently undulating terrains, soil spatial variability is not closely related to relief, and thus, digital soil mapping methods based on soil–landscape relationships often fail in these areas. It is necessary to find new environmental variables and methods to mapping soil attribute over the low-relief areas. Time series remotely sensed data, such as thermal imagery, provide possibilities for mapping SOM in such areas. In this study, Jiangsu Province was chosen as an example in eastern China and a total of 1519 soil samples (0 ~ 20 cm layer) were collected from the Second National Soil Survey of Jiangsu Province. 8-day composited land surface diurnal temperature difference (DTD) was extracted from the time series of MODIS 8-day composited land surface temperature. 8-day averaged DTD was mean of 8-day composited DTD in the same periods between 2002 and 2011. Analysis showed that SOM content was significantly negative correlated with 8-day averaged DTD of different periods, of which higher correlation was in vegetation sparse periods. Averaged DTD of many periods and averaged DTD of specific periods were selected as two group of independent variable dataset. Linear regression, regression kriging (RK), and linear mixed model (LMM) fitted by residual maximum likelihood were used to model and map SOM spatial distribution. Ordinary kriging was used as a baseline comparison. The root-mean-squared error, mean error, and mean absolute error calculated from independent validation were used to assess prediction accuracy. Results showed that LMM are the best predictions, of which LMM using DTD of specific periods and DTD cell statistics as variables performed best. RK were somewhat worse than LMM. Linear regression performed worst. This suggests that time series remotely sensed data can provide useful auxiliary variable for mapping SOM in low-relief agricultural areas and LMM improved mapping SOM spatial distribution, which provided an effective approach for improving DSM in the low-relief areas.