Multiple regression analysis is a valuable method to reduce information gaps in a sparse soil moisture data set by fusing its information content with those of densely mapped data sets. Regression ...analysis utilizing uncertain data results in an indeterminate regression model and indeterminate soil moisture predictions when applying the regression model. We employ an unsupervised multiple regression approaches, taking optimally located sparse soil moisture measurements directly as coefficients in a linear regression model. We propagate data uncertainties into our probabilistic soil moisture estimation results by embedding the regression in a Monte Carlo approach. The computed uncertainty defines the quantitative limit for information retrieval from the resultant ensemble of soil moisture maps. This raises doubts on the true presence of some prominent channel‐like features of increased soil moisture that are clearly visible in a previously and deterministically derived soil moisture map ignoring the presence of data uncertainty. The approach followed in this work is computationally simple and could be applied routinely to databases of similar size. Insufficient uncertainty communication by the data provider became the biggest obstacle in our efforts and led us to the insight that the geoscientific community may need to revise their standards with regard to uncertainty communication related to measured and processed data.
Core Ideas
Data uncertainty propagation through regression by means of a Monte Carlo approach.
Unsupervised nonlinear regression and its dependency on optimal sparse sampling.
Uncertainty communication for proper information retrieval.
Soil acidification is caused by natural paedogenetic processes and anthropogenic impacts but can be counteracted by regular lime application. Although sensors and applicators for variable-rate liming ...(VRL) exist, there are no established strategies for using these tools or helping to implement VRL in practice. Therefore, this study aimed to provide guidelines for site-specific liming based on proximal soil sensing. First, high-resolution soil maps of the liming-relevant indicators (pH, soil texture and soil organic matter content) were generated using on-the-go sensors. The soil acidity was predicted by two ion-selective antimony electrodes (RMSE
pH
: 0.37); the soil texture was predicted by a combination of apparent electrical resistivity measurements and natural soil-borne gamma emissions (RMSE
clay
: 0.046 kg kg
−1
); and the soil organic matter (SOM) status was predicted by a combination of red (660 nm) and near-infrared (NIR, 970 nm) optical reflection measurements (RMSE
SOM
: 6.4 g kg
−1
). Second, to address the high within-field soil variability (pH varied by 2.9 units, clay content by 0.44 kg kg
−1
and SOM by 5.5 g kg
−1
), a well-established empirical lime recommendation algorithm that represents the best management practices for liming in Germany was adapted, and the lime requirements (LRs) were determined. The generated workflow was applied to a 25.6 ha test field in north-eastern Germany, and the variable LR was compared to the conventional uniform LR. The comparison showed that under the uniform liming approach, 63% of the field would be over-fertilized by approximately 12 t of lime, 6% would receive approximately 6 t too little lime and 31% would still be adequately limed.
Liming agricultural fields is necessary for counteracting soil acidity and is one of the oldest operations in soil fertility management. However, the best management practice for liming in Germany ...only insufficiently considers within-field soil variability. Thus, a site-specific variable rate liming strategy was developed and tested on nine agricultural fields in a quaternary landscape of north-east Germany. It is based on the use of a proximal soil sensing module using potentiometric, geoelectric and optical sensors that have been found to be proxies for soil pH, texture and soil organic matter (SOM), which are the most relevant lime requirement (LR) affecting soil parameters. These were compared to laboratory LR analysis of reference soil samples using the soil’s base neutralizing capacity (BNC). Sensor data fusion utilizing stepwise multi-variate linear regression (MLR) analysis was used to predict BNC-based LR (LR
BNC
) for each field. The MLR models achieved high adjusted R
2
values between 0.70 and 0.91 and low RMSE values from 65 to 204 kg CaCO
3
ha
−1
. In comparison to univariate modeling, MLR models improved prediction by 3 to 27% with 9% improvement on average. The relative importance of covariates in the field-specific prediction models were quantified by computing standardized regression coefficients (SRC). The importance of covariates varied between fields, which emphasizes the necessity of a field-specific calibration of proximal sensor data. However, soil pH was the most important parameter for LR determination of the soils studied. Geostatistical semivariance analysis revealed differences between fields in the spatial variability of LR
BNC
. The sill-to-range ratio (SRR) was used to quantify and compare spatial LR
BNC
variability of the nine test fields. Finally, high resolution LR maps were generated. The BNC-based LR method also produces negative LR values for soil samples with pH values above which lime is required. Hence, the LR maps additionally provide an estimate on the quantity of chemically acidifying fertilizers that can be applied to obtain an optimal soil pH value.
Liming is an effective measure to increase the soil pH and to counterbalance soil acidification. Therefore, the liming recommendations (LRs) for agricultural practice consider two aspects: changing ...the initial pH to the desired pH and compensating for all pH decreases taking place within the liming interval. The separation of these aspects is essential to optimize LRs and to minimize lime losses to the environment. Therefore, we developed a pedotransfer function (PTF) to calculate the lime demand to change the initial pH to the desired pH and compared the results with the LRs for agricultural practice. Applying this PTF to a set of 126 soil samples that were analyzed for base neutralization capacity could explain approximately 78% of the variability in the pH changes after the addition of different amounts of Ca(OH)2. Consequently, the lime demand to change the initial pH to the desired pH increased by approximately one-sixth compared to the lime demand proposed by the liming recommendation scheme, which is commonly used in Germany. From the numerical difference between the lime demand according to the LRs and the PTF, we calculated the annual acidification rates based on the soil texture, organic matter content and initial pH. Decoupling the abovementioned two aspects of LRs might be helpful to optimize the LRs by adapting to different regions, diverse management strategies and a changing climate.
Despite being a natural soil-forming process, soil acidification is a major agronomic challenge under humid climate conditions, as soil acidity influences several yield-relevant soil properties. It ...can be counterbalanced by the regular application of agricultural lime to maintain or re-establish soil fertility and to optimize plant growth and yield. To avoid underdose as well as overdose, lime rates need to be calculated carefully. The lime rate should be determined by the optimum soil pH (target pH) and the response of the soil to lime, which is described by the base neutralizing capacity (BNC). Several methods exist to determine the lime requirement (LR) to raise the soil pH to its optimum. They range from extremely time-consuming equilibration methods, which mimic the natural processes in the soil, to quick tests, which rely on some approximations and are designed to provide farmers with timely and cost-efficient data. Due to the higher analytical efforts, only limited information is available on the real BNC of particular soils. In the present paper, we report the BNC of 420 topsoil samples from Central Europe (north-east Germany), developed on sediments from the last ice age 10,000 years ago under Holocene conditions. These soils are predominantly sandy and low in humus, but they exhibit a huge spatial variability in soil properties on a small scale. The BNC was determined by adding various concentrations of Ca(OH)2 and fitting an exponential model to derive a titration curve for each sample. The coefficients of the BNC titration curve were well correlated with soil properties affecting soil acidity and pH buffer capacity, i.e., pH, soil texture and soil organic matter (SOM). From the BNC model, the LRs (LRBNC) were derived and compared with LRVDLUFA based on the standard protocol in Germany as established by the Association of German Agricultural Analytic and Research Institutes (VDLUFA). The LRBNC and LRVDLUFA correlated well but the LRVDLUFA were generally by approximately one order of magnitude higher. This is partly due to the VDLUFA concept to recommend a maintenance or conservation liming, even though the pH value is in the optimum range, to keep it there until the next lime application during the following rotation. Furthermore, the VDLUFA method was primarily developed from field experiments where natural soil acidification and management practices depressed the effect of lime treatment. The BNC method, on the other hand, is solely based on laboratory analysis with standardized soil samples. This indicates the demand for further research to develop a sound scientific algorithm that complements LRBNC with realistic values of annual Ca2+ removal and acidification by natural processes and N fertilization.
Accurate characterization of spatial soil moisture patterns and their temporal dynamics is important to infer hydrological fluxes and flow pathways and to improve the description and prediction of ...hydrological models. Recent advances in ground-based and remote sensing technologies provide new opportunities for temporal information on soil moisture patterns. However, spatial monitoring of soil moisture at the small catchment scale (0.1-1 km2) remains challenging and traditional in situ soil moisture measurements are still indispensable. This paper presents a strategic soil moisture sampling framework for a low-mountain catchment. The objectives were to: (i) find a priori a representative number of measurement locations, (ii) estimate the soil moisture pattern on the measurement date, and (iii) assess the relative importance of topography for explaining soil moisture pattern dynamics. The fuzzy c-means sampling and estimation approach (FCM SEA) was used to identify representative measurement locations for in situ soil moisture measurements. The sampling was based on terrain attributes derived from a digital elevation model (DEM). Five time-domain reflectometry (TDR) measurement campaigns were conducted from April to October 2013. The TDR measurements were used to calibrate the FCM SEA to estimate the soil moisture pattern. For wet conditions the FCM SEA performed better than under intermediate conditions and was able to reproduce a substantial part of the soil moisture pattern. A temporal stability analysis shows a transition between states characterized by a reorganization of the soil moisture pattern. This indicates that, at the investigated site, under wet conditions, topography is a major control that drives water redistribution, whereas for the intermediate state, other factors become increasingly important.
Site-specific estimation of lime requirement requires high-resolution maps of soil organic carbon (SOC), clay and pH. These maps can be generated with digital soil mapping models fitted on covariates ...observed by proximal soil sensors. However, the quality of the derived maps depends on the applied methodology. We assessed the effects of (i) training sample size (5–100); (ii) sampling design (simple random sampling (SRS), conditioned Latin hypercube sampling (cLHS) and k-means sampling (KM)); and (iii) prediction model (multiple linear regression (MLR) and random forest (RF)) on the prediction performance for the above mentioned three soil properties. The case study is based on conditional geostatistical simulations using 250 soil samples from a 51 ha field in Eastern Germany. Lin’s concordance correlation coefficient (CCC) and root-mean-square error (RMSE) were used to evaluate model performances. Results show that with increasing training sample sizes, relative improvements of RMSE and CCC decreased exponentially. We found the lowest median RMSE values with 100 training observations i.e., 1.73%, 0.21% and 0.3 for clay, SOC and pH, respectively. However, already with a sample size of 10, models of moderate quality (CCC > 0.65) were obtained for all three soil properties. cLHS and KM performed significantly better than SRS. MLR showed lower median RMSE values than RF for SOC and pH for smaller sample sizes, but RF outperformed MLR if at least 25–30 or 75–100 soil samples were used for SOC or pH, respectively. For clay, the median RMSE was lower with RF, regardless of sample size.
Detailed knowledge of a soil’s lime requirement (LR) is a prerequisite for a demand-based lime fertilization to achieve the optimum soil pH and thus sustainably increasing soil fertility and crop ...yields. LR can be directly determined by the base neutralizing capacity (BNC) obtained by soil-base titration. For a site-specific soil acidity management, detailed information on the within-field variation of BNC is required. However, soil-base titrations for BNC determination are too laborious to be extensively applied in routine soil testing. In contrast, visible and near-infrared spectroscopy (visNIRS) is a time and cost-effective alternative that can analyze several soil characteristics within a single spectrum. VisNIRS was tested in the laboratory on 170 air-dried and sieved soil samples of nine agricultural fields of a quaternary landscape in North-east Germany predicting the soil’s BNC and the corresponding lime requirement (LR
BNC
) at a target pH of 6.5. Seven spectral pre-processing methods were tested including a new technique based on normalized differences (ND). Furthermore, six multivariate regression methods were conducted including a new method combining a forward stagewise subset selection algorithm with PLSR (FS-PLSR). The models were validated using an independent sample set. The best regression model for most target variables was FS-PLSR combined with the second Savitzky-Golay derivation as pre-processing method achieving R
2
s from 0.68 to 0.82. Finally, the performance of the direct prediction of LR
BNC
(R
2
= 0.68) was compared with an indirect prediction that was calculated by the predicted BNC parameters. This resulted in slightly higher correlation coefficients for the indirect method with R
2
= 0.75.
Core Ideas
The FCM SEA was used for sampling and spatial estimation of soil moisture patterns.
Multispectral remote sensing and terrain data were combined to guide the sampling and estimation.
...Selected vegetation patterns and terrain data provided reasonable estimates of soil moisture.
The FCM SEA was stable to explain about 50% of the total observed variance.
The FCM SEA was superior to an approach driven solely by terrain data.
Detailed information on the temporal and spatial evolution of soil moisture patterns is of fundamental importance to improve runoff prediction, optimize irrigation management and to enhance crop forecasting. However, obtaining representative soil moisture measurements at the catchment scale is challenging because of the dynamic spatial and temporal behavior of soil moisture. High‐resolution remote sensing data provide detailed spatial information about catchment characteristics (e.g., terrain and land use) that can be used as proxies to estimate soil moisture. We assessed the potential use of combined multitemporal multispectral remote sensing (RS) and terrain data for estimating spatial soil moisture patterns at the small catchment scale. The fuzzy c‐means sampling and estimation approach (FCM SEA) was applied to conduct a sensor (proxy) directed (guided) sampling and to reconstruct multitemporal soil moisture patterns based on time domain reflectometry measurements. A comprehensive soil moisture database for the Schäfertal catchment, located in central Germany, was used to test, validate, and compare the FCM SEA performances of the combined remote sensing data with those of a benchmark approach driven solely by terrain data. Results from the study show that a FCM SEA model that integrates bi‐temporal RS imagery and terrain data was more effective in estimating spatial soil moisture patterns relative to the benchmark model. It outperformed the benchmark model in 58% of the cases and was stable to explain about 50% of the total observed variance for a range of different catchment moisture conditions. This was achieved with only a small sample size (n = 30). The results of this study are promising because they highlight the importance of considering multitemporal RS and terrain data and demonstrate how in situ sensors can be optimally placed to enable cost‐efficient monitoring and prediction of spatial soil moisture patterns at the small catchment scale.