These are some of the activities which soil scientists regularly practice. This work is most often conducted in coordination with other professionals with lesser training and knowledge of soil systems.
Well-trained soil scientists are in high demand for a wide array of professional positions with public agencies or private firms. Here are some specific examples of positions currently held by soil science graduates from just one university over the past 10 years. Most soil scientists have earned at least a bachelor degree from a major agricultural university. At many universities, two choices are available for specialized training in soils. The Soil Science option prepares students to enter the agricultural sector as farm advisors, crop consultants, soil and water conservationists, or as representatives of agricultural companies.
The Environmental Soil Science option prepares soil scientists for careers in environmental positions dealing with water quality concerns, remediation of contaminants or for on-site evaluation of soil properties in construction, waste disposal, or recreational facilities.
Most of the major universities that still have a soil science program now offer soils courses only at the graduate level. Specific qualifications are listed for each vacancy. Search for the job series to find all soil scientist eligible vacancies. Natural Resources Conservation Service Soils. Stay Connected. Loading Tree Careers in Soil Science What is a soil scientist? The labels and positions of the trenches are presented, as follow: one pit at the top P10 , two in the middle third position P3 and P9 , one in the lower third P8 , three in the foothills P2, P5, and P7 and three in the lowland area P1, P4, and P6.
Both the soil profiles and the soil samples were described and collected according to Santos et al. Soil properties were measured according to Teixeira et al. Besides, undisturbed samples were collected to determine soil bulk density, total porosity, and macro and micropores. The profiles were then classified according to the Brazilian Soil Classification System Santos et al.
These soil profiles were used as modal profiles for the pedologists training. Once the soil was taxonomically classified, the group of tutors and apprentices begun to define the mapping units to be used in the soil map. The definition of soil mapping units is a key point in the definition of the soil map and its validation process.
The Algorithm for Quantitative Pedology APQ was used to evaluate the dissimilarity between soil profiles, according to Beaudette et al. Algorithms for quantitative pedology: A toolkit for soil scientists. Comput Geosci. The 10 soil profiles were submitted to the dissimilarity analysis using both the available data of the landscape and the soil attributes soil depth and sand, silt and clay content. This exercise allowed the tutors to show the relationship between soil and landscape and how this is associated with a given classification system, in this case, SiBCS Brazilian Soil Classification System.
As presented along item 2. However, the use of these covariates to develop soil prediction models depends on their effectiveness in explaining the soil variation along the landscape. Besides, some covariates can be redundant, that is, more than one of them represents similar relation with soil variation in the landscape and are highly correlated with each other. Considering these aspects and facilitating the choice of the environmental covariates to train the predictive model, a principal component analysis PCA was performed to verify the best explanations for soil variation in the landscape.
A correlation analysis was then performed between the previously selected covariates those more explainable covariates to eliminate redundant covariates and simplify the models. The same analysis was used to create environmental strata of the landscape, which was useful not only for the apprentices to develop their mental models of soil mapping but also to establish the sampling designs to validate both the conventional and the digital soil maps item 2.
The conventional soil mapping CSM was produced based on a soil survey and soil-landscape relationship. A lot of soil survey aspects were discussed during the training process, in theory and practice, always respecting the limitation feeling and time of each participant.
After theoretical and practical classes in the field and laboratory, the apprentices went to the part of conducting a detailed soil survey and mapping project. The first step office work consisted of gathering information about the study site that could help the understanding of the soil-landscape relationship of the area. The apprentices could evaluate each map of the relief terrain attributes and the land-use history through satellite images and indexes and ten complete soil profiles along the landscape.
After that, the soil profiles were colleted, analyzed, and classified. To define the soil MUs a quantitative method using dissimilarity analysis was carried out and discussed during the training process. Afterward, some boundaries of MUs were verified and adjusted during the field survey depending on the apprentice feeling.
In summary, the criteria used for the apprentices to build the mental model were: soil taxon, at first until the second level, and soil attributes of each profile. After discussions, the third and fourth categorical levels of SiBCS and the specific attributes and characteristics of these levels, such as E horizon thickness, texture, soil color, waxiness, and others, were also considered for the definition of the MUs.
In addition, characteristics such as position in the landscape, rock outcrop, presence of gravel pebbles, drainage, and others. The boundaries of the MUs were generated basically from the analysis of elevation, slope, contour lines, image bands, and overlapping theses maps.
The process of constructing the conceptual model of pedogenesis in the study site was guided by training process and prior knowledge of soil formation conditions of the study site; as all apprentices have studied at UFRRJ, they had some prior contact with landscapes presenting some similarities with the training study site. The use of DSM allowed the apprentices to get in contact with algorithms commonly used to predict soil MUs.
Besides, it was a good exercise to evaluate the advantages, difficulties, and limitations of each technique and compare DSM with conventional maps. The tutors decided to use the two most commonly used techniques for predicting soil MUs: Multinomial Logistic Regression and Random Forest.
Multinomial Logistic Regression MLR is a technique used exclusively for predicting categorical variables such as soil types. It is a parametric method that allows predicting the probability of occurrence of a response variable, considering the values of a series of independent variables environmental covariates that can be qualitative or quantitative. The logistic function is represented by:.
Categorical data analysis. Florida: Gainesville; The logistic model belongs to the family of generalized MLR and was used to model the relationships between the different soil types map units as categorical dependent variables and the environmental covariates as independent variables.
A model with previous covariates selected was fitted to predict the spatial distribution of soil types. Random Forest RF is an algorithm developed by Breiman, It is based on regression and classification trees, where it built various regression or classification trees with bootstrap sampling on the input variables and internal validation Grimm et al. Soil organic carbon concentrations and stocks on Barro Colorado Island - Digital soil mapping using Random Forests analysis.
Comparison of boosted regression tree and random forest models for mapping topsoil organic carbon concentration in an alpine ecosystem.
Ecol Indic. RF depends only on three user-defined parameters: the number of trees in the forest, the minimum number of data points on each terminal node nodesize , and the number of variables used to produce each tree mtry.
A comparative assessment of support vector regression, artificial neural networks, and random forests for predicting and mapping soil organic carbon stocks across an Afromontane landscape. As done for MLR, a model with the same previous covariates selected was fitted in RF using the parameters indicated by the literature.
In the literature, it is already established that the probabilistic sampling is recommended for the validation of spatial predictions Brus et al. Sampling for validation of digital soil maps. Sampling for digital soil mapping: A tutorial supported by R scripts.
The design of probabilistic sampling can influence the results of the map validation. According to Brus et al. Another important aspect of the sampling for map validation is the number and the type of field observations used. Before deciding on the number of field observations, an assessment was made of what kind of observation would be used to validate the map auger holes, mini trenches or regular trenches. Although auger holes are simpler sampling and allow greater yield in the field, these observations do not allow the ideal visualization of some important soil morphological aspects used to classify soils, such as the thickness of the horizons, shape, size, and grade of soil structure, quantity and size of mottles and clay films.
On the other hand, although ideal for soil evaluation, conventional pit are very laborious for opening and closing, resulting in less field yield. Thus, it was decided to use mini trenches with dimensions of 0. The mini pit dimensions can be adapted to a specific study site to attend the most important differentiation of soil types.
Considering mini trenches as field observation, the group realized that it was possible to open, manually, 60 mini pits along the study site, 30 of them for each sampling design. The SRS is the sampling technique where all the elements that compose the sample universe have the same probability of being selected for the sample, that is, the same likelihood of inclusion and are completely independent of each other.
It would be like making a fair draw among the individuals of the universe: In the specific case of soil science would be, for example, to select any pedon of a given type of soil class. In this type of sampling, only the number of samples n is defined. Figure 2 Sampling design applied to the study site.
Despite being a totally probabilistic method not biased , which is an ideal condition for validation, in this selected method, the spatial distribution of the mini trenches can be irregularity scattered along the study site. It means that there may be strong grouping of some locations sampling, in addition to the presence of large voids between sample sites. The SSRS approach was tested because if the subareas strata are internally homogeneous and heterogeneous between strata, the use of stratified random sampling reduces the sample error and may be somewhat more efficient for validation, since you certify that there are no empty spaces and that all soil types or soil attributes have at least one representative individual for validation.
When using SSRS, it is assumed that the environmental strata is related to different soils types, which could not have been perfectly sampled using SRS.
One important aspect using this sample design is the delineation of the strata. Commonly, the map unities are used as strata to distribute the validation points Brus et al.
In the specific case, the study site was subdivided into ten strata using an unsupervised classification, k-means algorithm, and previous selected covariates, which are the environmentally homogeneous areas Figure 2b.
Once the ten strata were defined, the 30 soil samples were distributed following a proportional stratification, where the number of points is proportional to the stratum area Figure 2b. All of them were based on the confusion matrix Brus et al. In the literature, overall accuracy is also called overall purity, map purity, global accuracy, and general accuracy. In which Eiu denotes the number of points mapped as the mapping unit u that is, the sum of the rows in the confusion matrix; and Eu are the classes correctly classified in that unit u , the main diagonal of the confusion matrix.
The complement of UA 1- UA is referred to as the error of commission inclusion , that is, the error ruled by the inclusion of pixels from other classes in the class in question. In which Eju denotes the number of points mapped as the mapping unit u, that is, the sum of the columns in the confusion matrix, and Eu are the classes correctly classified in that unit u , main diagonal of the confusion matrix. The complement of PA 1- PA is referred to as the omission errors exclusion , that is, when a pixel ceases to be classified correctly in that mapping unit and is incorrectly classified as another unit.
Where c is the number of classes on the matrix, E ij values on the row i and column j, Ei total on the row i and E j total on column j, and n the number of samples observations. Both user and producer accuracy are commonly calculated for each mapping unit. Thus, as in each map, the mapping units have different territorial expression, both indexes are weighted averages of user and producer accuracy. The weighting is done by multiplying this accuracy by the area of each mapping unit MU , divided by the total area of the map A.
The WUAI and WPAI indices allow us to know how the types of errors are distributed commission or omission, respectively in each map give an overview of these errors for a specific map. Thus, after comparing the global accuracy and the kappa index, before entering into the detailed evaluation of the types of errors per mapping unit which is conventionally done , we used the WUAI and WPAI indices to compare the relevance of commission and omission errors on each map 4 CSM generated by apprentices and 2 DSM generated by RF and MLR.
The index values range from 0 to 1, with 0 lack of accuracy and 1 maximum accuracy. In which Eju denotes the number of points mapped as the mapping unit u that is, the sum of the columns in the confusion matrix and Eu are the classes correctly classified in that unit u , main diagonal of the confusion matrix, a u is the surface area of the mapped unit u and A is the total surface area of the map.
In which Eiu denotes the number of points mapped as the mapping unit u that is, the sum of the rows in the confusion matrix and Eu are the classes correctly classified in that unit u , main diagonal of the confusion matrix, a u is the surface area of the mapped unit u, and A is the total surface area of the map. For simplicity, we will use the terms overall accuracy for global accuracy and user and producer accuracy for commission and omission errors, respectively.
The Spring 5. Spring: integrating remote sensing and gis by object-oriented data modelling. Comput Graph. The following packages were used: raster, rgdal, maptools, and RSAGA for data management, preparation, and visualization Brenning et al. Classification and Regression by randomForest R package. R News. Modern applied statistics with S. New York: Springer; Figure 3 presented the ten soil profiles and their respective position according to the toposequence.
As shown, the soil color has a strict relationship with the topography, which, in turn, is directly related to soil drainage and soil moisture characteristics.
Figure 3 Soil profiles and position on the landscape. Further information about each MU is presented in table 4. Figure 4 The dissimilarity of soil profiles based on the soil key properties and landscape relationship. Theses soils are associated with the lower third of the landscape. These soils P1, P4, and P6 are in the lower part of the landscape.
The MU3 P8 represents the soil pattern between the middle third and lower third of the landscape. The soil is well-drained, with a yellowish color and clayey subsurface horizon Argissolo Amarelo. The color is strongly related to the soil moisture and geology; in this case, there is a predominance of goethite as mineral.
The soils of MU4 P3 and P9 , despite belonging to the same taxonomic class Argissolo Vermelho as the soil of MU5 P10 - top of the landscape , differ in depth, presence of gravel, bases saturation. Both on MU4 and MU3, the presence of hematite is remarkable, resulting in redder colors to the soil which are directly related to the better drainage condition. Irrespective of the method used to generate soil maps DSM or CSM , one of the great challenges of soil mapping is not only how to select good environmental covariates among many available, but also how to make the soil mapping a simple way creating a parsimonious model, especially for learners.
Analyzing the PCA, we see that component 1 dimension 1 explains Component 2 dimension 2 features terrain attributes. Figure 5 Plot of loading corresponding to the first two principal components, the environmental covariates and soil mapping units related. The number corresponds to the profile number, as defined on the soil survey and the symbol corresponds the MU.
To simplify the model that will be used in the stratification of the area it was used only environmental variables with low correlation less than 0. Since DEM has high correlation with chnd greater than 0. Figure 6 Correlation analysis of the environmental covariates with a higher score on the PCA analysis.
Band 2, despite having a high correlation with SAVI, it was less than 0. In summary: the covariates of the relief selected were: DEM, rsp, and twi. From the satellite image organism soil factor band 2 and SAVI. In general, there is a clear difference in the patterns of the maps when comparing the conventional soil maps Figures 7a and 7d with the digital ones Figures 7e and 7f.
This pattern can be explained by the strong relationship between soil distribution and toposequence, which the apprentices observed during the fieldwork. That is, when the pedologist, in the learning process, connects theoretical knowledge with the distribution of soils in the landscape and manages to build a mental model of distribution and prediction of soil types in the landscape interpretation of soils in the landscape.
This cognitive process is subjective and quite different among pedologists since cognition includes functions, such as learning, attention, memory, language, reasoning, decision making, etc. Soil scientists travel the globe and take soil samples. Then they study the samples in a laboratory to determine the nutrients, composition, resistance to erosion, and ability to retain water.
These factors allow the soil scientist to grade and categorize the soils. Next they survey the land and take aerial photos. With this information they create soil maps that will show types of soils and will help predict erosion patterns. The maps are also used to determine land value and whether the land is best suited for growing crops or constructing buildings. The goal is to make the land as productive as possible.
Loaded with all sorts of soil information, soil scientists write reports and act as consultants to land owners, farmers, and businesses.
They may advise them on what to do with their land or recommend new fertilizers and techniques to increase soil nutrients. Soil scientists can help improve crop health by suggesting when and how to do crop rotations. Their knowledge is important to efficiently use the land. Besides consulting and scientific studies, soil scientists may find other work. Chemical evaluations of the soil are important for people trying to grow crops and plants.
Some soil scientists examine soils from the Moon or Mars. Others work on archeological sites to try to get a glimpse of lost times.
0コメント