James E. Burt

Purposive Sampling for Digital Soil Mapping for Areas with Limited Data

Digital soil mapping requires two basic pieces of information: spatial information on the environmental conditions which co-vary with the soil conditions and the information on relationship between the set of environment covariates and soil conditions. The former falls into the category of GIS/remote sensing analysis. The latter is often obtained through extensive field sampling. Extensive field sampling is very labor intensive and costly. It is particularly problematic for areas with limited data. This chapter explores a purposive sampling approach to improve the efficiency of field sampling for digital soil mapping. We believe that unique soil conditions (soil types or soil properties) can be associated with unique combination (configuration) of environmental conditions. We used the fuzzy c-means classification to identify these unique combinations and their spatial locations. Field sampling efforts were then allocated to investigate the soil at the typical locations of these combinations for establishing the relationships between soil conditions and environmental conditions. The established relationships were then used to map the spatial distribution of soil conditions. A case study in China using this approach showed that this approach was effective for digital soil mapping with limited data.

A Markov chain-based probability vector approach for Modeling spatial uncertainties of soil classes

Integrating livestock with cotton (Gossypium hirsutum L.) offers profitable alternatives for producers in the southeastern USA, but could result in soil water depletion and soil compaction. We conducted a 3-yr field study on a Dothan loamy sand (fine-loamy, kaolinitic, thermic Plinthic Kandiudult) in southern Alabama to develop a conservation tillage system for integrating cotton with winter-annual grazing of stocker cattle under rainfed conditions. Winter annual forages and tillage systems were evaluated in a strip-plot design where winter forages were oat (Avena sativa L.) and annual ryegrass (Lolium mutiflorum L.). Tillage systems included moldboard and chisel plowing and combinations of noninversion deep tillage (none, in-row subsoil, or paratill) with or without disking. We evaluated forage dry matter, N concentration, average daily gain, net returns from grazing, soil water content, and cotton leaf stomatal conductance, plant populations, and yield. Net returns from winter-annual grazing were between US

The Impact of Neighbourhood Size on Terrain Derivatives and Digital Soil Mapping

185 to US

An experiment using a circular neighborhood to calculate slope gradient from a DEM

200 ha - yr - . Soil water content was reduced by 15% with conventional tillage or deep tillage, suggesting that cotton rooting was increased by these systems. Oat increased cotton stands by 25% and seed-cotton yields by 7% compared with ryegrass. Strict no-till resulted in the lowest yields-30% less than the overall mean (3.69 Mg ha -1 ). Noninversion deep tillage in no-till (especially paratill) following oat was the best tillage system combination (3.97 Mg ha -1 ) but deep tillage did not increase cotton yields with conventional tillage. Integrating winter-annual grazing can be achieved using noninversion deep tillage following oat in a conservation tillage system, providing producers extra income while protecting the soil resource.

Neighborhood size and spatial scale in raster-based slope calculations

Slope gradient, slope aspect, profile curvature, contour curvature, and other terrain derivatives are computed from digital elevation models (DEMs) over a neighbourhood (spatial extent). This chapter examines the combined effect of DEM resolution and neighbourhood size on computed terrain derivatives and its impact on digital soil mapping. We employed a widely used regression polynomial approach for computing terrain derivatives over a user-specified neighbourhood size. The method first fits a least squares regression polynomial to produce a filtered (generalized) terrain surface over a user defined neighbourhood (window). Slope gradient, slope aspect, profile, and contour curvatures are then computed analytically from the polynomial. To examine the effects of resolution and neighbourhood, we computed terrain derivatives using various combinations of DEM resolution and neighbourhood size and compared those values with corresponding field observations in two Wisconsin watersheds. In addition, we assessed the effects of resolution and neighbourhood in the context of knowledge-based digital soil mapping by comparing soil class (series) predictions with observed soils. Our results show that a neighbourhood size of 100 feet produces the closest agreement between observed and computed gradient values, and that DEM resolution has little impact on the agreement. Both profile curvature and contour curvature are more sensitive to neighbourhood size than slope gradient, and sensitivity is much higher at small neighbourhood sizes than at large neighbourhood sizes. Because of the importance of terrain derivatives in the knowledge base, predictive accuracy using a digital soil mapping approach varies strongly with neighbourhood size. In particular, prediction accuracy increases as the neighbourhood size increases, reaching a maximum at a neighbourhood of 100 feet and then decreases with further increases in neighbourhood size. DEM resolution again does not seem to impact the accuracy of the soil map very much. This study concludes that, at least for knowledge-based soil mapping, DEM resolution is not as important as neighbourhood size in computing the needed terrain derivatives. In other words, assuming the DEM resolution is sufficient to capture information at the optimum neighbourhood size, there is no predictive advantage gained by further increasing DEM resolution.

Depicting classification uncertainty using perception-based color models

The traditional 3 � 3 cell neighborhood used in a focal operation on a raster layer has a square shape that results in a dimensional neighborhood of which the orientation is eventually arbitrary to the physical features represented. This paper presents an experiment using a circular neighborhood to calculate slope gradient. Comparisons of the results from a circular neighborhood with the results from some traditional methods show that (a) for a smooth surface, the result from a circular neighborhood is more accurate than that from a square neighborhood, (b) a circular neighborhood is generally more sensitive to noise in the input DEM than a square neighborhood, and (c) in a validation using field measurements, the circular neighborhood performs better than the square neighborhood when the ratio of user-specified neighborhood size to cell size is high.

Unification of soil feedback patterns under different evaporation conditions to improve soil differentiation over flat area

Raster-based slope estimation is routine in GIS. Like many other terrain attributes, the slope at a location is determined from elevations of surrounding cells. This spatial extent – ‘neighborhood size’ – is often treated as the ‘spatial scale’ of the calculation. In fact, neighborhood size and spatial scale are two connected yet different concepts, but few studies have investigated the relationship between them. The distinction is important because neighborhood size is under user control whereas spatial scale is merely implicit in the computational method. This article attempts to clarify and provide a more precise meaning of the two terms by considering slope operators from the standpoint of the frequency (or wavenumber) domain. This article derives analytical expressions for the amplitude response functions of four popular slope estimators. These are used to characterize the individual methods and also to show that the neighborhood size and spatial scale of a slope calculation are not numerically the same. In fact, because there is no single spatial scale that can be unambiguously associated with a given neighborhood size, neighborhood size cannot be an adequate indicator of spatial scale. Furthermore, this article shows that different indices of ‘scale’ yield different impressions about the action of a slope estimator and its response to changing neighborhood size. Therefore, it is necessary to examine the amplitude response function when investigating the spatial scale. The article also provides guidance for GIS practitioners when selecting a slope estimation method.

Bias-variance decomposition of errors in data-driven land cover change modeling

Fuzzy classification typically assigns a location or an area to a category with some estimated degree of uncertainty. There are strong incentives for depicting uncertainty along with category, and numerous authors have recommended that this be done using progressive desaturation of the entitys color with increasing uncertainty. This article shows that such recommendations cannot be naively applied using color models widely used in computer graphics because colors equally ‘saturated’ do not appear equally certain. We demonstrate that models based on color perception are preferred, particularly if one wishes to compare uncertainties across classes. We discuss geometrical complications arising with perceptual models that are not present with models closely tied to hardware. An algorithm for selecting colors is presented and illustrated using the model.

Data-Gap Filling to Understand the Dynamic Feedback Pattern of Soil

Abstract Detailed and accurate information on the spatial variation of soil types and soil properties are critical components of environmental research and hydrological modeling. Early studies introduced a soil feedback pattern as a promising environmental covariate to predict spatial variation over low-relief areas. However, in practice, local evaporation can have a significant influence on these patterns, making them incomparable at different locations. This study aims to solve this problem by examining the concept of transforming the dynamic patterns of soil feedback from the original time-related space to a new evaporation-related space. A study area in northeastern Illinois with large low-relief farmland was selected to examine the effectiveness of this idea. Images from MODIS in Terra for every April–May period over 12 years (2000–2011) were used to extract the soil feedback patterns. Compared to the original time-related space, the results indicate that the patterns in the new evaporation-related space tend to be more stable and more easily captured from multiple rain events regardless of local evaporation conditions. Random samples selected for soil subgroups from the SSURGO soil map show that patterns in the new space reveal a difference between different soil types. And these differences in patterns are closely related to the difference in the soil structure of the surface layer.

INTERPOLATION ERRORS AND SPATIAL RESOLUTION OF THE UNITED STATES SOLAR RADIATION NETWORK

ContextCareful model evaluation is essential for using the results of data-driven land cover change (LCC) models. A useful evaluation should help the analyst understand model behavior and improve model performance. Conventional error analysis methods provide limited insights on these two issues.ObjectivesWe propose the use of bias-variance error decomposition (BVD) for LCC model evaluation and investigate its value through a pilot study.MethodsWe examined the mathematical underpinnings of BVD and applied the approach to a model describing the expansion of development for a county in southeastern Michigan, USA.ResultsThe spatial structure of the BVD error component maps effectively informed the design of model improvement efforts. We found that bias maps can help detect and eliminate the negative influence of spatial non-stationarity in LCC models, and variance maps can shed light on optimal training sample allocation and ways to mitigate model instability. In our case study, model improvements suggested by BVD showed potential to substantially reduce estimation error. Our analysis also revealed that fast and slow developments can result from different mechanisms even when they are geographically near to each other, and different development types may be best analyzed with separate models.ConclusionsBVD has advantages over conventional error analysis and can deepen our understanding of the underlying LCC process and the sources of modeling error. The BVD insights can help effectively design and apply model improvement strategies. These results and the BVD approach have great generality and are applicable to many other geospatial models.