Sub-project I: Improved Acquisition and Usage of Thematic Soil Maps for Site-Specific Management

1. Introduction

Site-specific management (SSM) involves technology that allows varying agricultural inputs (fertilizer, seed, crop protection chemicals, etc.) in response to changing local conditions. Some techniques assist in assessing variability of the crop outputs (yield, quality, etc.), while others are dedicated to investigating reasons for the occurrence of such variation. Variation in soil properties is a key factor affecting the overall outcome of crop production. Accounting for changing soil conditions using variable rate technology may significantly reduce waste, increase profits, and minimize negative environmental impact.

Despite its intuitive promise, progress in successfully implementing SSM techniques and demonstrating significant benefits over conventional management of agricultural inputs has been mixed (Pierce and Nowak, 1999). There are several limitations that may have halted the adoption of site-specific nutrient management:

Insufficient accuracy of thematic soil maps. Destructive soil sampling followed by laboratory analysis is commonly used as the principal method to assess soil fertility and to make decisions for site-specific management. The accuracy of these soil maps depends on the sampling scheme. The high cost of sampling and analysis, however, limits the density of collected samples. Grid soil sampling for developing variable rate fertilizer application maps has largely proven unsuccessful (Pierce and Nowak, 1999). Defining “management zones” and adaptive soil sampling have been attempted to improve sampling efficiency. (Francis and Schepers, 1997). Different procedures have been tried (Wollenhaupt et al., 1997; Fraisse et al., 1999), but there is no reliable method to derive such zones from various sources of qualitative and quantitative spatial information, to conduct sampling within such zones, and to interpret the results. There is also limited understanding of the relationship between zones of permanent and variable soil properties causing variability in yield. Newly developed automated systems equipped with on-the-go soil sensors are expected to decrease the cost of soil sampling and improve the accuracy of soil nutrient maps, mainly by significantly increasing the spatial resolution of measured properties (Sudduth et al., 1997).

Inadequate algorithms for site-specific nutrient management. Currently, there are no accepted guidelines on how to implement site-specific management in response to spatial variation of yield and soil properties. Present algorithms are based on existing “Best Management Practices” or local expertise, but their adequate use for SSM requires further study. In Nebraska, for example, a continuous equation for nitrogen (N) is applied, which includes yield goal and N supply from soil and other sources. However, recent research on SSM concluded that spatial variation in yield goal and nutrient response within a field needs to be taken into account for spatially varying rates of pre-plant nutrient application (Ferguson et al., 2001). Yield goal is the most important variable driving the estimated amount of N to be applied and it should be related to more permanent soil and landscape features. This also requires better understanding of yield stability. In the case of nutrient such as P and K, a purely soil test-based sufficiency concept employs few categories of nutrient applications only. It is questionable whether such simple algorithms can provide sufficient fine-tuning of nutrients at elevated yield levels (Hergert et al., 1997). There is also no multivariate response function that can estimate the yield response to inputs, site characteristics, and varying plant density (Bullock et al., 1998).

Lack of dynamic adjustment of post-emergence field operations. Much of the previous work on nutrient management focused on spatially varying pre-plant fertilizer applications. For yield factors such as nitrogen, temporal variation during the growing season is probably as important as the early season spatial variation because it affects soil N mineralization as well as the attainable yield and crop N demand. To manage this, real-time N management concepts are needed that allow adjustment of N application to the seasonal variation in yield potential (Blackmer and Schepers, 1994; Varvel et al., 1997a; Varvel et al., 1997b; Shapiro, 1999; Schroeder et al., 2000; Schepers et al., 2000a). Few attempts have been made to use crop simulation models for real time precision farming (Matthews and Cosser, 1997; Sadler and Russel, 1997; Fraisse et al., 2000; van Alphen and Stoorvogel, 2000).

Application uncertainty. Inaccurate flow rate and material inconsistency are limiting factors affecting the precision of application rates. However, modern technology allows improvement of current sprayers and spreaders equipped with variable rate controllers as well as their guidance systems (Medlin et al., 2000).

Overall uncertainty. Successfully implementing SSM depends on how errors propagate through a typical SSM application, including data acquisition, interpretation, and application and how the overall error can be minimized to truly apply inputs in congruence with the projected variation in soil and crop demand. Applying empirical or mechanistic models to spatially imprecise data often results in substantial uncertainties (Leenhardt et al., 1994; Leenhardt, 1995; Heuvelink, 1998), but no error propagation analysis has been done for most SSM applications. There is also a lack of economic sensitivity analysis.

Because quantitatively important differences in soil properties, plant nutrient uptake and yield often occur at distances as short as 10 m or less (Beckett and Webster, 1971; Sadler et al., 1998), successful site-specific management depends on accurate maps of soil properties and other yield-limiting factors and on their adequate interpretation. This project will result in improved techniques to obtain and interpret soil property maps, including an assessment of the accuracy and cost associated with different approaches.

Our long-term goal is to develop the scientific basis and tools for improved soil mapping and interpretation of spatially variable data for site-specific management. Initial emphasis will be on irrigated corn-based systems. The specific objectives of this proposal are:

1. Data Acquisition: To develop cost-efficient technology and methods for accurate mapping of physical and chemical soil properties.

2. Interpretation: To develop validated procedures for integration of multiple layers of site-specific data and defining finite management elements (zones).

3. Evaluation: To quantify costs and errors associated with different soil mapping procedures and assess their impact on crop performance and profit.

4. Implementation: To validate the results of this project through a set of case studies and to disseminate the developed tools.

2. Relevance and Significance

Site-specific management is a way to increase the profitability of crop production and to minimize possible negative consequences of intensive agriculture. This project addresses several key issues that halt the adoption of SSM. Our goal is to develop innovative procedures for gathering and processing of geospatial information that are likely to improve the effectiveness of SSM. Special emphasis will be given to soil nutrient supply and crop response to applied nutrients in irrigated continuous corn and corn-soybean systems of Nebraska, but the techniques developed will be of general value for any site-specific crop management application. The integration of technology development and agronomic studies with economic profitability and risk analysis is essential for a successful research program. A multidisciplinary approach to look at SSM is the major strength of the proposed project,

This project proposal was developed through intensive discussion with farmers and crop consultants interested in precision farming (see Appendix). It will address the following practically relevant questions:

· How can we improve the spatial density and accuracy of thematic soil maps for site-specific crop management? What prior or auxiliary information can be used to optimize sampling schemes and/or improve the mapping process? For what soil properties do on-the-go ground sensors offer advantages over destructive sampling or remote sensing?

· How can we objectively integrate numerous layers of spatial information for making management decisions? What are minimum sets of management zones and what should be the components and spatial resolution of each? How can different procedures be validated?

· What is the cost and accuracy associated with different SSM approaches and how do errors propagate through them? When is variable rate application superior to a uniform application of inputs? How can we optimize the collection and use of spatial data to improve the agronomic and economic performance of SSM?

3. Research Approach

3.1. Overview of Proposed Work

Understanding spatial variation in yield limiting factors is the essential step towards SSM. In this project we will use the following framework of yield definitions:

Potential yield (Ymax) = Theoretically achievable yield solely determined by genetic characteristics and climate (solar radiation, temperature). Ymax varies among agroecological zones and, at a particular location, from year to year due to climatic fluctuations and differences in the date of planting. Knowing the average Ymax of a site is important for setting pre-season yield goals, whereas knowing the actual Ymax is useful for in-season adjustment of crop management decisions. Crop models can be used to obtain estimates of Ymax (Muchow et al., 1990). Within a field, spatial variation in Ymax is small and associated with terrain features that affect temperature and intercepted solar radiation.

Attainable yield (Ya) = Yield that can be achieved by minimizing abiotic and biotic stresses through the best available technology tailored to a given location within a production field. Ya is always smaller than Ymax due to some limitation of water and/or nutrient supply and varies form year to year depending on variation in Ymax. Within a field, Ya varies due to spatial variation in nutrient and/or water supply. Spatial variation in Ya is associated with permanent soil and landscape features that are difficult to change as well as factors that are associated with crop management. This variation, e.g., soil organic matter, pH, and available soil nutrients, soil compaction, is manageable. In irrigated systems water is not a limiting factor so that Ya mainly represents the attainable yield limited by nutrient supply. This includes soil physical constraints and their effect on root growth and nutrient uptake. In this case, Ya can be estimated using a crop simulation model running in nutrient-limited mode or using simple models that estimate Ya as a function of soil test values and Ymax (Janssen et al., 1990).

Actual yield (Y) = Yield that is actually achieved, i.e., Ya reduced by weeds, insects, diseases, mineral toxicities, or other constraints. At the same site, Y varies form year to year depending on the variation in Ya, and the quality of crop management. In the same year within a field, Y varies due to significant spatial variation in Ya and short-distance variation in yield-reducing factors such as pests. Only part of the spatial variation in Y is associated with variation in intrinsic field features, whereas other parts are more stochastic by nature and difficult to manage. This "random" variation overlies more structured spatial variation in Ya across major soil zones and needs to be separated from this to understand the causes of yield variation in a more systematic way.

In this project, we focus on improving the acquisition and usage of thematic soil maps that determine the spatial variation in Ya within large production fields because Ya is the key manageable determinant of spatial yield variation (Y). The proposed research consists of the following four essential tasks:

1. Data acquisition: Collect all available information (historical climate data, digital soil maps, digital elevation models, and digital orthophotography quarter quadrangles of bare soil). Other information may include several years of yield maps, soil analysis reports, remote sensing images, or spatially dense maps of soil properties measured with on-the-go sensors. For properties measured at coarser spatial density, use advanced geostatistical techniques that allow interpolation to a finer grid by also using one or more layers of correlated spatially dense auxiliary information. For several soil properties new prototype systems for automated mapping of physical and chemical soil characteristics will be developed in this project.

2. Interpretation: Several procedures for evaluation of spatial information will be designed and tested using real data sets and simulation models. A crop simulation model is used to estimate Ymax for the site based on the historical climate data, including an assessment of the annual variability in Ymax and the effect of different planting dates on it. This determines the window of optimal planting and the average upper limit to yield and its annual variation. Using the available layers of spatially dense information, different maps of functional management zones will be created using fuzzy classification.

3. Evaluation: Before proceeding to use the thematic maps for implementing SSM, their quality must be judged and decisions about future improvements must be made. Evaluation includes (a) assessment of uncertainties and cost associated with the data acquisition and interpretation procedures, (b) evaluation of the potential for using on-the-go sensors, and (c) assessment of the potential benefits of SSM practices.

4. Implementation: Gathered information will be used to (a) vary some inputs according to larger zones and variation in more permanent landscape features and (b) vary inputs such as fertilizers and seeds according to yield goal zones and short-distance variation in soil nutrients.

3.2. Experimental Sites

All experimental work will be done at four experimental locations in South-Central and Eastern Nebraska. Sites include a research field at the Agricultural Research and Development Center at Mead (1), and three production fields in Butler (2), Clay (3), and Hall (4) Counties of Nebraska. All sites are representative of loess-derived, silt loam soils of the Great Plains under continuous corn or corn-soybean systems. Sites 1 and 2 represent the agroclimatic conditions of eastern Nebraska, whereas sites 3 and 4 are in south-central Nebraska. At each site, one 1/4 section (65 ha) will be used as the principal experimental unit for the whole project duration. Differences among sites are mainly due to soil types and topography, whereas the climatic yield potential (Ymax) and cropping history are similar. Irrigation ensures full water availability at all sites. By holding climate and water constant and using a uniform on-farm research database we will be able to develop generic methodologies and relationships with a high extrapolation potential.

For each site, we will develop a standardized GIS database using previously collected information (remote sensing images, soil maps, historical climate data, a DEM, soil test data, maps of electrical conductivity, 3-5 years of yield maps, county yield statistics) and data to be collected during the proposed project. In years 1 to 3, replicated strip trails will be conducted in each field to gradually develop a database for testing the response of corn to different plant densities, site characteristics, and input levels (Bullock et al., 1998).

3.3. Data Acquisition: New Technology for High-Resolution Mapping

3.3.1. High-resolution Soil Mapping Using On-The-Go Sensing Techniques

Background and Previous Research. Ground-based, continuous or semicontinuous sensors developed to automatically measure soil properties have potential benefits through increased density of measurement points with a relatively low cost (Sonka et al., 1997). However, current technology is inapplicable to measure many essential soil properties, even though numerous attempts to develop such instrumentation have taken place (Sudduth et al., 1997; Morgan and Ess, 1997). Analysis of soil reflectance in the visual and near-infrared portion of the spectrum may provide valuable information applicable to assess soil OM content, CEC, moisture, texture and nitrate level (Shonk and Gaultney, 1989; Sudduth and Hummel, 1993; Shibusawa et al., 1999). Similar to traditional remote sensing, use of ground-based optical sensors is soil-sensitive and a great amount of work is needed to define recommended rates of seeds, fertilizers and pesticides based on this data only. Recent publications underline the significance of soil electrical properties (Colburn, 1999; Fritz et al., 1999). Soil conductivity might be useful for identifying management zones for adoptive soil sampling and/or suggest trends in variation of soil properties to be targeted with optical sensors.

Other automated soil nutrient sensors are still in the early development stage (Sudduth et al., 1997). Automated mapping of soil pH gives a cost-effective alternative to the routine manual soil sampling with laboratory analysis. A prototype automated soil-sampling system (Figure 1) allowed on-the-go measurements of soil pH on naturally moist samples using combination, flat-surface, ion-selective electrodes (Adamchuk et al., 1999). Although several experimental alternatives currently exist to estimate nitrate and potassium levels or soil pH, there are no on-the-go methods to directly measure soil phosphorous.

Figure 1. Major components of the automated soil sampling system

Implementation of site-specific crop management may also require detection of spatial variability of soil resistance, both horizontal and vertical. The need to measure soil strength on a systematic basis forced researchers to automate cone penetrometer measurements to allow stop-and-go mapping of agricultural fields (Larney et al., 1989; Clark, 1999). Even with automated penetrometers, soil strength data is limited to a few discrete points within a field. Maps interpolated from point data have an accuracy limitation due to the relatively small achievable measurement density. Therefore, several attempts have also been made to conduct continuous estimation of horizontal soil mechanical resistance at a particular depth on-the-go (Alihamsyah et al., 1990; Lui et al., 1996)