12.5.1 Soil Sampling

Taking a representative soil sample is both the first step and the largest source of error in soil fertility evaluation, because of the magnitude of extrapolation of the analytical results. When a 5-g subsample taken from a 500-g composite sample of a 2.5-ha field at 15-cm depth is used for phosphorus determination, it represents one-billionth (10^–9) of the total soil volume for which the analysis is made (Perur et al. Reference Perur, Subramanian, Muhr and Ray1974).

Soil scientists assume that the distribution of printed instructions and diagrams on how to take soil samples is sufficient to be assured of representative specimens. Unfortunately, this is not always the case. A representative soil sample is a composite of 10–20 samples of the root zone of a field with no major variation in slope, drainage, color or past fertilizer history. For deep-rooted plants, I recommend taking a 0–20-cm and 20–50-cm sample, and for the deepest rooted ones, a 50–100-cm and 100–300-cm sample. Non-representative areas, such as fence rows, termite mounds, those where straw has been burned and manure piles must be avoided. Appropriate information is also needed, including the name and address of the farmer, georeference, previous crop, and fertilizer practices. This is best provided with a web-based questionnaire in the local language that can be sent electronically to a central laboratory. An example for maize-growing in Mindanao, Philippines, has been successfully developed by the International Rice Research Institute (IRRI 2011).

For established pastures or permanent crops where fertilizers are not likely to be incorporated, sampling the top 20 cm is usually sufficient. For nitrate analysis, the 20–50-cm subsoil is usually sampled, and nitrate sampling tools and procedures exist. Subsoil samples, to at least 50 cm and beyond, are important in oxidic subsoils for estimates of anion exchange capacity, gypsum requirements and deep SOC.

Another important question is where to sample when phosphorus has been applied in bands. The answer is between the bands, if their locations are known. The same is true for conservation tillage. A third consideration is the time of the year. Soil samples should be taken a long time before planting so that the results will be available when the decision is made as to how much fertilizer is to be applied. In ustic soil moisture regimes, this means sampling during the dry season, when upward ion movement takes place. This may change some of the soil test results, principally in regard to potassium.

How often a field should be sampled depends on the intensity of fertilizer use and the economic value of the crop. For average management intensity, once every 3 years has been recommended by the International Fertilizer Evaluation and Improvement Program (ISFEIP; 1967–1974), while for very intensively managed areas, annual sampling is necessary. In my opinion, deep soil sampling in oxidic soils should be carried out once every 5 years, or at a changing point in a rotation. Deep sampling and related shallower sampling can estimate changes in soil carbon at that time.

As soil testing is a service program, the time between soil sampling and its analysis should be minimized. As emphasized in the first edition, delays in shipping, power failures and the absence of some reagents in many tropical laboratories are unfortunately very common. Several studies have been made on the effect of time in storage and drying on analytical results. This is still valid, four decades later, in much of the tropics.

After arrival at the laboratory, the soil sample is dried, then ground, either by pounding or by electric-powered grinders, passed through a 2-mm sieve, and assigned a laboratory number or stored until ready for analysis. Bouldin et al. (Reference Bouldin, Greweling and Lui1971) studied the effect of drying versus keeping several Oxisols and Ultisols moist for a period of months. Drying decreased the exchangeable aluminum ion (Al³⁺) content by 20 percent and increased the exchangeable ammonium (NH₄⁺) content when both were extracted by 1 M KCl. No changes in pH, available phosphorus, calcium or magnesium were observed.

12.5.2 Tackling Soil Variability

Spatial variability in soil properties is always an issue because soil is far from a homogeneous system and bears the footprints of generations of farmers who add temporal variability. The question is how wide-ranging is the variability. I have seen thousands of hectares of very uniform soils that follow a predictable pattern due to landscape position in Oxisols of South America, where large-scale farming is practiced. The uniformity of these soils results from the deep, weathered sediments that they are derived from; variability of these soils is minimal, and is found mainly along slopes. Local extension workers recommend sampling every tens or hundreds of hectares, every few years.

The mixing and leveling of flooded and puddled rice soils for decades or centuries in Asia acts essentially like the blender we use in our kitchens, making a pretty uniform slurry. Variability is thus minimized. At the other extreme, in smallholder farms in Africa, the human footprint is enormous. Lots of natural spatial variability exists in a mosaic of soils due to the topography and because the soils are derived from different kinds of rock, and also from management, as illustrated in Fig. 12.5. When people farm 1 or 2 hectares, they get to know their fields very well, and such knowledge helps in determining where to sample.

Fig. 12.5 In-field variability in a maize field in Koraro, Ethiopia.

Photo courtesy of Ray Weil

Fields close to the household (infields) are usually more fertile than those farther away (outfields) because smallholder households dump garbage and kitchens ashes, and their domestic animals often urinate and defecate in these nearby fields. This results in a fertility gradient, illustrated in Fig. 12.6, using a portable pH meter.

Fig. 12.6 Home garden soils are almost always much more fertile than soils in the larger fields away from home. Mbola, Central Tanzania.

Courtesy of Ray Weil

12.5.3 Laboratory Analysis

The most widely used way to assess the status of a nutrient in a soil is by an extraction in a wet chemistry laboratory. Soils are shaken with an extractant, which is supposed to indicate in a few minutes the amount of a nutrient a typical crop would take up from the soil over its lifetime. These values are variously known as available phosphorus, potassium, etc., usually including the name of the procedure’s author (Olsen, for example). Water is also used as an extractant, and resin, to capture ions in a soil slurry. All these extracted products are artifacts (like humic or fulvic acids), which makes them difficult to relate to reflectance spectrometry on the whole soil, but when they correlate with plant growth, they become useful metrics. Because of its mobility in soils, there are no similar soil tests for nitrate nitrogen (nitrate-N).

The wet chemistry laboratory is the backbone of a soil testing program. Unlike research laboratories, service laboratories must be geared to handle large numbers of samples, rapidly and accurately. A complete system of semi-automated apparatus was developed by the ISFEIP for tropical laboratories in the 1970s and 1980s. This system is summarized below, abstracted from the first edition of this book, when ISFEIP was being rapidly scaled up in Latin America and India. Other models are now in place.

Soil samples are measured volumetrically, eliminating the time-consuming weighing process, and then placed in multiple-unit trays, where the extracting and diluting solutions are added or transferred by specially designed diluter-dispensers. Other apparatus transfers the aliquots to spectrophotometers and pH-measuring units automatically. Shaking, stirring and cleaning apparatus are also semiautomatic and capable of handling 30 units at a time. A single technician can run 100 soil samples a day, measuring 10 determinations per sample. The pieces of apparatus are based on manual operation and use a minimum of electrical power. Expensive electronic equipment, such as atomic absorption spectrophotometers, is used only during the last stages. An additional advantage is that the same units can be used for plant tissue as for soil analysis.

The bulk of the work is done with two extractants: the dilute double acid (Mehlich) or the modified Olsen extractant for determining available phosphorus, potassium, calcium, magnesium, sodium, iron, manganese, zinc and copper; and the 1 M KCl extraction for aluminum. Routine tests for nitrogen, boron, sulfur and molybdenum could not be adapted to the system.

12.5.4 Soil Test Correlation

A soil test value per se is worthless; it is an empirical number that may or may not indirectly reflect nutrient availability, and as mentioned before, what is measured is an artifact of the extraction procedure. Soil test values become useful only when they are correlated with crop responses. Such correlations are usually conducted at two levels: an exploratory one in the greenhouse with a large number of widely diverse soils, and a more definite one in the field with fewer, but carefully selected, soils. Correlations can be done in many ways, using fairly complicated formulas, but in my view the simplest and most practical one is to plot relative yields versus the soil test levels as a scatter diagram. Then simply use a transparent overlay with a cross, placing it where the number of points in the upper left quadrant and in the lower right quadrant are at the minimum. This Cate–Nelson method (Cate and Nelson Reference Cate and Nelson1971) gives the soil critical level, as illustrated in Fig. 12.7. (I am sure a software program has been developed to do the same.)

Fig. 12.7 The Cate–Nelson method for determining the critical soil test level. In this case the critical level is 6 ppm available phosphorus by the Bray 1 method, for sugar cane, in the State of Pernambuco, Brazil. Each dot represents a field plot

(ISFEIP 1967)

This concept of a “critical value” for a given soil and farming system has important practical implications for efficient phosphorus use. Maintaining the soil at or close to the critical value has important benefits to the farmer (in terms of economic return) and to the environment, in terms of reducing the risk of phosphorus transfers to surface waters (Syers et al. Reference Syers, Johnston and Curtin2008).

The Cate–Nelson method also tells us the points with a high probability of a phosphorus response (the lower left quadrant), and those with a low probabiliy of phosphorus response (in the upper right quadrant). It does not tell you how much fertilizer to add, but it tells you whether you need it or not. This could be improved with better estimates of uncertainty.

If the soil in question is above the critical level, the decision is clearly not to apply that particular nutrient element to that crop. If the soil is below the critical level, then the decision becomes how much to apply. This requires yield response field trials. There are two main types of models to express yield response curves: the classic curvilinear model and the simpler linear reponse and plateau model.

12.5.5 Curvilinear Models

These classic models are based on the law of diminishing returns, where curvilinear functions (quadratic, square-root, logarithmic, Mitscherlich, etc.) are fitted to the yield response data. Statistical techniques determine which function fits the data best by providing the highest coefficient of determination (R²). The optimum fertilizer rate is determined at that point where marginal revenue equals the marginal cost (i.e. the point where the price of the last yield increment equals the cost of the last increment of fertilizer). The point can be determined mathematically or graphically by drawing a price:cost ratio line, expressed in the agronomic equivalents in the yield response diagram. The optimum yield is then determined at the point where a tangent of the price:cost ratio line intersects the response curve. When more than one element is deficient, regression models take this into account, as well as the interactions between these elements. The recommended rates are determined by solving simultaneous equations.

Curvilinear models are also used in a different manner to produce one function that takes into account soil test levels as variables, as well as other variables related to soil, climate and management properties, in an attempt to account for the uncontrolled variables. A yield equation relating potato yields in Peru needed 27 variables (Ryan and Perrin Reference Ryan and Perrin1973). Optimum fertilizer levels are calculated on the basis of levels of crop prices, fertilizer costs, expected rainfall and soil properties.

Such complex models are effective when there is adequate information about the variables involved and when prices are stable. They fail in the tropics where there are insufficient data to quantify all variables. The use of complex models in tropical regions is limited to after-the-fact analysis in areas with detailed information; they do not serve successfully as predictive tools. In addition, Anderson and Nelson (Reference Anderson and Nelson1975) found that quadratic models are biased when there is a marked response to the first fertilizer increment, followed by little or no response at higher rates. In such cases, optimum fertilizer recommendations became unrealistically high.

12.5.6 The Linear Response and Plateau Model

A series of studies conducted in England by Boyd (Reference Boyd1970) and in the United States by Bartholomew (Reference Bartholomew1972) summarized many fertilizer response functions over the world and concluded separately that in most instances fertilizer response curves can be characterized by a sharp linear increase followed by a flat horizontal line. Bob Cate and Larry Nelson in North Carolina then developed the Linear Response and Plateau model (LRP), which is also based on Liebig’s law of the minimum, and is a logical extension of the Cate–Nelson correlation model (Waugh et al. Reference Waugh, Cate, Nelson, Manzano, Bornemisza and Alvarado1975). Fertilizer response from a field, or a group of fields, is represented in this model by two straight lines for each nutrient. The first line represents the relatively steep response of an added nutrient until it ceases to be a major limiting factor. This is followed by a line showing a flat plateau, where further additions no longer increase yields. The fertilizer response curve so constructed consists of two main points. The “threshold yield” is the yield at the zero level of the nutrient in question, but not of all nutrients. The “plateau yield” is the yield at the point where the nutrient ceases to be a limiting factor; it is not the maximum yield because other factors may still limit yields. The fertilizer rate needed to reach the plateau yield is the recommended rate for the particular nutrient.

The comparison between the two approaches is shown in Fig. 12.8 for the same data set. The dotted lines indicate how the fertilizer recommendations are arrived at. The LRP model has only one optimum point, independent of cost and prices. The curvilinear model shows an optimum point based on the particular price:cost ratio at the time the experiments were conducted.

Fig. 12.8 Comparison between the LRP model and the quadratic model using the same field data from potatoes in Peru. The recommended rate is much lower using the first model.

Adapted from Waugh et al. (Reference von Liebig1973)

It is very important to note the wide differences in recommended rates between the two methods. The LRP model recommended a lower fertilizer rate (75 kg N/ha) to reach a yield plateau of about 19 t/ha of potatoes. This rate in effect provides nearly maximum yields, while preserving an efficient return per unit of fertilizer, because it is still along the increasing slope. The quadratic model in this case more than doubles the recommendation (160 kg N/ha) in order to obtain only an additional 1 t/ha of yield. This is in the relatively flat part of the curve, where variability is quite high. Small changes in yields result in large changes in recommended rates. In other comparisons, however, the difference in recommended rates might not be as large as in this example.

Which approach is right? I tried it myself with a set of field trials on the response to sulfur-coated urea by wetland rice in the Coast of Peru, where I used the quadratic method (Sanchez et al. Reference Waugh, Cate and Nelson1973). By using the graphic technique, plateau yields and recommended rates were computed. The average nitrogen recommendation in this very high-yielding environment was 224 kg N/ha according to the quadratic model, and 170 kg N/ha with the LRP model. The average gross return for fertilization at the recommended rates was not statistically significant between the two models. The net return per dollar invested in fertilizer was superior with the LRP model.

The LRP model is much simpler, and focuses on the range that provides the most bang for the buck, the linear portion at the threshold yield point. The quadratic models, in their search for the optimum value along the flat part of the response curve, are the least cost-effective. While the science is correct in their case (fertilizer response curves are indeed curves), the practicalities are not, because there is wide variability in the actual results under field conditions. As Bob Cate famously said “If you want to go (from North Carolina) to Tokyo you go via the Great Circle route, but if you want to go from this building to the next, you go in a straight line.” So let’s be as simple and practical as we can.

It is heartening to know that the LRP model is now widely used in US agriculture (Beegle Reference Beegle, Sims and Sharpley2005).

12.6.1 Spectrometry

The analysis of wavelengths of light is one of the simplest in physics, and not until the start of this century has it been applied to soil science at scale. Spectrometry has for decades been used as a standard in various industries, such as pharmaceuticals and paints. The near- and mid-infrared wavelengths can be correlated with many, but not all soil properties. In the tropics, the seminal papers by Keith Shepherd and Markus Walsh (Shepherd and Walsh Reference Shepherd and Walsh2002, Reference Shepherd and Walsh2007) showed the value of then near-infrared rudimentary spectrometers that can estimate many soil properties, just by passing light though a soil sample and making the necessary correlations (Fig. 12.9). The main properties that can be determined at present are: sand, silt, clay, organic carbon, exchangeable bases, cation exchange capacity, phosphorus retention and some soil minerals. Where spectrometry does not (yet) work is available phosphorus, potassium and sulfur, very critical soil fertility parameters. Spectrometry is also helpful in determining organic input quality and carbon mineralization rate (Shepherd et al. Reference Shepherd, Vanlauwe, Gachengo and Palm2005).

Fig. 12.9 Keith Shepherd (top left) using a portable spectrometer in the field in Kenya, and a technician using it in his Nairobi laboratory (bottom left). The resulting soil spectral signature (top right), and the close correlation between the predicted value by spectroscopy and the “actual” value of SOC from a wet chemistry laboratory (bottom right).

12.6.2 Digital Soil Mapping

Traditional soil maps are composed of polygons (mapping units) and their legends, showing mostly soil classification terms that are of very limited information value to anyone other than a pedologist. Such maps are static and tell you nothing about soil properties. Polygon maps assume that soils are uniform within the mapping unit, which is seldom the case. Interpretations made from them in terms of soil properties and agronomy are of course no better than the map.

To address some of these limitations, a digital soil-mapping working group of the International Union of Soil Sciences (IUSS) started work in 2006, and has since produced valuable information that can be found in books arising from their conferences (Lagacherie et al. Reference Lagacherie, McBratney and Voltz2007, Hartemink et al. Reference Hartemink, McBratney and Mendonça-Santos2008, Boettinger et al. Reference Boettinger, Howell, Moore, Hartemink and Kienast-Brown2010). In 2008, a group of soil scientists, agronomists and ecologists decided to produce a digital map of soil properties and soil functions of the world (Sanchez et al. Reference Sanchez, Ahamed, Carré, Hartemink, Hempel, Huising, Lagacherie, McBratney, McKenzie, Ml, Minasny, Montanarella, Okoth, Palm, Sachs, Shepherd, Vågen, Vanlauwe, Walsh, Winowiecki and Zhang2009, Fisher Reference Fisher2012), and the African portion of that is in progress.

Soil properties are estimated quantitatively with a statistical inference system using spatial (kriging, co-kriging) or non-spatial (regressions, neural networks) techniques, or both, and expressed as their probability of occurrence with uncertainty estimates. They are derived using quantitative relationships between the punctual soil measurements and the spatially continuous soil covariates. This results in maps of soil properties such as the ones selected by the Global Soil Map Consortium as their minimum data set, i.e. clay, organic carbon, bulk density, pH, effective cation exchange capacity, electrical conductivity and bulk density – bulk density is needed to express carbon and nutrients on a mass basis (t/ha) for biogeochemical modeling (Sanchez et al. Reference Sanchez, Ahamed, Carré, Hartemink, Hempel, Huising, Lagacherie, McBratney, McKenzie, Ml, Minasny, Montanarella, Okoth, Palm, Sachs, Shepherd, Vågen, Vanlauwe, Walsh, Winowiecki and Zhang2009). These are the first two actions in the process for developing the digital soil map, as shown in Fig. 12.10. Such properties are mapped on a pixel basis with a resolution as small as 30 × 30 m, which is great for smallholder farms.

Fig. 12.10 Conceptual flow of the soil digital map process.

Chart based on text by Sanchez et al. (Reference Sanchez, Ahamed, Carré, Hartemink, Hempel, Huising, Lagacherie, McBratney, McKenzie, Ml, Minasny, Montanarella, Okoth, Palm, Sachs, Shepherd, Vågen, Vanlauwe, Walsh, Winowiecki and Zhang2009)

The spatially inferred soil properties are then used to predict more difficult-to-measure soil functions (action 3, Fig. 12.10), such as available soil water storage, carbon density and phosphorus fixation, using pedotransfer functions. Rather than calibrating only individual soil properties to infrared spectra, multivariate associations are used to identify soil functions. The overall uncertainty of the prediction is assessed by combining the uncertainties of the input data with the uncertainties of the spatial inference and those of the soil functions, using global sensitivity analysis.

The key is to connect all these soil functions with fertilizer response or other georeferenced field data from well-conducted agronomic experiments (legacy data) as well as with digital maps of population density, roads, distance to markets, internet access and other socioeconomic digital maps (action 4, Fig. 12.10). Analyses of decisions need to be made, connecting the maps with actual on-the-ground experience. Then the final two actions (management recommendations and reaching the end user) can be achieved. Web-based digital soil maps can be made interactive. This is very much work in progress.

12.6.3 Bringing the Laboratory to the Field

Although soil nutrient replenishment in sub-Saharan Africa is widely recognized as a critical biophysical entry point to an agricultural transformation, the actual application of soil science is still minimal, and frustratingly so, in large part because of the delay in obtaining the necessary information from soils sampled in farmers’ fields and sent to a wet-chemistry, conventional laboratory. The data often take a year to come back (if at all) and the paper-based process is prone to errors of recording and transcribing, as well as delays in transmission, interpretation and visualization. Worse still, the data are usually hard to interpret, largely irrelevant and often of questionable quality. The poor infrastructure of Africa and the budget limitations of national institutions are the main reasons used to rationalize this useless exercise. This is not a problem in the more advanced tropical countries, but it is also a problem in many areas outside Africa as well.

The way to get around this is to get the data in situ and send it electronically to the central laboratory where senior people or algorithms make the recommendations and send it back electronically to the extension agent or farmer. The problem is that, up until recently, field soil test kits have proven to be highly inaccurate and not sufficiently precise (Vanlauwe Reference Vanlauwe2008). Weil (Reference Weil2011) of the University of Maryland and Columbia University has been developing a wet-chemistry “SoilDoc,” which has been proven to work well and is highly correlated with central wet-chemistry laboratories (Fig. 12.11). It enables extension workers to make on-the-spot diagnoses of soil constraints, allowing targeted recommendations to advise farmers in real time (Sanchez et al. Reference Sanchez, Weil and Rose2013).

Fig. 12.11 SoilDoc, a “lab-in-the-box” field test kit. Ray Weil is entering the data in a smart phone that transmits it to the cloud or to a central location. Arusha, Tanzania, September 2012.

This tool uses state-of-the-art battery-powered miniaturized instruments, originally created for other purposes such as the beer and swimming-pool testing industries. The soil fertility parameters analyzed include soil pH, electrical conductivity (indicative of general fertility levels as well as salinity issues), biologically active organic carbon, and 0.01 M calcium chloride (CaCl₂)-extractable nitrate-N, sulfate-S, phosphate-P, and potassium using specific sensors. Additionally, the kit includes tools to measure soil physical properties such as surface sealing strength, plow pans and volumetric soil water content. Soil texture is estimated by feel. SoilDoc essentially measures what a conventional wet-chemistry laboratory does at a similar level of accuracy (Fig. 12.12) plus biologically active SOC and several physical properties (Weil Reference Weil2011, Sanchez et al. Reference Sanchez, Weil and Rose2013). Like digital soil mapping, this too is very much work in progress.

Fig. 12.12 Correlations for pH (top), nitrate-N (middle), and available phosphorus (bottom), determinations by the SoilDoc kit and a wet-chemistry laboratory.

Reproduced with permission from Ray Weil, University of Maryland

At the time of this writing, we need to use the field-based wet chemistry of SoilDoc and similar systems in conjunction with near-infrared spectrometers, because spectrometry cannot determine available soil test levels since they are artifacts of the extractants used. Wavelength data cannot detect something that does not exist in the soil. To repeat, both approaches are works in progress and their combined use could result in positive synergies.

The combination of these approaches, as well as the rapidly developing soil scanning by satellites, shows tremendous potential for faster and more effective ways of evaluating soil fertility.