The delineation of agricultural management zones with high resolution remotely sensed data

Remote sensing (RS) techniques have been widely considered to be a promising source of information for land management decisions. The objective of this study was to develop and compare different methods of delineating management zones (MZs) in a field of winter wheat. Soil and yield samples were collected, and five main crop nutrients were analyzed: total nitrogen (TN), nitrate nitrogen (NN), available phosphorus (AP), extractable potassium (EP) and organic matter (OM). At the wheat heading stage, a scene of Quickbird imagery was acquired and processed, and the optimized soil-adjusted vegetation index (OSAVI) was determined. A fuzzy k-means clustering algorithm was used to define MZs, along with fuzzy performance index (FPI), and modified partition entropy (MPE) for determining the optimal number of clusters. The results showed that the optimal number of MZs for the present study area was three. The MZs were delineated in three ways; based on soil and yield data, crop RS information and the combination of soil, yield and RS information. The evaluation of each set of MZs showed that the three methods of delineating zones can all decrease the variance of the crop nutrients, wheat spectral parameters and yield within the different zones. Considering the consistent relationship between the crop nutrients, wheat yield and the wheat spectral parameters, satellite remote sensing shows promise as a tool for assessing the variation in soil properties and yield in arable fields. The results of this study suggest that management zone delineation using RS data was reliable and feasible.     

Geostatistical modelling of within-field soil and yield variability for management zones delineation: a case study in a durum wheat field

The paper proposes a geostatistical approach for delineating management zones (MZs) based on multivariate geostatistics, showing the use of polygon kriging to compare durum wheat yield among the different MZs (polygons). The study site was a durum wheat field in southern Italy and yield was measured over three crop seasons. The first regionalized factor, calculated with factorial cokriging, was used to partition the field into three iso-frequency classes (MZs). For each MZ, the expected value and standard deviation of yield were estimated with polygon kriging over the three crop seasons. The yield variation was only in part related to soil properties but most of it might be ascribable to different patterns of meteorological conditions. Both components of variation (plant and soil) in a cropping system should then be taken into account for an effective management of rainfed durum wheat in precision agriculture. The proposed approach proved multivariate Geostatistics to be effective for MZ delineation even if further testing is required under different cropping systems and management.     

Evaluation of site-specific management zones on a farm with 124 contiguous small paddy fields in a multiple-cropping system

The identification of homogeneous management zones (MZs) within a field is a basis for site-specific management (SSM). We assessed the method of defining MZs based on the spatio-temporal homogeneity of six soil properties and above-ground biomass data from paddy rice, winter wheat and soybean over 3 years on a farm with 124 contiguous small paddy fields. The soil data were recorded at 372 soil sampling sites on a rectangular grid over the farm. A non-hierarchical cluster analysis was applied to the soil data and the algorithm grouped the sites into three clusters with similar soil properties. These clusters represent soil fertility and soil drainage. The three clusters were not randomly distributed across the fields, but formed contiguous areas associated with landscape position. This was due to the spatial variation of the soil in the study area. We delineated five MZs based on the spatial structure of the soil heterogeneity of the study area. The validity of the MZs was evaluated using the biomass data from paddy rice, winter wheat and soybean in each MZ; this depended mainly on soil fertility when conditions were dry. When the growing season precipitation was greater than the 10-year average, the biomass of winter wheat and soybean depended on soil drainage. This suggested that the delineation of MZs for site-specific management in fields under a paddy-upland crop rotation system should be based on several soil properties. The biomass data from the three crops over 3 years was not effective for delimiting MZs.     

Spatial variation in tree characteristics and yield in a pear orchard

We examined the spatial structure of fruit yield, tree size, vigor, and soil properties for an established pear orchard using Moran’s I, geographically weighted regression (GWR) and variogram analysis to determine potential scales of the factors affecting spatial variation. The spatial structure differed somewhat between the tree-based measurements (yield, size and vigor) and the soil properties. Yield, trunk cross-sectional area (TCSA) and normalized difference vegetation index (NDVI, used as a surrogate for vigor) were strongly spatially clustered as indicated by the global Moran’s I for these measurements. The autocorrelation between trees (determined by applying a localized Moran’s I) was greater in some areas than others, suggesting possible management by zones. The variogram ranges for TCSA and yield were 30–45 m, respectively, but large nugget variances indicated considerable variability from tree to tree. The variogram ranges of NDVI varied from about 14–27 m. The soil properties copper, iron, organic matter and total exchange capacity (TEC) were spatially structured, with longer variogram ranges than those of the tree characteristics: 31–95 m. Boron, pH and zinc were not spatially correlated. The GWR analyses supported the results from the other analyses indicating that assumptions of strict stationarity might be violated, so regression models fitted to the entire dataset might not be fitted optimally to spatial clusters of the data.     

Regional model for nitrogen fertilization of site-specific rainfed corn in haplustolls of the central Pampas,Argentina

In semi-arid regions, soil water and nitrogen (N) are generally limiting factors for corn (Zea mays L.) production; hence, implementation of appropriate N fertilization strategies is needed. The use of precision agriculture practices based on specific site and crop properties may contribute to a better allocation of fertilizer among management zones (MZ). The aim of this study was to develop a model for diagnosis of N availability and recommendation of N fertilizer rates adjusted to MZ for dryland corn crops growing in Haplustolls. The model considered variability between MZ by including site-specific variables [soil available water content at sowing (SAW) and Available Nitrogen (soil available N-NO₃ at planting + applied N, Nd)] using spatial statistical analysis. The study was conducted in Córdoba, Argentina in Haplustolls and consisted in four field trials of N fertilizer (range 0–161 kg N ha⁻¹) in each MZ. The MZ were selected based on elevation maps analysis. Grain yields varied between MZ and increased with larger SAW and Nd at sowing. Grain responses to Nd and SAW in any MZ were not different between sites, allowing to fit a regional model whose parameters (Nd, Nd², SAW, SAW²) contributed significantly (p < 0.001) to yield prediction. Agronomical and economically optimum N rates varied among MZs. However, the spatial variability of optimum N rates among MZs within sites was not enough to recommend variable N fertilizer rates instead of a uniform rate. Variable N fertilizer rates should be recommended only if variability in SAW and soil N among MZ is greater than that found in this work.     

Delineating productivity zones in a citrus grove using citrus production,tree growth and temporally stable soil data

The productivity of a citrus grove with variation in tree growth was mapped to delineate zones of productivity based on several indicator properties. These properties were fruit yield, ultrasonically measured tree canopy volume, normalized difference vegetation index (NDVI), elevation and apparent electrical conductivity (EC_a). The spatial patterns of soil series, soil color and EC_a, and their correspondence with the variation in yield emphasized the importance of variation in the soil in differentiating the productivity of the grove. Citrus fruit yield was positively correlated with canopy volume, NDVI and EC_a, and yield was negatively correlated with elevation. Although all the properties were strongly correlated with yield and were able to explain the productivity of the grove, citrus tree canopy volume was most strongly correlated (r = 0.85) with yield, explaining 73% of its variation. Tree canopy volume was used to classify the citrus grove into five productivity zones termed as ‘very poor’, ‘poor’, ‘medium’, ‘good’ and ‘very good’ zones. The study showed that productivity of citrus groves can be mapped using various attributes that directly or indirectly affect citrus production. The productivity zones identified could be used successfully to plan soil sampling and characterize soil variation in new fields.     

An integrated approach to site-specific management zone delineation

Dividing fields into a few relatively homogeneous management zones (MZs) is a practical and cost-effective approach to precision agriculture. There are three basic approaches to MZ delineation using soil and/or landscape properties, yield information, and both sources of information. The objective of this study is to propose an integrated approach to delineating site-specific MZ using relative elevation, organic matter, slope, electrical conductivity, yield spatial trend map, and yield temporal stability map (ROSE-YSTTS) and evaluate it against two other approaches using only soil and landscape information (ROSE) or clustering multiple year yield maps (CMYYM). The study was carried out on two no-till corn-soybean rotation fields in eastern Illinois, USA. Two years of nitrogen (N) rate experiments were conducted in Field B to evaluate the delineated MZs for site-specific N management. It was found that in general the ROSE approach was least effective in accounting for crop yield variability (8.0%–9.8%), while the CMYYM approach was least effective in accounting for soil and landscape (8.9%–38.1%), and soil nutrient and pH variability (9.4%–14.5%). The integrated ROSE-YSTTS approach was reasonably effective in accounting for the three sources of variability (38.6%–48.9%, 16.1%–17.3% and 13.2%–18.7% for soil and landscape, nutrient and pH, and yield variability, respectively), being either the best or second best approach. It was also found that the ROSE-YSTTS approach was effective in defining zones with high, medium and low economically optimum N rates. It is concluded that the integrated ROSE-YSTTS approach combining soil, landscape and yield spatial-temporal variability information can overcome the weaknesses of approaches using only soil, landscape or yield information, and is more robust for MZ delineation. It also has the potential for site-specific N management for improved economic returns. More studies are needed to further evaluate their appropriateness for precision N and crop management.     

Data fusion techniques for delineation of site-specific management zones in a field in UK

Fusion of different data layers, such as data from soil analysis and proximal soil sensing, is essential to improve assessment of spatial variation in soil and yield. On-line visible and near infrared (Vis–NIR) spectroscopy have been proved to provide high resolution information about spatial variability of key soil properties. Multivariate geostatistics tools were successfully implemented for the delineation of management zones (MZs) for precision application of crop inputs. This research was conducted in a 18 ha field to delineate MZs, using a multi-source data set, which consisted of eight laboratory measured soil variables (pH, available phosphorus (P), cation exchange capacity, total nitrogen (TN), total carbon (TC), exchangeable potassium (K), sand, silt) and four on-line collected Vis–NIR spectra-based predicted soil variables (pH, P, K and moisture content). The latter set of data was predicted using the partial least squares regression (PLSR) technique. The quality of the calibration models was evaluated by cross-validation. Multi-collocated cokriging was applied to the soil and spectral data set to produce thematic spatial maps, whereas multi-collocated factor cokriging was applied to delineate MZ. The Vis–NIR predicted K was chosen as the exhaustive variable, because it was the most correlated with the soil variables. A yield map of barley was interpolated by means of the inverse distance weighting method and was then classified into 3 iso-frequency classes (low, medium and high). To assess the productivity potential of the different zones of the field, spatial association between MZs and yield classes was calculated. Results showed that the prediction performance of PLSR calibration models for pH, P, MC and K were of excellent to moderate quality. The geostatistical model revealed good performance. The estimates of the first regionalised factor produced three MZs of equal size in the studied field. The loading coefficients for TC, pH and TN of the first factor were highest and positive. This means that the first factor can be assumed as a synthetic indicator of soil fertility. The overall spatial association between the yield classes and MZs was about 40 %, which reveals that more than 50 % of the yield variation can be attributed to more dynamic factors than soil parameters, such as agro-meteorological conditions, plant diseases and nutrition stresses. Nevertheless, multivariate geostatistics proved to be an effective approach for site-specific management of agricultural fields.     

Use of soil nitrogen parameters and texture for spatially-variable nitrogen fertilization

Recent studies have demonstrated the potential importance of using soil texture to modify fertilizer N recommendations. The objective of this study was to determine (i) if surface clay content can be used as an auxiliary variable for estimating spatial variability of soil NO₃–N, and (ii) if this information is useful for variable rate N fertilization of non-irrigated corn [Zea mays (L.)] in south central Texas, USA across years. A 64 ha corn field with variable soil type and N fertility level was used for this study during 2004–2007. Plant and surface and sub-surface soil samples were collected at different grid points and analyzed for yield, soil N parameters and texture. A uniform rate (UR) of 120 kg N ha⁻¹ in 2004 and variable rates (VAR) of 0, 60, 120, and 180 kg N ha⁻¹ in 2005 through 2007 were applied to different sites in the field. Distinct yield variation was observed over this time period. Yield and soil surface clay content and soil N parameters were strongly spatially structured. Corn grain yield was positively related to residual NO₃–N with depth and either negatively or positively related to clay content depending on precipitation. Residual NO₃–N to 0.60 and 0.90 m depths was more related to corn yield than from shallower depths. The relationship of clay content with soil NO₃–N was weak and not temporally stable. Yield response to N rate also varied temporally. Supply of available N with depth, soil texture and growing season precipitation determined proper N management for this field.     

Estimation of yield zones using aerial images and yield data from a few tracks of a combine harvester

Yield maps derived from yield mapping systems are often erroneous not only due to limitations in measuring the yield precisely but due to insufficient consideration of the requirements of yield mapping systems in practice as well. Aerial images of cultivated crop fields at an advanced growth stage frequently provide a spatial pattern similar to that of yield maps. Therefore, the possibility of generating a yield map using aerial images and measured yield data of a few tracks was examined for a period of 2 years in two fields grown with cereals. Yield zones based on Visible Atmospherically Resistant Index (VARI) values were compared with yield zones based on measured yield data of the whole field. About half of the grid cells of a field were allocated to the same yield zones irrespective of the mode of yield determination. Using the Kruskal–Wallis test, the data sub-sets of measured yield within the yield zones based on the VARI values differed significantly for all tested yield zones. As a result, the approach was successful in the case of these experimental sites.     

Using spectral distance, spectral angle and plant abundance derived from hyperspectral imagery to characterize crop yield variation

Vegetation indices (VIs) derived from remote sensing imagery are commonly used to quantify crop growth and yield variations. As hyperspectral imagery is becoming more available, the number of possible VIs that can be calculated is overwhelmingly large. The objectives of this study were to examine spectral distance, spectral angle and plant abundance (crop fractional cover estimated with spectral unmixing) derived from all the bands in hyperspectral imagery and compare them with eight widely used two-band and three-band VIs based on selected wavelengths for quantifying crop yield variability. Airborne 102-band hyperspectral images acquired at the peak development stage and yield monitor data collected from two grain sorghum fields were used. A total of 64 VI images were generated based on the eight VIs and selected wavelengths for each field in this study. Two spectral distance images, two spectral angle images and two abundance images were also created based on a pair of pure plant and soil reference spectra for each field. Correlation analysis with yield showed that the eight VIs with the selected wavelengths had r values of 0.73–0.79 for field 1 and 0.82–0.86 for field 2. Although all VIs provided similar correlations with yield, the modified soil-adjusted vegetation index (MSAVI) produced more consistent r values (0.77–0.79 for field 1 and 0.85–0.86 for field 2) among the selected bands. Spectral distance, spectral angle and plant abundance produced similar r values (0.76–0.78 for field 1 and 0.83–0.85 for field 2) to the best VIs. The results from this study suggest that either a VI (MSAVI) image based on one near-infrared band (800 or 825 nm) and one visible band (550 or 670 nm) or a plant abundance image based on a pair of pure plant and soil spectra can be used to estimate relative yield variation from a hyperspectral image.     

A floating sensing system to evaluate soil and crop variability within flooded paddy rice fields

Continuous paddy rice cultivation requires fields to be flooded most of the time limiting seriously the collection of detailed soil information. So far, no appropriate soil sensor technology for identifying soil variability of flooded fields has been reported. Therefore, the primary objective was the development of a sensing system that can float, acquire and process detailed geo-referenced soil information within flooded fields. An additional objective was to determine whether the collected apparent electrical conductivity (EC_a) information could be used to support soil management at a within-field level. A floating sensing system (FloSSy) was built to record EC_a using the electromagnetic induction sensor EM38, which does not require physical contact with the soil. Its feasibility was tested in an alluvial paddy field of 2.7 ha located in the Brahmaputra floodplain of Bangladesh. The high-resolution (1 × 1 m) EC_a data were classified into three classes using the fuzzy k-means classification method. The variation among the classes could be attributed to differences in subsoil (0.15–0.30 m below soil surface) bulk density, with the smallest EC_a values representing the lowest bulk density. This effect was attributed to differences in compaction of the plough pan due to differential puddling. There was also a significant difference in rice yield among the EC_a classes, with the smallest EC_a values representing the lowest yield. It was concluded that the floating sensing system allowed the collection of relevant soil information, opening potential for precision agriculture practices in flooded crop fields.     

Nutrient Management Zones for Citrus Based on Variation in Soil Properties and Tree Performance

Site-specific soil management can improve profitability and environmental protection of citrus groves having large spatial variation in soil and tree characteristics. The objectives of this study were to identify soil factors causing tree performance decline in a variable citrus grove, and to develop soil-specific management zones based on easily measured soil/tree parameters for variable rate applications of appropriate soil amendments. Selected soil properties at six profile depths (0–1.5 m), water table depth, ground conductivity, leaf chlorophyll index, leaf nutrients and normalized difference vegetation index were compared at 50 control points in a highly variable 45-ha citrus grove. Regression analysis indicated that 90% of spatial variation in tree growth, assessed by NDVI, was explained by average soil profile properties of organic matter, color, near-infrared reflectance, soil solution electrical conductivity, ground conductivity and water table depth. Regression results also showed that soil samples at the surface only (0–150 mm) explained 78% of NDVI variability with NIR and DTPA-extractable Fe. Excessive available copper in low soil organic matter areas of the grove apparently induced Fe deficiency, causing chlorotic foliage disorders and stunted tree growth. The semivariograms of selected variables showed a strong spatial dependence with large ranges (varied from 230 m to 255 m). This grove can be divided into different management zones on the basis of easily measured NDVI and/or soil organic matter for variable rate application of dolomite and chelated iron to improve tree performance.     

Off-nadir hyperspectral measurements in maize to predict dry matter yield,protein content and metabolisable energy in total biomass

Sensor-based methods of analysis to assess dry matter yield and quality constituents of crops are time- and labour-saving, and can facilitate site-specific management. Nevertheless, standard nadir measurements of maize (Zea mays cv. Ambrosius), based on top-of-canopy reflectance, are difficult due to plant heights of more than three metres. This study was conducted to explore the potential of off-nadir field spectral measurements for the non-destructive prediction of dry matter yield (DM), metabolisable energy (ME) and crude protein (CP) in total biomass in a maize canopy. Plants were measured at five different heights (0–50, 50–100, 100–50, 150–200 and 200–250 cm above the soil) at three zenith view angles (60°, 75° and 90°, respectively). Modified partial least squares regression was used for analysis of the hyperspectral data (355–2300 nm and 620–1000 nm). Optimum combinations of angle and height as well as an optimum one-sensor-strategy were determined for DM yield, CP and ME in total biomass. Coefficients of determination for off-nadir measurements were compared to nadir measurements; the results showed improved prediction accuracies for DM yield and ME using off-nadir measurements, but not for CP for which nadir measurements were better.     

Automatic identification of crop and weed species with chlorophyll fluorescence induction curves

Automatic identification of crop and weed species is required for many precision farming practices. The use of chlorophyll fluorescence fingerprinting for identification of maize and barley among six weed species was tested. The plants were grown in outdoor pots and the fluorescence measurements were done in variable natural conditions. The measurement protocol consisted of 1 s of shading followed by two short pulses of strong light (photosynthetic photon flux density 1700 μmol m⁻² s⁻¹) with 0.2 s of darkness in between. Both illumination pulses caused the fluorescence yield to increase by 30–60% and to display a rapid fluorescence transient resembling transients obtained after long dark incubation. A neural network classifier, working on 17 features extracted from each fluorescence induction curve, correctly classified 86.7–96.1% of the curves as crop (maize or barley) or weed. Classification of individual species yielded a 50.2–80.8% rate of correct classifications. The best results were obtained if the training and test sets were measured on the same day, but good results were also obtained when the training and test sets were measured on different dates, and even if fluorescence induction curves measured from both leaf sides were mixed. The results indicate that fluorescence fingerprinting has potential for rapid field separation of crop and weed species.     

Evaluation of an algorithm for automatic detection of broad-leaved weeds in spring cereals

Lack of automatic weed detection tools has hampered the adoption of site-specific weed control in cereals. An initial object-oriented algorithm for the automatic detection of broad-leaved weeds in cereals developed by SINTEF ICT (Oslo, Norway) was evaluated. The algorithm (“WeedFinder”) estimates total density and cover of broad-leaved weed seedlings in cereal fields from near-ground red–green–blue images. The ability of “WeedFinder” to predict ‘spray’/‘no spray’ decisions according to a previously suggested spray decision model for spring cereals was tested with images from two wheat fields sown with the normal row spacing of the region, 0.125 m. Applying the decision model as a simple look-up table, “WeedFinder” gave correct spray decisions in 65–85% of the test images. With discriminant analysis, corresponding mean rates were 84–90%. Future versions of “WeedFinder” must be more accurate and accommodate weed species recognition.     

Estimation of economic and environmental potentials of variable rate versus uniform N fertilizer application to spring barley on morainic soils in SE Norway

Spring barley was grown for 4 years (2001–2004) in field trials at two sites on morainic soil in central SE Norway, with five N level treatments: 0, 60, 90, 120 and 150 kg N ha^-1. Regression analyses showed that a selection of soil properties could explain 95–98% of the spatial yield variation and 47–90% of the yield responses (averaged over years). A strategy with uniform fertilizer application of 120 kg N ha⁻¹ (U _N120) was compared with two variable-rate (VR) strategies, with a maximum N rate of either 150 kg N ha⁻¹ (VR_N150) or 180 kg N ha⁻¹ (VR_N180). These strategies were tested using either Norwegian prices (low price ratio of N fertilizer to yield value; P_N/P_Y), or Swedish prices (high P_N/P_Y). The VR_N180 strategy had the highest potential yield and net revenue (yield value minus N cost) at both sites and under both price regimes. Using this strategy with Norwegian prices would increase the profit of barley cropping as long as at least 40 and 31% of the estimated potential increase in net revenue was realized, respectively. Using Swedish prices, uniform application appeared to be as good as or even better economically than the VR methods, when correcting for extra costs of VR application. The environmental effect of VR compared with uniform application, expressed as N not accounted for, showed contrasting effects when using Norwegian prices, but was clearly favourable using Swedish prices, with up to 20% reduction in the amount of N not accounted for.     

Adoption of variable rate fertiliser application in the Australian grains industry: status, issues and prospects

Variable rate application of fertiliser (VR) is a practice underpinning a profitable grains industry in Australia. We updated the extent of VR adoption through a national survey (n = 1 130) covering all grain growing regions. Three smaller regional-based surveys (n = 39–102) collected detailed information on the nature and reasoning behind the use of various forms of the technology. We analysed the constraints to the adoption of each step using adoption theory. Surveys showed that 20% of grain growers have adopted some form of VR (varied from 11–35%), up significantly from <5% found 6 years earlier. Adopters are more than likely to have larger farms with a higher area in cropping. Many non-adopters were convinced of the agronomic and economic benefits of VR. A significant proportion of growers were managing within-field variability with manually-operated systems rather than more sophisticated VR technology, and have adopted some form of VR without yield maps, preferring to use soil tests, electro-magnetic induction or their own knowledge of soil and yield variation to define management. The rate of adoption is expected to continue to rise based on greater awareness of the benefits of the technology. The constraints to adoption were technical issues with equipment and software access to service provision and the incompatibility of equipment with existing farm operations.     

Site-specific early season potato yield forecast by neural network in Eastern Canada

Deterministic potato (Solanum tuberosum L.) growth models hardly rely on driving seasonal field variables that directly characterize spatial variation of plant growth. For example, the SUBSTOR model computes the leaf area index (LAI) as an auxiliary variable from meteorological conditions and soil properties. Empirical models may account for seasonal LAI functions and accurately predict potato yield. The objective was to evaluate multiple linear regression (MLR) and neural networks (NN) as predictive models of potato yield. Using data from several replicated on-farm experiments conducted over 3 years, model performance was evaluated for their capacity to forecast tuber yields 9, 10 and 11 weeks before harvest compared to SUBSTOR. A 3-input NN using LAI functions and cumulative rainfall yielded the most accurate estimations and forecasts of tuber yields. This NN showed that tuber yield of contrasting zones was mostly a function of meteorological conditions prevailing during the first 5–8 weeks after planting. Subsequent development of tubers was essentially controlled by biomass allocation to tubers. The NN models were more coherent than MLR and SUBSTOR for two reasons: (1) the use of seasonal LAI directly as input rather than computed as an auxiliary variable and (2) the non-linearity of the modeling process resulting in more accurate estimation of the temporal discontinuities of potato tuber growth. This model showed potential for application in precision agriculture by accounting for temporal and spatial real-time climatic and crop data.