Image masking for crop yield forecasting using AVHRR NDVI time series imagery

Time series of the Advanced Very High-Resolution Radiometer (AVHRR) Normalized Difference Vegetation Index (NDVI) have been used for crop yield forecasting since the 1980s. Image masking was a critical component of several of these yield forecasting efforts as researchers attempted to isolate subsets of a region’s pixels that would improve their modeling results. Approaches generally sought to identify cropped pixels and when possible, pixels that corresponded to the particular crop type under investigation. In this paper, the former approach will be referred to as cropland masking, and the latter approach will be called crop-specific masking.
The research presented in this paper examines the underlying assumptions made when image masking for the purpose of regional crop yield forecasting. An alternative statistical image masking approach (called yield-correlation masking) is proposed that is objective (i.e., can be automated) and has the flexibility to be applied to any region with several years of time series imagery and corresponding historical crop yield information. The primary appeal of yield-correlation masking is that, unlike cropland masking, no land cover map is required, yet we will show that NDVI models generated using the two methods demonstrate comparable predictive ability.

The goal of this research is very specific: We will establish the yield-correlation masking procedure as a viable image masking technique in the context of crop yield forecasting. To accomplish this objective, we empirically evaluate and compare cropland masking and yield-correlation masking for the purpose of crop yield forecasting. Cropland masking has been shown to benefit yield forecasting models, thus providing a practical benchmark. In the process, we present a robust statistical yield forecasting protocol that can be applied to any (region, crop)-pair possessing the requisite data, and this protocol is used to evaluate the two masking methods that are being compared.
This paper is developed as follows. A brief review of related research is presented. The two primary data sets, AVHRR NDVI time-series imagery and United States Department of Agriculture (USDA) regional crop yield data, are described. Details of the study regions, crop types, and time periods under investigation are specified. Three image masking procedures are discussed, two of which are evaluated in the research.
Finally, the modeling approach and performance evaluation framework are described, along with a summary of results and conclusions drawn from the analysis. Details of the modeling strategy used are described in the Appendix.
Related research
Traditionally, yield estimations are made through agrometeorological modeling or by compiling survey information provided throughout the growing season. Yield estimates derived from agro-meteorological models use soil properties and daily weather data as inputs to simulate various plant processes at a field level (Wiegand & Richardson, 1990; Wiegand et al., 1986). At this scale, agro-meteorological crop yield modeling provides useful results. However, at regional
scales these models are of limited practical use because of spatial differences in soil characteristics and crop growthdetermining factors such as nutrition levels, plant disease, herbicide and insecticide use, crop type, and crop variety, which would make informational and analytical costs excessive. Additionally, Rudorff and Batista (1991) indicated that, at a regional level, agro-meteorological models are unable to completely simulate the different crop growing conditions that result from differences in climate, local weather conditions, and land management practices. The scale of applicability of agrometeorological models is getting larger, though, but presently only through the integration of remotely sensed imagery. For instance, Doraiswamy et al. (2003) developed a method using AVHRR NDVI data as proxy inputs to an agro-meteorological model in estimating spring wheat yields at county and subcounty scales in the U.S. state of North Dakota.
In the past 25 years, many scientists have utilized remote sensing techniques to assess agricultural yield, production, and crop condition. Wiegand et al. (1979) and Tucker et al. (1980) first identified a relationship between the NDVI and crop yield using experimental fields and ground-based spectral radiometer measurements. Final grain yields were found to be highly correlated with accumulated NDVI (a summation of NDVI between two dates) around the time of maximum greenness
(Tucker et al., 1980). In another experimental study, Das et al. (1993) used remotely sensed data to predict wheat yield 85– 110 days before harvest in India. These early experiments identified relationships between NDVI and crop response, paving the way for crop yield estimation using satellite imagery.
Rasmussen (1992) used 34 AVHRR images of Burkina Faso, Africa, for a single growing season to estimate millet yield. Using accumulated NDVI and statistical regression techniques, he found strong correlations between accumulated NDVI and yield, but only during the reproductive stages of crop growth. The lack of a strong correlation between accumulated NDVI and yield during other stages of growth was attributed to the limited temporal profile of imagery used in the study and the high variability of millet yield in his study area. Potdar (1993) estimated sorghum yield in India using 14 AVHRR images from the same growing season. He was able to forecast actual yield at an accuracy of T15% up to 45 days before harvest. Rudorff and Batista (1991) used NDVI values
as inputs into an agro-meteorological model to explain nearly 70% of the variation in 1986 wheat yields in Brazil. Hayes and Decker (1996) used AVHRR NDVI data to explain more than 50% of the variation in corn yields in the United States Corn Belt. Each of these studies found positive relationships between crop yield and NDVI, but the strength of the relationships depended upon the amount and quality of the imagery used.
Some studies have used large, multi-year AVHRR NDVI data sets. Maselli et al. (1992) found strong correlations between NDVI and final crop yields in the Sahel region of Niger using 3 years of AVHRR imagery (60 images). In India, Gupta et al. (1993) used 3 years of AVHRR data to estimate wheat yields within T5% up to 75 days before harvest. The success of this study was dependent on the fact that over 80% of the study area was covered with wheat. In Greece, 2 years of AVHRR imagery were used to estimate crop yields (Quarmby et al., 1993). Actual harvested rice yields were predicted with an accuracy of T10%, and wheat yields were predicted with an accuracy of T12% at the time of maximum greenness. Groten (1993) was able to predict crop yield with a T15% estimation error 60 days before harvest in Burkina Faso using regression techniques and 5 years of AVHRR NDVI data (41 images).
Doraiswamy and Cook (1995) used 3 years of AVHRR NDVI imagery to assess spring wheat yields in North and South Dakota in the United States. They concluded that the most promising way to improve the use of AVHRR NDVI for estimating crop yields at regional scales would be to use larger temporal data sets, better crop masks, and climate data. Lee et al. (1999) used a 10-year, biweekly AVHRR data set to forecast  corn yields in the U.S. state of Iowa. They found that the most accurate forecasts of crop yield were made using a cropland mask and measurements of accumulated NDVI. Maselli and Rembold (2001) used multi-year series of annual crop yields and monthly NDVI to develop cropland masks for four Mediterranean African countries. They found that application of the derived cropland masks improved relationships between NDVI and final yield during optimal yield prediction periods. Ferencz et al. (2004) found yields of eight different crops in Hungary to be highly correlated with optimized, weighted seasonal NDVI sums using 1-km AVHRR NDVI from 1996 to 2000. They used non-forest vegetation masks and a novel time series interpolation approach and actually obtained their best results when using a greenness index equivalent to the numerator of the NDVI formula (NIR-RED; see Section 3). Additionally, many researchers have found that crop condition and yield estimation are improved through the inclusion of metrics that characterize crop development stage (Badhwar & Henderson, 1981; Groten, 1993; Kastens, 1998; Lee et al., 1999; Quarmby et al., 1993; Rasmussen, 1992). Ancillary data have been found useful as well. For example, Rasmussen (1997) found that soil type information improved the explanation of millet and ground nut yield variation using 3 years of AVHRR NDVI from the Peanut Basin in Senegal. In a later study, Rasmussen (1998) found that the inclusion of tropical livestock unit density further improved the explanation of millet yield variation in intensively cultivated regions of the Peanut Basin.

Based on the studies described, for the purpose of crop yield forecasting, longer time series of NDVI imagery are preferred to shorter ones. Also, few image masking techniques have been thoroughly and comparatively explored, likely due to the inherent complexities underlying this phase in any remote sensing-based yield forecasting methodology. Thus, to help achieve the goal of this project, an important objective of this research is to use historical yield information and historical time series AVHRR NDVI imagery to devise a thorough and robust statistical procedure for obtaining early to mid-season crop yield forecasts, with particular emphasis on image masking. The techniques described in this paper can be applied to any (region, crop)-pair that possesses sufficient historical yield information and corresponding time series NDVI imagery. Since few meaningful crop phenology metrics can be accurately derived at early points in the growing season, our research does not attempt to use this information. Also, no ancillary information is used, to prevent dependence on the availability of such data.

Description of data
The research presented in this paper relies on two data sets. The first is a time series of biweekly AVHRR NDVI compositeimagery from 1989 to 2000, obtained from the U.S. Geological survey Earth Resources Observation Systems (EROS) Data Center (EDC) in Sioux Falls, SD (Eidenshink, 1992). This data set was chosen because it is relatively inexpensive, reliable, and is updated in near real-time. NDVI is defined by the formula (NIR-RED)/(NIR +RED), where NIR is reflectance in the near-infrared spectrum (0.75 – 1.10 Am) and RED is reflectance in the red band of the visible spectrum (0.58 – 0.68 Am). Chlorophyll uses electromagnetic energy in the RED band for photosynthesis, and plant structure is reflective of energy in the NIR band. So, for vegetated surfaces, NDVI increases if plant biomass increases or if photosynthetic activity increases.
The NDVI data were received in unsigned 8-bit integer format, with the original NDVI range [1,1] linearly scaled to the integer range 0 –200. For analysis purposes, the integer data were rescaled to their native range of [1,1]. As a consequence of the limited precision of the original 8-bit data, the precision of the rescaled data is 0.01, so there is an implicit expected numerical error of 0.005 in the pixel-level NDVI values.
The NDVI data set is not without uncertainty, both temporal and spatial. From 1989 – 2000, two polar orbiting National Oceanic and Atmospheric Administration (NOAA) satellites (NOAA-11 [1989 – 1994] and NOAA-14 [1995 – 2000]) carried the AVHRR sensors that collected the data comprising our data set. The U.S. annually invariant target curve, with NDVI from periods 5– 21 (February 26 –October 21) shown for each year. This curve represents the average time series of nearly 3500 pixels selectively sampled from the 48 states in the conterminous U.S. to possess highly regular annual periodicity, thus exposing any artificial interannual NDVI value drift (Kastens et al., 2003). The NDVI data originating from NOAA-11 are fairly consistent over time. The data from NOAA-14 are less so, exhibiting a large artificial oscillation from 1997 to 2000. The range of the trend curve from 1989 to 2000 has width 0.0464. Comparing this width to
the overall effective range of the AVHRR NDVI data being used (which is approximately [0.05,0.95] for the full U.S. terrestrial range, but narrower in most practical situations), it follows that nearly 5% of the effective AVHRR NDVI data range can be attributed to artificial interannual NDVI value drift. In retrospect, we know that sensor orbit decay and sensor calibration degradation were the primary sources of the interannual NDVI value drift found in the NOAA-14 data.
Image resolution (1 km2 /pixel, or 100 ha/pixel) of the AVHRR NDVI data is also an issue because pixel size is more  than twice as large as the typical field size for soybeans and major grains in the U.S., which is roughly 40 ha (Kastens & Dhuyvetter, 2002). Furthermore, when considering spatial error of the image registration performed during the NDVI compositing process, the area of the region from which a single pixel’s values can be obtained grows to more than 4 km2, or more than 400 ha (Eidenshink, 1992). A combination of sensor factors (e.g., sensor stability, view angle, orbit integrity) and effects of image pre-processing and compositing induce this spatial variation.

The second data set is historical, final, state-level yield data, obtained from the USDA National Agricultural Statistics Service (NASS) through its publicly accessible website ( The database is updated annually for all crops, with each particular crop’s final regional yield estimates released well after harvest completion. Updates to the final regional yield estimates can occur up to 3 years after their initial release, but generally these changes are not large. No historical or expected error statistics for these estimates are published below the national spatial scale, but they are nonetheless accepted by the industry as the best widely available record for average regional crop yield in the U.S.

Description of crops, regions, and time periods under investigation
The crops and regions under investigation in the present research are corn and soybeans in the U.S. states of Iowa (IA) and Illinois (IL), winter wheat and grain sorghum in the state of Kansas (KS), and spring wheat and barley in the state of North Dakota (ND). Compared to other states, during 1989 –2000, Iowa ranked first in corn production (100.2 million mt/year; ‘‘mt’’=metric ton) and second in soybean production (26.9 million mt/year). Illinois ranked second in corn production (89.9 million mt/year) and first in soybean production (27.1 million mt/year). Kansas was the top-producing winter wheat state (25.5 million mt/year) and the top-producing grain sorghum state (14.2 million mt/year). North Dakota was the top-producer of both spring wheat (16.6 million mt/year) and barley (6.3 million mt/year).

For each crop, a six-period window of early to mid-season NDVI imagery is considered. The source data for these six biweekly composites span nearly 3 months of raw AVHRR imagery, corresponding to Julian biweekly periods 5 – 10 (approximately February 26 –May 20) for winter wheat, 9– 14 (approximately April 23 – July 15) for spring wheat and barley, and 11 –16 (approximately May 21–August 12) for corn, soybeans, and sorghum. Labeling the six biweekly periods 1 to 6, yields are modeled using data from periods 1– 4, 1 –5, and 1 – 6, with each of these three ranges providing a unique yield forecasting opportunity corresponding to a
different point in the growing season. To obtain the dates for the crop-specific ranges, the initial release dates of USDA NASS yield forecasts were considered. The first NDVI image generated after the release of the initial USDA state-level estimates for the season is assigned to period 6, which fixes periods 1– 5 as well. Initial release dates for USDA state level estimates are approximately May 11 for winter wheat, July 11 for spring wheat and barley, and August 11 for corn, soybeans, and grain sorghum. Hereafter, winter wheat will be classified as an early-season crop, spring wheat and barley as mid-season crops, and corn, soybeans, and grain sorghum as late-season crops. With this timing, forecasts generated at periods 4 and 5 for each crop are produced before any state-level USDA yield estimates are released.

Approaches to image masking in crop yield forecasting
The purpose of image masking in the context of crop yield forecasting is to identify subsets of a region’s pixels that lead to NDVI variable values that are optimal indicators of a particular crop’s final yield. A good image mask should capture the essence (i.e., salient features) of the present year’s growing season with respect to how the crop of interest is progressing. This growing season essence is a combination of climatic and terrestrial factors.
Crop-specific masking
In theory, the ideal approach to image masking for the purpose of crop yield forecasting would be to use crop-specific masking. This would allow one to consider only NDVI information pertaining to the crop of interest. However, when such masking is applied to multiple years of imagery, several difficulties are encountered. Principal among these is the widespread practice of crop rotation, which suggests that year-specific masks are needed rather than a single cropspecific mask that can be applied to all years. Regional trending in crop area (increase or decrease in the amount of a region’s area planted to a particular crop over time), if severe
enough, also may call for year-specific masking. Identifying a particular crop in the year to be forecasted presents even greater difficulties, as only incomplete growing season NDVI information is available. This is especially true early in the season when the crop has low biomass and does not produce a large
NDVI response. In addition to hindering crop classification, this low NDVI response of a crop early in its development also  stifles crop yield modeling efforts, as AVHRR NDVI measurements from pixels corresponding to immature crops are not very sensitive and are thus minimally informative (see Wardlow et al., in press, for an example of such insensitivity occurring with 250-m Moderate Resolution Imaging Spectroradiometer [MODIS] NDVI data, and MODIS has better radiometric resolution than AVHRR). Moreover, with the coarse-resolution (about 100 hectares/ pixel) AVHRR NDVI imagery used in this study, identifying monocropped pixels becomes an improbable task. This is particularly true in low-producing regions and in regions with sparse crop distribution. As noted, a single pixel covers an area well over twice the average field size, and when error of the image registration is considered, a pixel’s effective ground coverage can become more than 400 hectares/pixel, or roughly ten times the typical field size.

Cropland masking
A more feasible alternative to crop-specific masking is cropland masking, which refers to using pixels dominated by land in general agricultural crop production. Kastens (1998, 2000) and Lee et al. (1999) obtained some of their best yield modeling results using this approach. Rasmussen (1998) used a percent-cropland map to improve his yield modeling by splitting the data into two categories based on cropland density and building different models for the two classes. Maselli and Rembold (2001) used correlations between 13 years of monthly NDVI composites and 13- year series of national crop yields to estimate pixel-level cropland fraction for four Mediterranean African countries. Upon application of these derived cropland masks, the authors found improved relationships between NDVI and
final estimated yield.
Cropland masks usually are derived from existing land use/land cover maps (one exception being Maselli and Rembold (2001)). If relatively small amounts of land in a study area have been taken out of or put into agricultural crop production during a study period, a single mask can be obtained and applied to all years of data. Considering that all traditional agricultural crops are now lumped in the general class of ‘‘cropland,’’ heavily cropped pixels are more prevalent in heavily cropped regions, which allows for the construction of well-populated masks dominated by cropland. But the generation of such masks becomes difficult when low-producing regions are encountered, as well as in regions where cropland is widely interspersed with non-cropland.
As with crop-specific masking, cropland masking also can suffer the effects of minimally informative NDVI response early in a crop’s growing season (e.g., March for early-season crops, May for mid-season crops, and late May and early June for late-season crops). Many important agricultural regions are almost completely dominated by single-season crop types (e.g., Iowa produces predominately late-season crops). In such cases, cropland AVHRR NDVI from time periods early in the
particular growing season may not be very useful for predicting final yield.

By late May and June, some of the year’s terrestrial and weather-based growth-limiting factors may have already been established for some crops or regions. For instance, in the U.S., soil moisture has largely been set by this time, and soil moisture is an important determinant of crop yields in the four states comprising our study area. Such moisture information is not readily detectable in immature crops because it is usually not a limiting factor until the plant’s water needs become
significant and its roots penetrate deeper into the soil. On the other hand, available soil moisture can noticeably affect other regional vegetation that is already well developed, such as grasslands, shrublands, and wooded areas, and in some cases, early- and mid-season crops.

Yield-correlation masking
For the reasons noted above, we propose a new masking technique, which we call yield-correlation masking. All vegetation in a region integrates the season’s cumulative growing conditions in some fashion and may be more indicative of a crop’s potential than the crop itself. Thus, all pixels are considered for use in crop yield prediction. This premise is most sound early in a crop growing season (especially for mid- and late-season crops), when the NDVI response of the immature crop is not yet strong enough to be a useful indicator of final yield. Also, as noted, when the crop is in early growth stages, problems such as lack of subsoil moisture may not yet have impacted the immature crop while having already affected more mature nearby vegetation.
Each NDVI-based variable captures a different aspect of the current growing season. This aspect manifests itself in different ways within the region’s vegetation, suggesting that optimal masks for the different NDVI-based variables likely will not be identical. Thus, for each (region, crop)-pair, yield-correlation masking generates a unique mask for each NDVI variable. The technique is initiated by correlating each of the historical, pixel-level NDVI variable values with the
region’s final yield history, a strategy similar to the initial step of the cropland classification strategy presented in Maselli and Rembold (2001). The highest correlating pixels,  thresholded so that some pre-specified number of pixels is included in the mask (this issue is addressed later), are retained for further processing and evaluation of the variable at hand.
Though much more computationally intensive, the yieldcorrelation masking technique overcomes the major problems afflicting crop-specific masking and cropland masking. Unlike these approaches, yield-correlation masking readily can be applied to low-producing regions and regions possessing sparse crop distribution. Also, since yield correlation masks are not constrained to include pixels dominated by cropland, they are not necessarily hindered by the weak and insensitive NDVI
responses exhibited by crops early in their respective growing seasons. Furthermore, once the issue of identifying optimal mask size (i.e., determining how many pixels should be  included in the masks) is addressed, the entire masking/ modeling procedure becomes completely objective.

Description of cropland masks
For Iowa, Illinois, and North Dakota, the cropland masks used in this study were derived from the United States Geological Survey (USGS) National Land Cover Database (NLCD) (Vogelmann et al., 2001). The original 30-m resolution land cover maps can be obtained from the website After generalizing the classes to cropland and non-cropland, the data were aggregated to a 1-km grid corresponding to the NDVI imagery used in this study. All annual crops, as well as alfalfa, were assigned to the cropland category, and all other cover types
were classified as non-cropland. Pixel values in the resulting  map corresponded to percent cropland within the 1-km2 footprint of the pixel.

(Source –

Read more

Crop Monitoring & Damage Assessment

Crop health evaluation, as well as early detection of crop infestations, is critical in ensuring high agricultural productivity. Stress associated with, for example, moisture deficiencies, insects, fungal and weed infestations, must be detected early enough to provide an opportunity for the farmer to mitigate. This process requires that remote sensing imagery be provided on a frequent basis (at a minimum, weekly) and be delivered to the farmer quickly, usually within 2 days.

Also, crops do not generally grow evenly across the field and consequently crop yield can vary greatly from one spot in the field to another. These growth differences may be a result of soil nutrient deficiencies or other forms of stress. Remote sensing allows the farmer to identify areas within a field which are experiencing difficulties, so that he can apply, for instance, the correct type and amount of fertilizer, pesticide or herbicide. Using this approach, the farmer not only improves the productivity from his land, but also reduces his farm input costs and minimizes environmental impacts.

There are many people involved in the trading, pricing, and selling of crops that never actually set foot in a field. They need data regarding crop health worldwide to set prices and to negotiate trade agreements. Many of these people rely on products such as a crop assessment index to compare growth rates and productivity between years and to see how well each country’s agricultural industry is producing. This type of information can also help target locations of future problems, for instance the famine in Ethiopia in the late 1980’s, caused by a significant drought which destroyed many crops. Identifying such areas facilitates in planning and directing humanitarian aid and relief efforts.

Remote sensing has a number of attributes that lend themselves to monitoring the health of crops. One advantage of optical (VIR) sensing is that it can see beyond the visible wavelengths into the infrared, where wavelengths are highly sensitive to crop vigour as well as crop stress and crop damage. Remote sensing imagery also gives the required spatial overview of the land. Recent advances in communication and technology allow a farmer to observe images of his fields and make timely decisions about managing the crops. Remote sensing can aid in identifying crops affected by conditions that are too dry or wet, affected by insect, weed or fungal infestations or weather related damage. Images can be obtained throughout the growing season to not only detect problems, but also to monitor the success of the treatment.

Healthy vegetation contains large quantities of chlorophyll, the substance that gives most vegetation its distinctive green colour. In referring to healthy crops, reflectance in the blue and red parts of the spectrum is low since chlorophyll absorbs this energy. In contrast, reflectance in the green and near-infrared spectral regions is high. Stressed or damaged crops experience a decrease in chlorophyll content and changes to the internal leaf structure. The reduction in chlorophyll content results in a decrease in reflectance in the green region and internal leaf damage results in a decrease in near-infrared reflectance. These reductions in green and infrared reflectance provide early detection of crop stress. Examining the ratio of reflected infrared to red wavelengths is an excellent measure of vegetation health.  This is the premise behind some vegetation indices, such as the normalized differential vegetation index (NDVI). Healthy plants have a high NDVI value because of their high reflectance of infrared light, and relatively low reflectance of red light. Phenology and vigour are the main factors in affecting NDVI. An excellent example is the difference between irrigated crops and non-irrigated land. The irrigated crops appear bright green in a real-colour simulated image. The darker areas are dry rangeland with minimal vegetation. In a CIR (colour infrared simulated) image, where infrared reflectance is displayed in red, the healthy vegetation appears bright red, while the rangeland remains quite low in reflectance.

Examining variations in crop growth within one field is possible. Areas of consistently healthy and vigorous crop would appear uniformly bright. Stressed vegetation would appear dark amongst the brighter, healthier crop areas. If the data is georeferenced, and if the farmer has a GPS (global position satellite) unit, he can find the exact area of the problem very quickly, by matching the coordinates of his location to that on the image.

Detecting damage and monitoring crop health requires high-resolution imagery and multispectral imaging capabilities. One of the most critical factors in making imagery useful to farmers is a quick turnaround time from data acquisition to distribution of crop information. Receiving an image that reflects crop conditions of two weeks earlier does not help real time management nor damage mitigation. Images are also required at specific times during the growing season, and on a frequent basis.

Remote sensing doesn’t replace the field work performed by farmers to monitor their fields, but it does direct them to the areas in need of immediate attention.

Canada vs. International

Efficient agricultural practices are a global concern, and other countries share many of the same requirements as Canada in terms of monitoring crop health by means of remote sensing. In many cases however, the scale of interest is smaller – smaller fields in Europe and Asia dictate higher resolution systems and smaller areal coverage. Canada, the USA, and Russia, amongst others, have more expansive areas devoted to agriculture, and have developed, or are in the process of developing crop information systems (see below). In this situation, regional coverage and lower resolution data (say: 1km) can be used. The lower resolution facilitates computer efficiency by minimizing storage space, processing efforts and memory requirements.

As an example of an international crop monitoring application, date palms are the prospective subject of an investigation to determine if remote sensing methods can detect damage from the red palm weevil in the Middle East. In the Arabian Peninsula, dates are extremely popular and date crops are one of the region’s most important agricultural products. Infestation by the weevil could quickly devastate the palm crops and swallow a commodity worth hundreds of millions of dollars. Remote sensing techniques will be used to examine the health of the date crops through spectral analysis of the vegetation. Infested areas appear yellow to the naked eye, and will show a smaller near infrared reflectance and a higher red reflectance on the remotely sensed image data than the healthy crop areas. Authorities are hoping to identify areas of infestation and provide measures to eradicate the weevil and save the remaining healthy crops.

 (Source –

Read more

Comparison of the effect of liquid humic fertilizers

Maize (Zea mays L.) is one of the most highly consumed  crops, and the most important foodstuff after wheat and  rice around the world. The global production of maize is 604 million tons, with a planting area of up to 140 million hectares. Iran produces 2 million tons of maize on 350000 hectares of land. However, the production from hybrid maize seeds in Iran is highly limited (FAO, 2002).

This plant, photosynthetically, is of C4 type and thrives in tropical and semitropical climates (Emam, 2008) and is native for central and southern America (Khodabandeh, 1998). Based on its role in production of grain and forage and providing food for livestock, as well as its industrial use, maize has become an important crop in Iran, as well as in other parts of the world. Expanding the area under  maize cultivation in Iran in order to become self-sufficient is one the most important goal pursued by the government and as a result of implementing programs designed to increase grain maize production over the last few years, this crop has seen a very fast growth in terms of planting area and yield.

Humic substances (HS) are the result of organic decomposition of the natural organic compounds comprising 50 to 90% of the organic matter of peat, lignites, sapropels, as well as of the non-living organic matter of soil and water ecosystems. Authors believe that humic substances can be useful for living creatures in developing organisms (as substrate material or food source, or by enzyme-like activity); as carrier of nutrition; as catalysts of biochemical reactions; and in antioxidant activity (Kulikova et al., 2005). Yang et al. (2004) argued that humic substances can both directly and indirectly

affect the physiological processes of plant growth. Soil organic matter is one of the important indices of soil fertility, since it interacts with many other components of the soil. Soil organic matter is a key component of land ecosystems and it is associated with the basic ecosystem processes for yield and structure(Pizzeghello et al., 2001).

Classically, humic substances are defined as a general group of heterogeneous organic materials which occur naturally and are characterized by yellow through dark colors with high molecular weight (Kulikova et al., 2005).  Shahryari et al. (2011) experienced the effect of two types of humic fertilizers (peat and leonardite derived) on germination and seedling growth of maize genotypes. They reported that interaction of “genotype × solutions (peat and leonardite based humic fertilizers and control) was significant in terms of the length of primary roots.

Application of leonardite based humic fertilizer had a remarkably more effect on relative root growth of Single Cross 794 and ZP 434 than other genotypes. In their experiment, the relation between germination rate and primary roots was positively significant under the condition of application of both types of humic fertilizers; but there was not the same relation for control treatment.

They argued that all types of various humic substances as a biological fertilizer can have an either similar or different effect in early growth stages of maize, as peat and leonardite based fertilizers that they applied produced more seedling roots than control, however the length of coleoptiles was higher in treatment with application of leonardite based humic fertilizer and control than treatment with application of peat based humic fertilizer. They believe that if used in lower quantity these natural fertilizers can have a lot of effect on plant growth.

Hence, in order to recognize how effective they might be, investigations should be considered based on various amounts of humic fertilizers. Finally, they suggested that both peat and leonardite based humic fertilizers could be used to stimulate growth of primary roots in maize which are critical for an optimal establishment of maize in the field.

Gadimov et al. (2009) claimed that humic substances are natural technological products with a miraculous biological effect on crops and concluded that a scientific and practical program is required to make use of this technology in the world, particularly in developing countries. Also, Shahryari et al. (2009) concluded that potassium humate is a miraculous natural material for increasing both quantity and quality of wheat and can be used to produce organic wheat. Thus, application of biological products such as humic fertilizers to provide nutrition for crops can be one of the useful methods to achieve some of the objects of organic crop production.

In addition, Shahryari et al. (2011) studied the response of various maize genotypes against chlorophyll content of the leaves at the presence of the two types of humic fertilizers. In their experiment, solutions (two types of peat and leonardite based liquid humic fertilizers and control) and interaction of “genotypes × solutions” produced significant difference at 1% probability level in terms of chlorophyll content of the leaves. Genotypes such as Single Cross 704 and 505 had the highest index for chlorophyll content when treated by leonardite based humic fertilizer. Peat based humic fertilizer decreased the index for chlorophyll content in genotypes such as 500, OS499 and 505, while leonardite based humic fertilizer decreased the index for chlorophyll content of the leaves in genotypes such as Golden West and Single Cross 704. However, peat based humic fertilizer did not have such an effect on these two maize genotypes.

Meanwhile, leonardite based humic fertilizer had no effect on index for chlorophyll content of leaves in genotypes such as 500, OS499 and 505. Genotypes such as ZP677 and ZP434 produced no response against the application of the two types of humic fertilizers. This study was aimed to compare the effect of liquid peat and leonardite based humic fertilizers on the yield of maize genotypes in Ardabil Region.


This experiment was conducted at Agriculture Research Station of Islamic Azad University, Ardabil Branch (5 km west of Ardabil City) in 2009 – 2010 cropping year. The region has a semiarid and cold climate, where the temperature during winter season usually drops below zero. This region is located 1350 m above the sea level with longitude and latitude being 48.2°E and 38.15°N, respectively.

Average annual minimum and maximum temperatures are -1.98and 15.18°C, respectively; whereas maximum absolute temperature is 21.8°C; and mean annual precipitation has been reported to be 310.9 mm. The soil of the field was alluvial clay with a pH ranging from 7.8 to 8.2.

Vegetative materials included six maize genotypes prepared from the Agriculture and Natural Resources Research Center of Ardabil Province. The Experiment was conducted as split plot in the basisof randomized complete block design with three replications. The main factor included three conditions (peat based humic fertilizer; leonardite based humic fertilizer; without the application of humic fertilizer) and the sub factor included six maize genotypes (ZP677, Golden west, OS499, ZP434, Ns540 and Single Cross 704). Each of experimental blocks included 3 plots, 320 cm length in rows, with80 cm from each other containing plants at 20 cm distances.

Pretreatment of seeds were done on the basis of 220 ml per 10 L of water to be applied for 1 ton of seeds. Moreover, irrigation was done in flooding manner. Weed-fighting was done both mechanically and manually during all growth stages. Liquid humic fertilizer was prepared and applied based on 400 ml per 50 L of water for 1 ha of maize plantation. The prepared solution was sprayed upon the aerial part of the plants during 5th leaf stage, appearance of reproductive organs, flowering and grain filling stages. All the samples were taken randomly from competitive plants at middle rows. Study traits included grain number per ear row, number of grain row per ear, ear number, weight of 1000 grains, biological yield, vegetative yield and grain yield.

Statistical analysis

Analysis of variance of data and mean comparison of them was done using MSTATC and SPSS programs. Mean comparison was done using Duncan’s multiple range test, at 5% probability level. Due to abnormality of data for ear number and its high coefficient of variation, square root conversion was used to normalize it.


Results from analysis of variance for study traits suggest that there was a significant difference  between experimental conditions in terms of grain yield and biological yield at 1 and 5% probability levels, respectively. In addition, there was a nonsignificant difference between study genotypes in terms of all evaluated traits except for number of grain row per ear and wet biomass at 1% probability level. Furthermore, there was no difference observed between the interaction of genotype and experimental conditions for any trait being studied, and this was in agreement with the report of Shahryari et al. (2009). This means that under study genotypes had the same responses to potassium humate.

Moreover, results from mean comparison of data (Table 2) for studied genotypes indicate that genotype OS499 (110.70 g) had the highest 1000 grain weight, whereas genotype Single Cross (81.20 g) had the lowest 1000 grain weight on average. Based on mean comparison of 1000 grain weight, genotypes OS499 and ZP434 were placed in the same group as NS540, whereas genotype ZP677 was placed in the same group as Golden West. Genotype ZP677 (with a mean value of 15.48) and genotype ZP434 (with a mean value of 13.49) had the highest and lowest values of number per ear, respectively; and genotypes such as Golden West and Single Cross were placed in  the same group as NS540 and had no difference in terms of this trait. Genotype ZP677 (with a mean value of 20.89 ton/ha) and genotype OS499 (with a mean value of 16.93 ton/ha) had the highest and lowest biological yield respectively and genotype OS499 was placed in the same group as ZP434, whereas genotypes such as Golden West and Single Cross were placed in the same group as NS540. Genotype ZP677 (with a mean value of 108.68 ton/ha) was the best among other genotypes in terms of wet biomass, whereas ZP434 (with a mean value of 77.52 ton/ha) had the lowest value for wet biomass. ZP677 was placed in the same group as NS540, whereas genotypes Golden West and OS499 were placed in the same group as ZP434 and had no difference in terms of this trait.

Shahryari and Shamsi (2009a) reported that potassium humate increased the rate of biological yield of wheat from 3.26 to 3.72 g/plant; but it had no effect on harvest index. Also, they found that uses of potassium humate increased grain yield. Results from mean comparison of data  for experimental conditions being studied indicate that application of leonardite based liquid humic fertilizer produced the highest biological yield(21.99 ton/ha on average), whereas no application of humic fertilizer produced the lowest biological yield(14.97 ton/ha on average). In this respect, both types of applied humic fertilizers had similar effects. Application of leonardite based liquid humic fertilizer produced the highest grain yield (7.09 ton/ha on average) among the conditions being studied, whereas under the condition of without humic fertilizer obtained the lowest value(4.07 ton/ha).

Ayas and Gulser (2005) reported that humic acid leads to increased growth and height and subsequently increased biological yield through increasing nitrogen content of the plant. It has also been reported that application of humic acid in nutritional solution led to increased content ofnitrogen within aerial parts and growth of shoots and root of maize (Tan, 2003). In another investigation, the application of humic acid led to increased phosphorus and nitrogen content of bent grass plant and increased the accumulation of dry materials (Mackowiak et al.,2001). Humic acid leads to increased plant yield through positive physiological effects such as impact on metabolism of plant cells and increasing the

concentration of leaf chlorophyll (Naderi et al., 2002).

Also, spraying humic acid on wheat crop increased its yield by 24% (Delfine et al., 2002). In general, the results from this study indicate that the application of leonardite based humic fertilizer increased biological yield by 46.89% compared to control, whereas peat based humic fertilizer increased biological yield by 34.47% compared to control. Seyedbagheri (2008)evaluated commercial humic acid products derived from lignite and leonardite in different cropping systems from 1990 to 2008. The results of those evaluations differed as a result of the source, concentration, processing, quality, types of soils and cropping systems. Under their research, crop yield increased from a minimum of 9.4%to a maximum of 35.8%. However, application of humic fertilizer in this study increased the biological yield by 40.68% on average. Application of leonardite based humic fertilizer increased the grain yield of maize by 74%.

Also, peat based humic fertilizer increased the grain yield by 44.7%. Overall, the mean increase for grain yield under the condition of application of humic fertilizers was as high as 59.45%. Similar results were also presented by Shahryari et al. (2009b) on wheat. They reported increase of grain yield (by 45%) from 2.49 ton/ha to 3.61 ton/ha affected potassium humate derived from sapropel in normal irrigation conditions.


Results from this experiment indicate that the application of liquid humic fertilizer can positively affect the maize yield and some agronomic traits related to it. These desirable effects can be a consequence of its effect on the physiology of the maize. In general, application of humic acid can lessen the need for chemical fertilizers and subsequently reduce environmental pollution, and compared with other chemical and biological fertilizers, they are affordable. Finally, it can be said that application of humic fertilizer not only increases the yield of maize, but also can play a significant role in achieving the goals of sustainable agriculture

( Source

Read more

Potential monitoring of crop production using a satellite-based Climate-Variability Impact Index

Crop monitoring and early yield assessment are important for agriculture planning and policy making atregional and national scales. Numerous crop growth simulation models are generated using crop state variables and climate variables at the crop/soil/atmosphere interfaces to get the pre-harvest information on crop yields. However, most of these models are limited to specific regions/periods due to significant spatial–temporal variations of those variables. Furthermore, the limited network of stations and incomplete climate data make crop monitoring and yield assessment a daunting task. In addition, the meteorological data may miss important variability in vegetation production, which highlights the need for quantification of vegetation changes directly when monitoring climate impacts upon vegetation. In this sense, remotely sensed metrics of vegetation activity have the following advantages: a unique vantage point, synoptic view, cost effectiveness, and a regular, repetitive view of nearly the entire Earth’s surface, thereby making them potentially better suited for crop monitoring and yield estimation than conventional weather data. For instance, it has been shown that the application of remotely sensed data can provide more accurate crop acreage estimates at national/continental scales. Furthermore, numerous field measurements and theoretical studies have demonstrated the utility of remotely sensed data in studies on crop growth and production. These two applications suggest the feasibility of large-scale operational crop monitoring and yield estimation.

Empirical relationships between the remotely sensed data and crop production estimates have been developed for monitoring and forecasting purposes since the early 1980s. For instance, Colwell found a strong correlation between winter wheat grain yield and Landsat spectral data. However, these relationships did not hold when extended in space and time (Barnett and Thompson, 1983). Later, various other vegetation indices generated from Landsat data, such as the ratio of the reflectance at near infrared to red and the normalized difference vegetation index (NDVI) were used in yield estimation of sugarcane, wheat, and rice. The Landsat series have a spatial resolution of 30 m and can provide reflectance data from different spectral bands. However, these highresolution data require enormous processing effort, and  may not be applicable for surveys of large-area general crop conditions.

Vegetation indices derived from data from the Advanced Very High Resolution Radiometer (AVHRR) were also used for crop prediction, environmental monitoring, and drought monitoring/assessment. For example, found that millet yields in northern Burkina Faso are linearly correlated with the AVHRR NDVI integrated over the reproductive period. Similarly, Hochheim and Barber found that the accumulated AVHRR NDVI provided the most consistent estimates of spring wheat yield in western Canada. The Vegetation Condition Index (VCI) derived from AVHRR data is widely applied in real-time drought monitoring and is shown to provide quantitative estimation of drought density, duration, and effect on vegetation. The VCI can separate the short-term weather signals in the NDVI data from the long-term ecological signals. According to Domenikiotis , the empirical relationship between VCI and cotton yield in Greece are sensitive to crop condition well before the harvest and provide an indication of the final yield. Unfortunately, the AVHRR data are not ideally suited for vegetation monitoring.


1. The MODIS land-cover classification product identifies 17 classes of land cover in the International Geosphere–Biosphere Programme (IGBP) global vegetation classification scheme. This scheme includes 11 classes of natural vegetation, 3 classes of developed land, permanent snow or ice, barren or sparsely vegetated land, and water. The latest version of the IGBP land-cover map is used to distinguish croplands from the other biomes in this research.


The retrieval technique of the MODIS LAI algorithm is as follows. For each land pixel, given red and near infrared reflectance values, along with the sun and sensor-view angles and a biome-type designation, the algorithm uses model-generated look-up tables to identify likely LAI values corresponding to the input parameters. This radioactive transfer-based look-up is done for a suite of canopy structures and soil patterns that represent a range of expected natural conditions for the given biome type. The mean value of the LAI values found within this uncertainty range is taken as the final LAI retrieval value. In certain situations, if the algorithm fails to localize a solution either because of biome misclassification/mixtures, high uncertainties in input reflectance data or algorithm limitations, a backup algorithm is utilized to produce LAI values based upon the empirical relationship between NDVI and LAI (Myneni et al., 1997).

The latest version of MODIS global LAI from February 2000 to December 2004 was taken to characterize the crop activity in this study. The 8-day LAI products are distributed to the public from the Earth Observing System (EOS) Data Gateway Distributed Active Archive Center. The 8-day products also provide quality control variables for each LAI value that indicate its reliability. The monthly global product was composited across the 8-day products using only the LAI values with reliable quality. The monthly global products at 1-km resolution with Sinusoidal (SIN) projection are available at Boston University. In this paper, monthly LAI at 1-km resolution are used to generate our Climate-Variability Impact Index. As these will be compared with estimates of crop production reported at county/state-levels, the vegetation-based CVII fields were aggregated over the corresponding counties/states using the county bound arias 2001 map from the National Atlas of the United States.


AVHRR LAI is used as a substitute for the MODIS LAI to examine the temporal characteristics of vegetation activity over longer time periods. The AVHRR LAI is derived from the Global Inventory Modeling and Mapping Studies (GIMMS) NDVI produced by NASA GIMMS group. Monthly LAI from 1981 to 2002 at 0.258 were derived based on the empirical relationship between NDVI and LAI for different biomes. Literature works show that this empirical relationship might be different for the same biome at different locations. To eliminate this effect, models are generated for each pixel to calculate GIMMS LAI from GIMMS NDVI. The MODIS LAI and GIMMS NDVI overlapped from March 2000 to December 2002, which provides a basis for generating a piecewise linear relationship between these two products. Once the coefficients of the linear model are calculated, the whole range of GIMMS NDVI can be converted into GIMMS LAI, which is consistent with the MODIS products. Our preliminary results indicate a good agreement between GIMMS LAI and MODIS LAI at quarter degree resolution with less than 5% relative difference for each main biome (results not shown).


In this research, we also use model-generated estimates of Net Primary Production (NPP) from Nemani  as a predictor of crop production. This NPP is a monthly product from 1982 to 1999 at a spatial resolution of half degree. This global NPP product was generated as follows. GIMMS NDVI were first used to create LAI and FPAR with a 3D radiative transfer model and a land-cover map as described in Myneni. Then, NPP was estimated from a production efficiency model (PEM) using the following three components: the satellite-derived vegetation properties, daily climate data, and a biome specific look-up table of various model constants and variables. Further details can be found in Nemani et al. (2003).

5. Crop production

Crop production data from several sources are used in this research. We focus upon total production, as opposed to yield, for instance, because although the two are highly correlated with each other, total production is typically the parameter of interest for crop monitoring and yield prediction. In this paper, we will explicitly refer to ‘‘production’’ when discussing quantitative results, however for simple qualitative statements wesometimes retain the generic term ‘‘yield’’ as synonymous for ‘‘production’’. The country-level crop production from 1982 to 2000 in European countries is from FAOSTAT 2004 data set. The county-, district-, and state-level production data in United States are from the National Agricultural Statistics Service (NASS) at United States

Department of Agriculture (USDA) USDA provides two independent sets of county crop data: one is a census of agriculture, which is released every 5 years; the other one is annual county crop data, which is based on reports from samples. We used the annual crop estimates in this study. Due to the processing effort required for the fine resolution remotely sensed data, we studied two crops (corn and spring wheat) in two US states (Illinois and North Dakota) at county- and district-scales. At coarser scales, we expanded the regions to include Illinois (IL), Minnesota (MN), Michigan (MI), Iowa (IA), Indiana (IN), and Wisconsin (WI) for corn; to North Dakota (ND), Montana (MT), Minnesota (MN), and South Dakota (SD) for spring wheat; to Kansas (KS), Oklahoma (OK), Colorado (CO), and Nebraska (NE) for winter wheat. The county- and district-level data of Illinois and North Dakota are from 2000 to 2004; the state-level data are from 1982 to 1999.

(Source –

Read more

Vegetation Control: Profitable Investment

Agricultural complex has always been one of the most significant propelling forces of the Ukrainian economy. Agriculture provides for 8% of total GDP, covering about 71% of the territory and 17% of working population employment. Meanwhile, in comparison with their foreign peers, the performance indicators of domestic agrarian companies demonstrate disproportion in operating results. Thus, if we scrutinize developing countries with agricultural specialization, Ukraine will demonstrate the highest rate of territory used for crop production (0.71 hectares per citizen) with a relatively small contribution to the GDP (-3−8% to the other countries level).

This correlation is reflected in the grain crop yield, which ranks Ukraine far behind the majority of leading grain-producing countries. Extensive management methods have led to the degradation of the black soil and thus – to the increase in expenses required per unit of cultivated area, fertilizers in particular.

It is impossible to improve  productivity performance in Ukraine without a substantial growth in mineral fertilizers use. However, global (reduction of the discrepancy between demand and production capacity) and domestic (deficit of phosphate fertilizers, potash fertilizer production standstill, increasing natural gas price) factors lead to constant rise in fertilizers prices for Ukrainian farmers and as a result lead to the profitable investments.

In addition, it should be noted that prices for agricultural machinery, fuel and pesticides have significantly increased. Even despite the fact that crop prices in 2011 have risen on average by 15%, the increase in prices for machinery and other accompanying expenses was more substantial.

The human capital costs have changed in a similar manner: the average nominal wage of an agricultural worker has increased by 25.9%during the last year. Still the Ukrainian agriculture industry employs +5-10% more population as compared to the European countries, while every employee produces 2-5x times less quantity of the added value.

In 2011 the expenses for fertilizers, POL and wages constituted almost 40% of crop production cost, thus during only one year the price jump caused spending spree by 15% and even more for specific crops. Let alone the cost of spare parts and materials for the repair of machinery and buildings (+0.3% in the total cost structure) and the rising costs for seed grain. In other words, such a scenario allows to catch up with the price increase tendency, but arouses the necessity to renovate obsolete production capacities, agriculture machines fleet, which also become less affordable, especially in terms of “price-quality” ratio. For example, some items from leading international equipment producers which are available in Ukraine have raised in price by 10% and even more in comparison with the analogue machines from the CIS.

Thus the problem is to find internal sources of production cost reduction to compensate the cost increase, which is beyond the agricultural holding control (fuel and metal prices, fertilizers deficiency, etc.). Global agricultural products prices do not depend on production cost in Ukraine, so it is the agrarian’s business to reduce it by cultivation cost-cutting. For this reason, developed countries opt for the use of precision agriculture system based on computer analysis of remote crops sensing data (RSD). Some Ukrainian farms already use such agricultural technologies as geologic information systems (GIS) and global positioning system (GPS). But in such a limited format these technologies are rather used to control equipment fleet maintenance, fuel input rationality and adequate farm maps creation. In the course of our cooperation with the agricultural companies management, we have discovered that a maximum allowable innovation is considered to be the purchase of expensive foreign equipment, its GPS monitoring installation and the creation of interactive maps of soils of rather satisfying quality. But precision agriculture implies exactly the efficient usage of every single asset. Even a large fleet of tractors can not effectively cultivate the fields without additional instructions on problem areas, non-rational heavy fertilization can be harmful and interactive maps do not allow to understand the current field condition in a real-time mode. It proves to be a real problem for the large farms as they simply fail to control the vegetation on their fields, and thus to identify in time the causes of low crop yield in different regions.

The main condition for the high-efficient GPS and GIS deployment is close cooperation with the system of constant remote vegetation control of field crops. Altogether, this forms the organizational strategic units . This scheme enables to attract fewer workers to control vegetation, field works planning and maintenance of communication between individual units and subunits of agro-enterprises. The vegetation control system performs constant monitoring of agro-enterprises soils irrespective of the distance among the fields and of the crops planted. Upon the abnormal “spot” appearance on the field, the person in charge receives a message and the agronomist makes appropriate decisions regarding fertilization, irrigation or other cultivation arrangements. We have to admit that other methods of soil monitoring (driving around the fields, installation of special observing equipment on certain areas, taking soil pieces for laboratory analysis, etc.) are less informative but consume more time and funds. In addition, each of the observations is far more difficult to organize and to hold than to download all required current and historical data (with its automatic interpretation) from any computer connected to the Internet.

As the conducted research have revealed, the cost of the each service on average starts from $1.5 a year per hectare, depending on the total farm area the system maintains. At the same time, this service allows to save $3-5/ha and to make profit from efficiency performance increment (e.g. for winter wheat) starting from $13/ha. In other words, every invested dollar gives an opportunity to earn 18 times more by reducing costs and increasing the efficiency of crops cultivation. For instance, if 10 largest domestic agricultural holdings have a land domain of more than 150,000 ha each, the application of RSD satellite analysis will bring a profit measured in tens of millions US dollars.

Nowadays, only a few domestic companies in Ukraine render services of crop vegetation control. However this derives from a low demand due to conservatism of the most agrarians, their general aversion to high technologies and short period of presence of such services in the domestic market. On the other hand, internal trends in the agricultural sector indicate that the next steps of the businessmen in agricultural field will become optimization of assets and search for sources to improve their operation efficiency.

Tags: satellite, precise agriculture, GIS, GPS, crop, Ukraine, efficiency, wheat, sunflower, barley, legumes

Read more