| Project Title: | Advanced information systems
for improved soybean production |
| Project Leader: | Michael S. Cox, Department of
Plant and Soil Science, MAFES |
| Other Participants: | M. Alan Blaine, Department of
Plant and Soil Science, MCES David R. Shaw, Department of Plant and Soil Science, MAFES James Thomas, Department of Agricultural and Biological Engineering, MCES David Laughlin, Department of Agricultural Economics, MAFES |
A three-year study into the benefits of precision agriculture to soybean production was proposed. The objectives of this research were two fold. The first objective is to use a Global Positioning System (GPS) and a Geographical Information System (GIS) to gather, store, and manipulate site-specific, within-field management information (yield, fertility, physical characteristics, weed populations, etc.). The second objective is to analyze and use this information to develop and make site-specific management recommendations and measure their effect on yield.
In the fall of 1996 and spring of 1997 (the soil sampling season for the first year of this study), three producer fields were identified to participate in this study. These fields were chosen based on the accessibility of yield-monitor-equipped combines, management system, soil variability, and the willingness of the producer to cooperate in the study. The fields are located near Leland, Shaw, and Hernando, Mississippi. Management systems include two irrigated sites and one dryland site. Each field was divided into 1Ha cells. Geo referenced soil subsamples were collected from a 10 meter radius around the centerpoint of each cell. Subsamples were combined to represent one sample for each cell. Soils were analyzed for P, K, Ca, Mg, S, Zn, pH, percent organic matter (%OM), and cation exchange capacity (CEC). In 1998, the fields located near Leland and Shaw, MS were planted with soybean. The field located near Hernando, MS was planted with corn. Soybean and corn yield was monitored "on the go" with yield monitor equipped combines. Yield data was combined with soil data in a GIS to match sampling locations. Coefficients of variation were used to determine the variability of the soil factors. Pearson correlation coefficients and stepwise regression were used to determine the effect of the soil factors on yield.
In the Hernando field, CVs for the soil chemical and physical properties ranged from 0.04 (elevation) to 0.539 (Mg). Corn yield was highly variable (CV=0.63) probably due the drought conditions in this unirrigated field. Yield was positively correlated to Mg and CEC and negatively correlated with elevation. It is theorized that the correlations with CEC and elevation are representative of the hydrology of the field. Elevation and CEC were negatively correlated indicating that lower areas of the field had higher CEC. The combination of lower elevation and higher CEC would result in water being held in these areas longer than in areas with lower CEC and higher elevations. Since 1998 was a drought year, these relations significantly affected corn yield.
In the Leland field, CVs for the soil chemical and physical properties ranged from 0.04 (pH) to 0.37 (P). Coefficient of variation for the soybean yield in this field was 0.28. Yield was positively correlated with Mg, Na, and clay content and negatively correlated with P, Zn, and elevation. As with the corn yield in the Hernando field, we feel that the yield correlations with elevation and clay content are reflective of the hydrology of the field especially during the 1998 year even though this field was irrigated. Lower yields requiring less nutrient uptake can explain the negative correlation between yield and P and Zn. We believe the correlation between yield and Mg to be an anomaly. Stepwise backward regression resulted in the equation Yield = 225.07 +0.06 Mg - 10.25 Zn -16.68 Elevation -1.03 Clay (R2 = 0.53, p < 0.05).
In the Shaw field, CVs for the soil chemical and physical properties ranged from 0.07 (elevation) to 0.64 (Na). Soybean yield in this field had a CV of 0.27. This field is rotated with rice and precision leveled prior to each planting explaining the very low CV for elevation. Soybean yield in this field was negatively correlated with P and elevation. As with the other two fields, we believe the negative correlation with elevation is reflective of the field hydrology. The negative correlation with P is reflective of the lower growth of the plant requiring less nutrients and leaving more in the soil. Stepwise backward regression found the equation Yield = 36.44 + 0.03 Mg -0.30 P +4.70 Zn (R2 = 0.51, p-< 0.05).
During the proposal review process in 1997, we were asked to combine the above study with one proposed by D. Laughlin. This second study was conducted at the Black Belt Branch Experiment Station, Brooksville, MS. Prior to planting soybean in 1998, a 15 ha field was intensively soil sampled on a 0.4 ha grid. Soil samples were analyzed for pH, % OM, CEC, Ca, K, Mg, Na, and P. Variable rate technology for fertilizer application was not available, therefore recommendations from the participating soil scientist along with the Mississippi State Soil Testing Laboratory were followed. Six weeks after planting, natural pitted morningglory (Ipomoea lacunosa L.) populations within a 1-m2 area were counted on a 0.1 ha grid overlying the soil sampling grid. Kriging techniques were applied to the soil nutrient factors to obtain interpolated soil information at corresponding weed sampling locations. Results of weed population analysis with MSU-HERB indicated the presence of pitted morningglory, if present in a I -m2 sample, was above the economic threshold level. Therefore, a binary response variable was constructed with 1 meaning the weed is present and 2 meaning the weed is absent. The binary response variable and the soil nutrient data were analyzed using stepwise logistic regression analysis to construct a prediction model for pitted morningglory based on actual soil data on actual soil properties combined with interpolated soil data. Initial field mapping was done on transects corresponding to yield monitoring data collection utilizing GPS technology to record specific locations of each species and their respective populations. Mapping was done at all sites that are common for the other segments of this project. Several scouting methods were evaluated to determine the intensity of scouting necessary to develop accurate maps of weed species and populations. Weed population maps were developed from these data and integrated with soil data and integrated into a GIS database to correlate with variability in yield as measured by yield monitoring systems. Once accurate field maps were developed, herbicides were applied only in those areas requiring treatments on one-half the fields, thus reducing unnecessary pesticide inputs into the environment and decreasing pesticide input costs.
Soils:
In the south field, CVs for the soil chemical and physical properties ranged from 0.02 (Elevation) to 0.53 (Organic C). The CV for yield in this field was fairly high (0.42) indicating that yield was highly variable. However there were no correlations between yield and any of the soil properties measured. This indicates that, for this field, some other growth factor (possibly water, or weed growth) was influencing soybean yield.
In the east field, CVs ranged from 0.04 (Elevation) to 0.37 (Na). The yield CV in this field was high (0.51) indicating high variability. Yield was negatively correlated to pH and Ca. These correlations reflect the high amounts of CaCO3 in this field due to the marine deposited parent material. The deposits increase the pH to levels higher than optimum and resulting in possible micronutrient deficiencies. Iron deficiency symptoms were present throughout this field. Stepwise backward regression resulted in the equation Yield = -64.27 - 0.001 Ca +0.07 K +0.11 Na -9.97 CEC -0.98 Elevation (R2 = 0.68, p< 0.05).
In the north field, CVs ranged from 0.05 (Organic C) to 0.66 (Ca). Prior to 1998 this field had been in meadow hence leading to the extremely low variation in Organic C. Yield in this field was negatively correlated to Ca, CEC, and pH. As in the east field, these correlations are probably reflective of the soil's marine deposit parent material. The regression equation for this field was Yield = 20.36 - 0.03 Ca -0.04 K +6.32 CEC (R2= 0.61, p< 0.05).
Weeds:
Using the actual soil data the logistic regression model logit(p) = -0.36 - 1.82(pH) + 0.03(K) + 0. 10Na) was developed. The predicted probability (p) of a weed being present with certain soil pH, K, and Na values was then determined by p = elogit(p) / 1+elogit(p). A classification rule was then developed to predict weed presence if the predicted probability exceeds a predetermined threshold. Within a predicted probability range of 0.30 to 0.46, the logistic regression model correctly predicted pitted morningglory presence or absence at 76% of the sampled field locations. Under a given set of soil pH, K, and Na parameters, and within a predicted probability range of 0.30 to 0.46, this model will correctly predict pitted morningglory presence at least 84% of the time and correctly predict its absence at least 65% of the time. However, using the actual soil data combined with the interpolated values resulted in exclusion of the soil potassium concentration from the model. The logistic regression model formed was logit(p) = 15.39-2.48(pH) + 0.05(Na). Within a predicted probability range of 0.34 to 0.72, this second logistic regression model correctly predicted pitted morningglory presence or absence at 70% of the sampled field locations. Based on these results, kriging may prove to be a valuable intermediate step in the management of site-specific data.
Using a linear discriminate function for analysis the spectral reflectance patterns acquired from the August 11 images resulted in 12 out of 16 observations of pitted morningglory being classified by species with an error rate of 20%, entireleaf morningglory was classified 11 times correctly with an error rate of 35%, common cocklebur was classified correctly 6 times with an error rate of 63%, and sicklepod was classified correctly only 4 times with an error rate of 75%. When a cross-validation was utilized, 12, 10, 5 and 4 observations were classified correctly with error rates of 20%, 41%, 68%, and 75% for pitted morningglory, entireleaf morningglory, common cocklebur, and sicklepod, respectively. For the August 18 image, 13, 12, 11 and 8 observations were classified correctly with error rates of 19%, 30%, 27%, and 50% for sicklepod, entireleaf morningglory, pitted morningglory, and common cocklebur, respectively. Cross-validation resulted in 11 out of 16 observations correctly classified for entireleaf and pitted morningglory with error rates of 35% and 27%. Sicklepod was classified correctly 8 times with a 50% error, while common cocklebur was classified only 6 times with a 62% error rate. August 28 image resulted in 15 observations classified correctly for pitted morningglory with an error rate of only 6%, 11 observations for sicklepod were classified correctly with an error rate of 31%, 10 observation for entireleaf morningglory were classified correctly with an error rate of 38%, and only 9 of the common cocklebur were classified correctly with an error rate of 44%. When a cross-validation was utilized 13, 11, 9, and 8 observations were classified correctly with error rates of 19%, 31%, 43% and 50% for pitted morningglory, sicklepod, entireleaf morningglory and common cocklebur, respectively.
Studies were initiated in 1998 to determine the impact of sampling intensity on accuracy of weed population prediction in 3 soybean fields in the Mississippi Delta, and in 2 soybean fields at the Black Belt Branch Experiment Station. Fields were gridded on a 0.1-acre spacing, and 1 -m2samples were collected at each grid point. Data analysis underway includes the impact that varying grid spacing has on accuracy of predictions of populations. After sampling, the two fields at the Black Belt station were split, and one-half of each were treated based on a whole-field average weed population, the other half received site-specific postemergence herbicide treatments based on individual populations that were input into the MSU-HERB model. Analyses of results are underway.
|
Soybeans in
Mississippi |
|
| For information
about this page, contact OAC Webmaster
This site is made possible by the Mississippi Soybean Promotion Board and is maintained by the Office of Agricultural Communications at Mississippi State University. |