Problem Statement. Case Studies on the Accuracy of Soil ph and Lime Requirement Maps. Automated Soil Testing. Direct Soil Measurement.

Transcription

1 The th ECPA (Prague, Czech Republic) July 13, 2011 Case Studies on the Accuracy of Soil and Lime Requirement s Viacheslav I. Adamchuk Bioresource Engineering, McGill University Allison K. Jonjak,, Charlie S. Wortmann,, Richard B. Ferguson, Charlie A. Shapiro University of Nebraska-Lincoln Problem Statement The assessment of soil variability is one of the most important steps in site-specific management Conventional means to attain soil variability data are incapable of accurately identifying spatial inconsistency within a production field at an economically feasible cost There is a need to develop equipment for mapping soil attributes on-the-go Automated Soil Testing Automated Soil ping Systems Shank Soil cutters Purdue University (West Lafayette, Indiana) 1 Coring tube Mixing ISFET Electrode Water jet ~0 g soil core Add 20 ml DI H 2 O Cleaning Sample collection for calibration Purdue University (West Lafayette, Indiana) 1 US Patent No.,3,30 ping Direct Soil Measurement RTK-level dual-system GNSS receiver Purdue University (West Lafayette, Indiana) Veris Technologies, Inc. (Salina, Kansas) UNL (Lincoln, Nebraska) Ion-selective Electrodes Water Nozzle Soil Sampler Soil mapping unit Galvanic contact apparent electrical conductivity mapping unit

2 Example Soil ping ping Alternatives Directed Soil Sampling On-the-Go Sensing Soil Sampling Soil s of a Kansas Universal Directed Sampling Average Interpolated Grid Adjusted Shifted Universal = Intercept Universal + Slope Universal On-the-Go ping Conventional 1 ha Grid Sampling Adjusted MSP Shifted MSP = Intercept specific + Slope = Shift specific + specific Soil s Evaluation Antimony Electrode 1-ha Grid Raw MSP Adjusted MSP R 2 = 0. R 2 = 0.2 OK1 1:1 line Grid Sampling R 2 = 0.0 R 2 = 0.0 OK1 1:1 line Unprocessed R 2 = 0.1 R 2 = 0.1 OK1 1:1 line Adjusted Average Universal Shifted OK1 1:1 line Average OK1 1:1 line Universal MSP OK1 1:1 line Shifted MSP Sandy and stony soils Soil Antimony Electrodes R 2 = 0. SE = Soil Glass Electrodes Portable Probe Objectives To generate lime requirement maps using different mapping approaches Average application rate 1-ha grid sampling On-the-go sensor mapping To compare these maps in terms of the accuracy of soil, buffer, and lime requirement predictions UNL 200

3 Experimental Sites Soil Distributions 2 Andersen 20 Number of Samples 1 0 Gross-Rhode Gross-Rhode Andersen Measurement Soil Data Shallow ECa Soil Data Shallow ECa Data Collection Sampling/measuring field field field Grid-based samples (1-ha) samples samples On-the-go soil measurements On-the-go EC a measurements Soil Data Shallow ECa

4 Grid-Based Sensor-Based Grid-Based Sensor-Based Lime Requirement 0 = (.0 ) if if if.0.3 < <.0.3, Mg/ha 2 Grid-Based Sensor-Based Lab Prediction for Alkaline Soils Predictions = a 0 +a 1 is soil in a 1:1 soil-water solution a 0 and a 1 are regression parameters =.0+a(-.) of.0 corresponds to of. = of.0 corresponds to of. of.0 corresponds to of.0 = b0 + b1 Sensor 1) b 0 = 0 and b 1 = 1 original data 2) b 0 0 and b 1 = 1 data shift 3) b 0 = 0 and b 1 1 data scale ) b 0 0 and b 1 1 linear regression

5 Lab Prediction RMSE ( Prediction) Lab = c + c Sensor + c EC + c Sensor EC a 1) c 1 = 1, c 0 = c 2 = c 3 = 0 raw data 2) c 0 0, c 1 =1, c 2 = c 3 = 0 data shift 3) c 0 0, c 1 1, c 2 = c 3 = 0 linear regression without ECa ) c 1 1, c 0 c 2 0, c 3 = 0 linear regression with EC a but without the product of sensor and EC a ) c 1 1, c 0 c 2 c 3 0 full regression 2 a Partial Case Data Set calibration validation calibration validation calibration validation b 0 = 0 and b 1 = 1 (raw data) 2 b 0 0 and b 1 = 1 (data shift) 3 b 0 = 0 and b 1 1 (data scale) b 0 0 and b 1 1 (linear regression) RMSE (Lab Prediction) Partial Case Data Set calibration validation calibration validation calibration validation c0 = 0, c1 = 1, c2 = 0, and c3 = 0 (raw data) c0 0, c1 = 1, c2 = 0, and c3 = 0 (data shift) c0 0, c1 1, c2 = 0, and c3 = 0 (sensor regression) c0 0, c1 1, c2 0, and c3 = 0 (sensor + ECa regression) c0 0, c1 1, c2 0, and c3 0 (full regression)

6 Coefficients of Determination Dataset Grid Sensor Grid Sensor Grid Sensor Differences in s Lab Mean Absolute Error s Sensor Grid Average Sensor Grid Average 0.2 a 0. a 0.1 a 0.2 b 0.30 b 0.1 ab 1.20 b 0.0 c 1.00 b 0.0 a 0.22 a 0.3 a 0.33 a a 0.03 a 0.0 a 0.0 b 1.00 b 0.02 a 0.00 a 0. a 0.0 a 0. a 0.0 a 0.30 a 0.20 a 0.2 a 1.00 ab 0.2 a 1.00 b 0.3 a 1.13 a 0.3 a a 0.3 a 1. a 0.1 a 0.2 b 0.0 b 0.22 a 0.3 a 0.31 a 2.2 a 2.13 a 2.0 a 0.0 a 0. a 0.3 a 1.2 a 2.2 a 2.0 a 1.1 a 1.22 a 1.0 a Conclusions Lime application maps based on sensor data with ten calibration points provided better delineation of acidic soil areas that needed lime than grid sampling or field average methods When defining a site-specific relationship between corresponding sensor and lab / measurements, it is not always necessary to adjust each parameter of a corresponding regression model E:mail: viacheslav.adamchuk@mcgill.ca