RESUMO
A combination tillage with disks, rippers, and roller baskets allows the loosening of compacted soils and the crumbling of soil clods. Statistical methods for evaluating the soil tilth quality of combination tillage are limited. Light Detection and Ranging (LiDAR) data and machine learning models (Random Forest (RF), Support Vector Machine (SVM), and Neural Network (NN)) are proposed to investigate roller basket pressure settings on soil tilth quality. Soil profiles were measured using LiDAR (stop and go and on-the-go) and RGB visual images from a Completely Randomized Design (CRD) tillage experiment on clay loam soil with treatments of roller basket down, roller basket up, and no-till in three replicates. Utilizing RF, SVM, and NN methods on the LiDAR data set identified median, mean, maximum, and standard deviation as the top features of importance variables that were statistically affected by the roller settings. Applying multivariate discriminatory analysis on the four statistical measures, three soil tilth classes were predicted with mean prediction rates of 77% (Roller-basket down), 64% (Roller-basket up), and 90% (No till). The LiDAR data analytics-inspired soil tilth classes correlated well with the RGB image discriminatory analysis. Soil tilth machine learning models were shown to be successful in classifying soil tilth with regard to onboard operator pressure control settings on the roller basket of the combination tillage implement.
RESUMO
Fine-textured clayey soils dominate Asian rice fields that are kept either fallow or cultivated with non-rice crops after harvest of monsoon rice. Use of seeding machinery compatible with the principles of conservation agriculture on such soils, however, has not been promising. Under these conditions - which predominate the population and poverty dense areas of coastal South Asia - such machinery fails to open a furrow or throws excessive soil out of the tilled furrow during strip-till seeding. This results in a poor seed coverage at planting jeopardizing crop establishment. In response, this soil bin study investigated strip-tillage blade designs and settings to optimize rotary strip-till system for wet clay soil conditions common in South Asian rice fields. Three designs of C type rotary blade (conventional, medium and straight) and two blade settings (four and six blades per row; 50 and 100 mm cutting widths) were tested at three blade operating depths (50, 75, and 100 mm) using a tillage test rig and a soil bin, and a high-speed camera to understand the processes of soil cutting, throwing, backfilling, and creation of furrow seedbed. The soil bin soil consisted of a wet sandy-clay-loam soil with a moisture content of 28.2% (85% of field capacity) and was compacted to the bulk density of 1440 kg m-3. Using the test rig, rotary speed of the blades was maintained at 480 rpm and forward speed at 0.4 m s-1. At four blades per row setting, all blades created high amounts of optimum clods (1-20 mm size). The conventional and medium blades threw too much soil out of the strip-tilled furrow while the straight blade created adequate backfill at 75 and 100 mm operating depths. At 6 blades per row setting, all blades produced high amounts of backfill at any depths, but the straight blade also produced the highest amounts of optimum clods and a uniform furrow. Considering machine and energy costs, blade performance, and the necessity of minimizing soil disturbance in strip-tillage, our study indicates that the use of straight blades (four blades per row) operated at a depth of 75 or 100 mm are more ideal. These specifications are likely to enhance strip-tillage stand establishment in fine-textured soils with high moisture contents, though further work is needed under actual field conditions to confirm suitability of the proposed strip-till system for crop establishment in currently fallowed as well as the intensively cropped lands of Asia.