Spatial variability in soil, crop, and topographic features, combined with temporal variability between seasons can result in variable annual yield patterns within a paddock. The complexity of interactions between yield-limiting factors such as soil nutrients and soil water require specialist statistical processing to be able to quantify variability, and thus inform crop management practices. This study uses multiple linear regression models, Cubist regression and feed-forward neural networks to predict spatial maize-grain (Zea mays) yield at two sites in the Waikato Region, New Zealand. The variables considered were: crop reflectance data from satellite imagery, soil electrical conductivity, soil organic matter, elevation, rainfall, temperature, solar radiation, and seeding density. This exercise explores methods which may be useful in predicting yield from proximal and remote sensed data with higher resolution than traditional low spatial resolution point sampling using soil testing and yield response curves.
6 January 2021 in New Zealand Journal of Crop and Horticultural Science
Precision agriculture manages within‐field spatial variability by applying suitable inputs at the appropriate time, place, and amount. Delineation of field‐specific management zones (MZs), representing significantly different yield potentials prescribe the rates of a specific crop inputs within‐field. This paper examines multiple‐year maize grain yield maps (2014, 2015, 2017 and 2018) and their spatial and temporal variability of within‐field datasets (soil electrical conductivity [EC], soil organic matter [OM], and elevation) and climate data. The research was undertaken on a non‐irrigated field at New Zealand’s Foundation for Arable Research (FAR) in the Waikato region, to provide a simple, heuristic method to delineate dynamic MZs for crop inputs. Supervised statistical learning models (stepwise multiple linear regression [SMLR], feedforward neural network (FFNN), classification and regression tree (CART), random forest (RF), extreme gradient boosting (XGBoost) and Cubist regression) were implemented to predict spatial yield. Prediction accuracies of the trained models were evaluated by withholding one subset of data for testing. For internal ‘split‐sample' validation, CART, random forest and XGBoost produced slightly better statistical predictions (RMSE = 1.9–2.0 and R2 = 0.60–0.63) than Cubist and FFNN (RMSE = 2.1–2.2 and R2 = 0.52–0.57), whereas MLR produced the weakest prediction (RMSE = 2.3 and R2 = 0.51). Spatial yield prediction of individual years, were poor ( R2 = 0.07–0.36). Input data used is readily and inexpensive for small arable fields in New Zealand. The methods presented, could be applied to a wider range of arable crops for within‐field management inputs, to respond to spatially diverse soil texture distribution and variable rainfall patterns.