Evaluation of Random Forests (RF) for Regional and Local-Scale Wheat Yield Prediction in Southeast Australia

Abstract

Wheat accounts for more than 50% of Australia’s total grain production. The capability to generate accurate in-season yield predictions is important across all components of the agricultural value chain. The literature on wheat yield prediction has motivated the need for more novel works evaluating machine learning techniques such as random forests (RF) at multiple scales. This research applied a Random Forest Regression (RFR) technique to build regional and local-scale yield prediction models at the pixel level for three southeast Australian wheat-growing paddocks, each located in Victoria (VIC), New South Wales (NSW) and South Australia (SA) using 2018 yield maps from data supplied by collaborating farmers. Time-series Normalized Difference Vegetation Index (NDVI) data derived from Planet’s high spatio-temporal resolution imagery, meteorological variables and yield data were used to train, test and validate the models at pixel level using Python libraries for (a) regional-scale three-paddock composite and (b) individual paddocks. The composite region-wide RF model prediction for the three paddocks performed well (R2 = 0.86, RMSE = 0.18 t ha−1). RF models for individual paddocks in VIC (R2 = 0.89, RMSE = 0.15 t ha−1) and NSW (R2 = 0.87, RMSE = 0.07 t ha−1) performed well, but moderate performance was seen for SA (R2 = 0.45, RMSE = 0.25 t ha−1). Generally, high values were underpredicted and low values overpredicted. This study demonstrated the feasibility of applying RF modeling on satellite imagery and yielded ‘big data’ for regional as well as local-scale yield prediction.

Modelling, mapping and monitoring of forest cover changes, using support vector machine, kernel logistic regression and naive bayes tree models with optical remote sensing data

“…The results are consistent with those of [ 78 ]; who tested several classifiers for change monitoring and found that the SVM, which makes use of publicly available Landsat imagery, was the most effective. From multiple time-series of high-resolution remote-sensing images [ 79 ], implemented a change detection method using visual saliency and SVM. When it came to mapping forests, however, conventional classifiers actually had the opposite effect.…”

“…When compared to ML classifiers that used commercially accessible remote sensing data like LiDAR and WorldView-3, the accuracy of using publicly available Landsat images to recognize forests was on par [ [78] , [79] ]. SVM's total accuracy only reached 82%, despite the excellent spatial resolution of the LiDAR, WorldView, and Rapid Eye images [ 75 ].…”

“…With proper tuning, ML algorithms can improve classification quality and robustness [ 80 , 81 ]. Processing massive volumes of remote sensing data quickly and cheaply was made possible by the widespread availability and simple implementation of ML, as well as the free availability of Landsat images [ [78] , [79] ].…”

Methodologies Used in Remote Sensing Data Analysis and Remote Sensors for Precision Agriculture

“…High-quality research was published in this SI from researchers from various countries, including China, the USA, Slovenia, Spain, Germany, Brazil, Australia, and Singapore. The SI’s studies have been ordered following the application within the soil–plant–atmosphere continuum starting with the soil salinity precision monitoring using unmanned aerial vehicles (UAV) and multispectral imagery [ 1 ]; the evaluation of optical sensors for the diagnosis of nitrogen content for wheat plants [ 2 ]; the detection of root-knot nematode infestation in potato plants using hyperspectral imagery [ 3 ]; detection of powdery mildew using hyperspectral, thermal, and RGB imagery [ 4 ]; leaf area index estimations for wheat using hyperspectral reflectance data [ 5 ]; vineyard canopy characteristics and vigor assessment using UAV and satellite imagery [ 6 ]; estimation of crop vegetation parameters using satellite and UAV spectral remote sensing [ 7 ]; above-ground biomass estimation of oat plants using UAV remote sensing and machine learning [ 8 ]; wheat lodging estimation using multispectral UAV imagery and deep learning [ 9 ]; yield estimation for guinea grass using UAV remote sensing [ 10 ]; and wheat yield prediction from satellite imagery, meteorological data, and machine learning modeling [ 11 ].…”

“…Deep learning based on convolutional neural networks (CNN) with different architectures to analyze RGB from a UAV platform was used to estimate dry matter yield for guinea grass resulting in correlation coefficients of 0.79 < R < 0.62 [ 10 ]. Furthermore, RF algorithms were also used for wheat yield prediction based on satellite-based NDVI combined with meteorological data in Australia, resulting in 0.89 < R2 < 0.42 for different locations [ 11 ].…”