To Blend or Not to Blend? A Framework for Nationwide Landsat–MODIS Data Selection for Crop Yield Prediction

The Potential of Landsat NDVI Sequences to Explain Wheat Yield Variation in Fields in Western Australia

Long-term maps of within-field crop yield can help farmers understand how yield varies in time and space and optimise crop management. This study investigates the use of Landsat NDVI sequences for estimating wheat yields in fields in Western Australia (WA). By fitting statistical crop growth curves, identifying the timing and intensity of phenological events, the best single integrated NDVI metric in any year was used to estimate yield. The hypotheses were that: (1) yield estimation could be improved by incorporating additional information about sowing date or break of season in statistical curve fitting for phenology detection; (2) the integrated NDVI metrics derived from phenology detection can estimate yield with greater accuracy than the observed NDVI values at one or two time points only. We tested the hypotheses using one field (~235 ha) in the WA grain belt for training and another field (~143 ha) for testing. Integrated NDVI metrics were obtained using: (1) traditional curve fitting (SPD); (2) curve fitting that incorporates sowing date information (+SD); and (3) curve fitting that incorporates rainfall-based break of season information (+BOS). Yield estimation accuracy using integrated NDVI metrics was further compared to the results using a scalable crop yield mapper (SCYM) model. We found that: (1) relationships between integrated NDVI metrics using the three curve fitting models and yield varied from year to year; (2) overall, +SD marginally improved yield estimation (r = 0.81, RMSE = 0.56 tonnes/ha compared to r = 0.80, RMSE = 0.61 tonnes/ha using SPD), but +BOS did not show obvious improvement (r = 0.80, RMSE = 0.60 tonnes/ha); (3) use of integrated NDVI metrics was more accurate than SCYM (r = 0.70, RMSE = 0.62 tonnes/ha) on average and had higher spatial and yearly consistency with actual yield than using SCYM model. We conclude that sequences of Landsat NDVI have the potential for estimation of wheat yield variation in fields in WA but they need to be combined with additional sources of data to distinguish different relationships between integrated NDVI metrics and yield in different years and locations.

Detect, Consolidate, Delineate: Scalable Mapping of Field Boundaries Using Satellite Images

Digital agriculture services can greatly assist growers to monitor their fields and optimize their use throughout the growing season. Thus, knowing the exact location of fields and their boundaries is a prerequisite. Unlike property boundaries, which are recorded in local council or title records, field boundaries are not historically recorded. As a result, digital services currently ask their users to manually draw their field, which is time-consuming and creates disincentives. Here, we present a generalized method, hereafter referred to as DECODE (DEtect, COnsolidate, and DElinetate), that automatically extracts accurate field boundary data from satellite imagery using deep learning based on spatial, spectral, and temporal cues. We introduce a new convolutional neural network (FracTAL ResUNet) as well as two uncertainty metrics to characterize the confidence of the field detection and field delineation processes. We finally propose a new methodology to compare and summarize field-based accuracy metrics. To demonstrate the performance and scalability of our method, we extracted fields across the Australian grains zone with a pixel-based accuracy of 0.87 and a field-based accuracy of up to 0.88 depending on the metric. We also trained a model on data from South Africa instead of Australia and found it transferred well to unseen Australian landscapes. We conclude that the accuracy, scalability and transferability of DECODE shows that large-scale field boundary extraction based on deep learning has reached operational maturity. This opens the door to new agricultural services that provide routine, near-real time field-based analytics.