Corn Grain Yield Prediction Using UAV-based High Spatiotemporal Resolution Multispectral Imagery

Killeen, Patrick; Kiringa, Iluju; Yeap, Tet Hin

doi:10.1109/icdmw58026.2022.00137

Cited by 5 publications

(6 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For harvester speed and yield inlier removal, and for turn removal, we applied the forward-backward pass method proposed by Lyle et al [31]. Note that in our previous work [32] and in the present work, there was a parameter configuration error in the cleaning process, and as a result, the forward-backward pass method was effectively not applied correctly, meaning the yield datasets used in the experiments may have a few more outliers. We used the Vesper software version 1.6 to perform yield semivariogram and interpolation using the block kriging method with a block size 10 m × 10 m, an interpolation grid of 2.5 m × 2.5 m, and a local variogram with 30 lags, 50% lag tolerance, and a maximum distance of 55 m. We removed readings with high kriging variance.…”

Section: Yieldmentioning

confidence: 84%

“…We applied most of the yield cleaning steps mentioned in our previous work [32] by implementing the steps in Java version 18.0.1 (the project is open-source and can be found on GitHub https://github.com/patkilleen/geospatial, accessed on 23 January 2024). We did not remove samples from headlands due to small size of the fields.…”

Section: Yieldmentioning

confidence: 99%

“…Cleaning yield datasets is important before using them for analysis, since they tend to be noisy [30]. Our previous work [32] describes and identifies common yield cleaning steps applied in the literature. Mapping a yield dataset to a grid, the interpolation step, is commonly conducted after cleaning [14,33,34], and is usually performed using kriging and local variograms [35,36], which can be conducted using the Vesper software version 1.6 [14,37].…”

Section: Introductionmentioning

confidence: 99%

“…Model selection occurs when hyperparameters are being tuned and/or the optimal features are being selected, and this is typically conducted by training an ML model using training data and validating the performance using the validation dataset [49], p. 406.Hyperparameter tuning methodologies [43] and feature selection strategies [40] should also take the spatial structure into account. In fact, even though standard/random KF-CV is over-optimistic [44,50], most ML studies in the literature that use earth observation spatial data only apply KF-CV to evaluate their ML models [50], including our previous work [32]. We found that this trend is also observable in the UAV imagery-based yield prediction literature; only one paper (Baghdasaryan et al [6]) out of the 28 papers related to the present work (see Section 4) clearly performed spatial CV to evaluate model spatial generalizability.…”

Section: Introductionmentioning

confidence: 99%

“…The present work is an extension of a conference paper [32] and improves on the paper by: (a) using imagery from both the RGB and MS camera instead of only using MS camera imagery; (b) considering a larger dataset in the experiments; and (c) evaluating the spatial generalizability of the models by using spatial CV.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Corn Grain Yield Prediction Using UAV-Based High Spatiotemporal Resolution Imagery, Machine Learning, and Spatial Cross-Validation

Killeen,

Kiringa,

Yeap

et al. 2024

Remote Sensing

Self Cite

View full text Add to dashboard Cite

Food demand is expected to rise significantly by 2050 due to the increase in population; additionally, receding water levels, climate change, and a decrease in the amount of available arable land will threaten food production. To address these challenges and increase food security, input cost reductions and yield optimization can be accomplished using yield precision maps created by machine learning models; however, without considering the spatial structure of the data, the precision map’s accuracy evaluation assessment risks being over-optimistic, which may encourage poor decision making that can lead to negative economic impacts (e.g., lowered crop yields). In fact, most machine learning research involving spatial data, including the unmanned aerial vehicle (UAV) imagery-based yield prediction literature, ignore spatial structure and likely obtain over-optimistic results. The present work is a UAV imagery-based corn yield prediction study that analyzed the effects of image spatial and spectral resolution, image acquisition date, and model evaluation scheme on model performance. We used various spatial generalization evaluation methods, including spatial cross-validation (CV), to (a) identify over-optimistic models that overfit to the spatial structure found inside datasets and (b) estimate true model generalization performance. We compared and ranked the prediction power of 55 vegetation indices (VIs) and five spectral bands over a growing season. We gathered yield data and UAV-based multispectral (MS) and red-green-blue (RGB) imagery from a Canadian smart farm and trained random forest (RF) and linear regression (LR) models using 10-fold CV and spatial CV approaches. We found that imagery from the middle of the growing season produced the best results. RF and LR generally performed best with high and low spatial resolution data, respectively. MS imagery led to generally better performance than RGB imagery. Some of the best-performing VIs were simple ratio index(near-infrared and red-edge), normalized difference red-edge index, and normalized green index. We found that 10-fold CV coupled with spatial CV could be used to identify over-optimistic yield prediction models. When using high spatial resolution MS imagery, RF and LR obtained 0.81 and 0.56 correlation coefficient (CC), respectively, when using 10-fold CV, and obtained 0.39 and 0.41, respectively, when using a k-means-based spatial CV approach. Furthermore, when using only location features, RF and LR obtained an average CC of 1.00 and 0.49, respectively. This suggested that LR had better spatial generalizability than RF, and that RF was likely being over-optimistic and was overfitting to the spatial structure of the data.

show abstract

Section: Yieldmentioning

confidence: 84%

Section: Yieldmentioning

confidence: 99%