Comparison of Machine Learning Regression Algorithms for Cotton Leaf Area Index Retrieval Using Sentinel-2 Spectral Bands

Abstract

Leaf area index (LAI) is a crucial crop biophysical parameter that has been widely used in a variety of fields. Five state-of-the-art machine learning regression algorithms (MLRAs), namely, artificial neural network (ANN), support vector regression (SVR), Gaussian process regression (GPR), random forest (RF) and gradient boosting regression tree (GBRT), have been used in the retrieval of cotton LAI with Sentinel-2 spectral bands. The performances of the five machine learning models are compared for better applications of MLRAs in remote sensing, since challenging problems remain in the selection of MLRAs for crop LAI retrieval, as well as the decision as to the optimal number for the training sample size and spectral bands to different MLRAs. A comprehensive evaluation was employed with respect to model accuracy, computational efficiency, sensitivity to training sample size and sensitivity to spectral bands. We conducted the comparison of five MLRAs in an agricultural area of Northwest China over three cotton seasons with the corresponding field campaigns for modeling and validation. Results show that the GBRT model outperforms the other models with respect to model accuracy in average ( R 2 ¯ = 0.854, R M S E ¯ = 0.674 and M A E ¯ = 0.456). SVR achieves the best performance in computational efficiency, which means it is fast to train, and to validate that it has great potentials to deliver near-real-time operational products for crop management. As for sensitivity to training sample size, GBRT behaves as the most robust model, and provides the best model accuracy on the average among the variations of training sample size, compared with other models ( R 2 ¯ = 0.884, R M S E ¯ = 0.615 and M A E ¯ = 0.452). Spectral bands sensitivity analysis with dCor (distance correlation), combined with the backward elimination approach, indicates that SVR, GPR and RF provide relatively robust performance to the spectral bands, while ANN outperforms the other models in terms of model accuracy on the average among the reduction of spectral bands ( R 2 ¯ = 0.881, R M S E ¯ = 0.625 and M A E ¯ = 0.480). A comprehensive evaluation indicates that GBRT is an appealing alternative for cotton LAI retrieval, except for its computational efficiency. Despite the different performance of the ML models, all models exhibited considerable potential for cotton LAI retrieval, which could offer accurate crop parameters information timely and accurately for crop fields management and agricultural production decisions.

Sentinel-2 Data for Precision Agriculture?—A UAV-Based Assessment

“…In contrast to the findings of Richter et al [ 22 ] and Verrelst et al [ 39 ], introducing additional bands did not enhance the GAI-prediction ( Table 3 ). This could be examined further, considering the full spectrum of Sentinel-2 bands and calibration methods—for example, machine learning regression algorithms [ 12 , 42 ]. However, other studies already demonstrated good results applying simple VI-approaches to Sentinel-2 data [ 11 , 22 ] and it should be kept in mind that a simple but still relative well working VI-approach is easier to communicate and has the advantage of lower download- and processing time [ 39 ].…”

“…Yet, the analysis revealed several deficiencies of the NIR/RE-model, raising the topics of temporal and spatial resolution. Apparently, the frequent statement that the GAI-course over the season can be well mapped with Sentinel-2 data [ 3 , 11 , 12 , 21 , 29 , 37 , 40 , 41 , 42 , 43 , 44 , 45 ] is based on large differences and continuously increasing GAI-values between different sampling dates. Yet, the Sentinel-2 based single-date GAI-estimations are at some dates systematically biased and the true GAI-variation is in most cases considerably underestimated ( Figure 4 ).…”

“…This raises the question why so far, as to say based on the literature research carried out [ 3 , 11 , 12 , 21 , 29 , 37 , 40 , 41 , 42 , 43 , 44 , 45 ], the impact of date-specific effects when working with Sentinel-2 data is not further discussed. The common practice of distributing the sampling dates over different BBCH stages (e.g., [ 11 , 29 ]) and combining data of different locations, years and cultivars [ 40 , 41 , 43 , 44 , 45 ] might have disguised this problem so far—thereby, even if a certain shift between different satellite data acquisitions is detected, it cannot be clearly distinguished from other reasons, such as a starting VI-saturation at high GAIs, different VI-GAI-correlations at differing BBCH stages, or altering soil background.…”

Evaluation of Machine Learning predictions of a highly resolved time series of Chlorophyll-a concentration

“…In this study, to validate the performance of the ML models, the dataset was split in 80% (n = 3940) for model training, and 20% (n = 980) for model testing, so we could investigate the model generalization ability [32]. To eliminate the dimensional differences of the data and also improve the prediction ability of the models, we used the StandardScaler method from the Scikit-Learn package, which standardizes features by removing the mean and scaling to unit…”

Special Issue on Machine Learning Techniques Applied to Geoscience Information System and Remote Sensing

“…Mao et al [17] in their paper entitled "Comparison of Machine Learning Regression Algorithms for Cotton Leaf Area Index Retrieval Using Sentinel-2 Spectral Bands" compared the algorithm performance of five advanced machine learning regression algorithms, including ANN, support vector regression (SVR), Gaussian process regression (GPR), RF, and gradient boosting regression tree (GBRT), to retrieve cotton leaf area index (LAI) in a relatively comprehensive manner. Although the five models showed different performance, all of the models showed a potential for cotton LAI retrieval.…”