Combined Use of Landsat 8 and Sentinel 2A Imagery for Improved Sugarcane Yield Estimation in Wonji-Shoa, Ethiopia

Abstract

In this study, a support vector regression (SVR) approach based on a radial basis function was used for estimating sugarcane yield in the Wonji-Shoa sugarcane plantation (Ethiopia) combining Landsat 8 (L8) and sentinel 2A (S2A) data. Vegetation Indices(VIs) involving visible, near-infrared, and shortwave infrared bands were calculated from the L8 and S2A sensor observations, and seasonal cumulative values were computed for the period June to October in the 9th month and June to November in the 10th month of the year for 2016/17 to 2018/19 cropping seasons. Sugarcane yield was predicted using the SVR, Multilayer perceptron neural network (MLPNN), and Multiple linear regression (MLR) methods. Then, a tenfold cross-validation approach was implemented for the performance evaluation. The results showed significant correlations between sugarcane yield and cumulative values of VIs computed during the 10th month in the growing season. The results also revealed that the estimation accuracy of sugarcane was better using the combined L8 and S2A (RMSE = 12.95 t/ha, and MAE = 10.14 t/ha) than using the S2A data alone (RMSE = 14.71 t/ha, and MAE = 12.18 t/ha). Comparing SVR results with MLPNN and MLR disclosed that SVR outperforms the other two models in terms of prediction accuracy. Overall, this study demonstrated the successful application of the SVR in developing a model for Sugarcane yield estimation and it may provide a guideline for improving the estimations of sugarcane in the study area.

Wheat Yield Prediction with Machine Learning based on MODIS and Landsat NDVI Data at Field Scale

“…In the study based on crop development periods, a good prediction accuracy was obtained with the SVM model. Obtaining results at similar levels in previous studies supported the argument that the SVM model is one of the appropriate methods for estimating crop yield(Joshi et al, 2021;Ju et al 2021;Abebe et al, 2022;Li et al, 2022).Fig. 8 Model performances by years in producer lands…”

Modelos para la estimación del rendimiento de la caña de azúcar en Costa Rica con datos de campo e índices de vegetación

“…En este estudio los índices de vegetación calculados en los meses de septiembre, diciembre y enero permitieron obtener los mejores modelos de estimación del rendimiento agrícola, confirmando que utilizar datos de un único momento fenológico no es suficiente para realizar estimaciones acertadas (Som-Ard et al, 2021). Durante estos meses el cultivo atraviesa las etapas de gran crecimiento e inicio de la maduración, resaltando la necesidad de disponer de información en la etapa de máxima cantidad de biomasa y en el punto donde sucede el descenso del área foliar por la maduración (Abebe et al, 2022).…”

“…En investigaciones de estimación del rendimiento agrícola a partir de índices de vegetación existen diversos ejemplos del uso del índice SR como un predictor debido a su elevada correlación positiva con el contenido de clorofila (Simões et al, 2005;Sishodia et al, 2020;Abebe et al, 2022). En el caso específico de las plantaciones de caña de azúcar, al estar la clorofila foliar estrechamente relacionada con el contenido de nitrógeno y este con la producción de biomasa, se esperaría que las variaciones en clorofila ocasionen cambios en la productividad, lo cual sería fácilmente identificado por los índices SR (Haboudane et al, 2002;Narmilan et al, 2022).…”

“…Por su parte la investigación realizada por Rao et al (2002) determinó correlaciones positivas entre el índice de vegetación NDVI y rendimiento en plantaciones de caña de azúcar delimitadas mediante técnicas de clasificación supervisada. Además, el aumento en la cantidad y la calidad de información que proveen los sensores satelitales ha permitido abordar la estimación de cosechas en caña de azúcar utilizando series temporales de diversos índices de vegetación (Morel et al, 2014;Dubey et al, 2018;Rahman y Robson, 2020;Abebe et al, 2022).…”

Estimation of Wheat Yield on a Farm in Najaf/Iraq, using Principal Component Analysis of Multi Temporal Satellite Images Pairs

“…It has become possible to produce dense time series images with high spectral and spatial resolution for the study of surface land cover and environmental monitoring; with the continuous progress in remote sensing technology, it has become possible to provide a large number of images with different spatial and temporal spectrums [4]. There are eleven spectral channels (bands) for the Landsat (8 and 9) satellites, with the 1 st seven (in addition to the 9 th ) having a spatial resolution of 30 m, the 8 th having a panchromatic resolution of 15 m, and the remaining having a spatial resolution of 100 m (9)(10)(11). Landsat 9 on September 27, 2021, launched into orbit, joining Landsat 8.…”