ISPRS Journal of Photogrammetry and Remote Sensing 169 (2020) 180–194 Available online 24 September 2020 0924-2716/© 2020 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by Elsevier B.V. All rights reserved. Developing a machine learning based cotton yield estimation framework using multi-temporal UAS data Akash Ashapure a , Jinha Jung a, * , Anjin Chang b , Sungchan Oh a , Junho Yeom c , Murilo Maeda d , Andrea Maeda d , Nothabo Dube e , Juan Landivar e , Steve Hague f , Wayne Smith f a Purdue University, USA b Texas A&M University – Corpus Christi, USA c Gyeongsang National University, South Korea d Texas A&M AgriLife Extension, Lubbock, USA e Texas A&M AgriLife Research at Corpus Christi, USA f Texas A&M University, USA A R T I C L E INFO Keywords: Precision agriculture Cotton genotype selection UAS ANN ABSTRACT In this research a machine learning framework was developed for cotton yield estimation using multi-temporal remote sensing data collected from unmanned aircraft system (UAS). The proposed machine learning model was based on an artificial neural network (ANN) and used three types of crop features derived from UAS data to predict the yield, namely; multi-temporal features including canopy cover, canopy height, canopy volume, normalized difference vegetation index (NDVI), excessive greenness index (ExG); non-temporal features including cotton boll count, boll size and boll volume, and irrigation status as a qualitative feature. The model provided low residual values with predicted yield values close to the observed yield values (R 2 ~ 0.9). ANN model performance was compared with support vector regression (SVR) and random forest regression (RFR). Comparison results revealed that ANN model outperforms SVR and RFR. Additionally, redundant features were removed using correlation analysis, and an optimal subset of features was obtained that included canopy volume, ExG, boll count, boll volume and irrigation status. Moreover, the relative significance of each feature in the optimal input feature subset was determined using sensitivity analysis. It was found that canopy volume and ExG contributed around 50% towards the corresponding yield. Finally, further analysis was performed to investigate how early in the growing season the model can accurately predict yield. It was observed that even at 70 days after planting the model predicted yield with reasonable accuracy (R 2 of 0.72 over test set). This study revealed that UAS derived multi-temporal data along with non-temporal and qualitative data can be combined within a machine learning framework to provide a reliable estimation of crop yield and provide effective understanding for crop management. 1. Introduction With a total aerial coverage of approximately 6 million acres, cotton is amongst one of the leading cash crops in the state of Texas (Adhikari et al., 2017), which has led to an upsurge of the cotton breeding research in the state. One of the main objectives of cotton breeding research is to select genotypes suitable for specific environments. For example, in South Texas the focus is to develop cotton genotypes resistant to water stress due to the dry climate and high irrigation costs. Traditionally, cotton breeding research has focused on manual field-based evaluation approaches that require the entire cotton field to be harvested, and later the best performing genotypes are selected based on ranking the yield of the individual genotypes (Iqbal et al., 2008; Kazerani, 2012; Shaukat et al., 2013; Clement et al., 2014). As this process is cumbersome, the scale of the field experiment is constrained by the limited availability of resources. As a result, cotton breeding research is focusing on the development of automated genotype selection techniques which do not require the entire field to be harvested. Remote sensing-based crop yield estimation methods have the potential to help automate the genotype selection process. In the literature, satellite remote sensing data have been extensively utilized for crop yield estimation (Singh et al., 2002; Ferencz et al., 2004; Sayago and Bocco, 2018; Hunt et al., 2019; Meng * Corresponding author at: Lyles School of Civil Engineering, Purdue University, 550 W Stadium Ave, West Lafayette, IN 47907, USA. E-mail address: jinha@purdue.edu (J. Jung). Contents lists available at ScienceDirect ISPRS Journal of Photogrammetry and Remote Sensing journal homepage: www.elsevier.com/locate/isprsjprs https://doi.org/10.1016/j.isprsjprs.2020.09.015 Received 20 April 2020; Received in revised form 11 August 2020; Accepted 15 September 2020