UAV and field spectrometer based remote sensing for maize phenotyping, varietal discrimination and yield forecasting.

Abstract

Maize is the major staple food crop in the majority of Sub-Saharan African (SSA) countries. However, production statistics (croplands and yields) are rarely measured, and where they are recorded, accuracy is poor because the statistics are updated through the farm survey method, which is error-prone and is time-consuming, and expensive. There is an urgent need to use affordable, accurate, timely, and readily accessible data collection and spatial analysis tools, including robust data extraction and processing techniques for precise yield forecasting for decision support and early warning systems. Meeting Africa’s rising food demand, which is driven by population growth and low productivity requires doubling the current production of major grain crops like maize by 2050. This requires innovative approaches and mechanisms that support accurate yield forecasting for early warning systems coupled with accelerated crop genetic improvement. Recent advances in remote sensing and geographical information system (GIS) have enabled detailed cropland mapping, spatial analysis of land suitability, crop type, and varietal discrimination, and ultimately grain yield forecasting in the developed world. However, although remote sensing and spatial analysis afforded us unprecedented opportunities for detailed data collection, their application in maize in Africa is still limited. In Africa, the challenge of crop yield forecasting using remote sensing is a daunting task because agriculture is highly fragmented, cropland is spatially heterogeneous, and cropping systems are highly diverse and mosaic. The dearth of data on the application of remote sensing and GIS in crop yield forecasting and land suitability analysis is not only worrying but catastrophic to food security monitoring and early warning systems in a continent burdened with chronic food shortages. Furthermore, accelerated crop genetic improvement to increase yield and achieve better adaptation to climate change is an issue of increasing urgency in order to satisfy the ever-increasing food demand. Recently, crop improvement programs are exploring the use of remotely sensed data that can be used cost-effectively for varietal evaluation and analysis in crop phenotyping, which currently remains a major bottleneck in crop genetic improvement. Yet studies on evaluation of maize varietal response to abiotic and biotic stresses found in the target production environments are limited. Therefore, the aim of this study was to model spatial land suitability for maize production using GIS and explore the potential use of field spectrometer and unmanned aerial vehicles (UAV) based remotely sensed data in maize varietal discrimination, high-throughput phenotyping, and yield prediction. Firstly, an overview of major remote-sensing platforms and their applicability to estimating maize grain yield in the African agricultural context, including research challenges was provided. Secondly, maize land suitability analysis using GIS and analytical hierarchical process (AHP) was performed in Zimbabwe. Finally, the utility of proximal and UAV-based remotely sensed data for maize phenotyping, varietal discrimination, and yield forecasting were explored. The results showed that the use of remote sensing data in estimating maize yield in the African agricultural systems is still limited and obtaining accurate and reliable maize yield estimates using remotely sensed data remains a challenge due to the highly fragmented and spatially heterogeneous nature of the cropping systems. Our results underscored the urgent need to use sensors with high spatial, temporal and spectral resolution, coupled with appropriate classification techniques and accurate ground truth data in estimating maize yield and its spatiotemporal dynamics in heterogeneous African agricultural landscapes for designing appropriate food security interventions. In addition, using modern spatial analysis tools is effective in assessing land suitability for targeting location-specific interventions and can serve as a decision support tool for policymakers and land-use planners regarding maize production and varietal placement. Discriminating maize varieties using remotely sensed data is crucial for crop monitoring, high throughput phenotyping, and yield forecasting. Using proximal sensing, our study showed that maize varietal discrimination is possible at certain phenological growth stages at the field level, which is crucial for yield forecasting and varietal phenotyping in crop improvement. In addition, the use of proximal remote sensing data with appropriate pre-processing algorithms such as auto scaling and generalized least squares weighting significantly improved the discrimination ability of partial least square discriminant analysis, and identify optimal spectral bands for maize varietal discrimination. Using proximal sensing was not only able to discriminate maize varieties but also identified the ideal phenological stage for varietal discrimination. Flowering and onset of senescence appeared to be the most ideal stages for accurate varietal discrimination using our data. In this study, we also demonstrated the potential use of UAV-based remotely sensed data in maize varietal phenotyping in crop improvement. Using multi-temporal UAV-derived multispectral data and Random Forest (RF) algorithm, our study identified not only the optimal bands and indices but also the ideal growth stage for accurate varietal phenotyping under maize streak virus (MSV) infection. The RF classifier selected green normalized difference vegetation index (GNDVI), green Chlorophyll Index (CIgreen), Red-edge Chlorophyll Index (CIred-edge), and the Red band as the most important variables for classification. The results demonstrated that spectral bands and vegetation indices measured at the vegetative stage are the most important for the classification of maize varietal response to MSV. Further analysis to predict MSV disease and grain yield using UAV-derived multispectral imaging data using multiple models showed that Red and NIR bands were frequently selected in most of the models that gave the highest prediction precision for grain yield. Combining the NIR band with Red band improved the explanatory power of the prediction models. This was also true with the selected indices. Thus, not all indices or bands measure the same aspect of biophysical parameters or crop productivity, and combining them increased the joint predictive power, consequently increased complementarity. Overall, the study has demonstrated the potential use of spatial analysis tools in land suitability analysis for maize production and the utility of remotely sensed data in maize varietal discrimination, phenotyping, and yield prediction. These results are useful for targeting location-specific interventions for varietal placement and integrating UAV-based high-throughput phenotyping systems in crop genetic improvement to address continental food security, especially as climate change accelerates.