Site-specific seeding using proximal and remote sensors data fusion

Traditional agronomic management applies a uniform rate seeding (URS) density throughout the field assuming that the entire field is equally productive. This is a misconception since most agricultural soils are spatially and temporally heterogeneous, and hence a unique seed rate never be optimal throughout a field. Consequently, the URS allows to grow an improper number of plant populations in differently fertile zones in a field, which raises inter-crop competitions and reduces crop yield and thus production profit. Planting density also is linked with the application rate of other farming inputs, and therefore, the non-optimal seeding rates may increase production costs associated with fertilizer and pesticide application, and affect the soil-water environment. Ideally, in-field heterogeneity should be managed properly by implementing precision farming technology. Site-specific seeding (SSS) is such a precision agricultural operation that can ensure an optimized application of seeding rate for each fertility zone by considering the in-field soil variations. The aim of this study is to develop an optimized SSS approach based on the fusion of multiple soil and crop quality indicators estimated with state-of-the-art proximal and remote sensing technologies for maximizing crop yield and production margin. In the literature, there are two methods for implementing SSS i.e., map-based and sensor-based. The map-based SSS system allows adjusting seeding rates depending on the management zone (MZ) map delineated with reference soil and/or crop information. The sensor-based system adjusts seed rates based on an on-line measured soil fertility index (SFI), which is then used as input for real-time calculation and implementation of the recommended seeding rate. The SSS has been often practiced using map-based technology, frequently relying on yield or soil electrical conductivity (EC) data while ignoring many other important soil-related information. Hence, this study intended to move forward the map-based SSS using MZ map delineated with the fusion of multiple soil and crop data layers. It also focuses on identifying the key yield-limiting factors and thus proposing a set of proxies in the delineation of the MZ map accurately as well as in the development of an effective SFI. As the traditional laboratory-based analytical methods are slow and limited to low-resolution estimation of MZ proxies, this study intends to propose the best sensor and/or a combination of sensing technologies for estimating the key yield-limiting factors for delineating the MZ map. Moreover, the early works did not follow a systematic approach to optimize seed rate and they assigned the seed rate per MZ arbitrarily. Therefore, this research proposes an approach of SSS rate optimization. Most importantly, to date no research about sensor-based SSS is available, and hence this thesis will develop, for the first time, a sensor-based SSS technology. As a very limited studies evaluated SSS, the agronomic and economic benefits of SSS have not been largely explored. They showed inconsistent economic profits, which hinder the SSS adoption by end-users. Therefore, this study also focuses on evaluating the agronomic and economic performance of SSS in comparison with the URS for two major crops i.e., maize and potato. Multiple field experiments were conducted for evaluating the agronomic and economic benefits of map-based SSS for potato and maize in two cropping seasons. An on-line visible and near-infrared spectroscopy (vis-NIRS), electromagnetic induction (EMI) sensor, and Sentinel-2 image data were used for delineating three types of MZ maps, which are used for designing SSS treatments: 1)SSS based on apparent electrical conductivity (ECa) measured with an EMI sensor(EMI-SSS), 2) fusion of on-line vis-NIRS estimated soil properties with normalised difference vegetation index (NDVI) retrieved from Sentinel-2 based SSS (visNIRSen-SSS), and 3) another fusion-based SSS integrating data from EMI, on-line vis-NIRS, and Sentinel-2 retrieved NDVI data (EvisNIRSen-SSS). Five treatments of seed rate distributed over the MZ maps were tested to compare the performance of the “Kings” and “Robin Hood” methods for recommending the seed-to-seed spacing for both seed and consumption potatoes and seed rates for maize. The “Kings” method recommends the highest seeding rate for the highest fertile MZ and vice-versa while the “Robin Hood” follows the opposite principle by feeding most to the least fertile MZ. After all, the agro-economic performance of these five treatments was compared with those of URS. Results showed that the “Kings” approach outperformed the “Robin Hood” method both for potatoes and maize production. The SSS resulted in higher yield and gross margin by up to 597 € ha-1 in contrast to URS. The enhanced gross margin mainly emerged from increased crop yield instead of savings on the seed costs. The ECa-based SSS and the visNIRSen-SSS performed equally with a mild variation over the test sites. Since the EvisNIRSen-SSS revealed the highest gross margin for potato, the fusion of data obtained from EMI, vis-NIRS and Sentinel-2 seemed to be the most effective approach for MZ delineation in designing and implementing of the map-based SSS. Higher improvement in yield and the gross margin was observed in potato (380 to 597 € ha-1) than maize production (93 € ha-1). The highest profitability of SSS of up to 56.50 % was observed in the field with the lowest productivity. Despite increasing, crop yield and gross margin by SSS were statistically insignificant except in one test field with consumption potatoes. Since crop yields frequently illustrated high-positive correlations with soil pH, OC, P, K, Mg and MC, these were identified as key fertility indicators in developing the SFI that was used as input for the sensor-based SSS. They were also considered as the key yield-limiting factors and used as proxies for MZ delineation needed for map-based SSS. The on-line vis-NIRS sensor can estimate all these MZ proxies accurately with high-spatial resolution, hence is highly recommended for realizing the two methods of SSS. A fully automated sensor-based SSS system was developed and validated. It consists of hardware and software integrating an on-line vis-NIRS soil sensor, an on-line SFI prediction model, a seed rate calculation algorithm, and a variable rate planter machine. While the on-line soil sensor was set on the front end of a tractor, the maize planter actuates the seed rate according to the SFI measured by the on-line sensor. The system was tested in one field with silage maize when it revealed its potential for increasing yield by 1.4 t ha-1 and thus the gross margin by 91 € ha-1, compared to the URS. The study proved that the vis-NIRS sensor was the ideal sensing technology to assess soil fertility and MC for accurate calculations of seeding rate recommendations. In conclusion, SSS is found to be a promising precision agriculture solution to increase crop productivity and thus the economic margin by scientifically managing in-field heterogeneity through optimizing the input seeding rates according to the yield potential of different zones of a field. To succeed in SSS application, soil pH, OC, P, K, Mg and MC should be included in the MZ delineation for map-based SSS and in SFI determination for sensor-based application. This will allow for a correct seed rate calculation, which should then be implemented according to the “Kings” method (e.g., feeding the rich). It is highly suggested in future works to include soil clay and elevation data as MZ proxies and in the SFI derivation. The study shows interesting economic and agronomic benefits that could be harvested when SSS is adopted by farmers. It is therefore recommended to adopt SSS in the commercial farming system. This study has become a reference and thus opened a window for further development and implementation of SSS for other crops