From Leaf to Landscape Spatio-temporal monitoring of crops to improve decision support
OPEN ACCESS
Loading...
Author / Producer
Date
2022
Publication Type
Doctoral Thesis
ETH Bibliography
yes
Citations
Altmetric
OPEN ACCESS
Data
Rights / License
Abstract
In this century, agriculture faces the challenge of providing food in sufficient quality and quantity to nearly 8 billion people. Improving the management of crops by means of smart farming can help to intensify plant production in a sustainable way. Commonly, smart farming encompasses techniques of data acquisition, data analysis and evaluation, and precision application technologies. Remote sensing plays a key role in smart farming. Remote sensing applications help to monitor crops in the field. Another way to monitor plants and their environment is to utilize environmental sensors. Calibrated plant growth models or phytopathology models can interpret this data. Together, these tools can support the decision making of a farmer. The proof of concept of the benefits of these technologies has long been accomplished. And yet, in Switzerland the adaption rate of these technologies is still relatively low. Key aspects that are relevant to determine the economic benefit of investing in smart farming technologies, are e.g. the field size and the field heterogeneity. This thesis therefore focused on the monitoring of crop traits with smart farming methods, such as remote sensing applications or measurements of environmental variables. A particular emphasis was placed on the effects of scales.
The first study presented in Chapter 2 reveals the importance of various environmental variables for crop growth on different temporal scales. The study was conducted in Oensingen and aimed to investigate the relationship between highly resolved in-situ growth measurements of winter wheat, meteorological variables and CO2 fluxes. Leaf growth was taken as a proxy for vegetative biomass development because it can be assessed non-destructively. The study shows that on an hourly scale, gross primary production (GPP) and shortwave radiation explain most of the variance of leaf elongation rate (LER), however on a daily scale, air temperature is the main driver. The strongest immediate relationship was found between LER and GPP and incoming shortwave radiation; these are variables that are determining photosynthesis. In principle, LER also shows the same diurnal patterns as air temperature and soil temperature, however air and soil temperature were lagging behind LER. Multivariate growth models show that combinations with GPP or incoming shortwave radiation and air temperature perform best. These results indicate that short term growth processes in young wheat leaves in the field are mainly controlled by incoming shortwave radiation, while the magnitude of growth is controlled by temperature. These results could be used to update crop growth models. The results further stress the dependence of plants on their environment and importance of measuring environmental variables.
The second study, which is shown in Chapter 3, aims to investigate the effect of the selected spatial scales on apparent field heterogeneity. This study was conducted in Eschikon, Switzerland, and encompassed data captured from 17 fields, with multiple crop species monitored at 36 dates across two seasons. In the study, very high spatial resolution orthomosaics of 0.5 m ground sampling distance (GSD) were compared to spatially lower-resolved orthomosaics of 5 m, 10 m and 20 m GSD. The findings experimentally confirm the theoretical assumption that management of fields with higher heterogeneity would profit from higher spatially resolved information. Furthermore, the study reveals that compared to the baseline, relative differences by means of the relative root mean square error varies between 0.0 and 0.3 for the normalized difference vegetation index (NDVI) and between 0.0 and 0.23 for the normalized difference red edge index (NDRE) with decreasing GSD. The effects are most pronounced in maize and least pronounced in low intensity grassland. The study further shows how in-field heterogeneity varies throughout the season, across vegetation indices and crop species, which emphasises the importance of monitoring schemes that are adapted for each species.
The third study, which is presented in Chapter 4, aims to analyse how remote sensing data can be used as a decision support tool in agriculture. Therefore, the study follows the full workflow from remote sensing data to application maps that are in line with farm traffic orientation. The study focusses on analysing the impact of spatial scale and classification algorithms on the subsequent fertilizer recommendation. The study reveals that the precision of fertilisation recommendation decreases from 1.8% at 0.5 m in the best case to 12.8% root-mean-square error (RMSE) at 20 m GSD in the worst case compared to the baseline of 0.5 m GSD with no aggregation into classes. In general, classifications with four classes provide higher accuracy than classifications with three classes. When the actual management width of 11.7 to 14.7 m is taken into account, the RMSE is comparable between the different GSDs. Moreover, the effects are more pronounced in the beginning of the season (March, April), when most management measures for wheat happen, compared to the situation in May. Finally, the analysis shows that up to 40% of the field area cannot be covered by a GSD of 20 m due to the comparably small field sizes, which highlights the need for higher resolution data even in situations in which a decreased GSD has a minor effect.
Overall, these studies show the importance of monitoring schemes that are adapted to local circumstances. While satellite imagery might be beneficial in terms of lower cost, a lower resolved image might miss out a large part of the field. Considering the recent advances of alternative farming methods such as strip cropping or spot farming, only highly resolved images can offer the information that is needed to monitor these crops cultivated in such cropping systems adequately.
Permanent link
Publication status
published
External links
Editor
Contributors
Examiner : Walter, Achim
Examiner : Aasen, Helge
Examiner : Bareth, Georg
Book title
Journal / series
Volume
Pages / Article No.
Publisher
ETH Zurich
Event
Edition / version
Methods
Software
Geographic location
Date collected
Date created
Subject
smart farming; Remote sensing; UAV; field heterogeneity; Nitrogen management
Organisational unit
03894 - Walter, Achim / Walter, Achim
Notes
This research is supported by the Swiss National Science Foundation (SNSF), within the framework of the National Research Programme “Sustainable Economy: resource-friendly, future-oriented, innovative” (NRP 73), in the InnoFarm project, Grant-N° 407340_172433.