% IMPORTANT: The following is UTF-8 encoded. This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.
@PHDTHESIS{Hao:992344,
author = {Hao, Shirui},
othercontributors = {Hendricks Franssen, Harrie-Jan and Ryu, Dongryeol and
Western, Andrew},
title = {{I}ntegrating remotely sensed information into the {APSIM}
wheat model to improve crop yield prediction},
school = {Rheinisch-Westfälische Technische Hochschule Aachen},
type = {Dissertation},
address = {Aachen},
publisher = {RWTH Aachen University},
reportid = {RWTH-2024-08168},
pages = {1 Online-Ressource : Illustrationen},
year = {2024},
note = {Veröffentlicht auf dem Publikationsserver der RWTH Aachen
University 2025. - Cotutelle-Dissertation; Dissertation,
Rheinisch-Westfälische Technische Hochschule Aachen, 2024},
abstract = {In agriculture, monitoring crop growth, and predicting crop
yield in a timely manner are of great importance. Crop yield
modelling and forecasting provide information to test
various crop management options, guide crop breeding,
understand and explore mitigation of environmental impacts,
and optimise production. Process-based crop models, such as
the Agricultural Production Systems sIMulator (APSIM), have
been widely applied to simulate crop growth and predict
yield because they incorporate the current understanding of
complex crop-environment dynamics. However, models simplify
complex processes and predict yield with uncertainty. Remote
sensing technology provides spatially distributed and
reliable quantitative estimation of crop status near
real-time, which can be integrated with crop models by using
data assimilation methods to mitigate prediction
uncertainties and improve predictive performance. This
thesis (1) reviews the performance of the APSIM-Wheat model
and identifies the important factors that influence yield
prediction uncertainty; (2) undertakes sensitivity analyses
using the Sobol’ method to examine the sensitivity of crop
yield prediction to six influential parameters identified in
part (1); (3) determines the most suitable APSIM-wheat state
variables for data assimilation and develops an observation
operator to facilitate model updating with remote sensing
observations; and (4) develops a data assimilation scheme to
integrate remotely sensed crop information into the
APSIM-Wheat model. The proposed scheme improves the wheat
yield estimation performance and enables the model to better
simulate spatial variability in yield. Chapter 2 reviews the
performance of APSIM-Wheat, one of the most popular crop
modules in APSIM, and identifies the factors that influence
yield prediction uncertainty. Model evaluation results from
76 published studies across thirteen countries on four
continents were analysed. In addition, a meta-database of
modelled and observed yields was established from 30 papers
within these studies. The analysis indicates that with
site-specific calibration, APSIM predicts yield with an RMSE
typically smaller than 1 t/ha under a wide range of
environments. For rainfed wheat, the review and
meta-analysis found that estimated soil hydraulic
characteristics, soil water conditions, nitrogen
availability, heat and frost events, and some other abiotic
stresses (lodging and root disease) lead to larger yield
prediction residuals and uncertainty. Integrating satellite
observations into APSIM-Wheat is hypothesised to improve the
yield prediction accuracy and robustness. As an important
steppingstone towards the satellite-APSIM integration, a
global sensitivity analysis (using the Sobol’ method) was
conducted to rank the sensitivity of influential parameters
affecting yield prediction (Chapter 3). Results indicate
that precipitation, initial soil nitrogen content, and soil
parameters have the largest influence on yield variability.
APSIM’s yield prediction becomes more sensitive to a
factor when that factor becomes more limiting. These results
are used to guide the design of data assimilation schemes
for crop models. Green leaf area index (GLAI) was selected
as the variable to assimilate into the APSIM-Wheat model.
Sentinel-2 was chosen to produce the observations due to its
high spatial (10-20 m) and temporal (5 days) resolutions and
its relevance to estimating GLAI. An observation operator,
which provides the mathematical mapping from model state to
observation space (or vice versa), was developed to link
several vegetation indices with APSIM GLAI in Chapter 4. The
results show that Sentinel-2 derived chlorophyll index (CI)
calculated using red edge bands have the closest
relationship with APSIM simulated GLAI. The uncertainty of
the observation operator was also determined and used to
represent the background prediction uncertainty in the
observation space. Data assimilation (DA) is a set of
statistical techniques that can be employed to combine
external information (such as in situ measurements or
remotely sensed observations) with a model to improve model
prediction performance. It enables the updating of model
time-step state variables using available information. In
Chapter 5, a synthetic data assimilation experiment was
conducted to test the updated state variables, the updating
periods, the updating intervals, and the uncertainties added
to the model and observations. It was found that updating
all biomass components in the wheat module (grain, leaf,
stem, spike, and root) from the whole duration of
sowing-harvest at a daily frequency resulted in the best
yield prediction performance. A more realistic 5-day
updating interval still resulted in noticeable improvement.
The designed data assimilation strategy was validated for
eight scenarios representing high-, medium-, and low-yield
cases. The results show that updating all biomass states
every 5 days across the whole growing season effectively
corrected yield prediction residual by $51\%$ - $85\%,$ with
the residual decreasing from 230 – 2134 kg/ha to 93 –
533 kg/ha. The standard deviation was also decreased by
$41.7\%$ – $66.7\%.$ After the synthetic experiment, the
data assimilation scheme was applied to a rainfed winter
wheat field located in north-western Victoria, Australia
using Sentinel-2 and Planet Scope observations (Chapter 6).
The field was segmented into 58 patches characterising yield
spatial variability. Two open loop cases were used to assess
the robustness of the data assimilation performance. The
results show that for the high-yield open loop case, data
assimilation corrected yield prediction residual by $37\%$
– $97\%,$ with a median correction efficiency of $73\%.$
The uncertainty was decreased from 0.71 t/ha to a range of
0.52 – 0.60 t/ha. For the low yield open loop case, the
residual was corrected by $18\%$ – $94\%,$ and a more
significant uncertainty reduction was achieved, decreasing
from 0.76 t/ha to between 0.29 and 0.40 t/ha. The developed
data assimilation framework for the APSIM-Wheat model shows
efficiency and robustness for improving model yield
estimation. The improvement provides the potential to
deliver quality yield predictions with credibility, enabling
better planning of management practices and optimisation of
food production. This work also provides a pipeline for the
design process of a crop model data assimilation framework,
which can serve as a guide for estimating model uncertainty,
choosing appropriate observations, estimating their errors,
and determining the state updating strategy.},
cin = {532820 / 530000},
ddc = {620},
cid = {$I:(DE-82)532820_20140620$ / $I:(DE-82)530000_20140620$},
typ = {PUB:(DE-HGF)11},
doi = {10.18154/RWTH-2024-08168},
url = {https://publications.rwth-aachen.de/record/992344},
}
Gast ::