[Link].

com/scientificreports

OPEN Dynamic graph learning framework

based seasonal and trend
decomposition approach for potato
crop evapotranspiration prediction
Saad Javed Cheema1,2, Mohammed Diykh1,2,10, Mumtaz Ali4, Aitazaz A. Farooque1,2,3,
Raheleh Malekian1,2, Shoaib Rashid Saleem5, Lakshman W. Galagedara6, Rehan Sadiq7,
Gurjit S. Randhawa8 & Qamar Uz Zaman9
Efficient estimation of crop water requirements (ETc) is important for sustainable agricultural
water management, particularly under increasing climate variability. Traditional methods lack a
comprehensive analysis of dynamic patterns associated with crop evapotranspiration factors. To
address these limitations, we propose a dynamic graph-based Dual-Graph Semantic Fusion (DG-DGSF)
for ETc estimation. The multivariate time series is decomposed into trend and seasonal parts. This
decomposition enables us to attain two dynamic graphs, Seasonal Dynamic Graph (SDG) and Trend
Dynamic Graph (TDG), with their semantic characteristics extracted through Dual-Graph Semantic
Fusion (DGSF). Each model is incorporated with the Dynamic Graph Learner (DGL) model and Graph
Convolutional based on Recurrent Unit (GC-GRU) to analyse the trend and seasonal components. The
DGL receives the trend or seasonal information to produce dynamic graphs, while GC-GRU combines
the dynamic graph characteristics with the original series data. To effectively combine and extract the
semantic characteristics from the trend and seasonal parts, a contrastive learning model is designed,
followed by a supervised prediction model based on a multi-layer perceptron. The proposed DG-DGSF
model was tested on data collected over two years (2023–2024) in Prince Edward Island, Canada. Three
experimental locations were selected within the research farm: Location 1 consisted of loam, Location
2 featured sandy loam, and Location 3 contained loamy sand. The DG-DGSF model is compared with
state-of-the-art models, including BiLSTM, GRU, GCN, BiGRU, LSTNet, DGDL, TPA-LSTM, and GCN-
LSTM. The performance of the DG-DGSF is evaluated using numerous visual, statistical and probability
metrics. The results demonstrated that the DG-DGSF model outperformed the benchmark models
with the lowest forecasting error and highest ETc prediction rates, RMSE = 0.0469, MAPE = 0.120,
NRMSE = 0.0431, KGE = 0.977, NSE = 0.963.

Keywords Potato crop water requirement, Dynamic graph, Contrastive Learning, MLP, Soil moisture
dynamics, Evapotranspiration coefficient, Trend, Season decomposition

Climate change is one of the most complicated global issues which greatly impacts ecological systems and raises
considerable in estimating crop water requirements1–3. The sustainability and availability of water sources have
been limited due to changing precipitation patterns, increasing temperatures, and extreme weather events4–6.
These climatic conditions affect key factors like evapotranspiration (ET, soil moisture, humidity, and crop
maturity stages, lowering predictable irrigation water applications in crop fields7,8. As a result, advanced data-

1School of Climate Change and Adaptation, University of Prince Edward Island, Charlottetown, PE, Canada.
2Canadian Centre for Climate Change and Adaptation, University of Prince Edward Island, Charlottetown, St Peters

Bay,, PE, Canada. 3Faculty of Sustainable Design Engineering, University of Prince Edward Island, Charlottetown,
PE, Canada. 4UniSQ College, University of Southern Queensland, Springfield Campus, QLD 4301, Australia. 5Faculty
of Agricultural Engineering and Technology, PMAS Arid Agriculture University Rawalpindi, Rawalpindi, Pakistan.
6School of Science and the Environment, Memorial University of Newfoundland, Corner Brook, NL, Canada. 7School

of Engineering, University of British Columbia, Kelowna, BC, Canada. 8School of Computer Science, University of
Guelph, Guelph, ON, Canada. 9Engineering Department, Faculty of Agriculture, Dalhousie University, Truro, NS,
Canada. 10Scientific Research Center, Al-Ayen Iraqi University, Thi-Qar, Iraq. email: [Link]@[Link];
afarooque@[Link]

Scientific Reports | (2025) 15:45732 | [Link] 1

[Link]/scientificreports/

driven approaches are essential for delivering timely and precise water use recommendations to support efficient
irrigation systems9. Consequently, there is a high demand for accurate daily ET prediction systems to optimize
real-time irrigation scheduling and provide resilient agricultural water management strategies10–13.
Potato cultivation has great importance in Prince Edward Island (PEI), Canada, as it is the leading agro
economy of the province and contributes significantly to the overall economy and rural livelihoods14–19. PEI
is well known for producing high-quality potatoes due to its mild climate, favourable soil properties, and
regular precipitation during the growing period20–25. However, recent climate change, including droughts and
rising temperatures has negatively affected potato yield and quality26. As a result, supplementary irrigation
has become important in this region for the optimum moisture level of the soil, which minimizes water stress
during key growth periods and guarantees tuber quantity and quality of potato tuber27,28. In this context,
accurately estimating the water demand of potato production systems is a challenging task to improve irrigation
practices, sustainable agriculture, and sustainable water resource management in PEI under varied climatic
conditions25,29–37.
Precise estimation of ETc and crop coefficients (Kc) is particularly important for high water-demand
crops like potatoes38,39. Although traditional models require extensive meteorological inputs, they struggle to
simulate the nonlinear and dynamic behaviour of agro-climatic systems40. Emerging ML and deep learning (DL)
techniques offer more adaptable and efficient alternatives41,42. Several studies have explored DL approaches
for ETc estimation43. For example, Nayak et al44. compared several machines learning algorithms, including
LSTM, SVR, and RF models using limited weather data, for daily reference evapotranspiration (ETc) estimation.
Authors that LSTM outperformed classic approaches, with an R2value of up to 0.98. Similarly, El-Kenawy et al45.
tested both LSTM and Gated Recurrent Unit (GRU) models. Hybrid models integrating Convolutional Neural
Networks (CNNs) with recurrent algorithms have also been employed widely46,47. For example, a CNN-GRU-
LSTM model was designed to predict water stress in Saudi Arabia48. They achieved an average of R2 = 0. 99.
In addition, graph-based learning approaches have been increasingly adopted to investigate complex
relationships in water management systems49,50. Graph neural networks (GNNs) were applied by51to predict
water shortage52,53. Employed graph-based deep learning approaches to predict crop ETc and Kc using time
series data.
Despite the promising results of these studies, they often failed to generalize across different soil types for
monitoring Etc which play a crucial role in water management systems54,55,Z.-Y2,3.. Figure 1 demonstrates the
motivation for utilising dynamic relation-based graphs for ETc prediction. Figure 1 depicts how the relationships
among variables such as soil moisture, temperature, and rainfall change over time, as shown in different graph
structures across time intervals. Traditional prediction approaches assume static correlations among time
series variables; however, those models fail to extract such temporal variations54–56. In contrast, dynamic graph
base methods can represent and learn dependencies, leading to more accurate prediction. This adaptability is
important factor for predicting ETc, where variables interactions shift due to seasonal and trend factors. As a
result, predicting future values of ETc effectively requires understanding and modelling dynamic relationships
among time series data.
As a result, developing a dynamic graph-based Dual-Graph Semantic Fusion (DG-DGSF) method can
effectively address these limitations. This proposed model is designed to represent trend and seasonal information
as dynamic graphs, enabling it to capture varying dependencies among time series variables. Additionally, it can
learn both long-term and short-term patterns influencing these variables, thereby improving the accuracy of ETc
predictions across different soil types. Such advancements will contribute to more efficient and precise irrigation
practices for potato cultivation.
In contrast to traditional models that rely on meteorological data or fixed crop coefficients, our
approach incorporates key hydrological and soil parameters, including soil texture, seasonal variability, field
capacity, wilting point, soil moisture, rainfall, irrigation, and leaching soil moisture storage change. The
proposed approach is designed to capture both long-term and short-term evapotranspiration patterns, allowing

Fig. 1. Modelling time series as dynamic graphs.

Scientific Reports | (2025) 15:45732 | [Link] 2

[Link]/scientificreports/

it to track changes in water consumption. Its ability to model the dynamic interactions between soil and plant
processes, along with fluctuating weather conditions, is expected to significantly improve the accuracy and
resolution of ETc predictions.
A comprehensive investigation is conducted to examine the performance of the DG-DGSF model. To the
best of our knowledge, this is the first attempt to integrate LOESS (locally estimated scatterplot smoothing),
dynamic graph, and contrastive graph learning for estimating crop water in the context of soil–plant-atmosphere
continuum, specifically for potato production. This study is relevant under the variable climatic and hydrological
conditions typical of regions such as Prince Edward Island. This study specifically aims to:

• Capture the complex spatial and dynamic relationships among key hydrological and soil parameters, includ-
ing soil texture type, field capacity, wilting point, soil moisture, rainfall, irrigation, periodic fluctuations within
the growing season, leaching, and change in soil moisture storage that influences potato crop water demand.
• Apply graph-based contrastive learning to enhance feature representation and inter-variable dependency
modelling for improved generalization under variable field conditions.
• Incorporate the STL (Seasonal and Trend decomposition using Loess) model LOESS to decompose the time
series data into trend and season parts and use them as inputs to the proposed DG-DGSF model.
• Design a DGL and a GC-GRU to analyse the trend and season parts.
• Suggest the DGSF model to integrate and fuse the trend and season dynamic graphs. The DGSF combined
spatial and temporal dependencies extracted from trend and season information. The DGSF is used to bridge
this information of trend and season by regulating and aligning this semantic information across the seasonal
trend information.
• Leverage contrastive learning to align semantic characteristics of dynamic graphs, which are then processed
through an MLP for prediction.

Materials and methods

Experimental setup
This study was conducted over two years (2023–2024) at the Canadian Centre for Climate Change and Adaptation
in St. Peter’s Bay, Prince Edward Island, Canada, as shown in Fig. 2. The experimental period spanned the
typical potato growing season, starting on June 1st and ending on October 8th each year. The region has a humid
continental climate and is widely recognized for its extensive potato cultivation. Three experimental locations
were selected within the research farm to ensure coverage of the primary soil textures found in Prince Edward
Island and maintain consistency in environmental conditions. Each site was designated a unique, representative
soil type: Location 1 consisted of loam, Location 2 featured sandy loam, and Location 3 contained loamy sand.

Fig. 2. Experimental location.

Scientific Reports | (2025) 15:45732 | [Link] 3

[Link]/scientificreports/

Three replicate lysimeter plots were built at each soil type, producing nine samples across all soil textures. This
structure was created to comprehensively compare soil–water interactions and crop water demands across
diverse soil conditions.

Lysimeter design and crop planting

Lysimeters are designed to mimic the field situation in a controlled environment so that exact measurement of
water movement related to crops can be taken57. The dimensions of each lysimeter utilized in the investigation
were 122 cm in length, 61 cm in breadth, and 61 cm in depth. The lysimeters were filled with soil and
planted with potatoes to resemble the natural field situation as much as possible and at the same time make
accurate measurements of all water inputs and outputs.
The Russet Burbank potato cultivar, a widely cultivated variety in Prince Edward Island, was chosen to be
planted in the lysimeters. Before planting, representative soil samples were collected from each soil type for
laboratory analysis at the PEI Analytical Services. These composite samples were analysed to ascertain the bulk
density, organic matter levels, and soil texture. The recognized soil classifications were loam, sandy loam, and
loamy sand, exhibiting corresponding bulk density values of 1.36, 1.44, and 1.68 g/cm3. The analysis revealed
specific ratios of sand, silt, and clay and organic matter content to enhance comprehension of the soil’s physical
characteristics. The site was prepared by digging the soil in each plot to a depth of roughly 15–20 cm to improve
drainage and establish optimal conditions for root and tuber development. Certified seed tubers were utilized
to mitigate disease risk and maintain uniform crop quality. Each seed piece was pruned to a weight of 45–60 g
and included a minimum of one discernible bud. Planting took place on June 1st, 2023. Each lysimeter unit is
comprised of a single row of three plants, with individual seed pieces placed 40 cm apart and a 30 cm buffer from
the lysimeter edge to mitigate boundary effects.
This configuration facilitated optimal tuber development and was established in collaboration with local
potato cultivators and agricultural consultants in PEI58. Pre-plant tillage was used to control the impact of
weeds, and then regular cultivation was conducted. A visual representation of crop development stages within
the lysimeter units is provided in Fig. 3. The required irrigation depth was first converted into an equivalent
water volume, which was then measured using graduated cylinders. As shown in Fig. 3, water was precisely
quantified before application. The measured volumes were subsequently delivered manually to each lysimeter
(Fig. 3), ensuring uniform distribution and consistency across replicates. This method allowed precise control
over irrigation amounts and provided high accuracy in water balance calculations.

Data collection and monitoring

A Spectrum Technologies WatchDog 1400 Micro Station was installed to monitor soil conditions and facilitate
precise irrigation planning for the Russet Burbank potato variety. The moisture levels at various soil strata were
captured by embedding three WaterScout SM100 soil moisture sensors manufactured by Spectrum Technologies
Inc. in each lysimeter at depths of 15 cm, 22.5 cm, and 30 cm. Before data collection, the FieldScout 350 TDR
system and the manufacturer’s recommended procedures were implemented to enhance the accuracy of
measurements. Volumetric soil water content and weather information were automatically recorded every 30
min to monitor changes that influence root development and plant water absorption. To inform daily irrigation
decisions, soil moisture data for all lysimeter plots were examined each morning at 9:30 AM. These parameters
facilitated the assessment of the need for irrigation on that day.
Irrigation time and volumes were estimated based on the soil moisture retention curves developed for each
soil texture utilized in the study, as described by26. The irrigation depth was modified to correspond with the
projected active root zone during the development cycle to provide efficient water distribution. The objective
was to maintain soil moisture around the field capacity without creating water stress or leaching. Observations

Fig. 3. Crop development phases within the lysimeter units, St. Peter’s Bay, Prince Edward Island, Canada.

Scientific Reports | (2025) 15:45732 | [Link] 4

[Link]/scientificreports/

of potato crop progression were conducted in a proximate field utilizing uniform planting and management
techniques. This facilitated precise monitoring of growth stages, encompassing seedling emergence, initiation of
tuber formation, and physiological maturity.

Estimation of reference evapotranspiration (ET0)

Reference evapotranspiration (ET₀) denotes the potential water loss from a properly irrigated grass surface
under ideal conditions and is a benchmark for estimating crop water requirements. ET₀ was estimated by the
FAO Penman–Monteith equation59 from data collected at two meteorological stations. To this end, an automated
weather station (iMETOS IMT300-USW) was deployed on-site for the duration of the study. This station
recorded a full set of the meteorological variables, including precipitation, air temperature, relative humidity,
solar radiation, wind speed, dew point temperature, vapour pressure deficit, wind direction and gust intensity.
All sensors displayed data, which was sent to the FieldClimate platform without interruption during the trial.
Additional daily weather data were collected from a nearby weather station in St. Peter’s Bay (coordinates:
46.45°N, 62.58°W) located 350 m away from the main experimental plots. For each year, these datasets were
employed in the computing of ET₀ on daily basis (using standard PM-56 calculator)59.
( 900
)
0.408 ∗ ∆ ∗ (Rn − G) + γ ∗ T +273
∗ U2 ∗ (es − ea )
ETo = (1)
∆ + γ ∗ (1 + 0.34 ∗ U2 )
[ ]
where Rn − G is the net balance of energy available at the surface MJ
mm2 d
, (es − ea ) indicates the vapour
pressure deficit of air at the reference height kPa , ∆ reflects the slope of the relationship between saturation
vapour pressure and temperature at the mean air temperature Kpa/o C , γ is the psychometric constant Kpa/oC,
and U2 is wind speed m/s at 2 m height.

Estimation of crop evapotranspiration (ETc)

ETc is the total amount of water that can be lost from the crop to the atmosphere by the combined processes
of soil evaporation and crop transpiration. This value varies according to the crop type, phenological stage,
and current weather conditions. In this study, ETc for potato was directly calculated from data obtained by the
lysimeter blocks, according to the following formula60,61:

ET c = I + R − DP ± ∆S (2)

where ETc = crop evapotranspiration (mm/day), I = irrigation applied (mm/day), R = rainfall received (mm/day),
DP = water lost through deep percolation (mm/day), ΔS = change in soil water storage (mm/day). The water
entering the lysimeter from rain or irrigation was closely monitored. The surplus water from precipitation or
excessive irrigation that extends beyond the root zone was collected in a container at the base of each lysimeter,
known as deep percolation. This approach enabled accurate plant water uptake assessment for different stages
of plant growth, soil types and climatic conditions.

Problem definition
The proposed model in this study aims to predict ETc. The problem of detection is defined as follows:
a multivariate time series is defined as X ∈ RN xM where T refers to the length of time series data, M is
the number of variables { irrigation, rainfall, water lost, change in soil water storage} in X. All variable M are
defined as G = (V, E, A), where V denotes the number of nodes, E is the number of edges, and A refers to the
adjacency matrix. Suppose each variable M is represented by s a sequence of datapoints, so we referred to the
values of all variables at time t as S t .
Over time, a sequence of graphs is created the evolving graph forms a sequence of graphs:
G = {G1 , G2 , ..., GT } . The learning mapping function is defined as
F ξ : G → ET C

Using the dynamic graphs graph deep learning, the model ξ learns from network behaviour to predict ET C .

Proposed framework for ETc prediction

Figure 4 depicts the proposed model for ETc prediction. The collected multivariate time series X is decomposed
into seasonal parts and trend parts Xst , XTt , using the STL model LOESS. This procedure of decomposition
enables us to attain two dynamic graphs, the Seasonal Graph and the Trend Graph, with their semantic
characteristics through DGSF. The proposed DGSF model is divided into three parts: trend dynamic graph and
seasonal dynamic graph, denoted as SDG and TDG, respectively, and contrastive learning based MLP. Each
part of SDG, and TDG includes two modules: DGL and GC-GRU. The DGL receives the trend or seasonal
information to produce dynamic graphs, while GC-GRU combines the dynamic graph characteristics with
the original series data. To effectively combine and extract the semantic characteristics from the trend and
seasonal time series data, a contrastive learning model is designed, followed by a supervised prediction model.
The contrastive learning is employed to align semantic characteristics, and then they are sent to an MLP for
prediction.

Time series decomposition

As the agriculture time series exhibits a high nonlinearity and complex behaviour, several machine learning
models fail to capture the hidden and influential characteristics. Consequently, in recent years, much work has

Scientific Reports | (2025) 15:45732 | [Link] 5

[Link]/scientificreports/

Seasonal dynamic graph (SDG) Trend dynamic graph (TDG)

Seasonal part DGL

DGL Trend part
1

Original data GC-GRU

DGSF 2

Seasonal part DGL

2

Original data GC-GRU GC-GRU Original part

Seasonal part DGL

1 DGL Trend part
Original data GC-GRU
DGSF
Seasonal part DGL

Original data GC-GRU GC-GRU Original part

Constructive learning
CPM

= + .
=[ || ]

MLP Predictor

Fig. 4. The proposed model for ETC prediction.

Scientific Reports | (2025) 15:45732 | [Link] 6

[Link]/scientificreports/

involved decomposition models to pre-process time series data. Those decomposition models can effectively
reveal the important and hidden characteristics of the data as well as improve the predictive accuracy of the
model. In the present study, we adopted a seasonal and trend decomposition model named STL model LOESS
(locally estimated scatterplot smoothing) to decompose the time series data into trends, seasonality, and
residuals62. For more details, the LOESS model is explained in Liu et al63.. We applied STL model to capture the
seasonal changes and long-term trends in potato crop evapotranspiration time series. Suppose a multivariate
time series is defined as X ∈ RN xM where T refers to the length of time series data, M is the number of
variables in X Each variable Y = {y1 , y2 , . . . yT } where T is the number of datapoints. The LOESS model
decomposes the y into trends, seasonality, and residuals.

Yt = Tt + St + Rt (31)

where: Yt = the original time series, Tt refers to the trend component, St denotes the seasonal component, Rt
is the residual/noise (random fluctuations). Figure 5 shows that Leaching, rainfall, irrigation, and change in
storage are decomposed into trend and season components. The data for the year 2023 is taken as an example.

DG-DGSF framework
In this section, a dynamic graph learner (DGL) is introduced to analyse trend and season parts. To process both
current time features and historical information, multi-head attention layer (MHAL) is employed in this paper.
The DGL model is derived as:

Ḧ t = M HAL (Q, F, V )(2)

( )
Q = F = V = concate ∗ H (t−1) , Γt (5)

where Ḧ t ∈ RN xh , h is the hidden size, V, Q, F are key, value, query matrices, Γt is either Xst , XTt season or
trend features extracted in Sect. 4.1.
The Ḧ t is further fused with node embedding N E (t−1) ∈ RN xe , where e represents the number of
embedding dimensions. A gated fusion approach proposed by64,65 is adopted in this paper to integrate the
spatial and temporal features. The gated fusion approach could learn important information from different data
sources. The gated fusion approach is applied as

Fig. 5. Time series is being decomposed into trend and season parts.

Scientific Reports | (2025) 15:45732 | [Link] 7

[Link]/scientificreports/

( )
rt = σ Wj Ḧ t + Uj N E (t−1) (6)
( )
zt = σ Wf Ḧ t + Uf N E (t−1) (7)

ct = tanh(Wc Ḧ t + Uc (rt ⊛N E (t−1) ))(8)

N E t = zt ⊙ rt + (1 − zt ) ⊙(9)

where σ denotes the Sigmoid activation function, ⊙ Hadamard product, z, r, c are the update gate, the reset gate,
and the memory cell.

Adjacency matrix producer

To construct an adjacency matrix for each dynamic graph, we followed the study in66. To obtain the Adjacency
matrix, the learnable weight tensor and node embedding are integrated using the following formula:
( )
M1 = tanh αN E t ⊛1 (10)
( )
M2 = tanh αN E t ⊛2 (11)
( )
Adj t = ReLU (tanh α(M1 M2T − M2 M1t (12)

where ⊛1 , ⊛1 are the training parameters, α is regulated parameter for ReLU

Dual-graph semantic fusion (DGSF)

In ETc Prediction, a short-term fluctuation due to rainfall, temperature changes, and long-term trends associated
with soil moisture depletion play significant roles. These characteristics evolve at different temporal scales;
however, they are inherently interdependent. To effectively extract this information, cross-scale dependencies,
we suggest DGSF to improve the prediction ability of the proposed model. The DGSF is employed to integrate
spatial and temporal dependencies extracted from trend graphs and season graphs. As we decomposed the
multivariate time series into seasonal and trend features and transferred them into dynamic graphs, the DGSF
is suggested to bridge this information by regulating and aligning this semantic information across the seasonal
trend information66. Firstly, the proposed DGSF identified nodes with global or local spatial information in each
graph by computing mutual information (MI) among aligned nodes’ embeddings over time,
The Adj t calculated value in the previous section reflects the correlation between the node vt [i] and other
nodes in the graphs. This matrix can be represented as a one-to-many relationship to depict the spatial semantics
of node v[i] at time t. As a result, each entity in Adj t is mapped as the rest (OvR) relationship to generate a vector
of OvR for each node. Then, the matrix AA is generated to represent OvR relationship of all nodes. Then, AA and
Adj t are concatenated to form spatial semantic state matrix (SSM ).
Ovr_matrix1 = σ(F CS (AA)(13)
( )
Ovr_matrix1 = σ(F CS Adj t (14)

SSM = concat (Ovrmatrix1 , Ovrmatrix2 )(15)

where F CS denotes the full connected layer, each entity in SSM [i] refers to the semantic relationship between
node i and other nodes. In ETc modelling, some variables, such as rainfall and irrigation, influence the spatial
behaviour of the demand for water over time. These impacts are not static; however, some nodes exhibit stable
behaviour.
In other words, the trend and seasonal information evolve separately; however, in fact, the seasonal patterns
and trend patterns interact. Learning them in isolation phases could miss these dependencies. To solve this issue,
we calculated the cosine similarity between these two semantic states SSM 1 , SSM 2 is calculated to measure
how each node’s spatial semantics change over time.
Nodes with the Top K highest similarities are considered to produce global spatial patterns, while the
remaining nonT opK nodes are assumed to exhibit local, time-sensitive behaviours. The mutual information is
calculated as follows:
n
T opK =
2 ∗ lin(n) (16)

M = T opK(cos(SSM 1season , SSM 2season ))(17)

Lmax = M I(cos(SSM 1season [M ], SSM 2trend [M ]))(18)

Lmin = M I(cos(SSM 1season [M ], SSM 2trend [M ]))(19)

where M refers to the index mask, M I computing approach, then the Lmax , andLmax are maximized and
minimized M I during the training phase. The MI measures the similarity between two variables by identifying
how much information is shared between them.

Scientific Reports | (2025) 15:45732 | [Link] 8

[Link]/scientificreports/

Graph convolutional based on GRU (GC-GRU)

Graph convolutional neural network models have been widely applied in time series prediction issues. Recently,
several studies have effectively integrated GCN with LSTM to improve the ability of GCN to capture and learn
complex representations of the multivariate time series data, leading to a more advanced prediction model
compared to using either LSTM or GCN. As temporal and spatial dependencies vary across different domains of
multivariate time series, and for better integration between temporal dependencies and spatial dependencies, in
this paper, the GCN is integrated with GRU67.

ϑ∗G (Hin ) = H k = βHin + (1 − β) AH (k−1) (20)

( ( ))
rt = σ(ϑr ∗G cocat X t , H t−1 (21)
( ( ))
zt = σ(ϑz cocat X t , H t−1
(22)
( ( ( )))
ct = tanh ϑc ∗G cocat X t , rt ⊛H t−1 (23)

H t = zt ⊛H t−1 + (1 − zt ) ⊛ct (24)

where ⊙ Hadamard product, ϑ∗G denotes the propagation layer, β refers to the hyperparameters which is used
to adjust the percentage of temporal information to the spatial information, σ is the Sigmoid activation function,
z, r, c are the update gate, the reset gate, and the memory cell.

Dual-view contrastive learner with MLP (CLM-MLP)

To improve the semantic interaction between SSM trend and SSM season characteristics, we designed a
contrastive learning model (CLM). This proposed CPM model aims at aligning the local and global semantics
characteristics extracted from the seasonal and trend graphs. The model receives the node embedding N Eseason
t

and N Etrend
t
extracted from the season graph and trend graph. This contrastive learning model keeps highly
impacted feature representations. The main formula of the CLM is:
let Xseason , and Xtrend represent the node embeddings N Eseasont
and N Etrend
t
obtained from SG and
TG
( )
Xseason = avP ool N Eseason
t
(25)
( )
Xtrend = avP ool N Etrend
t
(26)

These two features are aligned using CLM using Inf o_loss function
( coss(Xseason ,Xtrend ) )
exp
Constlearning = −log ∑ τ
( sim(Xseason ,Xtrend )
) (27)
j
exp τ

where coss is the cosine similarity, τ ∈ {0.1, 0.2, 0.5} denotes the hyperparameters, then the fused
representation F usX is defined:
F usX = [Xseason | |Xseason ](28)

The fused features are transferred into a prediction MLP to generate the final target value ẍt :

ẍt = M LP (F usX)(29)

The total training objective Ltotal is computed as the sum of the supervised losses and contrastive.
Ltotal = Lsup + ζ.Cons(30)
 
Lsup = ẍt − xt (31)

Model Development and optimisation

The proposed DG-DGSF model aims at ETc prediction which is defined as the total amount of water that
can be lost from the crop. The data was collected from the period of 2023 to 2024. The collected dataset was
carefully divided into the training set, which includes the data during the growing season from June 1, 2023, to
October 8, 2023, and the testing and validation sets, which contain the data from the growing season from June
1, 2023, to October 8, [Link] proposed model (DG-DGSF) was evaluated against GRU68, BiGRU67, BiLSTM
and graph-based methods including Graph Convolutional network (GCN) (He et al., 2020), GCN based LSTM
(GCN-LSTM)69, Dynamic Based Graph Deep Learning (DGDL)65 ), Long-Short-term Network (LSTNet)66,
TPA-LSTM66. All models were implemented using Python 3.13. All simulation results were conducted on a
computer with the following specification: Intel(R) Iris(R) Xe Graphics, CPU 12th Gen Intel(R) Core (TM)
i7-1260U, 1100 MHz, 10 Core(s), 12 Logical Processor(s). The model DG-DGSF development process includes
six phases, ensuring a comprehensive model and accurate prediction abilities based on two years data collected
from three fields.

Scientific Reports | (2025) 15:45732 | [Link] 9

[Link]/scientificreports/

• Phase 1: Data collection The data was collected over two years (2023–2024) at the Canadian Centre for Climate
Change and Adaptation in St. Peter’s Bay, Prince Edward Island, Canada. Figure 3 shows the data collection
process. Three experimental locations were chosen in this research. The three experimental locations were se-
lected within the research farm to ensure coverage of the primary soil textures found in Prince Edward Island
and maintain consistency in environmental conditions. Each location was designated a unique, representative
soil type: Location 1 consisted of loam, Location 2 featured sandy loam, and Location 3 contained loamy sand.
• Phase 2: Data decomposition- in this paper a seasonal and trend decomposition model called STL model
LOESS (locally estimated scatterplot smoothing) was employed to decompose the time series data into trends,
seasonality, and residuals components. The STL model LOESS can improve the performance of the predic-
tion model by exhibiting the nonlinearity and complex behaviours in crop time series data. Each variable in
the time series data was passed through the STL model LOESS model and three components were extracted
including trends, seasonality, and residuals representations. Figure 5 shows the decomposition process of
different variables.
• Phase 3: DG-DGSF model The DG-DGSF model was designed to analysis trend and season parts extracted
in phase 4 and predict ETc values. The trend and seasonal components were transferred into dynamic graph
(DG). As a result, two dynamic graph models were constructed named trend dynamic graph model and sea-
sonal dynamic graph model. Each model was included dynamic graph learner (DGL), and Graph Convolu-
tional based on GRU (GC-GRU) to extract representative information from trend and season components. To
combined dynamic graph representation extracted from trend dynamic graph and seasonal dynamic graphs,
we developed Dual-Graph Semantic Fusion (DGSF). The DGSF integrates spatial and temporal representa-
tions extracted from trend graphs and season graphs. The DGSF produced SSM trend and SSM season
characteristics to predict ETc.
• Phase 4: prediction phase: Dual-View Contrastive Learner with MLP (PCM-MLP) To improve the semantic
interaction between the extracted features in Phase 3, a contrastive learning model was suggested to merge the
SSM trend and SSM season characteristics, This contrastive learning model aims at aligning the local and
global semantics characteristics extracted from the seasonal and trend graphs. The model receives the node
embedding N Eseason
t
and N Etrend
t
extracted from the season graphs and trend graphs. Then it fused final
features into the prediction MLP model to generate the final target value ẍt :
• Phase 5: Training models: the proposed DG-DGSF model and all benchmark models were trained using the
time series data from June 1, 2023, to October 8, 2023, and they were tested and validated using the data from
the growing season from June 1, 2024, to October 8, 2024. Different metrics were used to evaluate the perfor-
mance of the proposed model against the benchmark models. In this paper, we ensured that all the models
were trained on a comprehensive set of data, integrating extensive historical data to make accurate predictions
about future ETc values.
• Phase 6: Optimisation and parameters selections All models parameters were selected carefully and optimized
during the validation phase. All model parameters were selected carefully and optimized during the training
phase. The hyperparameters of the proposed model and benchmark models were reported in Tables 1, and 2.
The baseline models were divided into two classes: classic models named GRU BiGRU, BiLSTM and graph-
based methods including Graph Convolutional network (GCN), GCN based LSTM (GCN-LSTM), Dynamic
Based Graph Deep Learning (DGDL), Long-Short-term Network (LSTNet), TPA-LSTM.

Evaluation metrics
We employed several probability and statistical metrics to evaluate the proposed model. according to the previous
studies, the following metrics were considered the most effective evaluation metrics: Root Mean Square Error
(RM SE ), Correlation coefficient (r), and Normalised RMSE (N RM SE ), Mean Absolute Percentage Error
(M AP E ), Normalised Root Mean Square Error (N RM SE ), Kling–Gupta Efficiency (KGE ), Nash–Sutcliffe
Efficiency (N SE ), Coefficient of Variation (CV ), Fractional Bias (F B ), Prediction Interval Normalized Average
Width metric (PINAWM) and Winkler Score (WS).

Hyperparameters Location 1 Location 2 Location 3

Learning rate 0.001 0.001 0.001
Batch size 32 64 32
Epochs 200 150 200
Hidden size 64 64 64
Weight decay rate 0.0001 0.001 0.0001
Dropout (MLP) 0.1–0.3 0.1–0.3 0.1–0.3
Optimiser Adam Adam RAdam
Constrictive weight loss 0.5 0.5 0.5
Temperature contractive τ 0 0.1 0.1
Over matrix dimension OvR σ 16 16 16

Table 1. Hyperparameters of the proposed DG-DGSF model.

Scientific Reports | (2025) 15:45732 | [Link] 10

[Link]/scientificreports/

Location Model Hidden Units

BiLSTM HU = 64, LR = 0.001, BS = 32, Dropout = 0.3, optimiser = Adam, epoch number = 100
BiGRU HU = 32, LR = 0.001, BS = 32, Dropout = 0.3, optimiser = Adam, epoch number = 100
GRU HU = 64, LR = 0.001, BS = 32, Dropout = 0.2, optimiser = Adam, epoch number = 80
GCN HU = 64, LR = 0.001, BS = 64, Dropout = 0.2, optimiser = Adam, epoch number = 150
Location 1
GCN-LSTM HL = 64, LR = 0.001, BS = 64, Dropout = 0.3, optimiser = Adam, epoch number = 120
LSTNet HU = 64, LR = 0.001, BS = 64, Dropout = 0.25, optimiser = Adam, epoch number = 150
DGDL HU = 64, LR = 0.0005, BS = 64, Dropout = 0.25, optimiser = Adam, epoch number = 150
TPA-LSTM HU = 64, LR = 0.001, BS = 64, Dropout = 0.3, optimiser = Adam, epoch number = 150
BiLSTM HU = 128, LR = 0.001, BS = 32, Dropout = 0.3, optimiser = Adam, epoch number = 100
BiGRU HU = 64, LR = 0.001, BS = 32, Dropout = 0.25, optimiser = Adam, epoch number = 100
GRU HU = 64, LR = 0.001, BS = 32, Dropout = 0.2, optimiser = Adam, epoch number = 80
GCN HU = 32, LR = 0.0005, BS = 64, Dropout = 0.2, optimiser = Adam, epoch number = 150
Location 1
GCN-LSTM HU = 64, LR = 0.001, BS = 64, Dropout = 0.3, optimiser = Adam, epoch number = 120
LSTNet HU = 64, LR = 0.001, BS = 64, Dropout = 0.2, optimiser = Adam, epoch number = 150
DGDL HU = 64, LR = 0.0005, BS = 64, Dropout = 0.25, optimiser = Adam, epoch number = 150
TPA-LSTM HL = 64, LR = 0.001, BS = 64, Dropout = 0.25, optimiser = Adam, epoch number = 150
BiLSTM HU = 128, LR = 0.0005, BS = 32, Dropout = 0.4, optimiser = Adam, epoch number = 100
BiGRU HU = 64, LR = 0.001, BS = 32, Dropout = 0.3, optimiser = Adam, epoch number = 100
GRU HU = 64, LR = 0.001, BS = 32, Dropout = 0.2, optimiser = Adam, epoch number = 80
GCN HU = 32, LR = 0.001, BS = 64, Dropout = 0.2, optimiser = Adam, epoch number = 150
Location 3
GCN-LSTM HU = 64, LR = 0.001, BS = 64, Dropout = 0.3, optimiser = Adam, epoch number = 120
LSTNet HU = 64, LR = 0.001, BS = 64, Dropout = 0.2, optimiser = Adam, epoch number = 150
DGDL HU = 64, LR = 0.0005, BS = 64, Dropout = 0.25, optimiser = Adam, epoch number = 150
TPA-LSTM HU = 64, LR = 0.001, BS = 64, Dropout = 0.3, optimiser = Adam, epoch number = 150

Table 2. Hyperparameters selection for three locations for the state-of-the-art models.

Experimental results
Numerical assessment
In this section, we quantify predictive performance using RMSE (mm·day⁻1), Root Mean Square Error
(RM SE ), Mean Absolute Percentage Error (M AP E ), Correlation coefficient (r), and Normalised RMSE
(N RM SE ). This study focuses on predicting ETc, which refers to the total water loss from a crop through both
soil evaporation and plant transpiration. The proposed model (DG-DGSF) was compared against nine models,
including graph-based models and standard DL models, including BiLSTM, GRU, GCN, BiGRU, LSTNet,
DGDL, TPA-LSTM, GCN-LSTM to predict Crop evapotranspiration (ET c). Table 3 reports the results in terms
of N RM SE , M AP E , N RM SE , and r.
Obviously, the proposed DG-DGSF model demonstrated superior performance against the graph-based
approaches and standard models, achieving the highest prediction rates across all four locations categorized by
the highest r = 0.993 and lowest RM SE = 0.0669 values, compared to other graph deep learning and standard
models. The proposed model outperformed the graph deep learning models, for example, the DGDL achieved
lower prediction results r, andRM SE , compared to our DG-DGSF model the r = 0.920 RM SE = 0.173.
The GCN-LSTM, and GCN also scored lower results than the proposed model for GCN-LSTM r = 0.8943
RM SE = 0.321, and for GCN r = 0.863 RM SE = 0.432. The results confirmed that the performance of the
proposed DG-DGSF model outperformed the standard dynamic graph deep learning techniques.
The standard deep learning models GRU, BiLSTM, BiGRU, TPA-LSTM, and LSTNet showed lower
performance compared to graph-based deep learning models. The LSTNet, and TPA-LSTM recorded the highest
RM SE and r among the standard deep learning models for TPA-LSTM and LSTNet r = 0.903, and0.887
respectively. These results of all standard models underscore the superior performance of the proposed model
(DG-DGSF). In addition, there was no noticeable fluctuation in the r values of the proposed DG-DGSF model
for all three locations. The proposed DG-DGSF attained r values of 0.992, 0.979, 0.982 for location, location 2,
and location 3, respectively. Compared to BiLSTM, GRU, GCN, BiGRU, LSTNet, DGDL, TPA-LSTM, GCN-
LSTM, the DG-DGSF model delivered more accurate ET c predictions for three locations, highlighting its
efficacy over these models. The noteworthy improvement in ET c prediction accuracy confirmed the advantage
of the proposed DG-DGSF.
Table 3 further reports the performance of all prediction models in terms of Mean Absolute Percentage
Error (M AP E ) and Normalised Root Mean Square Error (N RM SE ). The N RM SE calculates the standard
deviation of prediction errors relative to the range of observed values, making it an effective metric for
comparing prediction model performance across different data scales. While M AP E evaluates the average
absolute percentage difference between actual and precited values, delivering a clear percentage-based indication
of accuracy. Lower values of both N RM SE and M AP E metrics reflect a higher predictive accuracy. The
obtained results based on N RM SE and M AP E revealed that the proposed DG-DGSF model showed a

Scientific Reports | (2025) 15:45732 | [Link] 11

[Link]/scientificreports/

Models RMSE (mm d−1) MAPE r NRMSE

Location 1
LSTNet 0.0798 0.343 0.8871 0.0876
GCN-LSTM 0.0821 0.342 0.8941 0.0743
GCN 00.932 0.421 0.8631 0.0795
GRU 0.0975 0.474 0.8432 0.0987
BiGRU 0.0903 0.276 0.0891 0.0876
BiLSTM 0.0804 0.269 0.8875 0.0865
DGDL 0.0773 0.212 0.920 0.0626
TPA-LSTM 0.0976 0.234 0.903 0.0587
The proposed DG-DGSF mode 0.0469 0.120 0.993 0.0431
Location 2
LSTNet 0.0912 0.352 0.865 0.0888
GCN-LSTM 00.965 0.357 0.887 0.0797
GCN 0.0965 0.441 0.857 0.0781
GRU 0.0998 0.483 0.842 0.0982
BiGRU 0.0973 0.280 0.885 0.0885
BiLSTM 0.0955 0.277 0.879 0.0877
DGDL 0.0884 0.234 0.911 0.0656
TPA-LSTM 0.0876 0.254 0.900 0.0599
DG-DGSF 0.0521 0.145 0.979 0.045
Location 3
LSTNet 0.092 0.375 0.865 0.0890
GCN-LSTM 00.967 0.398 0.886 0.0799
GCN 0.0986 0.487 0.844 0.0797
GRU 0.0745 0.499 0.834 0.0993
BiGRU 0.0521 0.294 0.876 0.0887
BiLSTM 0.0877 0.297 0.886 0.0883
DGDL 0.0799 0.267 0.900 0.0671
TPA-LSTM 0.0698 0.279 0.896 0.0622
DG-DGSF 0.0556 0.126 0.982 0.0491

Table 3. Performance evaluation of the proposed model DG-DGSF against the state-of-the-art in terms of
RM SE , M AP , r, and N RM SE .

consistent and significant improvement in performance compared to previous models. Compared DG-DGSF
N RM SE and M AP E values for three locations, it can be noticed that there was a decrease in both values
compared to BiLSTM, GRU, GCN, BiGRU, LSTNet, DGDL, TPA-LSTM, and GCN-LSTM. The proposed DG-
DGSF scored the lowest values of N RM SE = 0.043, 0.0452, 0.0475. and M AP E = 0.120, 0.112, 0.125 for
location 1, location 2 and location 3 respectively. The obtained results highlighted the robustness and efficacy of
the proposed model as a superior tool for ET c prediction.
For further assessment, we adopted four metrics named Kling–Gupta Efficiency (KGE ), Nash–Sutcliffe
Efficiency (N SE ), Coefficient of Variation (CV ), and Fractional Bias (F B ). Table 4 summarises the perfume
evaluation of the proposed DG-DGSF model compared to previous approaches using KGE , CV , N SE , F B .
Our DG-DGSF model achieved a remarkable KGE = 0.976, 0.967, 0.975, and N SE = 0.985, 0.966, 0.976
for three locations confirming its superior capability to capture the dynamic patterns of observed data compared
to state-of-the-art models. These are high values of KGE , and N SE showed a significant advantage in extracting
trends and behaviours associated with the observed data that are not visible for other models. However, the
graph-based deep learning models scored a range of N SE values: GCN (0.851), DGDL (0.932), GCN-LSTM
(0.889) for location one, GCN (0.841), DGDL (0.928), GCN-LSTM (0.865) for location two, and GCN (0.837),
DGDL (0.910), GCN-LSTM (0.876) for location three. While the graph-based deep learning models showed
reasonable performance, the DG-DGSF model consistently outperformed them in all three locations. The
variation in performance intensifies further with the standard model GRU (0.831), BiGRU (0.865), BiLSTM
(0.873). The KGE further emphasize the DG-DGSF model exceptional performance. The DG-DGSF model
achieved the highest KGE for three stations, highlighting its efficiency in capturing the important patterns of
observed data. The consistently high KGE, and NSE values across three locations confirmed its ability to capture
trends and seasonal patterns.
Further evaluation using CV, and FB delivers complementary analysis into potential variability and biases in
the model’s predictive results. The DG-DGSF model scored a lower F B = 0.0081 and CV = 0.342 compared
to the state of the models, demonstrating good performance in terms of reducing both variability and bias. The

Scientific Reports | (2025) 15:45732 | [Link] 12

[Link]/scientificreports/

Models KGE N SE CV FB
Location 1
LSTNet 0.895 0.887 0.321 0.052
GCN-LSTM 0.885 0.874 0.354 0.062
GCN 0.832 0.822 0.312 0.068
GRU 0.842 0.832 0.431 0.0932
BiGRU 0.886 0.875 0.310 0.061
BiLSTM 0.882 0.874 0.332 0.043
CDGDL 0.912 0.901 0.299 0.041
TPA-LSTM 0.901 0.893 0.298 0.018
The proposed DG-DGSF mode 0.976 0.963 0.211 0.007
Location 2
LSTNet 0.884 0.875 0.355 0.055
GCN-LSTM 0.863 0.865 0.362 0.068
GCN 0.828 0.811 0.331 0.069
GRU 0.839 0.828 0.464 0.097
BiGRU 0.877 0.867 0.331 0.067
BiLSTM 0.881 0.874 0.354 0.048
DGDL 0.901 0.899 0.300 0.055
TPA-LSTM 0.891 0.888 0.301 0.019
DG-DGSF 0.965 0.963 0.245 0.008
Location 3
LSTNet 0.873 0.865 0.357 0.089
GCN-LSTM 0.855 0.855 0.368 0.088
GCN 0.810 0.800 0.345 0.087
GRU 0.820 0.814 0.474 0.099
BiGRU 0.865 0.845 0.345 0.079
BiLSTM 0.864 0.865 0.365 0.058
DGDL 0.893 0.884 0.366 0.065
TPA-LSTM 0.888 0.875 0.376 0.022
DG-DGSF 0.953 0.957 0.247 0.009

Table 4. Performance evaluation of the proposed model DG-DGSF against the state-of-the-art in terms of
RM SE , M AP , r and N RM SE .

lower CV , and F B values of the DG-DGSF model highlight its outstanding facility to produce predictions with
low bias and variability, in addition, confirming its robustness as a ET c prediction tool.

Visual evaluation of the proposed model

While the results in Tables 3 and 4 provided valuable quantitative evaluation of prediction model performance,
those metrics might not always expose limitations and potential shortcomings. To provide a more thorough
understanding of the proposed DG-DGSF mode behaviour for the ET c prediction. In this experiment, several
plots were utilized as a complementary analysis approach. These visual plots offer valuable evaluation that
numerical measures might ignore. Figure 6 displays scatterplots for all models, comparing the predicted with
actual ET c values. Each plot includes the coefficient of determination (R2 ∈ [0,1] ) with the linear fit equation
(y = mx + c). A higher R2 value indicates stronger relationship between the actual values and the predicted
model. Figure 6 shows that the proposed model DG-DGSF mode behaves very closely to the actual values.
It scored the highest R2 compared to the state-of-the-art models. The visual results in Fig. 6 align with the
previous observations from Tables 3 and 4, where the graph-based models and standard models achieved lower
performance values compared to the DG-DGSF model. The scatterplot for the DG-DGSF mode produced
patterns that are closely equivalent to the y = x line. The results indicated there was a strong alignment between
the DG-DGSF model and actual ET c values. These visual results support our findings in Tables 3 and 4 to
demonstrate the exceptional performance of the proposed model in predicting ET c. Figure 7 employs violin
plots to report the absolute forecasting Error (FE) for DG-DGSF model and the state-of-the-art models.
The violin plots utilize an inverted probability density function to show the error distribution of prediction
models. The DG-DGSF model violin plots demonstrated slight bias|. The limit range of errors in the DG-DGSF
model plot refers to its strength in handling various data points, even for complex predictions. Presenting the
distribution of errors using violin plots supports the findings attained by previous metrics and proves the DG-
DGSF model’s suitability for ET c prediction. The superior performance of the proposed DG-DGSF in reducing
prediction errors made it a suitable choice for practical applications in this agricultural domain.

Scientific Reports | (2025) 15:45732 | [Link] 13

[Link]/scientificreports/

Fig. 6. Scatter plots of each location for models’ performance comparison.

Scientific Reports | (2025) 15:45732 | [Link] 14

[Link]/scientificreports/

Fig. 7. Prediction Errors using Violin plots generated by the proposed model against comparing models.

Scientific Reports | (2025) 15:45732 | [Link] 15

[Link]/scientificreports/

The efficiency of DG-DGSF was assessed using visualized Taylor plots in Fig. 8. Taylor plots are a graphical
measure which provides a summary of how well the predicted model matches the actual values in terms of
correlation, root mean square difference and ratio of variance. Figure 8 shows the Taylor diagrams of the
proposed model as well as the baselines. From the results, it can be observed that the behaviour of the proposed
DG-DGSF model was satisfactory and very close to the actual data. The proposed DG-DGSF demonstrated
a strong relationship with the ETc across three locations. In addition, the DGDL model also showed a high
performance and scored high regression values compared to other graph-based models.

Evaluation based on statistical approaches

Statistical assessment was adopted for further evaluation of the proposed models. In this experiment, the
Diebold Mariano (DM) approach was employed to evaluate the proposed DG-DGSF model against the state-of-
the-art methods. Using this approach, a statistical examination was conducted to identify whether the prediction
errors of the two approaches differ significantly. In other words, DM verify if the predictions of the proposed
DG-DGSF model are statistically more accurate than other approaches. Figure 9 displays a heatmap for three
locations using DM. A negative number in a cell means that the model in the row performs better than models
in the column by producing lower prediction errors, while a positive value indicates that the column model
outperforms the row model. The results demonstrate that the proposed DG-DGSF model has a compelling
advantage over the state of the art. The proposed DG-DGSF model consistently produced negative values in the
entire heatmap, showing its efficiency compared to other models. Conversely, some standard and graph-based
models, such as GRU and GCN models, produced positive values toward other models, signifying that their
predictions are statistically not accurate enough.
For further evaluation, we applied a residual-based bootstrapping, which is a statistical measure employed to
assess the distribution of possible prediction errors of the predictive models. This approach involves resampling
the residuals from the training sample. Then, it subsequently evaluates models using these resampled samples.
In this paper, we resampled the time series data 100 times, and the resulting residuals were kept calculating the
probabilistic measures shown in Fig. 9. The average of PICP for the three locations was presented in Fig. 10.

Fig. 8. Taylor diagrams of all three locations generated by the proposed model against comparing models.

Scientific Reports | (2025) 15:45732 | [Link] 16

[Link]/scientificreports/

(a) Location 1 (b) Location 2

697

(c) Location 3

Fig. 9. Heating map of DM metric for all three locations generated by the proposed model against comparing
models.

The proposed DG-DGSF model showed extraordinary performance, scoring PICP of 0.99, proving that
approximately 0.99% of true values lay within its predicted intervals. The results demonstrated that the proposed
DG-DGSF model showed a high degree of confidence by capturing most of the actual values. Furthermore, the
graph-based models such as DGDL and GCN-LSTM also showed promising performances, and they achieved
an average of PICP of 0.93.
Prediction Interval Normalized Average Width metric (PINAWM) and Winkler Score (WS) are also adopted
in this paper, which calculates the average width of the data’s variability relative to prediction intervals. Figures
10, 11 and 12 present a comparison evaluation in terms of P IN AW M, P IN AW M, W S . The model with
lower values of PINAWM refers to narrower intervals and accurate predictions. Based on the results, the
proposed DG-DGSF model outperforms the state-of-the-art models, scoring PINAWM of 0.15. Although
DGDL and GCN-LSTM models recorded slightly lower PINAWM 0.18 and 0.22, respectively, they scored lower
PICP compared to the proposed DG-DGSF model, obtaining an average of 0.93.
Based on the obtained results using PINAWM, WS, and PICP, the proposed model delivered itself as a
superior model for ETc prediction. As it showed reliable predictions with lower uncertainty. This exceptional
performance of the proposed model in Table 5 supports the findings in Figs. 9, 10, 11 and 12, showing the
proposed DG-DGSF model demonstrated a strong prediction capability across different evaluation metrics.
In the section, a deeper investigation was made by employing the following metrics: F-Index, Coverage Width
(CW ), and Average Coverage Error (ACE ). The proposed DG-DGSF model demonstrated strong performance
across these measures. The proposed DG-DGSF model scored the lowest value of CW =17.34F-Index = 25.22,

Scientific Reports | (2025) 15:45732 | [Link] 17

[Link]/scientificreports/

Fig. 10. Model evaluation using PCIP criteria

Fig. 11. The PINAWM score for model evaluation

and ACE =4.56 for location 1. The results indicated that the proposed DG-DGSF model significantly produced
narrower prediction intervals compared to the state-of-the-art models. This designed DG-DGSF model offers
more accurate predictions.

Performance evaluation based on nested cross validation

We also adopted nested, chronological cross-validation to further evaluate and assess the generalisation of the
proposed DG-DGSF model. The time-series data were divided into folds such that each outer fold held out a
strictly future test block. Within each outer fold, hyperparameters and early stopping were tuned only on earlier
training data using rolling or blocked validation. The future test block was not used for tuning. All preprocessing

Scientific Reports | (2025) 15:45732 | [Link] 18

[Link]/scientificreports/

Location 1 Location 2 Location 3

Models CW ACE F-Index CW ACE F-Index CW F-Index ACE
LSTNet 23.41 10.34 30.32 23.73 10.55 30.67 23.56 10.98 30.99
GCN-LSTM 21.11 9.24 27.54 21.56 9.74 27.56 21.77 9.87 27.64
GCN 20.34 9.21 26.45 20.65 9.22 26.34 20.44 9.11 26.44
GRU 29.34 12.34 44.43 29.56 12.56 44.67 29.79 12.54 44.72
BiGRU 24.12 10.14 28.45 24.52 10.14 28.76 24.45 10.76 28.34
BiLSTM 25.71 9.77 27.87 25.87 9.44 27.09 25.89 9.78 27.90
DGDL 20.61 9.11 25.12 20.74 9.54 25.44 20.74 9.34 25.54
TPA-LSTM 23.11 8.71 25.22 23.16 8.65 25.66 23.67 8.87 25.76
DG-DGSF 17.34 4.56 13.45 17.6 4.75 13.54 17.63 4.71 13.68

Table 5. Statistical analysis of each model for all three locations.

Fig. 12. Model assessment based on WS

steps were fitted on training windows only and applied forward to validation and test samples to eliminate
information leakage. We report performance only on outer test folds, using RM SE(mmday−1 ),
M AP E (%), and N SE (%), KGE , CV (%).
Across the five outer test folds, the model maintains stable predictive accuracy with RM SE = 0.05 ±
0.002 mm·day⁻1 and M AP E = 0.128 ± 0.003, KGE =0.9622 ± 0.006, N SE =0.9502 ± 0.008 demonstrating
consistent performance under a time-aware protocol that mirrors deployment (train on past → predict future).
The results indicate that predicted and observed ETc values were very close across all folds. Overall, the nested,
chronological cross-validation confirms that the proposed model generalises well across the evaluated time
periods (Table 6).

Conclusion, future direction, and limitations

This study introduces a novel dynamic graph-based methodology (DG-DGSF), designed to enhance the
prediction accuracy of potato crop evapotranspiration amidst changing climatic conditions, particularly
pertinent for agricultural water management in Prince Edward Island. The suggested methodology employs
seasonal and trend decomposition using LOESS smoothing (STL), augmenting the model’s ability to identify
intricate temporal patterns in multivariate agricultural datasets. The integration of DGL and GC-GRU, along
with dual-view semantic fusion and contrastive learning frameworks, facilitated the thorough modelling of
complex spatio-temporal interactions in soil–water-plant relationships.

Scientific Reports | (2025) 15:45732 | [Link] 19

[Link]/scientificreports/

Fold number RM SE(mmday−1 ) MAPE KGE NSE CV

1 0.0489 0.121 0.971 0.961 0.212
2 0.0516 0.132 0.957 0.941 0.232
3 0.0487 0.127 0.962 0.952 0.213
4 0.0497 0.129 0.967 0.957 0.216
5 0.0511 0.131 0.954 0.940 0.231
M ean ± SD 0.05 ± 0.002 0.128 ± 0.003 0.9622 ± 0.006 0.9502 ± 0.008 0.2208 ± 0.008

Table 6. Performance evaluation based on nested cross validation.

While our validation on PEI showed promising results, wider applicability requires testing under varied soils,
climates, and management practices. We are currently working on collecting a multi-site dataset that covers
Atlantic maritime, humid temperate, Prairie semi-arid, and continental interior conditions, with a range of
soil textures and farming systems cross Canada. Future work will use region-held-out and climate-stratified
evaluation protocols. We will also evaluate the proposed model on additional crops to assess crop-specific
transferability. Together, these steps aim to demonstrate the model’s robustness and universality beyond the
current setting.
The DG-DGSF model was evaluated comprehensively over three soil textures (loam, sandy loam, and loamy
sand), exhibiting the variety of agricultural circumstances within the research area. Quantitative evaluations
indicated that DG-DGSF surpassed multiple benchmark models, including advanced graph-based methods
such as GCN-LSTM and DGDL, as well as traditional deep learning techniques like BiLSTM and GRU. Metrics
such as RMSE, MAPE, NSE, KGE, and correlation coefficients repeatedly demonstrated the enhanced predictive
capabilities of the proposed model. The quantitative findings were supported by extensive visual analyses,
including scatterplots, Taylor diagrams, violin plots, and Diebold-Mariano statistical tests, all affirming the
model’s efficacy in identifying underlying patterns and reducing prediction mistakes and biases.
The proposed DG-DGSF model substantially outperformed the state-of-the-art models in ETc prediction.
The obtained results based on visual approaches, such as forecasting error plots and Taylor plots, demonstrated
that the DG-DGSF model has a lower prediction error compared to all benchmark models. Also, based on Taylor
and scatter plots, the results showed that the proposed DG-DGSF model produced very close values to actual
values.
The DG-DGSF model has considerable potential, although several constraints must be recognized to clarify
its applicability and scope. The main limitation is the reliance on high-quality, regularly monitored data from
lysimeters and automated meteorological stations. Moreover, although this model has undergone rigorous
validation under specific experimental conditions in Prince Edward Island, its applicability to significantly
diverse climatic zones, soil types, or agricultural systems remains minimally examined. Therefore, careful
application is advised when evaluating differences from the examined contexts.
An additional significant factor is the intrinsic complexity and computing demands of the DG-DGSF system.
The complex amalgamation of dynamic graph structures, semantic fusion processes, and contrastive learning
modules requires substantial computational resources. Moreover, the intricacy of the model may pose challenges
to stakeholders and practitioners, potentially hindering acceptance or practical utility unless accompanied by
enhanced interpretability mechanisms.
The model’s complexity may present difficulties for stakeholders and practitioners, potentially obstructing
acceptance or practical use unless supported by improved interpretability mechanisms. Practices include variable
fertilizer applications, diverse cropping systems, rotating techniques, and targeted pest or disease management
protocols that may affect evapotranspiration rates and soil–water dynamics, introducing additional factors not
comprehensively incorporated into the existing framework.
To improve deplorability of the proposed model under limited resources, we will consider model tuning
as a multi-objective issue: keep prediction error low while also reducing time complexity including FLOPs,
latency, and memory usage. We will examine several metaheuristic policies such as Genetic Algorithms (GA)
and Particle Swarm Optimisation (PSO) to discover the key settings of the proposed model including window
length, decomposition, and graph parameters. In parallel, we will investigate the effects of model compression on
the results with knowledge distillation to produce compact student models that run faster with less memory. For
hyperparameter selection, our plan will be considered Bayesian search to minimise tuning time while maintaining
accuracy. Together, these techniques aim to retain accuracy at low cost and deliver robust, lightweight models
suitable for stakeholders with constrained hardware.
Future research must focus on overcoming these constraints while enhancing the robustness and application
of the DG-DGSF model. A prompt approach involves the incorporation of extensive datasets that include
various geographic regions, climate variables, and agricultural management practices. Comprehensive cross-
validation of worldwide datasets, especially from areas with diverse soil types and climatic variations, would
enhance the model’s robustness and universality. Additional research into sophisticated optimization methods
and computational simplifications may help reduce computational limitations. Methods such as metaheuristic
optimization, ensemble forecasting, and model pruning may reduce processing requirements while maintaining
prediction accuracy. These strategies would improve the model’s accessibility, especially for stakeholders with
constrained computational resources, thereby expanding their usefulness.

Scientific Reports | (2025) 15:45732 | [Link] 20

[Link]/scientificreports/

Later studies could examine the effects of climate change scenarios on model efficacy and irrigation
methodologies. Integrating climate projection data from global climate models (GCMs) across different
emission scenarios (e.g., Representative Concentration Pathways—RCPs) would facilitate evaluating future
evapotranspiration patterns and irrigation requirements. Such assessments would be essential for long-term
agricultural planning and resource distribution, enabling proactive initiatives for climate change adaptation and
resilience.
Notwithstanding its constraints, the suggested framework signifies a substantial progression in agricultural
water management, presenting encouraging opportunities for additional improvement. Future research will
enhance the DG-DGSF model’s utility as a decision-support tool for sustainable agricultural water management
by tackling computational challenges, broadening data diversity, improving model interpretability, and
integrating socio-economic and climate change factors.

Data availability
The datasets generated and/or analysed during the current study are not publicly available as the authors are
not allowed to share the data on public domains but are available from the corresponding author on reasonable
request.

Received: 12 August 2025; Accepted: 11 November 2025

References
1. Dzvene, A. R., Zhou, L., Slayi, M. & Dirwai, T. L. A scoping review on challenges and measures for climate change in arid and
semi-arid agri-food systems. Discov. Sustain. 6, 151 (2025).
2. Yang, Y. et al. Characterization of greenhouse gas emissions and water requirement of farmland in China’s main grain-producing
areas under future climate scenarios. Agric. Syst. 225, 104293 (2025).
3. Yang, Z.-Y. et al. A comprehensive review of deep learning applications in cotton industry: from field monitoring to smart
processing. Plants 14, 1481 (2025).
4. AlZaatiti, F., Halwani, J. & Soliman, M. R. Climate change impacts on flood risks in the Abou Ali river basin, Lebanon: A
hydrological modeling approach. Results Eng. 25, 104186 (2025).
5. Borah, G. Urban water stress: Climate change implications for water supply in cities. Water Conserv. Sci. Eng. 10, 20 (2025).
6. Granata, F., Di Nunno, F. Financing the future of water: unlocking investment, innovation, and governance for resilient
infrastructure in a changing climate. Earth Syst. Environ. 1–25 (2025).
7. Adebayo, O., Singh, A., Bista, P., Angadi, S., Ghimire, R. Compost addition improves soil water storage and crop water productivity
in cover crop integrated sorghum production system under a limited irrigation management. Irrig. Sci. 1–15 (2025).
8. Tang, Z. et al. Farmland mulching and optimized irrigation increase water productivity and seed yield by regulating functional
parameters of soybean (Glycine max L.) leaves. Agric. Water Manage. 298, 108875 (2024).
9. Togneri, R., Prati, R., Nagano, H. & Kamienski, C. Data-driven water need estimation for IoT-based smart irrigation: A survey.
Expert Syst. Appl. 225, 120194 (2023).
10. Et-taibi, B. et al. Enhancing water management in smart agriculture: A cloud and IoT-Based smart irrigation system. Results Eng.
22, 102283 (2024).
11. Cheema, S.J., Karbasi, M., Randhawa, G.S., Liu, S., Esau, T.J., Grewal, K.S., Abbas, F., Zaman, Q.U., Farooque, A.A. A state-of-the-
art novel approach to predict potato crop coefficient (Kc) by integrating advanced machine learning tools. Smart Agric. Technol.
100896 (2025).
12. Bhatti, A. Z. et al. Climate change impacts on rainfed agriculture and mitigation strategies for sustainable agricultural management:
A case study of Prince Edward Island, Canada. World Water Policy 8, 142–179. [Link] (2022).
13. Farooque, A. A. et al. How can potatoes be smartly cultivated with biochar as a soil nutrient amendment technique in Atlantic
Canada?. Arab. J. Geosci. 13, 1–9 (2020).
14. Bhatti, A. Z. et al. An overview of climate change induced hydrological variations in Canada for irrigation strategies. Sustain
[Link] (2021).
15. Bhatti, A. Z. et al. Climate change impacts on precipitation and temperature in Prince Edward Island, Canada. World Water Policy
7, 9–29. [Link] (2021).
16. Bhutto, R.A., Khanal, S., Wang, M., Iqbal, S., Fan, Y., Yi, J. Potato protein as an emerging high-quality: Source, extraction,
purification, properties (functional, nutritional, physicochemical, and processing), applications, and challenges using potato
protein. Food Hydrocoll. 110415 (2024).
17. Doorenbos, J., Pruitt, W.O. Guidelines for predicting crop water requirements (1977).
18. Economou, F. et al. Life cycle assessment of potato production in insular communities under subtropical climatic conditions. Case
Stud. Chem. Environ. Eng. 8, 100419 (2023).
19. Hu, T., Zhang, X., Khanal, S., Wilson, R., Leng, G., Toman, E.M., Wang, X., Li, Y., Zhao, K. Climate change impacts on crop yields:
A review of empirical findings, statistical crop models, and machine learning methods. Environ. Model. Softw. 106119 (2024).
20. Ierna, A. Water management in potato, in: Potato Production Worldwide. Elsevier, pp. 87–100 (2023).
21. Ishak, N. F. & Mazlan, Z. Key challenges and potentials of potato (Solanum tuberosum L.) farming in Malaysia: A mini review.
Potato Res. 1–16 (2025).
22. King, B.A., Stark, J.C., Neibling, H. Potato Irrigation Management. Potato Prod. Syst. 417–446. (2020) ​h​t​t​p​s​:/​ ​/​d​o​i​.​o​r​g​/1​ ​0​.​1​0​0​7​/​9​7​
8​-​3​-​0​3​0​-​3​91​ ​5​7​-​7​_​1​3​​​​
23. Nayak, L., Barik, M., Tiwari, R.K., Kumar, R., Kumar, A., Lal, M.K. Overview of underground vegetable crops, in: Abiotic Stress in
Underground Vegetables. Elsevier, pp. 3–11 (2025).
24. Nyawade, S. O., Karanja, N. N., Gachene, C. K. K., Schulte-Geldermann, E. & Parker, M. Effect of potato hilling on soil temperature,
soil moisture distribution and sediment yield on a sloping terrain. Soil Tillage Res. 184, 24–36 (2018).
25. Piekutowska, M. & Niedbała, G. Review of methods and models for potato yield prediction. Agriculture 15, 367 (2025).
26. Cheema, S. J. et al. A comprehensive analytical and computational assessment of soil water characteristics curves in Atlantic
Canada: Application of a novel SelectKbestbased GEP model. Agric. Water Manag. 298, 108868 (2024).
27. Adekanmbi, T. et al. Assessing future climate change impacts on potato yields—A case study for prince Edward island. Canada.
Foods 12, 1176 (2023).
28. Danielescu, S. et al. Crop water deficit and supplemental irrigation requirements for potato production in a temperate humid
region (Prince Edward Island, Canada). Water 14, 2748 (2022).
29. Rajendran, S., Domalachenpa, T., Arora, H., Li, P., Sharma, A., Rajauria, G. Hydroponics: Exploring innovative sustainable
technologies and applications across crop production, with Emphasis on potato mini-tuber cultivation. Heliyon (2024).

Scientific Reports | (2025) 15:45732 | [Link] 21

[Link]/scientificreports/

30. Tekle, S. L., Bonaccorso, B. & Naim, M. Simulation-based optimization of water resource systems: a review of limitations and
challenges. Water Resour. Manag. 39, 579–602 (2025).
31. Velten, B. & Stegle, O. Principles and challenges of modeling temporal and spatial omics data. Nat. Methods 20, 1462–1474 (2023).
32. Zeghina, A., Leborgne, A., Le Ber, F., Vacavant, A. Deep learning on spatiotemporal graphs: A systematic review, methodological
landscape, and research opportunities. Neurocomputing 127861 (2024).
33. Ren, W., Jin, N. and OuYang, L. Phase space graph convolutional network for chaotic time series learning. IEEE Trans. Ind. Inform.
(2024).
34. Yang, Q., Yao, W., Liu, W. and Liu, H. An Enhanced TPA-LSTM Method for PMU Data Recovery and Prediction. In 2024 21st
International conference on harmonics and quality of power (ICHQP) (pp. 614–618). IEEE (2024).
35. Zheng, W. & Chen, G. An accurate GRU-based power time-series prediction approach with selective state updating and stochastic
optimization. IEEE Trans. Cybern. 52(12), 13902–13914 (2021).
36. She, D. & Jia, M. A BiGRU method for remaining useful life prediction of machinery. Measurement 167, 108277 (2021).
37. Kim, J. and Moon, N. BiLSTM model based on multivariate time series data in multiple field for forecasting trading area. J. Ambient
Intell. Human. Comput, pp.1–10 (2019).
38. Sharma, D. N. & Tare, V. Assessment of irrigation requirement and scheduling under canal command area of Upper Ganga Canal
using CropWat model. Model. Earth Syst. Environ. 8, 1863–1873 (2022).
39. Zhang, F. et al. Coupling effects of irrigation amount and fertilization rate on yield, quality, water and fertilizer use efficiency of
different potato varieties in Northwest China. Agric. Water Manag. 287, 108446 (2023).
40. Fu, C. et al. Combining the FAO-56 method and the complementary principle to partition the evapotranspiration of typical
plantations and grasslands in the Chinese Loess Plateau. Agric. Water Manag. 295, 108734 (2024).
41. Ajith, S., Vijayakumar, S. & Elakkiya, N. Yield prediction, pest and disease diagnosis, soil fertility mapping, precision irrigation
scheduling, and food quality assessment using machine learning and deep learning algorithms. Discov. Food 5, 1–23 (2025).
42. Hailegnaw, N. S. et al. Integrating machine learning and empirical evapotranspiration modeling with DSSAT: Implications for
agricultural water management. Sci. Total Environ. 912, 169403 (2024).
43. Xue, Y., Zhang, Z., Li, X., Liang, H., Yin, L. A review of evapotranspiration estimation models: advances and future development.
Water Resour. Manag. 1–17 (2025).
44. Nayak, A.K., Sarangi, A., Pradhan, S., Panda, R.K., Jeepsa, N.M., Satpathy, B.S., Kumar, M. Estimation of daily reference
evapotranspiration using machine learning and deep learning techniques with sparse meteorological data (2024).
45. El-Kenawy, E.-S.M., Alhussan, A.A., Khodadadi, N., Mirjalili, S., Eid, M.M. Predicting potato crop yield with machine learning
and deep learning for sustainable agriculture. Potato Res. 1–34 (2024).
46. Gündüz, A., Orman, Z. Hyperspectral image classification using a hybrid RNN-CNN with enhanced attention mechanisms. J.
Indian Soc. Remote Sens. 1–17 (2024).
47. Mahmoud, A. & Mohammed, A. Leveraging hybrid deep learning models for enhanced multivariate time series forecasting.
Neural Process. Lett. 56, 223 (2024).
48. Elabd, E., Hamouda, H. M., Ali, M. A. & Fouad, Y. Climate change prediction in Saudi Arabia using a CNN GRU LSTM hybrid
deep learning model in al Qassim region. Sci. Rep. 15, 1–19 (2025).
49. Han, D., Wang, P., Tansey, K., Zhang, Y. & Li, H. A graph-based deep learning framework for field scale wheat yield estimation. Int.
J. Appl. Earth Obs. Geoinf. 129, 103834 (2024).
50. Saravanan, K. S. & Bhagavathiappan, V. Innovative agricultural ontology construction using NLP methodologies and graph neural
network. Eng. Sci. Technol. an Int. J. 52, 101675 (2024).
51. Ghayekhloo, M. & Nickabadi, A. Supervised contrastive learning for graph representation enhancement. Neurocomputing 588,
127710 (2024).
52. Li, X., Wang, Y., Wang, Y. & An, X. Graph contrastive learning for recommendation with generative data augmentation. Multimed.
Syst. 30, 170 (2024).
53. Xia, J., Wu, L., Chen, J., Hu, B. & Li, S. Z. Simgrace: A simple framework for graph contrastive learning without data augmentation.
Proc ACM Web Conference 2022, 1070–1079 (2022).
54. Wang, L., Chen, Z., Liu, W. & Huang, H. A temporal-geospatial deep learning framework for crop yield prediction. Electronics 13,
4273 (2024).
55. Wang, D. et al. Dynamic travel time prediction with spatiotemporal features: using a GNN-based deep learning method. Ann.
Oper. Res. 340(1), 571–591 (2024).
56. Huang, D., Liu, H., Bi, T. & Yang, Q. GCN-LSTM spatiotemporal-network-based method for post-disturbance frequency
prediction of power systems. Global Energy Interconnection 5(1), 96–107 (2022).
57. Zsembeli, J., Czellér, K., Sinka, L., Kovács, G., Tuba, G. Application of lysimeters in agricultural water management. Creat. a Platf.
to address Tech. used Creat. Prot. Environ. Econ. Manag. Water Soil 5–21 (2019).
58. Strange, P. C. & Blackmore, K. W. Effect of whole seed tubers, cut seed and within row spacing on potato (cv. Sebago) tuber yield.
Aust. J. Exp. Agric. 30, 427–431 (1990).
59. Allen, R. G., Pereira, L. S., Raes, D. & Smith, M. FAO Irrigation and drainage paper No 56. Rome Food Agric. Organ. United Nations
56, e156 (1998).
60. Rana, G. & Katerji, N. Measurement and estimation of actual evapotranspiration in the field under Mediterranean climate: A
review. Eur. J. Agron. 13, 125–153 (2000).
61. Srinivas, B. & Tiwari, K. N. Determination of crop water requirement and crop coefficient at different growth stages of green gram
crop by using non-weighing lysimeter. Int. J. Curr. Microbiol. Appl. Sci. 7, 2580–2589 (2018).
62. He, R., Zhang, L. & Chew, A. W. Z. Modeling and predicting rainfall time series using seasonal-trend decomposition and machine
learning. Knowl.-Based Syst. 251, 109125 (2022).
63. Liu, X. and Zhang, Q. Combining seasonal and trend decomposition using LOESS with a gated recurrent unit for climate time
series forecasting. IEEE Access (2024).
64. Liang, Z., Li, W., Wang, Z., Zheng, X. and Pang, B. SSSLN: Multivariate time series forecasting via collaborative dynamic graph
learning. Neural Netw, p.107485 (2025).
65. Islam, M.I.K., Saifuddin, K.M., Hossain, T. and Akbas, E. Dygcl: Dynamic graph contrastive learning for event prediction. In 2024
IEEE International Conference on Big Data (BigData) (pp. 559–568). IEEE (2024).
66. Georgousis, S., Kenning, M. P. & Xie, X. Graph deep learning: State of the art and challenges. IEEe Access 9, 22106–22140 (2021).
67. Shang, C., Chen, J. and Bi, J. Discrete graph structure learning for forecasting multiple time series. (2021) arXiv preprint
arXiv:2101.06861.
68. Zhang, Z., Cui, P. & Zhu, W. Deep learning on graphs: A survey. IEEE Trans. Knowl. Data Eng. 34(1), 249–270 (2020).
69. Bhatti, U. A., Tang, H., Wu, G., Marjan, S. & Hussain, A. Deep learning with graph convolutional networks: An overview and latest
applications in computational intelligence. Int. J. Intell. Syst. 2023(1), 8342104 (2023).

Acknowledgements
Authors would also like to thank the Sustainable Agriculture Research Group at UPEI’s Canadian Center for
Climate Change and Adaptation for their assistance during experimentation.

Scientific Reports | (2025) 15:45732 | [Link] 22

[Link]/scientificreports/

Funding
This research was supported by the Natural Sciences and Engineering Research Council of Canada (NSERC)
Alliance Sustainable Agriculture Research Initiative Grant.

Additional information
Correspondence and requests for materials should be addressed to M.A. or A.A.F.
Reprints and permissions information is available at [Link]/reprints.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives
4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in
any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide
a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have
permission under this licence to share adapted material derived from this article or parts of it. The images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated
otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence
and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to
obtain permission directly from the copyright holder. To view a copy of this licence, visit ​h​t​t​p:​ ​/​/​c​r​e​a​t​i​ve​ ​c​o​m​m​o​
n​s​.​or​ ​g​/​l​i​c​e​n​s​es​ ​/​b​y​-​n​c​-​n​d​/​4​.​0​/​​​​.​​

© The Author(s) 2025

Scientific Reports | (2025) 15:45732 | [Link] 23