Skip to main content
NCBI home page
Springer logoLink to Springer
. 2026 Aug 18;70(9):238. doi: 10.1007/s00484-026-03285-3

Weather data sources influence predictive performance of disease forecasting models

Theophilus Abonyi Mensah 1,✉, Neil White 1,2, Vivian Rincón-Flórez 1, Olufemi A Akinsanmi 1,✉
PMCID: PMC13486128  PMID: 42611104

Abstract

Weather data from fine- and coarse-scale sources are increasingly used in disease forecasting models to support management decisions, yet comparisons between fine- and coarse-scale sources remain limited, largely due to the difficulty of acquiring fine-scale datasets across locations and seasons. Differences among weather sources can influence model outputs, resulting in variation in predictive performance and the timing of management decisions. This study evaluated four weather data sources, grouped as fine-scale (in-canopy and on-farm stations) and coarse-scale (regional gridded and point-based data from SILO, and Bureau of Meteorology, BOM, stations), across two production regions in Australia. Macadamia orchards were used as a case study to assess spatial variability in 10 weather variables associated with fungal infection over three seasons (2022–2024). Heterogeneous trends and associations were identified among the weather sources, primarily driven by differences in minimum temperature and minimum relative humidity. On-farm and SILO data showed high spatial variability, whereas in-canopy and BOM data were comparatively stable. Predictive performance of disease models based on each weather source was evaluated using airborne conidia dynamics for the fungal pathogens Botrytis, Cladosporium and Neopestalotiopsis/Pestalotiopsis, which cause flower blight in macadamia. In-canopy sources produced significantly lower predictive performance (p < 0.05) than BOM sources, while on-farm, SILO and BOM sources showed comparable performance (p > 0.05) across all pathogens. These findings indicate that the choice of weather data source, rather than spatial scale alone, is critical for reliable disease forecasting.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s00484-026-03285-3.

Keywords: Airborne conidia, Disease forecasting, Flower blight, Microclimate, Weather-based models

Introduction

Disease forecasting models offer a valuable tool to optimise management strategies and reduce fungicide wastage (Perria et al. 2022; Shtienberg 2013). Models may predict infection risks and disease epidemics for specific crops and regions, providing early warning signals that enable targeted fungicide applications based on actual disease risk rather than calendar schedules. In potatoes, the application of models such as TOMCAST and other infection-based forecasting systems are used for managing potato early blight (Alternaria species), which have reduced fungicide application frequency and quantity by more than 20%, while increasing profit margin by up to 9% compared with a fixed-schedule program (Abuley et al. 2023). Several predictive models have also been developed for timing infection and controlling diseases caused by Diaporthe (Phomopsis cane and leaf spot of grapevine), Botryosphaeria (larch shoot blight), Botrytis (grapevine bunch rot) and Cladosporium (tomato powdery mildew) (Gonzalez-Dominguez et al. 2022; Jensen et al. 2025; Zhao et al. 2024; Zhou et al. 2024). Accurate forecasting depends on selecting suitable predictor variables, and includes knowledge of the pathogen life cycle, host phenology and environmental variables required for infection and disease development (Jensen et al. 2025).

Weather variables are the most commonly used predictors of disease epidemiology because they directly influence pathogen development and infection processes and are relatively easy to obtain (Jensen et al. 2025). Air temperature, dew point temperature, relative humidity, rainfall and leaf wetness data have been associated with the progression of several crop diseases (Rodríguez-Moreno et al. 2020). A generic forecasting framework for various pathosystems was developed by scaling leaf wetness duration to a temperature response function involving maximum, minimum, mean and optimum temperatures for pathogen infection (Magarey et al. 2005). In macadamia, relative humidity and rainfall have been identified as major weather variables associated with dynamics of Phomopsis husk rot and Botryosphaeria branch dieback pathogens, respectively (Mohankumar et al. 2023; Prasannath et al. 2022a). Preliminary regression analyses suggest that these variables could potentially explain variation in airborne conidia abundance for both pathosystems (Mohankumar et al. 2023; Prasannath et al. 2022a).

The source and quality of weather data used as inputs in disease models may influence predictive accuracy. Weather data are generally obtained from two spatial scales: farm-level (fine-scale) and regional-level (coarse-scale) sources. Fine-scale weather sources include in-canopy stations, with sensors that capture microclimatic conditions located within the crop canopy, and on-farm stations, located in open areas outside the canopy. Fine-scale sources are often expensive to install and maintain, and their placement may disrupt farm operations. Their accuracy may also be compromised by improper sensor installation or operator inexperience (Gleason et al. 2008). In contrast, coarse-scale data from government-maintained networks such as the Australia Bureau of Meteorology (BOM) and the Scientific Information for Land Owners (SILO) are more accessible and widely used for large-scale modelling (Brinkhoff and Robson 2021). BOM provides point-based measurements with a spatial resolution of up to 60 km, whereas SILO offers point-based and 5 km gridded datasets derived from BOM, providing finer coverage (Wedd et al. 2022; Zaman et al. 2024). Assessing predictive models, Brinkhoff and Robson (2021) noted that predictions from finer scale models tend to be more accurate when evaluated at coarser scales. However, Deines et al. (2021) emphasised that model accuracy is scale-dependent. To identify the most informative and reliable weather data source for disease forecasting tools applicable across regions, it is essential to assess how potential variability between fine- and coarse-scale data affects the predictive performance of disease models. In this study, we used macadamia as a case study.

Macadamia is a high value commercial tree nut crop cultivated in tropical and subtropical regions worldwide. China (32%), South Africa (28%), Australia (13%), Kenya (10%) and the United States (4%) accounts for over 82,000 metric tonnes (kernel basis) with supply value of more than $640 million (INC 2025). Despite increasing production trends, macadamia cultivation is challenged by fungal diseases that cause significant yield losses, including flower blight, husk rot, husk spot and Botryosphaeria branch dieback (Akinsanmi and Drenth 2017; Akinsanmi et al. 2007; Mohankumar et al. 2023; Prasannath et al. 2023). These diseases develop across nearly all phenological stages and are mostly managed using integrated approaches, including fungicide application. In macadamia, phenological development begins with budbreak in early winter, followed by raceme elongation in mid-winter and flowering from mid-winter to early spring (Bright 2025). Flowering comprises four distinct stages: punching bag (stage 1), style emergence (stage 2), peak anthesis (stage 3) and swollen ovary (stage 4). The swollen ovary develops into the matchhead and pea-size nut set stages during spring. Fruit development then progresses to the full-size nut stage, characterised by shell hardening and oil accumulation, before reaching physiological maturity in summer. Previous studies have established the critical role of weather conditions in disease development across these phenological stages (Akinsanmi and Drenth 2010, 2017; Mohankumar et al. 2023).

Flower blight is the most frequent and economically significant disease affecting flowers and can severely reduce pollination efficiency, causing over 99% fruit set failure and 80% yield loss (Prasannath et al. 2021, 2022a). Flower blight is caused by a complex of fungal pathogens of contrasting lifecycles, including several species of Pestalotiopsis, Neopestalotiopsis, Botrytis and Cladosporium, and risk periods (Prasannath et al. 2023). These fungal pathogens produce abundant airborne conidia that are primarily dispersed by wind. The peak risk periods are from budbreak to swollen ovary for Pestalotiopsis and Neopestalotiopsis, elongated raceme to swollen ovary for Cladosporium and punching bag to peak anthesis for Botrytis. Disease onset and severity are strongly influenced by weather conditions, including leaf wetness, temperature, rainfall and vapour pressure deficit (Molitor et al. 2020; Navi et al. 2005; Prasannath et al. 2022a). Hence, improved knowledge on the specific timing and duration of infection risk is crucial for effective disease control, making forecasting a useful early diagnostic tool.

The quantitative nature of fine- and coarse-scale weather data inputs in disease forecasting remain poorly understood, limiting their harmonisation for practical on-farm application. Weather data from fields within 15 km can result in different infection risk predictions (El Jarroudi et al. 2017; Rodríguez-Moreno et al. 2020). To evaluate the predictive performance of fine- and coarse-scale weather data for disease forecasting in macadamia orchards, we hypothesised that fine-scale sources provide greater predictive accuracy than coarse-scale sources. Specifically, this study aimed to (i) explore the relationship between weather variables from fine-scale and coarse-scale sources, (ii) assess the predictive performance of VPD-based disease models across different growing regions and (iii) determine the impact of weather data source on predicting fungal flower blight in macadamia.

Materials and methods

Sources of weather data from case study sites

The case study sites included 10 commercial macadamia orchards in two agroecologically distinct regions, with five orchards each in the Central Queensland (CQ) and Northern Rivers (NR), New South Wales in Australia (Fig. 1). The regions and orchards were selected based on the availability of fine-scale weather sources and records (up to 3 years). A summary of the orchard characteristics is shown in Table 1. The tree ages ranged from 4 to 22 years, with mean canopy diameters of about 5 m in CQ and 8 m in NR (Table 1). For each orchard, weather data were collected from two fine-scale and two coarse-scale sources. The fine-scale sources included in-canopy and on-farm stations. Data from both fine-scale sources were accessed from Davis weather stations through the WeatherLink platform (https://www.weatherlink.com/). The coarse-scale sources comprised BOM (http://www.bom.gov.au/climate/data) and SILO (www.longpaddock.qld.gov.au/silo/point-data/). BOM and SILO provide regional data and were regarded as coarse-scale based on their distances from the orchards (Table 1), calculated using the Haversine formula (Basyir et al. 2017) as

graphic file with name d33e427.gif 1

Fig. 1.

Fig. 1

Open in a new tab

Location of orchards and their corresponding weather sources used in the study. Each circle indicates an orchard and its associated weather sources in Central Queensland (solid circles): Orchard 1 (black), Orchard 2 (orange), Orchard 3 (brown), Orchard 7 (blue) and Orchard 8 (pink); and Northern Rivers, New South Wales (dashed circles): Orchard 4 (black), Orchard 5 (orange), Orchard 6 (brown), Orchard 9 (blue) and Orchard 10 (pink)

Table 1.

Characteristics and distances (km) of selected commercial macadamia orchards to nearest coarse-scale weather data sources in Central Queensland (CQ) and Northern Rivers (NR), with corresponding number and range of observation years

Region Orchard characteristics Distance to coarse-scale sources (km) Range of observation years
Label Tree age (years) Mean canopy diameter (m) SILO BOM (station ID)
CQ Orchard 1 8 5.96 2.94 19.87 (39128) 2022–2024
Orchard 2 4 2.44 2.31 23.68 (39128) 2022–2024
Orchard 3 5 5.30 2.82 37.91 (39128) 2022–2024
NR Orchard 4 20 10.32 2.63 19.22 (58216) 2022–2024
Orchard 5 22 9.21 3.05 11.75 (58216) 2022–2024
Orchard 6 13 8.47 2.11 11.37 (58214) 2022–2024
CQ Orchard 7 8 5.87 2.58 15.27 (39128) 2023–2024
Orchard 8 17 6.33 2.47 15.09 (39128) 2023–2024
NR Orchard 9 5 4.95 2.57 15.71 (58198) 2024
Orchard 10 15 8.03 3.61 13.00 (58198) 2024
Open in a new tab

In-canopy and on-farm stations were located within the orchards (distance to orchard = 0 km). BOM: Bureau of Meteorology (Australian Government); SILO: Scientific Information for Land Owners (Queensland Government)

where Inline graphic is the distance (km), r is the radius of the earth (6371 km), and Inline graphic, Inline graphic and Inline graphic, Inline graphic are the latitudes and longitudes (radians) of an orchard and coarse-scale source.

Weather variables

Weather data from July (winter) to December (summer) in the 2022 to 2024 production seasons were used (Table 1). This period coincides with the flowering and fruit development stages for macadamia in Australia. Weather variables common across the four sources were selected, including maximum temperature (Inline graphic, °C), minimum temperature (Inline graphic, °C), maximum relative humidity (Inline graphic, %), minimum relative humidity (Inline graphic, %) and rainfall (mm). Rainfall data were not available for the in-canopy stations. Therefore, in-canopy stations were excluded from rainfall-related analyses. Five weather variables commonly used in disease models were derived for each weather source, including VPD (kPa), leaf wetness duration (LW, days), maximum dew point temperature (Inline graphic, °C), minimum dew point temperature (Inline graphic, °C) and thermal oscillation (Inline graphic, °C) (Abuley et al. 2023; Manstretta and Rossi 2015; Sadyś et al. 2018; Vélez-Pereira et al. 2019). Inline graphic was calculated as the difference between Inline graphic and Inline graphic while LW was calculated using the fixed RH threshold approach, with a day considered wet if RH ≥ 90% (Sentelhas et al. 2008). The fixed RH threshold provides an indirect estimate of LW and may not fully capture the actual LW measurements taken using sensors (Rowlandson et al. 2015). VPD, Inline graphic and Inline graphic were calculated (Allen et al. 1998; Lawrence 2005) as

graphic file with name d33e721.gif 2
graphic file with name d33e725.gif 3
graphic file with name d33e729.gif 4

where  exp is exponential function,  ln is logarithmic function and Inline graphic = 0.6108, Inline graphic = 17.27,  C = 237.3, Inline graphic = 17.625 and Inline graphic = 243.04 are constants.

Coarse-scale sources (BOM and SILO) recorded daily data, while fine-scale sources (in-canopy and on-farm) recorded data at 5–60 min intervals depending on the orchard. Fine-scale data were standardised to daily resolution using the same procedure for coarse-scale data (BOM 2004). Some missing and aberrant records were observed in fine-scale sources and were excluded from the analyses.

All analyses and graphical visualisations were performed in Python version 3.12.7 using the packages; pandas (McKinney 2010), numpy (Harris et al. 2020), matplotlib (Hunter 2007), seaborn (Waskom 2021), pymannkendall (Hussain and Mahmud 2019), scipy (Virtanen et al. 2020), statsmodels (Seabold and Perktold 2010), scikit-learn (Pedregosa et al. 2011) and pca (Taskesen 2020).

Comparison of weather variables from fine- and coarse-scale sources

To assess variability in weather variables, six orchards, three in CQ (Orchards 1–3) and three in NR (Orchards 4–6) were used. These orchards had data in all sources for the 2022–2024 periods. Data were expressed as 14-day moving averages to visualise temporal trends. Short-term temporal dependence was quantified using lag-1 serial correlations (Inline graphic) (Seabold and Perktold 2010). Serially correlated variables were pre-whitened using the trend-free pre-whitening (TFPW) procedure (Yue et al. 2002) to generate serially independent data for trend analysis. Trend magnitude was estimated using Theil-Sen’s nonparametric method (Sen 1968; Theil 1992). All tests were performed at a 5% significance level. Monotonic trends and directional changes at four phenological stages were then quantified using the standardised Mann-Kendall statistic (Inline graphic) (Kendall 1948; Mann 1945).

To examine inter-source similarity, bivariate (pairwise correlations) and multivariate (clustering) approaches were used. Pairwise correlations were computed using Spearman’s rank correlation coefficient (ρ) to assess direct variable-to-variable agreement between sources, as the serially independent datasets significantly deviated from normality (Shapiro-Wilk test, p < 0.05) (Shapiro and Wilk 1965; Spearman 1904). Component-based clustering was performed on principal component scores using k-means and Gaussian mixture model to capture overall source-level variability (Ssegane et al. 2012). Silhouette analysis suggested varied optimal cluster numbers across orchards; therefore, a fixed number of four clusters, corresponding to the number of weather sources per orchard, was used. Clustering performance was evaluated using classification metrics, including confusion matrix, precision, recall, F1-score and overall accuracy. Precision is the proportion of correctly assigned sources within a cluster, recall is the proportion of true sources correctly assigned and F1-score is their harmonic mean. Accuracy reflects the proportion of correctly classified sources across all clusters. All metrics range from 0 to 1, with 1 indicating optimal performance (Vujović 2021).

Predictive performance of weather variables for flower blight pathogens

The predictive performance of source-specific weather variables was assessed by comparing predicted airborne flower blight conidial abundance to field observations from orchard traps. Data from four orchards in the 2023 and 2024 seasons were used. In CQ, Orchards 7 and 8, and in NR, Orchards 9 and 10 were selected because of available spore trapping systems. Airborne conidia of flower blight pathogens (Botrytis, Cladosporium and Pestalotiopsis/Neopestalotiopsis) were trapped using passive spore samplers constructed from rectangular plastic bottles assembled into opposing L-shaped sections (15.0 cm × 5.5 cm). Vaseline-coated Melinex tape (10.0 cm × 1.9 cm) was affixed to the internal surfaces. Traps were installed on branches at 1.8 m above the orchard floor to ensure uniformity (González-Fernández et al. 2020). Melinex tapes were collected bi-weekly from flowering (winter) to shell hardening and oil accumulation (summer). The number of conidia was quantified using multiplex quantitative PCR (qPCR) assay as described by Prasannath et al. (2022b).

Multivariate regression models were developed to relate conidial abundance of each pathogen during the spore-trapping period for each weather source. However, different predictor variables were identified as optimal among the weather data sources. For direct comparison of predictive performance across the weather sources, univariate regression models were developed using VPD from each weather data source. VPD was selected to standardise comparisons among weather sources rather than to maximise prediction accuracy, based on the previously reported polynomial association between flower blight conidial abundance and VPD (Prasannath et al. 2022a). Different temporal VPD structures were identified among the weather data sources. Hence, contemporaneous mean VPD from each weather source was used as the standardised predictor. Polynomial (p0 = 0.5, 1; maxfev = 10000) and simple linear regression models were evaluated, and the best-performing model type across weather sources was selected based on their predictive performance. Model development and evaluation were performed using the leave-one-out cross-validation (LOO-CV) method (De Cól et al. 2024), where each observation was iteratively held-out as the validation sample while the remaining observations were used as the training set (Hastie et al. 2009; Maleki et al. 2020). Predicted and observed conidial abundance from all iterations were aggregated to compute overall performance metrics, including the coefficient of determination (Inline graphic), Lin’s concordance correlation coefficient (LCCC), coefficient of residual mass (CRM) and normalised root mean square error (NRMSE), describing variance explained, agreement, bias and error (Ibrahim et al. 2022; Jadon et al. 2024; Lin 1989). For NRMSE, the observed conidia range was used as the normalising factor.

Performance metrics were min-max normalised to an identical scale (0–1) to ensure equal weighting among metrics and minimise scale-related bias during comparison of predictive performance among the weather sources across orchards. Predictive performance among weather sources was compared using the Wilcoxon signed rank (W) and Kruskal-Wallis (H) tests (Kruskal and Wallis 1952; Wilcoxon 1945). Pairwise comparisons were performed using Dunn’s post hoc test with Bonferroni correction (Dunn 1964). To identify the best-performing weather source, a multi-criteria prediction robustness index (PRI) was calculated as

graphic file with name d33e923.gif 5

where Inline graphic is the mean of the normalised performance metrics representing predictive performance, and Inline graphic is the standard deviation of the normalised metrics across orchards representing performance variability. The weighting parameter (Inline graphic) functions as a penalty term to control the trade-off between maximising predictive performance and minimising variability in performance across orchards. Subtracting (Inline graphic) from Inline graphic penalises weather sources that showed high cross-orchard variability to reduce potential effects of isolated orchard-specific performance extremes on overall PRI rankings. To select the Inline graphic value, a sensitivity analysis was performed using Inline graphic values ranging from 0 to 1. The resulting PRI values were compared across weather sources for the tested Inline graphic range. A final Inline graphic value (0.1), corresponding to the minimal penalty, was selected for the robustness analysis.

Results

Variability in weather variables based on sources

Temporal analysis revealed similar trends in most of the weather variables across the four sources (Fig. 2). Serial dependence between consecutive daily values of variables was detected across all sources, except for BOM, where rainfall (Orchards 1–3) and LW (Orchards 4–5) were largely independent (Table S1). A more stable microclimate was observed at in-canopy sources (mean Inline graphic = 0.74) compared with on-farm (mean Inline graphic = 0.71), SILO (mean Inline graphic = 0.71) and BOM (mean Inline graphic = 0.69) sources. A relative increase in Inline graphic, Inline graphic, Inline graphic, Inline graphic and VPD (Inline graphic = 2.02–4.64, p < 0.05) with similar magnitudes (mean slope = 0.01) was observed beginning from the budbreak (Fig. 2f–j, Fig. S1, Table S2). In contrast, Inline graphic, Inline graphic and Inline graphic were more source-specific, with orchard-specific increases, decreases or no clear trends (Fig. 2a–c, Fig. S1).

Fig. 2.

Fig. 2

Open in a new tab

Moving averages (14-day) of 10 weather variables from fine-scale (in-canopy, IC and on-farm, OF) and coarse-scale (SILO and BOM) sources for six commercial macadamia orchards in Central Queensland (Orchards 1–3) and Northern Rivers (Orchards 4–6) across four developmental stages (BB: budbreak, FL: flowering, NS: Nut set and spring flush, and SH: Shell hardening and oil accumulation) in 2022 to 2024, with starting day set to 1 July for each season. Inline graphic: thermal oscillation; Inline graphic: maximum relative humidity; Inline graphic: minimum relative humidity; LW: leaf wetness; Inline graphic: maximum temperature; Inline graphic: minimum temperature; Inline graphic: maximum dew point temperature; Inline graphic: minimum dew point temperature; VPD: vapour pressure deficit. Rainfall was not applicable for in-canopy stations. Numbers in the legend correspond to orchard labels

Correlations across all variables were generally stronger between SILO and BOM (ρ = 0.44–1.00, p < 0.001) than between in-canopy and on-farm (ρ = 0.40–0.97, p < 0.001) sources, with no correlation in Inline graphic at Orchard 5 (Fig. S2). Inter-scale comparisons showed both coarse-scale sources correlated better with on-farm (ρ = 0.20–0.96, p < 0.01) than with in-canopy (ρ = 0.17–0.89, p < 0.05) sources (Fig. 3).

Fig. 3.

Fig. 3

Open in a new tab

Spearman’s correlation coefficient between 10 weather variables from (a) SILO and fine-scale (in-canopy, IC; on-farm, OF) sources and (b) BOM and fine-scale (in-canopy, IC; on-farm, OF) sources across six commercial macadamia orchards in (a1–2 and b1–2) Central Queensland (Orchards 1–3) and (a3–4 and b3–4) Northern Rivers (Orchards 4–6) across three seasons (2022–2024). Statistically significant coefficients are indicated by *** (p < 0.001), ** (p < 0.01) or * (p < 0.05). Inline graphic: maximum temperature; Inline graphic: minimum temperature; Inline graphic: maximum relative humidity; Inline graphic: minimum relative humidity; Inline graphic: thermal oscillation; Inline graphic: maximum dew point temperature; Inline graphic: minimum dew point temperature; VPD: vapour pressure deficit; LW: Leaf wetness. Crosses indicate not applicable

Inline graphic (0.40–0.49), Inline graphic (0.47) and Inline graphic (0.53–0.66) significantly contributed to 71.92–87.07% of the total variance, as captured by principal component (PC) 1 (44.10–68.56%) and PC2 (18.51–34.53%) (Fig. 4). Inline graphic and Inline graphic were primarily associated with in-canopy (Fig. 4b–c, e–f), whereas Inline graphic was primarily associated with on-farm (Fig. 4a–b, d–f). The PC3 explained additional 11.33–18.66% of the variance, with the highest contributions from Inline graphic (0.61–0.82), LW (0.70) and Inline graphic (0.67), primarily associated with SILO and BOM (Fig. 4b–d, f). Fine-scale sources commonly showed mixed clustering, with on-farm sources predominating across orchards (Fig. S3a–b, d–f). Notably, distinct clustering was observed in Orchards 5 and 6 (Fig. S3e–f, Table 2), where data from in-canopy sources separated completely (100%) from other sources across all variables (precision = 1.00, recall = 1.00, F1-score = 1.00, overall accuracy = 0.50–0.56).

Fig. 4.

Fig. 4

Open in a new tab

Principal component analysis (PCA) biplot of 10 weather variables from fine-scale (in-canopy, IC and on-farm, OF) and coarse-scale (SILO and BOM) sources across six commercial macadamia orchards in (a–c) Central Queensland (Orchards 1–3) and (d–f) Northern Rivers (Orchards 4–6) across three seasons (2022–2024). Variables with highest loadings are indicated in the biplot. Inline graphic: minimum relative humidity; Inline graphic: maximum relative humidity; Inline graphic: minimum temperature; LW: leaf wetness; Inline graphic: maximum temperature; Inline graphic: thermal oscillation; PC: principal component

Table 2.

Clustering performance metrics for fine-scale (in-canopy and on-farm) and coarse-scale (SILO and BOM) sources of weather data for six commercial macadamia orchards in two distinct agroecological regions, Central Queensland (Orchards 1–3) and Northern Rivers (Orchards 4–6) in 2022 to 2024 seasons

Metric Weather data source cluster Central Queensland Northern Rivers
Orchard 1 Orchard 2 Orchard 3 Orchard 4 Orchard 5 Orchard 6
Precisiona In-canopy 0.24 0.80 0.00 0.12 1.00 1.00
On-farm 0.31 0.43 0.37 0.01 0.31 0.54
SILO 0.23 0.31 0.06 0.37 0.23 0.42
BOM 0.22 0.32 0.29 0.37 0.48 0.37
Recallb In-canopy 0.29 0.97 0.00 0.08 1.00 1.00
On-farm 0.07 0.12 0.10 0.01 0.22 0.23
SILO 0.21 0.34 0.07 0.53 0.28 0.52
BOM 0.36 0.44 0.39 0.43 0.52 0.49
F1-scorec In-canopy 0.26 0.88 0.00 0.09 1.00 1.00
On-farm 0.11 0.19 0.15 0.01 0.25 0.32
SILO 0.22 0.32 0.07 0.43 0.25 0.46
BOM 0.27 0.37 0.33 0.40 0.50 0.42
Accuracyd 0.23 0.47 0.14 0.26 0.50 0.56
Open in a new tab

aPrecision is the proportion of correctly assigned sources within a cluster relative to the total number of sources assigned to that cluster

bRecall is the proportion of correctly assigned sources within a cluster relative to the total number of true sources belonging to that cluster

cF1-score is the harmonic mean of precision and recall

dAccuracy is the proportion of correctly classified sources across all clusters

Higher values (range = 0–1) indicate greater differences in orchard microclimate conditions captured by a source relative to the other sources of each orchard

Predictive performance of weather variables based on sources

Predictive models were fitted for each weather source in the four selected orchards (Figs. 5 and 6). In the 2023 season, Botrytis conidia were not detected and were excluded from further analyses, whereas there were rapid increases in conidia dynamics in two Orchards (7 and 8) for Cladosporium (Fig. 5b1 and 2) and Neopestalotiopsis/Pestalotiopsis (Fig. 5c1 and 2). The relationships were best described by exponential models (Table 3). In the 2024 season, linear models best described the gradual accumulation of the fungal conidia in all orchards, except for Botrytis in Orchard 8, which followed an exponential trend (Fig. 5a4).

Fig. 5.

Fig. 5

Open in a new tab

Regression of airborne conidial abundance for (a1–4) Botrytis, (b1–4) Cladosporium and (c1–4) Neopestalotiopsis/Pestalotiopsis against vapour pressure deficit (VPD) from in-canopy (IC, yellow), on-farm (OF, green), SILO (red) and BOM (blue) sources of weather data for two commercial macadamia orchards in Central Queensland (Orchards 7 and 8) during two production seasons (2023 and 2024). Solid lines and points represent the fitted regression and observed data, respectively. VPD values represent biweekly mean corresponding to the spore trapping periods

Fig. 6.

Fig. 6

Open in a new tab

Regression of airborne conidial abundance for (a1–2) Botrytis, (b1–2) Cladosporium and (c1–2) Neopestalotiopsis/Pestalotiopsis against vapour pressure deficit (VPD) from in-canopy (IC, yellow), on-farm (OF, green), SILO (red) and BOM (blue) sources of weather data for two commercial macadamia orchards in Northern Rivers (Orchards 9 and 10) during the 2024 production season. No conidia data was collected during the 2023 season. Solid lines and points represent the fitted regression and observed data, respectively. VPD values represent biweekly mean corresponding to the spore trapping periods

Table 3.

Regression equations of Inline graphic(number of conidia + 1) (y) against vapour pressure deficit (VPD) (x) from fine-scale (in-canopy and on-farm) and coarse-scale (SILO and BOM) sources of weather data across four commercial macadamia orchards in two agroecologial regions during the 2023 and 2024 seasons

Season Region Orchard label Source of weather data Fungal genus Regression type
Botrytis Cladosporium Neopestalotiopsis/Pestalotiopsis
2023 Central Queensalnd Orchard 7 In-canopy NA Inline graphic Inline graphic Exponential
On-farm NA Inline graphic Inline graphic
SILO NA Inline graphic Inline graphic
BOM NA Inline graphic Inline graphic
Orchard 8 In-canopy NA Inline graphic Inline graphic Exponential
On-farm NA Inline graphic Inline graphic
SILO NA Inline graphic Inline graphic
BOM NA Inline graphic Inline graphic
2024 Central Queensalnd Orchard 7 In-canopy  y=-7.509x+15.745  y=-1.390x+8.632  y=-1.417x+9.601 Linear
On-farm  y=4.203x+1.151  y=0.418x+6.276  y=0.073x+7.537
SILO  y=-6.154x+11.095  y=-1.305x+7.931  y=-1.367x+8.922
BOM  y=-4.795x+9.747  y=-0.997x+7.627  y=-1.078x+8.635
Orchard 8 In-canopy Inline graphic  y=-1.284x+8.445  y=-1.132x+9.087 Exponential and linear
On-farm Inline graphic  y=0.264x+6.365  y=0.386x+7.107
SILO Inline graphic  y=-0.441x+7.044  y=-0.223x+7.691
BOM Inline graphic  y=-0.085x+6.699  y=0.107x+7.372
Northern Rivers Orchard 9 In-canopy  y=-20.114x+22.025  y=5.080x+1.449  y=5.748x+1.667 Linear
On-farm  y=-6.152x+8.581  y=11.097x-1.031  y=11.375x-0.411
SILO  y=-19.288x+18.344  y=7.918x+0.238  y=8.457x+0.650
BOM  y=-14.320x+15.513  y=7.809x-0.045  y=8.273x+0.398
Orchard 10 In-canopy  y=-28.911x+31.696  y=-6.214x+11.485  y=-5.397x+11.558 Linear
On-farm  y=-14.238x+15.694  y=2.091x+4.875  y=3.292x+4.907
SILO  y=-32.820x+30.345  y=-4.150x+9.123  y=-3.034x+9.099
BOM  y=-29.213x+28.797  y=-3.632x+8.881  y=-2.610x+8.887
Open in a new tab

No conidia data was collected for Orchards 9 and 10 during the 2023 season. NA not applicable

Model directionality varied among the weather sources, particularly for Cladosporium and Neopestalotiopsis/Pestalotiopsis in Orchards 7, 8 and 10 in the 2024 season, where on-farm-based models (Table 3) showed positive relationships, whereas in-canopy, SILO and BOM-based models showed negative relationships (Fig. 5b3-4, 5c3-4, 6b2 and 6c2), which may reflect differences in microclimatic humidity dynamics among weather sources. Subsequently, predictive performance differed significantly across pathogens, orchards and seasons (H(3) = 11.09, p = 0.01), with agreement, bias and prediction errors for airborne conidia showing wide ranges (Fig. S4). The PRI rankings of the weather sources were consistent across the tested Inline graphic values (Table S3). Therefore, PRI values for Inline graphic = 0.1, which applied the minimal penalty, were used to compare the robustness of weather source predictions. BOM-based models (PRI = 0.55) outperformed the in-canopy models (PRI = 0.29; p < 0.05) but were similar to on-farm (PRI = 0.52) and SILO (PRI = 0.55) models (p > 0.05; Fig. 7). In-canopy, on-farm and SILO models showed comparable predictive performance (p > 0.05).

Fig. 7.

Fig. 7

Open in a new tab

Probability (p) values from Dunn’s post hoc comparisons of predictive metrics for VPD-based models derived from fine-scale (in-canopy and on-farm) and coarse-scale (SILO and BOM) sources of weather data for flower blight conidia across four commercial macadamia orchards during two production seasons (2023 and 2024). Values in brackets indicate predictive robustness index (PRI) for each weather source. Statistically significant difference is indicated by p < 0.05 (*)

Discussion

This study identified variability in weather variables from fine (in-canopy and on-farm) and coarse (SILO and BOM) spatial scale sources and their accuracy in predicting flower blight fungal conidia dynamics. Inter-scale correlations ranged from weak to strong, and differences among sources were mainly associated with Inline graphic and Inline graphic. In-canopy sources showed more temporal stability and formed distinct clusters in 33% of orchards, whereas SILO and on-farm were more variable and often clustered with BOM sources. These source-specific characteristics was associated with differences in predictive performance, with on-farm, SILO and BOM sources outperforming in-canopy sources in predicting Botrytis, Cladosporium and Neopestalotiopsis/Pestalotiopsis conidia dynamics.

Weather-based prediction models are crucial for effective disease management (Fenu and Malloci 2021; Jensen et al. 2025). However, variations in microclimate due to different weather sources can be associated with the accuracy of these predictions (El Jarroudi et al. 2017; Rodríguez-Moreno et al. 2020). This study identified a more stable microclimate using in-canopy sources compared with on-farm, SILO and BOM sources, primarily characterised by reduced temporal variability in Inline graphic, Inline graphic and VPD. This stability possibly arises from a buffering effect of the canopy that moderates atmospheric fluctuations (Murakami et al. 2022; Spronken-Smith and Oke 1998). Depending on conditions such as cloud cover and sunlight, canopy temperatures can be 2 °C to 5 °C cooler and relative humidity 15% to 25% higher than ambient conditions outside the canopy (van Westreenen et al. 2020).

Variables from SILO and BOM were strongly correlated across orchards despite being 9 km to 35 km apart, reflecting SILO’s derivation from BOM point-based and gridded (0.05° × 0.05°) datasets (Zaman et al. 2024). Rainfall remained relatively source-specific and consistent with previous studies (Rivera et al. 2018; Shah et al. 2019). This is likely due to differences in local topography and precipitation events, as rainfall is highly location-specific and may not be fully captured by grids (El Jarroudi et al. 2017; Schleiss and Smith 2016). Conversely, in-canopy and on-farm sources, though located within the same orchard, showed lower correlations with better alignment in CQ than in NR. These differences may be associated with canopy density and light interception gradients (Schreiber et al. 2025; van Westreenen et al. 2020), where trees in NR orchards were denser, older canopies (> 8 m inter-row, > 12 years) whereas CQ orchards were younger and sparser (< 7 m inter-row, < 12 years). Inline graphic and LW showed the strongest and weakest spatial correlations among sources, respectively, consistent with previous studies (Ashraf et al. 1997; Gleason et al. 2008). It is important to note that spatial distance had a minimal effect on correlations between fine-scale and coarse-scale weather sources. These findings suggest that local microclimatic conditions may play a greater role than spatial distance alone in explaining the observed associations.

Inline graphic and Inline graphic were the primary contributors to variation among the sources, reflecting differences in microclimate, with in-canopy sources aligned with Inline graphic and on-farm and coarse-scale sources aligned with Inline graphic. This contrast has been demonstrated in previous studies (Ashraf et al. 1997; Murakami et al. 2022), and likely reflects canopy depth effects (De Frenne et al. 2019). In-canopy RH varies substantially across canopies, with temperature dominating RH variations in the upper canopy and water vapour pressure influencing RH in the lower canopy (van Westreenen et al. 2020). Consequently, Inline graphic is maintained primarily within the canopy (Schreiber et al. 2025), consistent with its stronger correlation with in-canopy sources. Clearer differences among the sources were observed from the clustering analysis, consistent with earlier research (Sönmez and Kömüşcü 2011; Yeo et al. 2020). In-canopy sources consistently formed fully or near-distinct clusters (> 97% agreement) in 50% of orchards, highlighting their distinctive spatial variability in microclimatic patterns, in contrast to BOM, which were more representative of overall orchard-level microclimate. Data from SILO and on-farm often clustered with BOM or in-canopy sources, reflecting their greater similarity in microclimates.

While on-farm and coarse-scale data often provide a reasonable approximation of orchard-level conditions, the distinctive orchard specific microclimates captured by in-canopy sources may be associated with pathogen inoculum dynamics and host colonisation rates, making them valuable for disease prediction (de Mol et al. 2018; Pangga et al. 2011). However, the contrasting VPD-conidia relationship directions observed between on-farm, and in-canopy and regional weather sources in some orchards may reflect differences in spatial representation of orchard-specific microclimatic conditions among weather sources. Exploratory analysis of VPD component variables suggested that more persistent Inline graphic conditions were captured by on-farm weather source, which showed lower Inline graphic variability compared with in-canopy, SILO and BOM weather sources. Differences in Inline graphic variability among the weather sources may have contributed to the contrasting VPD-conidia relationship directions between on-farm, and in-canopy, SILO and BOM weather sources, possibly due to the strong effect of moisture availability on conidial abundance (De Cól et al. 2024; Mata et al. 2026; Meno et al. 2022).

Conversely, coarse-scale weather data provided more consistent and generalisable predictions for flower blight fungal conidia, whereas fine-scale data exhibited more orchard-specific and variable predictive performance across orchards. This may suggest a trade-off between local microclimatic specificity and model generalisability. Coarse-scale weather sources may smooth orchard-specific microclimatic fluctuations and provide a more stable representation of weather conditions across orchards, which may contribute to more consistent predictive performance (De Cól et al. 2024). In contrast, fine-scale weather sources may capture localised weather variability that is biologically relevant within individual orchards but inconsistent across orchards, thereby increasing predictive variability and reducing model generalisability (Wang et al. 2000). The higher predictive performance of on-farm sources than in-canopy sources may be associated with canopy structural characteristics (Tivoli et al. 2013). However, attributes such as inter-row canopy diameter (Inline graphic = 0.19) and tree age (Inline graphic = 0.18) were only weakly associated with in-canopy predictive accuracy, indicating that structural attributes alone may not adequately explain the predictive performance of in-canopy weather data (Chang et al. 2007).

The relatively higher and more stable VPD conditions (mean = 1.29 kPa) observed within canopies were likely suboptimal for conidia development by the pathogens (O’neill et al. 1997; Prasannath et al. 2022a), contributing to the reduced conidia predictive performance of in-canopy weather data. However, the mean VPD of on-farm (0.87 kPa), SILO (0.94 kPa) and BOM (0.93 kPa) sources aligns closely with the VPD range for conidial development of all flower blight pathogens, potentially accounting for their higher and more consistent predictive performance relative to the in-canopy source. The relatively larger VPD range observed for the in-canopy source (1.13 kPa) compared with the on-farm (0.87 kPa), SILO (0.83 kPa) and BOM (0.78 kPa) sources may also underlie the greater inconsistency in the predictive performance of in-canopy VPD across orchards. These observed in-canopy VPD characteristics suggest that additional factors may be associated with microclimate beyond canopy diameter (Inline graphic = 0.04) and age (Inline graphic = 0.07). Cultivar characteristics such as leaf area index, together with sensor-related factors including height and cardinal orientation, are associated with vapour pressure dynamics (Murakami et al. 2022; Schreiber et al. 2025; van Westreenen et al. 2020). It should be noted that other weather variables from in-canopy sources, aside from VPD, may provide better predictions of flower blight fungal conidia, as demonstrated in previous study (Rodríguez-Moreno et al. 2020).

The comparable predictive performance between in-canopy, on-farm and SILO (< 4 km apart), in contrast to BOM (> 13 km apart), suggests that in-canopy and open-air microclimates share partial spatial similarity, consistent with Tobler (1970). Overall, these findings emphasise that the choice of weather data source is strongly associated with model performance, often more than variable selection or analytical methods, as highlighted by previous studies (Jensen et al. 2025; Shah et al. 2019). Accordingly, users of weather-driven disease models should carefully consider input data sources to ensure robust management decisions. Weather data from on-farm and coarse-scale sources may be more suitable for predictive models intended for broader applicability across sites, whereas models based on in-canopy weather data may be more appropriate for site-specific applications due to the highly localised nature of in-canopy microclimates.

Limited historical data from fine-scale stations restricted the study to 10 orchards with complete two- to three-season records. While the study demonstrated the impact of variability among weather data sources on disease prediction accuracy, this was limited to flower blight pathogens and univariate VPD-based regression models. Although flower blight dynamics are associated with multiple interacting environmental variables, the univariate modelling approach may not fully capture the complexity of disease dynamics. Additionally, LOO-CV does not preserve temporal ordering and may provide optimistic estimates when applied to time-dependent data. Therefore, future forecasting studies should consider blocked or rolling-origin validation procedures.

Conclusion

This study evaluated fine (in-canopy and on-farm) and coarse (regional: SILO and BOM) spatial scale weather sources across two major crop production regions in Australia to assess their reliability for weather-based disease prediction. Weather variables differed considerably, with the variation driven more strongly by data source than by spatial scale. The contrasting relationship directions among weather sources are particularly interesting and potentially represent one of the most novel findings of this study. Predictive performance of models for conidial abundance of the fungal pathogens Botrytis, Cladosporium and Neopestalotiopsis/Pestalotiopsis varied markedly among weather sources, with regional sources generally providing the most reliable input for disease forecasting. These findings provide a practical framework for improving weather-based disease management across cropping systems.

Supplementary Information

Below is the link to the electronic supplementary material.

Supplementary Material 1 (DOCX 702 KB) (703.1KB, docx)

Acknowledgements

Theophilus A. Mensah is a recipient of the University of Queensland Research Training Scholarship. This research was funded by Hort Innovation using the macadamia research and development levy and funds from the Australian Government, project no. MC21001. We are thankful to all the macadamia growers and managers in Central QLD and Northern Rivers, New South Wales who generously offered their orchards for use in this study.

Author contributions

Theophilus Abonyi Mensah: Methodology, Investigation, Formal analysis, Visualisation, Validation, Data curation, Writing – original draft, review and editing. Neil White: Investigation, Data curation, Resources, Validation, Visualisation, Writing - review and editing. Vivian Rincón-Flórez: Investigation, Resources, Writing - review and editing. Olufemi A. Akinsanmi: Conceptualisation, Methodology, Data curation, Resources, Validation, Visualisation, Writing - review and editing.

Funding

Open Access funding enabled and organized by CAUL and its Member Institutions

Data availability

All the datasets and code used for the analyses in this study are available in The University of Queensland Research Data Manager (UQRDM). Further inquiries can be directed to the corresponding authors.

Declarations

Competing interests

The authors declare that they have no conflict of interest.

Footnotes

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Contributor Information

Theophilus Abonyi Mensah, Email: t.mensah@uq.net.au.

Olufemi A. Akinsanmi, Email: o.akinsanmi@uq.edu.au

References

  1. Abuley IK, Hansen JG, Fariñas LM (2023) Evaluation of models based on a generic infection model for controlling early blight in potatoes. Crop Protect 169:106229. 10.1016/j.cropro.2023.106229 [DOI] [Google Scholar]
  2. Akinsanmi OA, Drenth A (2010) Spatial pattern and the effects of climatic factors on husk spot disease in macadamia. Australas Plant Pathol 39:125–131. 10.1071/AP09071Australasian [DOI] [Google Scholar]
  3. Akinsanmi OA, Drenth A (2017) Characterisation of husk rot in macadamia. Ann Appl Biol 170:104–115. 10.1111/aab.12320 [DOI] [Google Scholar]
  4. Akinsanmi OA, Miles AK, Drenth A (2007) Timing of fungicide applications for control of husk spot caused by Pseudocercospora macadamiae in macadamia. Plant Dis 91:1675–1681. 10.1094/PDIS-91-12-1675 [DOI] [PubMed] [Google Scholar]
  5. Allen RG, Pereira LS, Raes D, Smith M (1998) Crop evapotranspiration-Guidelines for computing crop water requirements-FAO Irrigation and drainage paper 56. FAO, Rome [Google Scholar]
  6. Ashraf M, Loftis JC, Hubbard KG (1997) Application of geostatistics to evaluate partial weather station networks. Agric Meteorol 84:255–271. 10.1016/S0168-1923(96)02358-1 [DOI] [Google Scholar]
  7. Basyir M, Nasir M, Suryati S, Mellyssa W (2017) Determination of nearest emergency service office using Haversine formula based on android platform. EMITTER Int J Eng Technol 5:270–278. 10.24003/emitter.v5i2.220 [DOI] [Google Scholar]
  8. BOM (2004) Some notes about daily weather observations. IDCJDW0000. Bureau of Meteorology, Australia [Google Scholar]
  9. Bright J (2025) Macadamia plant protection guide 2025–26. DPIRD, NSW, Australia [Google Scholar]
  10. Brinkhoff J, Robson AJ (2021) Block-level macadamia yield forecasting using spatio-temporal datasets. Agric Meteorol 303:108369. 10.1016/j.agrformet.2021.108369 [DOI] [Google Scholar]
  11. Chang KF, Ahmed HU, Hwang SF, Gossen BD, Howard RJ et al (2007) Impact of cultivar, row spacing and seeding rate on ascochyta blight severity and yield of chickpea. Can J Plant Sci 87:395–403. 10.4141/P06-067 [DOI] [Google Scholar]
  12. De Cól M, Coelho M, Del Ponte EM (2024) Weather-based logistic regression models for predicting wheat head blast epidemics. Plant Dis 108:2206–2213. 10.1094/PDIS-11-23-2513-RE [DOI] [PubMed] [Google Scholar]
  13. De Frenne P, Zellweger F, Rodríguez-Sánchez F, Scheffers BR, Hylander K et al (2019) Global buffering of temperatures under forest canopies. Nat Ecol Evol 3:744–749. 10.1038/s41559-019-0842-1 [DOI] [PubMed] [Google Scholar]
  14. de Mol F, Winter M, Gerowitt B (2018) Weather determines the occurrence of wheat stem base diseases in biogas cropping systems. Crop Protect 114:1–11. 10.1016/j.cropro.2018.07.017 [DOI] [Google Scholar]
  15. Deines JM, Patel R, Liang S-Z, Dado W, Lobell DB (2021) A million kernels of truth: Insights into scalable satellite maize yield mapping and yield gap analysis from an extensive ground dataset in the US Corn Belt. Remote Sens Environ 253:112174. 10.1016/j.rse.2020.112174 [DOI] [Google Scholar]
  16. Dunn OJ (1964) Multiple comparisons using rank sums. Technometrics 6:241–252. 10.2307/1266041 [DOI] [Google Scholar]
  17. El Jarroudi M, Kouadio L, El Jarroudi M, Junk J, Bock C et al (2017) Improving fungal disease forecasts in winter wheat: a critical role of intra-day variations of meteorological conditions in the development of Septoria leaf blotch. Field Crops Res 213:12–20. 10.1016/j.fcr.2017.07.012 [DOI] [Google Scholar]
  18. Fenu G, Malloci FM (2021) Forecasting plant and crop disease: an explorative study on current algorithms. Big Data Cogn Comput 5:2. 10.3390/bdcc5010002 [DOI] [Google Scholar]
  19. Gleason ML, Duttweiler KB, Batzer JC, Taylor SE, Sentelhas PC et al (2008) Obtaining weather data for input to crop disease-warning systems: leaf wetness duration as a case study. Sci Agric 65:76–87. 10.1590/S0103-90162008000700013 [DOI] [Google Scholar]
  20. Gonzalez-Dominguez E, Caffi T, Paolini A, Mugnai L, Latinović N et al (2022) Development and validation of a mechanistic model that predicts infection by Diaporthe ampelina, the causal agent of Phomopsis cane and leaf spot of grapevines. Front Plant Sci 13:872333. 10.3389/fpls.2022.872333 [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. González-Fernández E, Piña-Rey A, Fernández-González M, Aira M, Rodríguez-Rajo FJ (2020) Identification and evaluation of the main risk periods of Botrytis cinerea infection on grapevine based on phenology, weather conditions and airborne conidia. J Agric Sci 158:88–98. 10.1017/S0021859620000362 [DOI] [Google Scholar]
  22. Harris CR, Millman KJ, Van Der Walt SJ, Gommers R, Virtanen P et al (2020) Array programming with NumPy. Nature 585:357–362. 10.1038/s41586-020-2649-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Hastie T, Tibshirani R, Friedman J (2009) The elements of statistical learning: data mining, inference, and prediction. Springer, New York, USA [Google Scholar]
  24. Hunter JD (2007) Matplotlib: a 2D graphics environment. Comput Sci Eng 9:90–95. 10.1109/MCSE.2007.55 [DOI] [Google Scholar]
  25. Hussain MM, Mahmud I (2019) pyMannKendall: a python package for non parametric Mann Kendall family of trend tests. J Open Source Softw 4:1556. 10.21105/joss.01556 [DOI] [Google Scholar]
  26. Ibrahim RA, Elsheikh AH, Abd Elaziz ME, Al-qaness MAA (2022) Basics of artificial neural networks. In: Elsheikh AH, Elaziz MEA (eds) Artificial neural networks for renewable energy systems and real-world applications. Academic, London, pp 1–10. 10.1016/B978-0-12-820793-2.00002-1 [DOI] [Google Scholar]
  27. INC (2025) Nuts & dried fruits statistical yearbook. INC, Reus, Spain [Google Scholar]
  28. Jadon A, Patil A, Jadon S (2024) A comprehensive survey of regression-based loss functions for time series forecasting. In: Sharma N, Goje AC, Chakrabarti A, Bruckstein AM (eds) Data management, analytics & innovation. Springer, Singapore, pp 117–147. 10.1007/978-981-97-3245-6_9 [DOI] [Google Scholar]
  29. Jensen A, Brown P, Groves K, Morshed A (2025) Next generation crop protection: a systematic review of trends in modelling approaches for disease prediction. Comput Electron Agric 234:110245. 10.1016/j.compag.2025.110245 [DOI] [Google Scholar]
  30. Kendall MG (1948) Rank correlation methods. Charles Griffin, London [Google Scholar]
  31. Kruskal WH, Wallis WA (1952) Use of ranks in one-criterion variance analysis. J Am Stat Assoc 47:583–621. 10.2307/2280779 [DOI] [Google Scholar]
  32. Lawrence MG (2005) The relationship between relative humidity and the dewpoint temperature in moist air: a simple conversion and applications. Bull Am Meteorol Soc 86:225–234. 10.1175/BAMS-86-2-225 [DOI] [Google Scholar]
  33. Lin L, I-K (1989) A concordance correlation coefficient to evaluate reproducibility. Biometrics 45:255–268. 10.2307/2532051 [DOI] [PubMed] [Google Scholar]
  34. Magarey RD, Sutton TB, Thayer CL (2005) A simple generic infection model for foliar fungal plant pathogens. Phytopathology 95:92–100. 10.1094/PHYTO-95-0092 [DOI] [PubMed] [Google Scholar]
  35. Maleki F, Muthukrishnan N, Ovens K, Reinhold C, Forghani R (2020) Machine learning algorithm validation: from essentials to advanced applications and implications for regulatory certification and deployment. Neuroimag Clin N Am 30:433–445. 10.1016/j.nic.2020.08.004 [DOI] [PubMed] [Google Scholar]
  36. Mann HB (1945) Nonparametric tests against trend. Econometrica 13:245–259. 10.2307/1907187 [DOI] [Google Scholar]
  37. Manstretta V, Rossi V (2015) Effects of weather variables on ascospore discharge from Fusarium graminearum perithecia. PLoS ONE 10:e0138860. 10.1371/journal.pone.0138860 [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Mata CI, Echeverria G, López-Cristoffanini C, Smets T, Vanwalleghem T et al (2026) Quantification of Botrytis cinerea airborne conidia in blackberry and raspberry greenhouses and rain covers by qPCR and its latent presence in flowers and fruits. J Plant Dis Protect 133:1–10. 10.1007/s41348-026-01271-w [DOI] [Google Scholar]
  39. McKinney W (2010) Data structures for statistical computing in Python. In: van der Walt S, Millman J (eds) Proceedings of the 9th Python in Science Conference. Austin, pp 56–61. 10.25080/Majora-92bf1922-00a [DOI]
  40. Meno L, Escuredo O, Abuley IK, Seijo MC (2022) Importance of meteorological parameters and airborne conidia to predict risk of Alternaria on a potato crop ambient using machine learning algorithms. Sensors 22:7063. 10.3390/s22187063 [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Mohankumar V, Dann EK, Akinsanmi OA (2023) Seasonal dynamics of inoculum of Botryosphaeriaceae in macadamia orchards in Australia. Plant Pathol 72:1160–1170. 10.1111/ppa.13730 [DOI] [Google Scholar]
  42. Molitor D, Baus O, Didry Y, Junk J, Hoffmann L et al (2020) BotRisk: simulating the annual bunch rot risk on grapevines (Vitis vinifera L. cv. Riesling) based on meteorological data. Int J Biometeorol 64:1571–1582. 10.1007/s00484-020-01938-5 [DOI] [PubMed] [Google Scholar]
  43. Murakami M, Nunes Ramos F, Durand M, Ashton R, Batke SP (2022) Quantification and variation of microclimatic variables within tree canopies-considerations for epiphyte research. Front Glob Change 5:828725. 10.3389/ffgc.2022.828725 [DOI] [Google Scholar]
  44. Navi SS, Bandyopadhyay R, Reddy RK, Thakur RP, Yang X-B (2005) Effects of wetness duration and grain development stages on sorghum grain mold infection. Plant Dis 89:872–878. 10.1094/PD-89-0872 [DOI] [PubMed] [Google Scholar]
  45. O’neill TM, Shtienberg D, Elad Y (1997) Effect of some host and microclimate factors on infection of tomato stems by Botrytis cinerea. Plant Dis 81:36–40. 10.1094/PDIS.1997.81.1.36 [DOI] [PubMed] [Google Scholar]
  46. Pangga IB, Hanan J, Chakraborty S (2011) Pathogen dynamics in a crop canopy and their evolution under changing climate. Plant Pathol 60:70–81. 10.1111/j.1365-3059.2010.02408.x [DOI] [Google Scholar]
  47. Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B et al (2011) Scikit-learn: Machine learning in Python. J Mach Learn Res 12:2825–2830 [Google Scholar]
  48. Perria R, Ciofini A, Petrucci WA, D’Arcangelo MEM, Valentini P et al (2022) A study on the efficiency of sustainable wine grape vineyard management strategies. Agronomy 12:392. 10.3390/agronomy12020392 [DOI] [Google Scholar]
  49. Prasannath K, Shivas RG, Galea VJ, Akinsanmi OA (2021) Neopestalotiopsis species associated with flower diseases of Macadamia integrifolia in Australia. J Fungi 7:771. 10.3390/jof7090771 [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Prasannath K, Galea VJ, Akinsanmi OA (2022a) Influence of climatic factors on dry flower, grey and green mould diseases of macadamia flowers in Australia. J Appl Microbiol 132:1291–1306. 10.1111/jam.15241 [DOI] [PubMed] [Google Scholar]
  51. Prasannath K, Galea VJ, Akinsanmi OA (2022b) Sources, detection, and inoculum quantification of flower blight pathogens in macadamia. Phytopathology 112:2151–2158. 10.1094/PHYTO-08-21-0365-R [DOI] [PubMed] [Google Scholar]
  52. Prasannath K, Galea VJ, Akinsanmi OA (2023) Diversity and pathogenicity of species of Botrytis, Cladosporium, Neopestalotiopsis and Pestalotiopsis causing flower diseases of macadamia in Australia. Plant Pathol 72:881–899. 10.1111/ppa.13707 [DOI] [Google Scholar]
  53. Rivera JA, Marianetti G, Hinrichs S (2018) Validation of CHIRPS precipitation dataset along the Central Andes of Argentina. Atmos Res 213:437–449. 10.1016/j.atmosres.2018.06.023 [DOI] [Google Scholar]
  54. Rodríguez-Moreno VM, Jiménez‐Lagunes A, Estrada‐Avalos J, Mauricio‐Ruvalcaba JE, Padilla‐Ramírez JS (2020) Weather‐data‐based model: an approach for forecasting leaf and stripe rust on winter wheat. Meteorol Appl 27:e1896. 10.1002/met.1896 [DOI] [Google Scholar]
  55. Rowlandson T, Gleason M, Sentelhas P, Gillespie T, Thomas C et al (2015) Reconsidering leaf wetness duration determination for plant disease management. Plant Dis 99:310–319. 10.1094/PDIS-05-14-0529-FE [DOI] [PubMed] [Google Scholar]
  56. Sadyś M, Kaczmarek J, Grinn-Gofron A, Rodinkova V, Prikhodko A et al (2018) Dew point temperature affects ascospore release of allergenic genus Leptosphaeria. Int J Biometeorol 62:979–990. 10.1007/s00484-018-1500-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Schleiss M, Smith JA (2016) Two simple metrics for quantifying rainfall intermittency: the burstiness and memory of interamount times. J Hydrometeorol 17:421–436. 10.1175/JHM-D-15-0078.1 [DOI] [Google Scholar]
  58. Schreiber J, Pouska V, Macek P, Thom D, Bässler C (2025) Effects of canopy-mediated microclimate and object characteristics on deadwood temperature. Agric Meteorol 362:110378. 10.1016/j.agrformet.2024.110378 [DOI] [Google Scholar]
  59. Seabold S, Perktold J (2010) Statsmodels: econometric and statistical modeling with python. In: van der Walt S, Millman J (eds) Proceedings of the 9th Python in Science Conference. Austin, pp 92–96. 10.25080/Majora-92bf1922-011 [DOI]
  60. Sen PK (1968) Estimates of the regression coefficient based on Kendall’s tau. J Am Stat Assoc 63:1379–1389. 10.2307/2285891 [DOI] [Google Scholar]
  61. Sentelhas PC, Dalla Marta A, Orlandini S, Santos EA, Gillespie TJ et al (2008) Suitability of relative humidity as an estimator of leaf wetness duration. Agric Meteorol 148:392–400. 10.1016/j.agrformet.2007.09.011 [DOI] [Google Scholar]
  62. Shah DA, De Wolf ED, Paul PA, Madden LV (2019) Functional data analysis of weather variables linked to Fusarium head blight epidemics in the United States. Phytopathology 109:96–110. 10.1094/PHYTO-11-17-0386-R [DOI] [PubMed] [Google Scholar]
  63. Shapiro SS, Wilk MB (1965) An analysis of variance test for normality (complete samples). Biometrika 52:591–611. 10.2307/2333709 [DOI] [Google Scholar]
  64. Shtienberg D (2013) Will decision-support systems be widely used for the management of plant diseases? Annu Rev Phytopathol 51:1–16. 10.1146/annurev-phyto-082712-102244 [DOI] [PubMed] [Google Scholar]
  65. Sönmez İ, Kömüşcü AÜ (2011) Reclassification of rainfall regions of Turkey by K-means methodology and their temporal variability in relation to North Atlantic Oscillation (NAO). Theor Appl Climatol 106:499–510. 10.1007/s00704-011-0449-1 [DOI] [Google Scholar]
  66. Spearman C (1904) The proof and measurement of association between two things. Am J Psychol 15:72–101. 10.2307/1412159 [DOI] [PubMed] [Google Scholar]
  67. Spronken-Smith RA, Oke TR (1998) The thermal regime of urban parks in two cities with different summer climates. Int J Remote Sens 19:2085–2104. 10.1080/014311698214884 [DOI] [Google Scholar]
  68. Ssegane H, Tollner E, Mohamoud Y, Rasmussen T, Dowd J (2012) Advances in variable selection methods I: Causal selection methods versus stepwise regression and principal component analysis on data of known and unknown functional relationships. J Hydrol 438:16–25. 10.1016/j.jhydrol.2012.01.008 [DOI] [Google Scholar]
  69. Taskesen E (2020) pca: a python package for principal component analysis
  70. Theil H (1992) A rank-invariant method of linear and polynomial regression analysis. In: Raj B, Koerts J (eds) Henri Theil’s contributions to economics and econometrics Advanced studies in theoretical and applied econometrics. Dordrecht, pp 345–381. 10.1007/978-94-011-2546-8_20 [DOI]
  71. Tivoli B, Calonnec A, Richard B, Ney B, Andrivon D (2013) Current knowledge on plant/canopy architectural traits that reduce the expression and development of epidemics. Eur J Plant Pathol 135:471–478. 10.1007/s10658-012-0066-6 [DOI] [Google Scholar]
  72. Tobler WR (1970) A computer movie simulating urban growth in the Detroit region. Econ Geogr 46:234–240. 10.2307/143141 [DOI] [Google Scholar]
  73. van Westreenen A, Zhang N, Douma JC, Evers JB, Anten NPR et al (2020) Substantial differences occur between canopy and ambient climate: quantification of interactions in a greenhouse-canopy system. PLoS ONE 15:e0233210. 10.1371/journal.pone.0233210 [DOI] [PMC free article] [PubMed] [Google Scholar]
  74. Vélez-Pereira AM, De Linares C, Canela M-A, Belmonte J (2019) Logistic regression models for predicting daily airborne Alternaria and Cladosporium concentration levels in Catalonia (NE Spain). Int J Biometeorol 63:1541–1553. 10.1007/s00484-019-01767-1 [DOI] [PubMed] [Google Scholar]
  75. Virtanen P, Gommers R, Oliphant TE, Haberland M, Reddy T et al (2020) SciPy 1.0: fundamental algorithms for scientific computing in Python. Nat Methods 17:261–272. 10.1038/s41592-019-0686-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  76. Vujović Ž (2021) Classification model evaluation metrics. Int J Adv Comput Sci Appl 12:599–606. 10.14569/IJACSA.2021.0120670 [DOI] [Google Scholar]
  77. Wang E, Ryley M, Meinke H (2000) Prediction of sorghum downy mildew risk in Australia using daily weather data. Australas Plant Pathol 29:108–119. 10.1071/AP00019 [DOI] [Google Scholar]
  78. Waskom ML (2021) Seaborn: statistical data visualization. J Open Source Softw 6:3021. 10.21105/joss.03021 [DOI] [Google Scholar]
  79. Wedd R, Alves O, de Burgh-Day C, Down C, Griffiths M et al (2022) ACCESS-S2: the upgraded Bureau of Meteorology multi-week to seasonal prediction system. J South Hemisph Earth Syst Sci 72:218–242. 10.1071/ES22026 [DOI] [Google Scholar]
  80. Wilcoxon F (1945) Individual comparisons by ranking methods. Biometrics Bull 1:80–83. 10.2307/3001968 [DOI] [PubMed] [Google Scholar]
  81. Yeo MH, Frei A, Gelda RK, Owens EM (2020) A stochastic weather model for generating daily precipitation series at ungauged locations in the Catskill Mountain region of New York state. Int J Climatol 40:687–705. 10.1002/joc.6230 [DOI] [Google Scholar]
  82. Yue S, Pilon P, Cavadias G (2002) Power of the Mann–Kendall and Spearman’s rho tests for detecting monotonic trends in hydrological series. J Hydrol 259:254–271. 10.1016/S0022-1694(01)00594-7 [DOI] [Google Scholar]
  83. Zaman W, Evans A, Frost A, Trewin B, Jones D (2024) A new gridded daily temperature analysis scheme for Australia. Bureau of Meteorology, Australia [Google Scholar]
  84. Zhao X, Liu Y, Huang Z, Li G, Zhang Z et al (2024) Early diagnosis of Cladosporium fulvum in greenhouse tomato plants based on visible/near-infrared (VIS/NIR) and near-infrared (NIR) data fusion. Sci Rep 14:20176. 10.1038/s41598-024-71220-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  85. Zhou H, Zhang S, Xie M, Li X, Chen Y et al (2024) A stacking-based model for the spread of Botryosphaeria laricina. J Res 35:131. 10.1007/s11676-024-01768-w [DOI] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Material 1 (DOCX 702 KB) (703.1KB, docx)

Data Availability Statement

All the datasets and code used for the analyses in this study are available in The University of Queensland Research Data Manager (UQRDM). Further inquiries can be directed to the corresponding authors.

Articles from International Journal of Biometeorology are provided here courtesy of Springer

RESOURCES