How Climate Change Will Shape Pesticide Application in Quebec’s Golf Courses: Insights with Deep Learning Based on Assessing CMIP5 and CMIP6

Abstract

The accelerating impact of climate change on golf course conditions has led to a significant increase in pesticide dependency, underscoring the importance of innovative management strategies. The shift from Coupled Model Intercomparison Project Phase 5 (CMIP5) to the latest CMIP6 phase has drawn the attention of professionals, including engineers, decision makers, and golf course managers. This study evaluates how climate projections from CMIP6, using Canadian Earth System Models (CanESM2 and CanESM5), impact pesticide application trends on Quebec’s golf courses. Through the comparison of temperature and precipitation projections, it was found that a more substantial decline in precipitation is exhibited by CanESM2 compared to CanESM5, while the latter projects higher temperature increases. A comparison between historical and projected pesticide use revealed that, in most scenarios and projected periods, the projected pesticide use was substantially higher, surpassing past usage levels. Additionally, in comparing the two climate change models, CanESM2 consistently projected higher pesticide use across various scenarios and projected periods, except for RCP2.6, which was 27% lower than SSP1-2.6 in the second projected period (PP2). For all commonly used pesticides, the projected usage levels in every projected period, according to climate change models, surpass historical levels. When comparing the two climate models, CanESM5 consistently forecasted greater pesticide use for fungicides, with a difference ranging from 65% to 222%, and for herbicides, with a difference ranging from 114% to 247%, across all projected periods. In contrast, insecticides, growth regulators, and rodenticides displayed higher AAIR values in CanESM2 during PP1 and PP3, showing a difference of 28% to 35.6%. However, CanESM5 again projected higher values in PP2, with a difference of 1.5% to 14%.

1. Introduction

Climate change (CC) refers to alterations in the climate occurring over time, driven by both natural processes and human-induced factors [1]. These changes affect various sectors, including agriculture, biodiversity, urban planning, and recreation, such as golf courses [2,3,4]. The increasing temperatures, erratic weather patterns, and extreme events like hurricanes and wildfires [5,6] resulting from climate change affect ecosystems and pesticide use on golf courses [7,8]. As global temperatures rise, golf courses experience increased pest infestations [9], leading to higher reliance on pesticides and raising environmental concerns such as water pollution and biodiversity loss [10,11]. There is an urgent need to develop sustainable management practices to mitigate these impacts.

Golf course management requires substantial effort and resources, often relying on pesticides to maintain the desired turf quality [12]. In Québec, golf course managers have been legally required since 2003 to submit a “pesticide reduction plan” (PRP) to the Ministry of Environment. These PRPs outline the specific quantities of pesticides used and include objectives for reducing pesticide usage over time [13]. However, weather conditions significantly influence pest occurrences, making it challenging to predict pesticide usage and set realistic reduction goals.

Machine learning (ML) models have been increasingly applied in predicting pesticide usage in agriculture. Several studies have highlighted the effectiveness of ML models, such as K-nearest neighbor, random forests (RFs), and convolutional neural networks (CNNs), in optimizing pesticide application [14,15,16,17]. In particular, Grégoire et al. [16] developed a hybrid machine learning model (RF-SVM-GOA) to predict pesticide use on Quebec golf courses, showing high accuracy compared to individual methods. To address the constraints of the prior research, Grégoire et al. [17] introduced a more advanced model combining CNN with RF. Key variables like total precipitation and average temperature are identified as the most influential factors in forecasting pesticide use.

Climate change models, including the Coupled Model Intercomparison Project (CMIP) phases, are critical tools in predicting climate impacts. The CMIP5 and CMIP6 models, using Canadian Earth System Models (CanESM2 and CanESM5), provide valuable data for understanding future climate scenarios [18,19,20,21,22]. The main differences between CMIP5 and CMIP6 lie in their use of Representative Concentration Pathways (RCPs) and Socioeconomic Pathways (SSPs) for projecting climate changes [10,21,23]. Studies comparing CMIP5 and CMIP6 across various regions, including Canada, have demonstrated notable differences in climate projections, particularly in terms of temperature anomalies and precipitation changes [24,25,26,27].

Multiple investigations have contrasted CMIP5 and CMIP6 across various geographical areas. Sobie et al. [24] compared climate change model projections focusing on Canada. The researchers noted that the CMIP6 ensemble projections demonstrate a slightly more pronounced scaling pattern for temperature anomalies in northern Canada. This behavior is also observed for specific indices related to extreme and moderate events. Hamed et al. [25] compared CMIP5 and CMIP6 models over Southeast Asia (SEA) using 13 global climate models (GCMs) from both CMIP simulations. The authors revealed that CMIP6 exhibited comparable performance to CMIP5 across most aspects, with the exception of rainfall. Rainfall witnessed a more substantial rise, whereas temperature experienced a diminished increase under CMIP6 as opposed to CMIP5. Given the findings from CMIP6, the authors advised contemplating new climate change mitigation policies explicitly tailored to the SEA region. In a study by Li et al. [26], the impacts of worldwide climate change on both the agricultural yield and hydrological cycle were explored, specifically within a heavily irrigated management context. Their investigation revealed that by the conclusion of the 21st century, the anticipated alterations in the hydrological cycle demonstrated resemblances under the high-emission scenarios of both CMIP5 and CMIP6.

Furthermore, the researchers indicated that when considering two scenarios from CMIP5 and CMIP6, the evolving patterns in the daily biomass and leaf area index of winter wheat and summer maize during their respective growing seasons exhibited parallel trends. Ebtehaj and Bonakdari [10] carried out an extensive analysis comparing the spatio-temporal variability of long-term flood susceptibility between CMIP5 and CMIP6 models. The researchers determined that the computed flow discharge distributions from both CMIP5 and CMIP6 models were nearly indistinguishable, showing insignificant variations for monthly and seasonal peak discharges. While the provided maps highlight notable disparities in monthly and seasonality peak discharge across various stations, it remains challenging to definitively conclude which model offers a more pronounced monthly peak discharge across all scenarios (stations and months). Therefore, considering the differing model outputs in various regions, it is advisable to examine both CMIP5 and CMIP6 models when conducting studies to capture a comprehensive view of climate impacts [27]. The persistence of CMIP5 models despite the advancements in CMIP6 can be attributed to several reasons. Firstly, the gradual transition between CMIP phases necessitates time due to ongoing CMIP6 development [28], while CMIP5 models are widely accepted. Research projects and evaluations are based on CMIP5, demanding the shift. Secondly, CMIP5 models serve as a baseline for historical climate comparisons, offering reliable data for past climate assessments. For long-term research requiring consistent data, CMIP5’s adoption is favored to avoid mid-project inconsistencies [29]. Comparative analysis across CMIP phases aids model comprehension. Researchers may opt for CMIP5 due to specific research inquiries or historical data compatibility. Resource constraints and ongoing CMIP6 evaluation further influence the choice, emphasizing the nuanced decision-making process between CMIP5 and CMIP6 models.

The main objective of this study was to examine how projected changes in temperature and precipitation, based on the CMIP5 and CMIP6 models using CanESM2 and CanESM5 climate simulations, will influence pesticide usage on Quebec’s golf courses. The actual active ingredient rate (AAIR) (kg/100 m²) was projected using three different climate change scenarios based on both CanESM2 and CanESM5 models across three distinct projected periods. Using the calculated AAIR, total pesticide use was calculated (kg). The present study aimed to (1) compare the projected climate variables (temperature and precipitation) between CanESM2 (CMIP5) and CanESM5 (CMIP6) models; (2) quantify the impact of these projected climate changes on pesticide application rates, particularly in terms of the active actual ingredient rate (AAIR) across three projected periods (2023–2048, 2049–2074, and 2075–2100); and (3) investigate differences in pesticide use projections under different climate change scenarios (RCP and SSP pathways) from both CanESM2 and CanESM5.

The tested hypothesis was that the climate projections from the CanESM5 model (CMIP6) would lead to significantly higher pesticide usage on Quebec’s golf courses than the CanESM2 model (CMIP5) due to increased temperatures and reduced precipitation. Additionally, it was expected that the projected increases in pesticide use would be particularly pronounced for certain pesticide categories (such as fungicides and herbicides) and during the later projected periods (2075–2100). The study also hypothesized that CanESM5 would project a more intense impact on pesticide use than CanESM2 due to its higher projected temperature increases and more moderate precipitation decline.

The remainder of this paper is structured as follows: Section 2 describes the materials and methods, including the study area, pesticide use calculation, and climate change models, while Section 3 presents the results, covering meteorological variables and pesticide use. Section 4 provides a detailed discussion, and Section 5 concludes the paper with the conclusions.

2. Materials and Methods

2.1. Study Area

In Québec, Canada, golf course managers have been legally required since 2003 to submit a pesticide reduction management plan to the Ministry of the Environment every three years. These plans must include the golf course’s identification details, total area, quantity of pesticides applied during the preceding three years, measures to prevent pesticide migration, strategies for non-pesticide pest control, and reduction objectives for the forthcoming three-year period.

These plans must be approved by a certified agronomist licensed by the Ordre des Agronomes du Québec. This regulatory framework allowed Québec’s Ministry of the Environment to compile a database of pesticide applications from 380 golf courses between 2003–2018. Figure S1 in the Supplementary Materials section shows the geographic locations of these golf courses, with most located in densely populated southern regions and a few scattered across the less populated eastern and western areas of the province.

2.2. Pesticides Use Calculation

Employing the dataset encompassing all the golf courses depicted in Figure S1, Grégoire et al. [17] introduced a novel approach for computing active actual ingredient rates (AAIR). The calculation is performed according to the following method:

AAIR = (Q × C)/(Treated Area)

(1)

where C is the concentration of the active ingredient in the applied pesticide (%), Q is the pesticide quantity applied (kg), and the Treated Area is presented in square meters (m²). The pesticide density was assumed to be equivalent to 1 kg/L, enabling easy conversion of liters into kilograms for pesticides that are distributed in liquid form.

To assess the impact of climate change on pesticide utilization in golf courses, the methodologies developed by Grégoire et al. [17] with a data bank of more than 40,000 samples, which integrate random forest and convolutional neural networks (RFCNN), were employed to predict the active actual ingredient rate (AAIR). This technique integrates the random forest and convolutional neural network (RFCNN). Various inputs, including coordinates, pesticide type, number of holes, average temperature, and total precipitation, are fed into the model to compute the AAIR, as presented in Figure S2 in the Supplementary Materials section. Data on average temperature and total precipitation were obtained from the Google Earth engine, while constant factors like coordinates and the number of holes remained unchanged. For each pesticide type applied in the past (i.e., fungicides (F), herbicides (H), insecticides (I), growth regulators (RC), rodenticides (Ro), and others (A)), projected total precipitation and average temperature derived from climate change models were utilized to compute the AAIR. The AAIR (kg/100 m²) was converted into total pesticide usage (kg) to calculate total pesticide use. Furthermore, using historical data, a coefficient was determined for each type of pesticide such that the sum of all coefficients equaled one. These coefficients were calculated based on the total pesticide usage in historical data. These coefficients were then multiplied by the projected pesticide use for each type.

2.3. Climate Change Models

To project temperature and precipitation patterns up to 2100, the outcomes from the second and fifth generations of the Canadian Earth System Model versions (CanESM2 and CanESM5) were employed. CanESM2 contributes to CMIP5, while CanESM5 is a part of CMIP6. The second iteration of the Earth System Model, CanESM2 [30,31,32], represents the fourth generation of the coupled global climate model developed by the Canadian Centre for Climate Modelling and Analysis (CCCma) within Environment and Climate Change Canada. CanESM2’s contribution was significant as part of the Canadian modeling community’s involvement in the IPCC Fifth Assessment Report (AR5). CanESM5, the latest version of the model, stands as an enhanced iteration of CanESM2, designed for CMIP6, as detailed by Swart et al. [33].

Global climate models currently exhibit spatial resolutions spanning about 100 to 300 km, which are generally coarse and lack the ability to capture intricate regional nuances influencing nearby weather patterns. To address this limitation, both statistical and dynamic downscaling techniques come into play [34]. These methodologies offer localized climate data with finer details while maintaining the border climate trends derived from global models. This study employs statistical downscaling, a widely recognized approach in climate change research [7,35,36,37,38]. Statistical downscaling methods encompass the utilization of empirical statistical frameworks to describe the interconnections between extensive-scale climatic data from global models and localized effects, such as the variations in temperature or rainfall at a specific site. Statistical downscaling carries key benefits, primarily rooted in aligning statistical models with observations [35]. This eradicates biases during their implementation in current climate simulations.

Additionally, spatial extent and precision are dictated by the scope and density of the observation network, as opposed to computational capacity, as seen in global and regional climate models [36]. It is important to note that the entirety of the computations pertaining to the downscaling of CanESM2 and CanEM5 in the current study were executed by Environment and Natural Resources Canada (https://climate-scenarios.canada.ca/?page=CanDCS6-data accessed on 30 May 2023). To analyze future pesticide usage, all projected pesticide uses are assessed across three defined periods: PP1 (2023–2048), PP2 (2049–2074), and PP3 (2075–2100). Indeed, the climate change models (CanESM2 and CanESM5) were applied from 2023 to 2100, resulting in the analysis of a total of 2,134,080 samples (78 years × 12 months × 2 models × 3 scenarios × 380 golf courses).

It is important to note that the spatial analysis and visualization of the results were conducted using ArcGIS, enabling detailed mapping of temperature, precipitation, and pesticide usage patterns, while all calculations were performed using MATLAB 9.9.

3. Results

3.1. Temperature and Precipitation

Figure 1 illustrates a spatial comparison between historical data and climate change projections for total precipitation and average temperature during the province of Quebec golf season (May to November). The meteorological variables considered in this study, namely average temperature and total precipitation, are integral components of the hybrid model developed by Grégoire et al. [17]. This model serves the purpose of estimating pesticide usage on golf courses. The hybrid model’s integration of climatic data, precisely average temperature and total precipitation, provides valuable insights into potential changes in future pesticide requirements for golf courses. It is important to emphasize that the presented values for these meteorological variables are averages derived from various scenarios employed for the CanESM2 (RCP2.6, RCP4.5, and RCP8.5) and CanESM5 (SSP1-2.6, SSP2-4.5, and SSP5-8.5) climate models.

A comparison was made between historical precipitation data and its projected values based on the CanESM2 and CanESM5 models, revealing that the highest levels of both historical and projected precipitation are concentrated around Quebec City (Figure 1). Interestingly, the observed and predicted data show that higher total precipitation amounts occur in the area located north of Quebec City. The minimum historical total precipitation occurred in the western part of the study area (Figure 1). The results from CanESM5 were consistent with historical precipitation patterns, showing minimal precipitation in the western part of the region (Figure 1). However, when the projected precipitation based on the CanESM2 model was examined, it was noted that the minimum precipitation shifted towards the southern areas, deviating from historical patterns (Figure 1). Notably, as one moves from the western to the central areas, a decrease in precipitation intensity was projected by CanESM5 (Figure 1) compared to historical data (Figure 1). In the southern portion of the study area, the distribution of precipitation in all three scenarios—historical, CanESM2, and CanESM5—is quite similar. However, it is worth noting that historical precipitation records showed values approximately 30% higher than those projected by CanESM2 and roughly 20% higher than those projected by CanESM5 (Figure 2A).

In scrutinizing the average season temperature (AST), the historical record reveals a minimum AST of 6.17 °C (Figure 1), while corresponding values for CanESM2 and CanESM5 stand at 9.26 °C (Figure 1) and 10.79 °C (Figure 1), respectively. The presence of this disparity directs attention to the projections of AST levels for the future, as dictated by the CanESM2 and CanESM5 models, indicating an increase of 50.08% and 74.87% (respectively) in forthcoming AST. The comparison of these temperature metrics further illustrates an average difference of approximately 10.9% between the two climate models. Furthermore, the highest historical AST on record registers at 12.11 °C (Figure 1). In contrast, the corresponding values for the CanESM2 and CanESM5 models amount to 15.26 °C (Figure 1) and 16.55 °C (Figure 1), respectively. This incongruity underscores the foreseeable escalation in future AST levels foreseen by the CanESM2 and CanESM5 models, signifying an increase of 26.01% and 36.66% (respectively) in the predicted maximum AST. Additionally, when these AST magnitudes are placed side by side, it becomes evident that CanESM5 projects an 8.45% increase in maximum AST compared to its CanESM2 counterpart. As per the depictions in the historical and projected AST maps, it is discernible that the highest AST is situated in the southern regions of the study area and the southwestern vicinity of Quebec City. Advancing towards the northern regions, a consistent decline in temperature is evident, as substantiated by both historical and projected datasets. Specifically, for Quebec City, the historical AST figure stands at 8.52 °C, whereas CanESM2 and CanESM5 envision this variable to reach 13.11 °C and 14.41 °C, respectively, in their projections. In conclusion, as per the historical and projected data analysis concerning the average season temperature (AST), it becomes evident that a significant surge in temperature (50.08% and 74.87% for CanESM2 and CanESM5, respectively) is anticipated across all sections of the study area. This temperature increase is poised to yield noteworthy ramifications for disease and insect pressure on golf courses within this region, thus affecting pesticide use.

Figure 2 presents the distribution of total precipitation and AST during the golf season in the province of Quebec (May to November). This presentation encompasses a range of climate change models and scenarios. The boxplots were applied to visually summarize the central tendency, spread, and skewness of the total precipitation and AST and identify potential outliers. A comparison between two different climate change models, CanESM2 and CanESM5, reveals significant differences. CanESM2 assesses future climate change impacts based on various greenhouse gas emission trajectories, while CanESM5 considers both emissions and a broader set of socioeconomic factors. This comparison indicates that total precipitation projections from CanESM2 are considerably lower than those projected by CanESM5, resulting in relative errors of 7.46% for the minimum and 5.4% for the maximum projected total precipitation.

The interquartile section of the boxplot, which represents the middle 50% of all samples, clearly demonstrates that historical total precipitation significantly exceeds the projections made by climate change models, particularly CanESM2 (Figure 2A). To illustrate this difference, attention can be drawn to the first quartile (Q1) in the provided box plot for historical total precipitation, which is higher than the third quartile (Q3) for CanESM2. This means that over 75% of recorded historical total precipitation values surpass the 25% mark of total precipitation projected by CanESM2. These observations indicate that, due to climatic effects, total precipitation in the coming years is anticipated to decrease when compared to the historical values recorded in the study area. Specifically, upon comparing the projected total precipitation values by CanESM2 to the historical records, it becomes evident that there is a decrease in the range of 11–14% for the minimum, Q1, mean, Q3, and maximum values of total precipitation. The values for the various components of these boxplots are provided in

Table 1. Summary statistics for total precipitation (mm) and average seasonal temperature (AST) for historical records and projections by climate change models CanESM2 and CanESM5.

Index	Total Precipitation (mm)			Total Precipitation (mm)
	Historical	CanESM2	CanESM5	Historical	CanESM2	CanESM5
Minimum	616.0	527.0	566.3	6.1	9.2	10.8
1st Quantile (Q1)	718.4	636.1	683.3	9.2	12.5	13.8
Mean	754.7	662.9	704.4	10.6	13.6	15.0
3rd Quantile (Q3)	825.3	710.7	765.3	11.4	14.5	15.8
Maximum	987.4	857.8	904.2	12.1	15.3	16.6
Interquartile (IQR)	106.9	74.6	82.1	2.1	2.0	2.0

. A comparison of the distribution of historical total precipitation with the projections made by CanESM5 indicates that all the presented characteristics, including minimum, Q1, average, Q3, maximum, and interquartile range, are higher for historical total precipitation, with relative errors of 8%, 4.9%, 6.66%, 7.27%, 8.42%, and 23.23%, respectively. Indeed, based on the CanESM5 projections, the average total precipitation is expected to decrease by 6.66%.

For the AST, the historical temperature ranges between 6.14 °C and 12.14 °C, while it is 9.22 °C to 15.32 °C degrees Celsius for CanESM2 and 10.76 °C to 16.61 °C degrees Celsius for CanESM5 (Figure 2B). It is evident that both climate change models project temperature distributions significantly higher than historical values. This suggests a substantial temperature increase in the study area according to two different climate change models, each based on distinct scenarios. The temperature rise associated with climate change can have complex and diverse effects on pesticide use in agriculture, impacting factors such as pest populations, pesticide efficacy, environmental concerns, and the overall sustainability of pest management practices [37,38,39]. A more comprehensive scrutiny of the historical and projected data reveals substantial deviations. Specifically, when comparing the historical AST to CanESM2 and CanESM5, the disparities are pronounced, amounting to 28.67% and 41.5% (respectively) relative error. Moreover, an analysis of the AST distribution unveils a noteworthy pattern. The first quartile (Q1) associated with the AST projected by climate change models surpasses the Q3 value linked to historical records. Indeed, less than 25% of the total AST values projected by climate change models align with the values recorded historically. AST exhibits a more substantial increase in CanESM5 than CanESM2. Consequently, the average AST value associated with CanESM5 surpasses that of CanESM2 by approximately 10%.

Table 2. ANOVA results comparing historical total precipitation and average temperature data with CanESM2 and CanESM5 climate models.

Climate Variables	Statistics	Historical vs. CanESM2	Historical vs. CanESM5	CanESM2 vs. CanESM5
Temperature	F-Statistics	1057.834	2105.13	198.438
	p-value	4.67 × 10−143	4.14 × 10−215	6.05 × 10−40
Precipitation	F-Statistics	273.9509	73.47208	105.2281
	p-value	1.65 × 10−52	6.03 × 10−17	3.92 × 10−23

shows the ANOVA results for comparing historical temperature and precipitation data with climate projections from the CanESM2 and CanESM5 models. These results reveal significant differences between historical observations and the projections from both models, as well as notable discrepancies between them, providing insights into how each model interprets future climate conditions.

For temperature, the F-statistics and p-values indicate significant differences between historical data and CanESM2 and CanESM5 projections. When comparing the historical temperature data to CanESM2, the F-statistic is 1057.834, with a p-value of 4.67 × 10⁻¹⁴³. This implies that CanESM2 anticipates a substantial departure from historical temperature trends. The significant variation points to likely increases in temperature as forecasted by CanESM2, which may be attributed to the model’s assumptions regarding the impact of greenhouse gases and other climate factors. The comparison between historical temperature data and CanESM5 shows an even higher F-statistic of 2105.13 with an extremely small p-value of 4.14 × 10⁻²¹⁵. This suggests that CanESM5 projects even greater temperature shifts than CanESM2, highlighting that the CanESM5 model foresees more extreme temperature changes. In this case, the higher F-statistic suggests that CanESM5 is more sensitive to climate drivers, possibly incorporating more updated factors or feedback mechanisms that result in larger temperature projections. When comparing the CanESM2 and CanESM5 models directly, the F-statistic for temperature is 198.438, with a p-value of 6.05 × 10⁻⁴⁰. While the difference between the two models is smaller than the difference between each model and historical data, it still indicates a significant discrepancy. This shows that, although both models predict rising temperatures, they diverge on the magnitude of the change, with CanESM5 indicating a stronger warming trend than CanESM2. These differences may stem from the models’ different approaches to climate system components or their treatment of emissions scenarios.

The ANOVA results for precipitation also demonstrate significant differences between historical data and the climate models, but these differences are less pronounced than temperature. The F-statistic for comparing historical precipitation and CanESM2 is 273.9509, with a p-value of 1.65 × 10⁻⁵². This suggests that CanESM2 projects notable changes in precipitation patterns, though the magnitude of the changes is less dramatic than for temperature. The results imply that CanESM2 forecasts altered precipitation levels, which could affect regional hydrological cycles and water availability. The comparison between historical precipitation data and CanESM5 yields an F-statistic of 73.47208 and a p-value of 6.03 × 10⁻¹⁷. Although the differences between historical precipitation and CanESM5 projections are statistically significant, they are smaller than those observed with CanESM2. This indicates that CanESM5, while still predicting deviations from historical precipitation trends, projects more moderate precipitation changes than CanESM2. When comparing the two models’ precipitation projections, the F-statistic is 105.2281, with a p-value of 3.92 × 10⁻²³. This shows that similar to temperature, the CanESM2 and CanESM5 models differ significantly in their precipitation forecasts. However, the discrepancy between the models for precipitation is smaller than for temperature, indicating that both models are more aligned in their precipitation projections, though differences still exist. These differences could be due to variations in how the models simulate atmospheric processes or regional climate effects, particularly in response to shifts in temperature and other climatic factors.

3.2. Pesticide Use

Figure 3 compares the pesticide use (PU) calculated based on the CanESM2 and CanESM5 across three scenarios and three projected periods (i.e., PP1, PP2, and PP3), as well as historical PU as the reference. The comparisons presented in this figure pertain to annual average values. This means that the average pesticide usage in a given year corresponds to the respective time period for all golf courses in the study area.

For the RCP2.6 as a scenario of the CanESM2, the projected PU in all projected periods is much higher than those for the historical ones, so the predicted PU is expected to be 2.66, 2.77, 4.10, and 3.17 times higher than those of the historical period for the PP1, PP2, PP3, and total period, respectively. Based on the other two CanESM2 scenarios, the projected PU values also exhibit significant increases. When comparing different projected periods, it is notable that, except for RCP2.6, which shows the maximum relative difference compared to historical values in PP3, the highest relative difference in the other two scenarios is related to PP1. Regarding the projected PU values across different CanESM2 scenarios, no distinct trend emerges regarding increasing or decreasing PU values based on various projected periods. For instance, in RCP2.6, the projected PU is highest for PP3, followed by PP2 and PP1. Conversely, in RCP4.5, the PU values are ranked highest in PP1, followed by PP2 and PP3. Similarly, for RCP8.5, PU values in PP1 are the highest, followed by PP2 and PP3. For PP1, according to the CanESM2 model, in the most optimistic scenario (scenario RCP2.6, PP1 period), the amount of pesticide used on Quebec Province’s golf courses is expected to increase by over 166%. The most pessimistic scenario (scenario RCP8.5, PP1 period) could rise to as much as 219%.

When comparing the relative differences between PU projections based on CanESM5 and historical PU, it is evident that all projected values significantly exceed historical records, with relative differences ranging from 10% to 251%. The maximum relative difference occurs in SSP5-8.5 for PP1 and SSP1-2.6 for PP2 and PP3. In the SSP1-2.6 scenario of CanESM5, the projected PU values in all projected periods significantly exceed historical records, with relative differences of more than 57.53%, 250%, and 115% for PP1, PP2, and PP3, respectively (Figure 3). In the SSP2-4.5 scenario, the projected PU values are also higher than historical values but notably lower than those in SSP1-2.6. Relative differences between historical PU and projected values for SSP2-4.5 are more than 48%, 22%, and 20% for PP1, PP2, and PP3, respectively. Under the SSP5-8.5 scenario, PU projections in all projected periods, except PP3, exceed historical values, with absolute relative differences of more than 125%, 10%, and 26% for PP1, PP2, and PP3, respectively. Comparing CanESM2 with CanESM5, the projected PU values in all scenarios and projected periods, except for the optimistic scenarios (SSP1-2.6 and RCP2.6) in PP2, show higher values for CanESM2. This results in a minimum and maximum relative difference of 0.66% and 53.31%, respectively.

Table 3. Results of the ANOVA analysis comparing historical pesticide usage data with climate projections from the CanESM2 and CanESM5 models across various scenarios and projected periods.

Model	Scenario	Projected Period	Historical vs. CanESM2
			F-Statistics	p-Value
Historical vs. CanESM2	RCP2.6	PP1	1.19 × 102	1.05 × 10−25
		PP2	1.17 × 102	2.72 × 10−25
		PP3	9.33 × 101	8.77 × 10−21
		Total	1.10 × 102	6.34 × 10−24
	RCP2.6	PP1	1.09 × 102	1.03 × 10−23
		PP2	1.43 × 102	5.74 × 10−30
		PP3	1.41 × 102	1.10 × 10−29
		Total	1.31 × 102	8.74 × 10−28
	RCP2.6	PP1	1.09 × 102	1.06 × 10−23
		PP2	1.50 × 102	2.21 × 10−31
		PP3	1.46 × 102	1.52 × 10−30
		Total	1.35 × 102	1.64 × 10−28
Historical vs. CanESM5	RCP2.6	PP1	1.43 × 102	4.08 × 10−30
		PP2	1.04 × 102	9.01 × 10−23
		PP3	1.31 × 102	8.58 × 10−28
		Total	1.26 × 102	6.83 × 10−27
	RCP2.6	PP1	1.45 × 102	2.36 × 10−30
		PP2	1.49 × 102	3.51 × 10−31
		PP3	1.48 × 102	5.56 × 10−31
		Total	1.47 × 102	7.44 × 10−31
	RCP2.6	PP1	1.28 × 102	2.87 × 10−27
		PP2	1.48 × 102	5.72 × 10−31
		PP3	1.54 × 102	4.07 × 10−32
		Total	1.44 × 102	3.78 × 10−30
CanESM2 vs. CanESM5	RCP2.6	PP1	7.13 × 102	1.82 × 10−16
		PP2	2.37 × 102	1.40 × 10−6
		PP3	2.57 × 102	2.69 × 10−49
		Total	5.36 × 102	7.00 × 10−13
	RCP2.6	PP1	1.79 × 102	1.84 × 10−36
		PP2	5.27	2.20 × 10−2
		PP3	5.94	1.50 × 10−2
		Total	4.72 × 101	1.47 × 10−11
	RCP2.6	PP1	4.72 × 101	1.46 × 10−11
		PP2	7.40 × 10−1	3.90 × 10−1
		PP3	1.38 × 101	2.23 × 10−4
		Total	1.94 × 101	1.26 × 10−5

reveals the results of the ANOVA analysis comparing historical pesticide usage data with climate projections from the CanESM2 and CanESM5 models across various scenarios and projected periods. For the CanESM2 model, the ANOVA results show significant differences between the historical pesticide usage data and the model projections across various periods and scenarios. For example, under the RCP2.6 scenario, the F-statistics for periods PP1, PP2, and PP3 are consistently high (e.g., 1.19 × 10² for PP1, 1.17 × 10² for PP2, and 93.3 for PP3), with extremely low p-values, indicating substantial deviations from historical data. These results suggest that CanESM2 anticipates a substantial change in pesticide usage over time, particularly for later projected periods, likely driven by changes in climate conditions affecting agricultural practices. The pattern continues in other periods, with F-statistics consistently around the same magnitude (e.g., 1.10 × 10² for total in RCP2.6), reinforcing the model’s prediction of increased pesticide usage due to evolving climatic factors. This is particularly pronounced in PP2 and PP3, where the differences are more marked, hinting at increasing divergence from historical usage trends in later projections.

Similar trends are observed in the comparison between historical data and the CanESM5 model, with significant differences across periods and scenarios. Under the RCP2.6 scenario, the F-statistics in PP1, PP2, and PP3 are again high (e.g., 1.43 × 10² for PP1, 1.04 × 10² for PP2, and 1.31 × 10² for PP3), with p-values indicating strong statistical significance. The CanESM5 model, like CanESM2, projects significant increases in pesticide usage in future periods. However, the F-statistics for CanESM5 tend to be slightly higher than those for CanESM2, suggesting that CanESM5 forecasts even larger deviations from historical pesticide usage patterns, possibly due to the model’s assumptions about future climate impacts. Notably, the differences are most pronounced in later periods, such as PP3, where the deviations from historical data are most significant. This aligns with the general expectation that climate change impacts will intensify over time, leading to more incredible changes in pesticide usage as agricultural conditions evolve.

The comparison between CanESM2 and CanESM5 models reveals notable discrepancies between the two projections. In specific periods such as PP3, the F-statistics are exceptionally high (e.g., 2.57 × 10²), indicating significant differences in how each model projects future pesticide usage. The p-values reinforce the statistical significance of these differences, with CanESM5 generally showing higher projected pesticide usage compared to CanESM2. In some cases, such as PP2 and PP3, the differences between the two models are more pronounced, reflecting varying assumptions and sensitivities in how each model accounts for future climate change and its impact on pesticide use. This highlights the uncertainty in climate model projections, particularly regarding agricultural practices like pesticide application, where small differences in projected climate variables can lead to larger discrepancies in outcomes.

Figure 4 shows the spatial relative differences between historical pesticide use and the projected use based on the CanESM2 and CanESM5 across three scenarios for PP1. In the 2023–2048 period (PP1), a comparison between the two scenarios, RCP2.6 in CanESM2 and SSP1-2.6 in CanESM5, reveals that in most regions, except the northern and western areas, the relative differences between historical pesticide use and the projected use based on the CanESM2 are higher than those for the CanESM5. Specifically, the minimum relative differences between historical pesticide use and the projected use based on the CanESM2 are observed in the areas south and southwest of Quebec City, while it is south and west of Quebec City for the CanESM5. Both scenarios for PP1 indicate a rise in relative differences between the historical and projected pesticide use from south to north within the study area, with the most significant relative difference observed in the north and some parts of the east of the study area. A comparison of the relative differences between historical pesticide use and the projected use based on both models reveals that the average value of this index for CanESM2 is 724%, while for CanESM5, it stands at 391%. Comparison of the CanESM2 (scenarios RCP4.5) and CanESM5 (Scenario SSP2-4.5) for the PP1 reveals that the range depicted in the maps for both figures shows that there is no significant difference between the minimum and maximum values of the RD calculated by two models. However, the difference in calculated RD values between the two models varies by location. In the southwest and south of the study area, the calculated RD based on the projected pesticide use by CanESM2 is higher than CanESM5, with up to a 300% difference. In other locations, the RD values do not show a consistent trend; in various parts of the study area, the RD value for CanESM2 is sometimes higher than that of CanESM5, and at other times, it is the other way around. The maximum value calculated using the CanESM5 model is 700% higher than that derived from another climate change model. These differences are primarily observed in the east, west, and central areas of the study area. A comparison of the RCP8.5 scenario using the CanESM2 model and the SSP5-8.5 scenario using the CanESM5 model reveals that for PP1, the range of calculated RD values is generally consistent between both models, except that the maximum RD value is higher for CanESM2. However, their distribution varies significantly. The RD values estimated by the CanESM5 are much higher in the west and north of the study area than in CanESM2. In central areas of the south of the study area, CanESM2 consistently provides higher values than CanESM5, with their relative difference reaching about 300% around the Saint Laurent River. Originating from Lake Ontario, the Saint Laurent River flows northeastward, passing by Montreal and Quebec City before reaching the Gulf of St. Lawrence.

Figure 5 shows the spatial relative differences between historical pesticide use and the projected use based on the CanESM2 and CanESM5 across three scenarios for PP2 (2049–2074). In the second projected area, the distribution of the relative difference (RD) between the historical pesticide use and projected use based on both climate change models demonstrates that the RD in the east of the study area is higher than in the west. Notably, in most eastern regions, the RD is higher for CanESM2 than for CanESM5. Conversely, the situation is reversed in the western areas, with CanESM5 showing higher RD values than CanESM2. The minimum RD values for both models are observed in the southwest of Quebec City and the southern part of the study area, which is the region where most golf courses are located. The expected RD values increase from south to north, consistent with the previous period’s projection. An analysis of the variance between past and forecasted pesticide usage, according to two different models, shows that the average RD for CanESM2 is 769%, whereas for CanESM5, it is notably higher at 1004%.

For PP2, the maximum value of projected AAIR corresponds to CanESM2, which is about 2000% higher than the maximum value of estimated RD by CanESM5. The minimum calculated RD value by both climate change models is observed in the south and southwest of Quebec City and generally the south of the study area. The highest estimated RD value for the CanESM5 model occurs in the northern region of the study area, where there are fewer golf courses. In contrast, for the CanESM2 model, the maximum RD value is observed in the eastern and some parts of the western and central parts of the area. The rate at which the RD increases from the south to the north of the study area is rapid for the CanESM2 model, whereas it is gradual for the CanESM5 model. A comparison of the RCP8.5 scenario using the CanESM2 model and the SSP5-8.5 scenario using the CanESM5 model reveals that for PP2, a portion of the north and central regions exhibit a very high relative difference, with CanESM5 projecting higher pesticide use than CanESM2.

Figure 6 shows the spatial relative differences between historical pesticide use and the projected one based on the CanESM2 and CanESM5 across three scenarios for PP3. In the final projected periods of the most optimistic scenario, the distribution of the RD in both models diverges significantly. In CanESM5, there is no significant difference in RD between the east and west of the study area. However, in CanESM2, there is a remarkable difference, with RD values in the west and southwest of the study area being considerably lower than those in the east. In CanESM2, only a few areas in the southern region have the minimum RD values, rapidly increasing to reach the maximum RD. In contrast, CanESM5 exhibits a different pattern. In PP1 (Figure 4) and PP2 (Figure 5), many areas in the southern and southwestern parts of Quebec (south of the study area) have minimum RD values that gradually increase. A comparison between the two climate change models reveals that CanESM5 provides higher RD values in the western and northwestern regions of the study area compared to the other model, with significant differences in RD values for other locations within this region when compared to CanESM2, so in the southern and northwestern areas of Quebec, the calculated RD values by CanESM2 are approximately twice as high as those projected by CanESM5. An examination of the disparities between past pesticide usage and future projections, as per two distinct models, indicates that the average RD value for CanESM2 is 1175% in contrast to 553% for CanESM5. In PP3, similar to PP2 (Figure 5), the south and west of Quebec City are observed to have the minimum estimated RD. However, as the movement towards the center of the study area occurs, more drastic changes in RD values based on CanESM2 are exhibited compared to CanESM5, with the highest values of CanESM2 being in some parts of the north and center of the study area (i.e., whole southern province), while the highest values of CanESM5 are shown in the north of the study area. Additionally, the maximum RD calculated based on the projected pesticide use by CanESM5 is approximately 350% higher than that in CanESM2. When comparing the two models in different areas, the RD values based on the CanESM5 are higher than those estimated by CanESM2 only in some parts of the north, northwest, and southwest of the study area, whereas CanESM2 provides larger values in other areas. A comparison of the relative differences between historical pesticide use and the projected use based on both models reveals that the average value of this index for CanESM2 at PP1 (Figure 4), PP2 (Figure 5), and PP3 (Figure 6) is 926%, 414%, and 420% (respectively), while for CanESM5, it stands at 367%, 286%, and 292% (respectively). A comparison of the RCP8.5 scenario using the CanESM2 model and the SSP5-8.5 scenario using the CanESM5 model reveals that for PP3, the trend is reversed, with the RD values calculated using CanESM2 being significantly higher than those calculated with CanESM5, often exceeding a difference of 1000%. Indeed, the RD values calculated based on CanESM5 in PP3 are the lowest across all scenarios and projection periods. Analyzing the variances in pesticide usage historically and in future projections according to both models, it is shown that the mean index values for CanESM2 at PP1 (Figure 4), PP2 (Figure 5), and PP3 (Figure 6) are 939%, 273%, and 315%, respectively, in contrast to CanESM5’s values, which are 590%, 286%, and 203% at the same points, respectively.

Figure 7 shows the average recorded and projected pesticide use based on the two tested climate change models through three projected periods for different pesticide types (the average of all three scenarios). Both climate models project a significant increase in pesticide use for all pesticide types. The trend across different projected periods indicates that CanESM2 projects the highest pesticide usage for all pesticide types in PP1, followed by PP3 and PP2 in the second and third positions, respectively. Analysis of projected pesticide use based on CanESM5 for all pesticide types reveals that the maximum values are associated with PP2, followed by PP1 and PP3.

Comparing the two climate change models reveals distinct patterns in projected pesticide use. Specifically, based on the CanESM2 model, pesticide type A as well as pesticide types I, RC, and Ro in projected periods PP1 and PP3 exhibit more considerable pesticide use compared to the CanESM5 model. For pesticide type A, the relative difference between the projected AAIR values by CanESM2 and CanESM5 across all projected periods (PP1, PP2, and PP3) exceeds 90%. In the case of pesticide types I, RC, and Ro, this relative difference falls within the range of [28%, 35.3%] for PP1 and PP3 and [1.5%, 14.4%] for PP2. Conversely, for pesticide types F and H, where the CanESM5 model projects higher pesticide use than CanESM2, the relative differences span [65.6%, 222.9%%] and [114%, 247%], respectively.

Table 4. ANOVA results comparing historical pesticide usage data with CanESM2 and CanESM5 climate models across different periods (PP1, PP2, and PP3).

Model	PT	Statistics	All	PP1	PP2	PP3
Historical vs. CanESM2	A	F-Statistics	8.26	4.59 × 10	1.43 × 102	3.06 × 10
		p-value	4.30 × 10−3	2.62 × 10−11	4.31 × 10−30	4.43 × 10−8
	F	F-Statistics	1.84 × 103	1.46 × 10	5.25 × 10	1.40 × 10
		p-value	0.00	1.44 × 10−4	1.11 × 10−12	1.95 × 10−4
	H	F-Statistics	2.09 × 102	1.72	6.95	2.34
		p-value	7.06 × 10−47	1.91 × 10−1	8.56 × 10−3	1.26 × 10−1
	I	F-Statistics	1.41 × 102	3.41	1.48 × 10	5.86
		p-value	5.25 × 10−32	6.51 × 10−2	1.29 × 10−4	1.57 × 10−2
	RC	F-Statistics	9.30 × 102	2.83	1.52 × 10	8.86
		p-value	3.16 × 10−162	9.29 × 10−2	1.06 × 10−4	3.01 × 10−3
	Ro	F-Statistics	1.90 × 102	2.59	1.43 × 10	8.77
		p-value	1.13 × 10−37	1.08 × 10−1	1.69 × 10−4	3.17 × 10−3
Historical vs. CanESM5	A	F-Statistics	3.93	1.19 × 10−1	6.87 × 10−3	1.42
		p-value	4.81 × 10−2	7.31 × 10−1	9.34 × 10−1	2.34 × 10−1
	F	F-Statistics	6.95 × 103	1.34	9.23 × 10−1	5.30 × 10
		p-value	1.00 × 10−7	2.47 × 101	3.37 × 10−1	8.96 × 10−13
	H	F-Statistics	1.24 × 103	5.34 × 10−4	6.06	3.10 × 10
		p-value	2.20 × 10−254	9.82 × 10−1	1.41 × 10−2	3.62 × 10−8
	I	F-Statistics	72.2	1.80 × 10−1	2.86 × 10−1	6.85 × 10
		p-value	2.69 × 10−17	6.71 × 10−1	5.93 × 10−1	9.04 × 10−3
	RC	F-Statistics	5.94 × 102	4.10 × 10−2	9.44 × 10−1	7.97
		p-value	2.49 × 10−112	8.40 × 10−1	3.32 × 10−1	4.88 × 10−3
	Ro	F-Statistics	1.12 × 102	6.03 × 10−2	1.68	4.04
		p-value	3.92 × 10−24	8.06 × 10−1	1.95 × 10−1	4.48 × 10−2
CanESM2 vs. CanESM5	A	F-Statistics	9.41 × 102	8.48 × 102	3.76 × 102	1.11 × 103
		p-value	1.96 × 10−132	1.72 × 10−123	1.25 × 10−67	3.56 × 10−147
	F	F-Statistics	1.09 × 102	5.58 × 10	2.62 × 102	3.14 × 10
		p-value	7.25 × 10−24	2.36 × 10−13	2.12 × 10−50	2.98 × 10−8
	H	F-Statistics	1.22 × 102	7.71 × 10	1.99 × 102	6.56 × 10
		p-value	3.35 × 10−26	1.19 × 10−17	5.20 × 10−40	2.42 × 10−15
	I	F-Statistics	6.03	1.21 × 10	5.14 × 10−1	2.09 × 10
		p-value	1.43 × 10−2	5.21 × 10−4	4.74 × 10−1	5.81 × 10−6
	RC	F-Statistics	2.85	8.09	3.47	1.64 × 10
		p-value	9.16 × 10−2	4.57 × 10−3	6.28 × 10−2	5.80 × 10−5
	Ro	F-Statistics	6.33	1.64 × 10	8.37 × 10−1	1.74 × 10
		p-value	1.21 × 10−2	5.80 × 10−5	3.61 × 10−1	3.44 × 10−5

shows the ANOVA results (F-statistics and corresponding p-values) for six pesticide types (A, F, H, I, RC, and Ro) across all years and specific periods. The ANOVA results indicate highly significant differences between the historical data and the CanESM2 model across most pesticides and periods, particularly for pesticide F and pesticide A. For instance, pesticide A shows a large variance in PP2, with an F-statistic of 1.43 × 10² and a very low p-value of 4.31 × 10⁻³⁰, indicating substantial deviation in the projected usage compared to historical values. Similarly, pesticide F shows extremely significant differences across all periods, with an exceptionally high F-statistic (1.84 × 10³) for all years, confirming strong discrepancies between historical and CanESM2 projections. These results suggest that CanESM2 predicts substantial increases in pesticide usage for certain types, especially during the later periods, likely due to changing climate conditions under this model.

In addition to pesticides A and F, other pesticide types, such as RC and Ro, show notable differences under the CanESM2 model. For example, pesticide RC demonstrates a substantial variance across all periods, particularly in PP2, with an F-statistic of 1.52 × 10¹ and a p-value of 1.06 × 10⁻⁴, indicating that the CanESM2 model projects a significant rise in its usage. These widespread differences across multiple pesticide types suggest that CanESM2 consistently predicts higher pesticide usage under future climate scenarios, likely due to increased temperature and altered precipitation patterns.

In contrast to CanESM2, the comparison between historical data and CanESM5 reveals fewer and less consistent significant differences across pesticides and periods. The only notable variance for pesticide A is for all years, with an F-statistic of 3.93 (p-value = 4.81 × 10⁻²), while other periods show non-significant differences. However, pesticide F demonstrates a significant difference for all years, with a very high F-statistic (6.95 × 10³) and a low p-value (10⁻⁷), although it appears less significant in individual periods. Interestingly, pesticide H, in contrast to pesticide F, shows significant differences in later periods, such as PP2 and PP3, reflecting potential increases in its usage under CanESM5 projections. These results suggest that the CanESM5 model, while still increasing some discrepancies, projects overall milder deviations from historical data than CanESM2. However, certain pesticides such as F and H still exhibit considerable increases in projected usage during specific periods.

The comparison between CanESM2 and CanESM5 reveals striking and widespread differences across all pesticides and periods, with particularly high significance for pesticide A and pesticide F. For instance, pesticide A shows large differences in PP3, where the F-statistic reaches 1.11 × 10³ with a p-value of 3.56 × 10⁻¹⁴⁷, suggesting that CanESM5 projects significantly higher usage than CanESM2. Similarly, pesticide F demonstrates major variances across all periods, especially in PP2, with an F-statistic of 2.62 × 10². The consistently low p-values across multiple pesticide types and periods indicate that the two models offer considerably different projections of future pesticide usage. This discrepancy is particularly important in understanding how each climate model interprets the impacts of climate change on pesticide use but still shows significant spikes in usage for certain pesticides in specific periods. These differences are critical, as they reflect the divergence in climate impact projections between the two models, with CanESM5 consistently predicting higher usage in later periods, especially in high-impact pesticide categories such as F and H.

4. Discussion

This study provides the first comprehensive assessment of how climate change may influence pesticide use on golf courses in Quebec, utilizing projections from both CMIP5 (CanESM2) and CMIP6 (CanESM5) models. Unlike previous studies that focused primarily on agricultural settings [40], this research highlights the impact of climate projections specifically on golf courses, a critical area of recreational management. No other studies have specifically examined the relationship between climate change and pesticide use in golf courses within this region. Thus, these findings are particularly significant for local stakeholders such as golf course managers, policymakers, and environmental planners who are directly involved in pest management and sustainable agriculture.

Significant differences in temperature and precipitation projections were observed between the two models, which directly translated into variations in pesticide application needs. CanESM5 projected higher temperatures and more moderate declines in precipitation compared to CanESM2, which is consistent with findings from previous studies in the literature [33,41,42]. Moreover, it confirms that the CanESM5 model predicts conditions that would likely increase pest pressures on turfgrass. This is consistent with global studies, such as Sobie et al. [24], who observed more pronounced warming patterns in CMIP6 models across Canada, further validating the trend observed in our research. These climate variations are crucial, as higher temperatures are associated with increased pest pressures, increasing pesticide application rates. This aligns with broader research that links rising temperatures to heightened pest activity [43,44], thus emphasizing the heightened urgency for adaptive management strategies specific to the Quebec region [42,43]. However, the application of this knowledge specifically to golf courses in Quebec had not been explored before, making our findings novel in this context.

Although this study is the first for Quebec’s golf courses, the results align with global trends seen in climate change research. For example, global agricultural studies often demonstrate the strong influence of temperature on pest proliferation [45,46], which is a similar pattern observed in our study but within a recreational setting. This highlights the importance of addressing both agricultural and non-agricultural pest management when considering climate impacts. Other studies comparing CMIP5 and CMIP6 have demonstrated similar patterns, where CMIP6 projections tend to predict more intense temperature increases. For example, our findings parallel Sobie et al. [24], where CMIP6 models projected more extreme warming across Canada. This reinforces the likelihood that the higher temperatures projected by CanESM5 will exacerbate pest issues on golf courses.

In contrast to global agricultural studies that often focus on crops, this research makes a critical contribution by filling a gap in understanding pesticide use in a recreational context. The unique environmental conditions of Quebec, including the region’s specific pest challenges and pesticide regulations, make this research particularly relevant. For instance, under Quebec’s pesticide reduction regulations, golf courses are already required to monitor and reduce pesticide usage. However, the projections in this study suggest that these reduction targets could become more difficult to achieve as pest pressures rise due to climate change. Specifically, the findings highlight that pest pressure will likely increase, increasing pesticide use even more despite existing regulations.

The augmented need for pesticide use, as projected in this study, could have several significant impacts. Our analysis reveals that increased pesticide application, driven by climate-related pest pressures, may lead to heightened environmental risks. These risks include potential contamination of water sources, adverse effects on non-target organisms, and increased human health concerns, particularly for golf course workers who are regularly exposed to these environments [47,48]. The environmental ramifications could be particularly severe in areas where golf courses are in proximity to residential zones or sensitive ecosystems, highlighting the need for rigorous mitigation strategies. Furthermore, as pest pressures rise, golf course managers may find it challenging to maintain turf quality without resorting to higher pesticide use, potentially leading to regulatory conflicts given the stringent pesticide reduction policies in Quebec. This could place golf course managers in a difficult position, balancing turf health with legal and environmental responsibilities.

One of the significant strengths of this study is its innovative application of climate models to forecast future pesticide use in a specific regional and recreational context. The integration of advanced climate models (CanESM2 and CanESM5) with pesticide usage data provides a detailed picture of how climate change could affect pesticide needs over the coming decades under different climate scenarios [43,49]. This offers an invaluable resource for stakeholders planning for the future of golf course management in Quebec. However, there are limitations as well. The predictions are based on current pesticide formulations and application rates, which may evolve as new pesticide technologies are developed. Additionally, the models assume that future pest dynamics will respond to temperature and precipitation changes in ways that mirror historical trends, which may not account for emerging pests or other unforeseen ecological shifts [50]. Given these limitations, future studies should include more dynamic models that account for potential changes in pest behavior and pesticide technology to improve the accuracy of these projections.

To address these rising challenges, policymakers and golf course managers should use this information to develop adaptive strategies that balance pest control needs with environmental and health considerations [51,52]. Adopting more sustainable and innovative approaches, such as investing in less toxic pesticides and enhancing integrated pest management systems, is essential. This may include enhancing monitoring systems to detect pest outbreaks early and employing alternative pest control methods. For instance, integrated pest management strategies focusing on biological control methods could mitigate the rising pesticide usage [53] projected in this study, helping golf courses meet regulatory requirements while also protecting local ecosystems. Additionally, policy adjustments may be necessary to accommodate the changing dynamics of pesticide use [54], ensuring regulations remain effective in protecting environmental and public health [55] while allowing golf courses to adapt to the changing climate. More adaptable, climate-sensitive regulatory frameworks could assist in balancing the need for increased pesticide use with the overarching goal of environmental preservation [56].

Future studies should aim to refine these projections by incorporating potential advancements in pesticide technology and alternative pest management strategies, such as integrated pest management [57]. Moreover, research into the specific impacts of climate change on different pest species prevalent in Quebec’s golf courses would be beneficial. Targeting specific pests that are most likely to be affected by climate change would allow for more precise pesticide application, reducing both environmental impacts and costs. Additionally, future studies should assess the implications of newer pesticide generations, which may have lower application rates, and evaluate the potential risks to both human health and the environment resulting from these anticipated changes in pesticide use on golf courses. Finally, investigating the broader socioeconomic impacts of increased pesticide usage [58], such as costs for golf course management and potential effects on tourism, would provide a more holistic understanding of the issue.

5. Conclusions

This study provided a comparative analysis of the CanESM2 (CMIP5) and CanESM5 (CMIP6) climate models to assess their projections of total precipitation, average temperature, and pesticide use on Quebec’s golf courses from 2023 to 2100. The results indicate that both models project significant reductions in total precipitation, with CanESM2 showing a more pronounced decrease than CanESM5 (up to 14% and 8%, respectively). This trend is especially critical given its potential to exacerbate drought conditions, impacting turf management on golf courses. Conversely, CanESM5 projects higher average temperature increases than CanESM2, particularly in the far-future period (2075–2100), with temperatures rising up to 36.7%. These changes in climate variables are expected to significantly affect pest pressures and, consequently, pesticide use.

The study revealed substantial increases in pesticide use under both climate models compared to historical data, with projected increases ranging from 30% to 141%, depending on the scenario. CanESM2 consistently predicted higher pesticide use for insecticides, growth regulators, and rodenticides, with the most significant increases observed under RCP2.6 in the far future (up to 35.6%). In contrast, CanESM5 projected significantly higher usage for fungicides and herbicides, with relative increases ranging from 65% to 247%, particularly under the SSP5-8.5 scenario in the near future (2023–2050). When comparing the two models, CanESM2 generally projected higher pesticide use across all periods and scenarios, except for RCP2.6 in PP3, where its predictions were 27% lower than SSP1-2.6. Across all pesticide types (fungicides, herbicides, insecticides, growth regulators, and rodenticides), the projected pesticide use exceeded historical levels, with relative differences ranging from 1.7% to 157%.

These findings underscore the significant impact of climate change on pesticide use in Quebec’s golf courses, necessitating adaptive management strategies to address the growing demands for pest control and to mitigate environmental and health risks associated with increased pesticide application.