Open AccessArticle

Advanced Forecasting of Drought Zones in Canada Using Deep Learning and CMIP6 Projections

Keyvan Soltani

¹,

Afshin Amiri

²,

Isa Ebtehaj

^1,*

Hanieh Cheshmehghasabani

³,

Sina Fazeli

⁴,

Silvio José Gumiere

and

Hossein Bonakdari

^5,*

Department of Soils and Agri-Food Engineering, Université Laval, Québec, QC G1V 0A6, Canada

Department of Remote Sensing and GIS, University of Tehran, Tehran 14155-6619, Iran

Department of Civil Engineering, Razi University, Kermanshah 67146, Iran

⁴

Department of Irrigation & Reclamation Engineering, Faculty of Agriculture Engineering & Technology, University of Tehran, Tehran 14155-6619, Iran

⁵

Department of Civil Engineering, University of Ottawa, 161 Louis Pasteur Private, Ottawa, ON K1N 6N5, Canada

Authors to whom correspondence should be addressed.

Climate 2024, 12(8), 119; https://doi.org/10.3390/cli12080119

Submission received: 30 May 2024 / Revised: 12 July 2024 / Accepted: 7 August 2024 / Published: 10 August 2024

Download

Browse Figures

Figure 1
The relief map of the study area. The black dots show the distribution of sample points across Canada. "> Figure 2
The Schematic of the CNN’s structure. "> Figure 3
Research flowchart. "> Figure 4
(a) Classification Accuracy for 1022 ELM models. (b) Area Under the Curve (AUC) for 1022 ELM models. "> Figure 5
Zoning of Projected Average Annual Precipitation Anomalies in Canada (2024–2100) Compared to the Observed Period (1983–2023) using CanESM5 from CMIP6. "> Figure 5 Cont.
Zoning of Projected Average Annual Precipitation Anomalies in Canada (2024–2100) Compared to the Observed Period (1983–2023) using CanESM5 from CMIP6. "> Figure 6
Zoning of Projected Average Annual Temperature Anomalies in Canada (2024–2100) Compared to the Observed Period (1983–2023) using CanESM5 from CMIP6. "> Figure 6 Cont.
Zoning of Projected Average Annual Temperature Anomalies in Canada (2024–2100) Compared to the Observed Period (1983–2023) using CanESM5 from CMIP6. "> Figure 7
Canada’s Percentage of Drought-Affected Area, Pr, and T under the SSP126 Scenario: A Comparative Analysis from 2003–2023 to 2024–2100, Examining Historical Trends and Future Projections Using CanESM5 within the CMIP6 Framework. "> Figure 8
Canada’s Percentage of Drought-Affected Area, Pr, and T under the SSP245 Scenario: A Comparative Analysis from 2003–2023 to 2024–2100, Examining Historical Trends and Future Projections Using CanESM5 within the CMIP6 Framework. "> Figure 9
Canada’s Percentage of Drought-Affected Area, Pr, and T under the SSP370 Scenario: A Comparative Analysis from 2003–2023 to 2024–2100, Examining Historical Trends and Future Projections Using CanESM5 within the CMIP6 Framework. "> Figure 10
Canada’s Percentage of Drought-Affected Area, Pr, and T under the SSP585 Scenario: A Comparative Analysis from 2003–2023 to 2024–2100, Examining Historical Trends and Future Projections Using CanESM5 within the CMIP6 Framework. "> Figure 11
Policy Implications for Canada in Addressing Climate Change and Drought Conditions. ">

Versions Notes

Abstract

This study addresses the critical issue of drought zoning in Canada using advanced deep learning techniques. Drought, exacerbated by climate change, significantly affects ecosystems, agriculture, and water resources. Canadian Drought Monitor (CDM) data provided by the Canadian government and ERA5-Land daily data were utilized to generate a comprehensive time series of mean monthly precipitation and air temperature for 199 sample locations in Canada from 1979 to 2023. These data were processed in the Google Earth Engine (GEE) environment and used to develop a Convolutional Neural Network (CNN) model to estimate CDM values, thereby filling gaps in historical drought data. The CanESM5 climate model, as assessed in the IPCC Sixth Assessment Report, was employed under four climate change scenarios to predict future drought conditions. Our CNN model forecasts CDM values up to 2100, enabling accurate drought zoning. The results reveal significant trends in temperature changes, indicating areas most vulnerable to future droughts, while precipitation shows a slow increasing trend. Our analysis indicates that under extreme climate scenarios, certain regions may experience a significant increase in the frequency and severity of droughts, necessitating proactive planning and mitigation strategies. These findings are critical for policymakers and stakeholders in designing effective drought management and adaptation programs.

Keywords:

climate change; CanESM5; Canada; Canadian Drought Monitor; convolutional neural network; drought; precipitation; shared socioeconomic pathways (SSP); temperature

1. Introduction

Drought is one of the most disastrous natural phenomena occurring under all climate regimes [1,2]. Sometimes, it can persist for long periods and adversely affect ecosystems, society, agriculture, and water supply worldwide. Therefore, accurate prediction is vital in improving drought resilience, managing water resources, and reducing drought consequences [3]. A Precipitation (Pr) deficit primarily causes drought, while in some cases, it may originate from anomalies in other variables such as Temperature (T) or evapotranspiration [4,5]. Generally, drought can be classified into four categories: meteorological, agricultural, hydrological, and socioeconomic [6]. Regarding the impact of drought on the economic and environmental sectors, scientific concerns related to the effects of climate change on the frequency, duration, and intensity of future droughts have been expressed in different regions of the world, including Canada [7]. Climate change plays a vital role in causing drought. So, climate change and the resulting droughts are the biggest challenges in the present and future worldwide [8,9]. Following climate change, there is a shift in the trend of meteorological parameters, especially the trend of Pr [10,11]. Therefore, the study of their possible future variations is of great importance.

Climate change models are essential for understanding and predicting climate behavior at different time scales [12]. The models examine the extent to which observed climate change is due to natural variability, human activities, or a combination of both. For example, Global Circulation Models (GCM) are one of the most essential tools for identifying climate systems and are widely applied to simulate past, present, and future climate changes. To facilitate comparisons between different models around the world, the World Climate Research Program (WCRP) created the Coupled Model Intercomparison Project, which is dedicated to providing standardized climate simulation outputs for comparison [13]. Salimi et al. [14] have shown that the CanESM5 model of CMIP6 accurately forecasts the rainfall data of a rain gauge station in Quebec City, Canada. Sobie et al. [15] showed that climate model simulations predict more intense warming globally and in Canada. The CanESM5 model, part of the Sixth Coupled Model Intercomparison Project (CMIP6), has remarkably predicted these changes, particularly in temperature and extreme precipitation indices in Canada [15].

To help decision-makers reduce the impacts of drought, improving monitoring tools that provide relevant and timely information supporting drought mitigation decisions is crucial. One of Canada’s most popular means of monitoring drought is the Canadian Drought Monitor (CDM). CDM was developed in 2002 to monitor and report drought across Canada [16]. It combines a suite of in situ, model-based, and satellite-based indicators along with expert commentary to achieve assimilation of evidence on drought. It also measures meteorological, agricultural, and hydrological droughts, their impacts, and whether the impacts are short-term or long-term [17]. Tadesse et al. [18] analyzed Canada’s CDM and Vegetation Drought Response Index results (VegDRI-Canada). A 15-year evaluation (2000–2014) showed VegDRI-Canada’s high correlation (r > 0.5) with canola yields in non-irrigated croplands, aligning well with the CDM in identifying drought severity patterns. Mardian et al. [19] implemented a novel technique using Bayesian Neural Networks (BNNs) to predict wheat yield with high resolution and associated uncertainties, utilizing CDM and satellite-based Normalized Difference Vegetation Index.

Artificial Intelligence (AI) models have shown remarkable progress in drought prediction in recent years. The growing popularity of such techniques is due to the low computational cost, rapid model training and testing, targeted point applicability, and competitive performance [20,21]. In recent years, hydrologists have widely used the CNN model to model various hydrological phenomena [22,23,24,25,26,27,28]. The CNN capability for Standardized Pr Anomaly Index (SPAI) drought index prediction was performed by Maity et al. [29] in India. Hao et al. [30] developed forecasts for the Standardized Pr Evapotranspiration Index (SPEI) drought index in China by employing long short-term memory neural networks (LSTM), CNN, and categorical boosting (CatBoost). The results reveal that CNN consistently outperformed LSTM and CatBoost models. Danandeh Mehr et al. [31] evaluated five different models—CNN, LSTM, artificial neural networks, genetic programming, and a hybrid of CNN and LSTM (CNN-LSTM) to estimate the SPEI drought index in Turkey. The findings indicated that the CNN-LSTM model was the most effective. Chen et al. [32] applied the CNN algorithm for modeling drought in China and reported that the results demonstrate the effectiveness of the CNN model in predicting drought events. Liu et al. [33] proposed a CNN model for forecasting summer drought in China. The results showed that the CNN model can significantly predict drought. Khan and Maity [34] simulated hydrological drought using LSTM, CNN, support vector regression, and feedforward neural networks in India. The results affirm both the superiority and benefits of CNN.

The main aim of the current study is to present a comprehensive approach to drought zoning in Canada, integrating data from the CDM, ERA5-Land, and an advanced deep learning technique. These resources provide an in-depth analysis of historical drought patterns and projections for future trends up to 2100. Enhanced data utilization was achieved by combining data from the CDM provided by the Government of Canada with daily data from ERA5-Land, creating a robust time series of average monthly precipitation and air temperature data for 199 sample locations in Canada from 1979 to 2023. The gaps in the data labeled “drought not analyzed” by the Canadian government were filled using the deep learning model presented in this research, and the future values for these points were calculated. A CNN model was employed to calculate CDM values based on the collected data, crucially filling gaps in the data provided by the Government of Canada and enabling a comprehensive representation of historical drought points across Canada. Future precipitation and temperature trends until 2100 were projected using the 6th Intergovernmental Panel on Climate Change (IPCC) report and four climate change scenarios derived from the CanESM5 model, and these projections were integrated into the CNN model to forecast future CDM values, enabling drought zoning based on four distinct future scenarios. This approach facilitates informed decision-making for future planning and policy development to mitigate and adapt to drought conditions. A robust framework for understanding and preparing for the impacts of climate change on drought occurrences in Canada is provided by the utilization of the CNN model and climate change scenarios from the 6th IPCC report. Through the delineation of potential drought zones under various future scenarios, this research aims to support policymakers, resource managers, and stakeholders in implementing proactive measures to mitigate the effects of drought and ensure the resilience of Canadian communities and ecosystems in the face of changing environmental conditions.

2. Materials and Methods

2.1. Study Area

The study area is located between Long (longitudes) 141° 01′ W–52° 37′ W and Lat (latitudes) 41° 40′ N–83° 8′ N, which includes the 10 Canadian provinces—Newfoundland and Labrador (NL), Nova Scotia (NS), Prince Edward Island (PE), New Brunswick (NB), Quebec (QC), Ontario (ON), Manitoba (MB), Saskatchewan (SK), Alberta (AB), and British Columbia (BC)—and the three territories—Northwest Territories (NT), Nunavut (NU), and Yukon (YT) (Figure 1). A total of 199 sample points were selected across Canada, and the input data for climate change models were prepared for these points. Figure 1 shows the distribution of data collection points in Canada.

2.2. Data Gathering

The daily data from ERA5-Land were used to create the time series of average monthly Pr and T data for 199 sample locations in Canada from 1979 to 2023. ERA5-Land is a reanalysis dataset offering climatic variables over decades with a finer-scale resolution (9 km) than ERA5 [35,36]. In the ArcMap 10.3 software environment, the network of sample points was created using the Fishnet tool. A total of 199 sample sites were included, with a distance of 250 km between each point. We used a systematic sampling plan to ensure consistent and orderly coverage of an area, thereby avoiding the limitations of random sampling [37]. The data were collected from the GEE environment by developing a JavaScript code. By determining JavaScript functions, the daily time series of Pr and T data were converted to the monthly average and then exported as a CSV file for each point.

2.3. Climate Change

The IPCC’s Sixth Assessment Report (AR6) provides the most up-to-date and comprehensive analysis of climate change to date. It synthesizes scientific understanding of the climate system, including past, present, and future climate changes, as well as the human causes, observed impacts, and potential risks of climate change. The report highlights a significant increase of 1.1 degrees Celsius in Earth’s surface temperature since the pre-industrial era, signaling severe consequences such as intensified heat waves, prolonged droughts, erratic precipitation patterns, and the perilous depletion of ice sheets and glaciers. Additionally, the escalating threat of rising sea levels and ocean acidification exacerbates environmental degradation on a global scale. The AR6 findings indicate that specific impacts of climate change, notably sea level rise, are irreversible over centuries to millennia, emphasizing the urgent need for concerted global action to mitigate these effects. The report ominously predicts heightened risks to both human and natural systems if climate change remains unabated, including widespread species extinction, destabilization of ecosystems, food and water insecurity, deepened poverty, and significant repercussions on human health and well-being.

The CMIP6, a collaborative initiative coordinated by the WCRP, is at the forefront of scientific efforts to understand and address these challenges. CMIP6 brings together climate modeling centers worldwide to advance climate modeling by developing more precise, comprehensive, and higher-resolution models. These models simulate diverse climate variables and phenomena, such as temperature and precipitation patterns, sea level rise, and ocean currents. A crucial strength of CMIP6 lies in its adoption of standardized experimental protocols, allowing for rigorous comparison and synthesis of model outputs from different centers. This systematic approach enhances the robustness of assessments in reproducing observed climate patterns and trends. CMIP6 also emphasizes improving models for regional-scale studies and refining our understanding of the role of biogeochemical cycles within the climate system.

A prominent participant in CMIP6 is the Canadian Earth System Model version 5 (CanESM5), developed by the Canadian Centre for Climate Modeling and Analysis. CanESM5 significantly contributes to elucidating various climate-related phenomena, including global temperature changes, precipitation variations, and intricate oceanic circulation dynamics. By simulating the Earth’s intricate climate system under different emission scenarios, CanESM5 provides essential insights for policymakers, guiding the formulation of effective climate policies. CanESM5 has a resolution of approximately 2.8 degrees, equivalent to around 310 km. This high resolution enables detailed regional climate projections, aiding in more precise climate impact assessments. Its comprehensive representation of interactions between atmospheric, oceanic, terrestrial, and cryospheric systems distinguishes itself, offering a nuanced understanding of climate dynamics. Furthermore, its relatively high resolution captures crucial regional climate nuances, enhancing our ability to comprehend the localized impacts of climate change. Beyond its scientific utility, CanESM5 is a powerful tool for exploring the potential repercussions of climate change on ecosystems, agriculture, and human health. This information is invaluable for devising strategies to mitigate and adapt to the evolving climate, making CanESM5 a significant asset for policymakers, researchers, and stakeholders worldwide [38,39,40,41,42,43,44].

2.4. Canadian Drought Monitor

CDM is essential for understanding and managing drought across Canada. By providing accurate and up-to-date assessments of drought, this tool helps with proactive measures to reduce its effects on agriculture, ecosystems, and communities. CDM is also essential in informing policy decisions and resource allocation during drought. Developed by Agriculture and Agri-Food Canada, the CDM provides a clear and up-to-date picture of the severity and extent of droughts and helps in effective water resource management and policy-making. This tool provides a complete view of drought conditions by combining various data, including Pr, T, soil moisture, and river flow. The opinion of experts is also included in this tool. This multi-parameter approach ensures a robust and accurate assessment of drought severity. CDM classifies drought severity using a standard scale from D0 (abnormal drought) to D4 (exceptional drought). This classification helps policymakers, stakeholders, and the public understand the severity of drought conditions. The information provided by the CDM is used by various sectors, including agriculture, water management, and emergency response, to make informed decisions about drought mitigation and adaptation strategies [17,19,45].

This study developed an approach to streamline calculations, considering the intricate and time-consuming nature of computing the CDM index. A model was devised to conduct computations with minimal input effortlessly. In stark contrast to the CDM, a system reliant on expert judgment and the utilization of various drought indices, thereby encountering significant challenges in the calculation process, this pioneering research endeavors to revolutionize drought severity assessment by applying cutting-edge deep learning (DL) techniques. While the CDM’s methodology is undeniably valuable, incorporating expert insights and diverse drought indices, its effectiveness is inevitably hindered by the complexities inherent in manual calculation processes. By harnessing the power of DL, this study seeks to transcend these limitations, offering a more robust and efficient approach to evaluating drought severity. While the CDM methodology effectively delivers periodic drought reports, it introduces subjectivity and potential biases due to its dependence on expert judgment.

Moreover, the absence of a clearly defined calculation methodology and limited transparency in data source selection and interpretation may compromise the reliability and reproducibility of reported drought conditions. To address these limitations, AI-based algorithms were utilized in this research to estimate drought severity. By employing Pr and T data as input variables, the DL models autonomously discern intricate patterns and relationships within the data, offering a more objective and data-driven approach to drought severity assessment. Unlike conventional manual methods prone to human and subjective errors, the DL-based approach furnishes a systematic and transparent framework for drought analysis. The integration of DL techniques enhances the accuracy and reliability of drought severity estimation and facilitates the interpretation and comprehension of fundamental factors influencing drought dynamics. By furnishing a transparent and replicable methodology, this research aims to advance drought monitoring and management efforts in Canada, providing valuable insights for stakeholders and policymakers. Furthermore, the transparent and data-driven nature of the DL-based approach enhances the credibility and applicability of the findings, ultimately contributing to more informed decision-making in Canada’s drought-prone regions. Data from the Canadian site were utilized to acquire the CDM values utilized as output for the AI model. Monthly maps depicting drought classes for the CDM model were downloaded, and study points within each class were determined. These classes were then designated as the output of the DL models at each respective point.

The Canadian Drought Monitoring Maps (CDMM) are generated monthly to assess the extent and intensity of drought nationwide. Utilizing an ordinal scale, the CDMM categorizes drought severity from No Drought (ND) to various levels, including abnormally dry (D0), moderate drought (D1), severe drought (D2), extreme drought (D3), and exceptional drought (D4) [19]. This research investigates future drought conditions, specifically at droughts classified as D1 to D4, representing Moderate Drought to Exceptional Drought levels. While D0 conditions can indicate the onset or recovery from drought, and scenarios with no drought indicate normal conditions, the more severe classifications (D1-D4) pose substantial risks and challenges.

2.5. Convolutional Neural Network

The approach to problems in computer vision and beyond has been revolutionized by Convolutional Neural Networks (CNNs). CNNs are a class of deep neural networks most commonly applied to analyzing visual imagery. Originating from the study of the brain’s visual cortex, the way humans recognize and process visual information is mimicked by CNNs, making them exceptionally good at tasks that involve understanding content from images. This ability stems from their unique architecture, which allows spatial hierarchies of features from input images to be automatically and adaptively learned by them [46].

In a CNN, images are processed through a series of specialized layers, each with a distinct role. The initial convolutional layers use filters to scan over input images, generating feature maps that highlight characteristics like edges, textures, and patterns [47]. During training, the network adjusts these filters to minimize the difference between actual outputs and the model’s predictions. Non-linearity is introduced using activation functions, typically the Rectified Linear Unit (ReLU) [48]. Pooling layers then reduce the spatial dimensions of the feature maps, with max pooling being a common method. Max pooling selects the highest value from a set of pixels within a feature map [49], reducing noise, enhancing feature detection, and lowering computational and memory requirements. This method is also simpler than average pooling, making the network more efficient and robust in pattern recognition. Before making predictions, the features extracted by the convolutional and pooling layers are flattened into a single vector and passed through fully connected layers. The final layer, typically a softmax layer for classification tasks, outputs the probabilities for each class, with the highest probability indicating the model’s prediction. CNN training involves using a large set of labeled images. The predicted outputs are compared to the actual labels during backpropagation, and the loss is calculated. The gradient of this loss is used to update the network’s weights to minimize it. This process is repeated over many iterations, monitoring performance on a validation set to prevent overfitting.

CNN offers several advantages, making it a popular choice across various applications. Automatic feature extraction [50] is facilitated by CNN, reducing the need for domain expertise and manual intervention and streamlining the development process. Hierarchical Feature Learning [51] is another strength of CNN, enabling the networks to learn complex patterns from simpler ones through their hierarchical structure. This hierarchical learning allows CNN to understand input data comprehensively. The translation invariance of CNN [52], where a feature learned in one image location can be recognized in any other, is particularly beneficial for object detection and classification tasks. Reduced parameter count [53] leads to lower memory requirements and computational costs, making it feasible to train deeper networks. The versatility of CNN [54] extends beyond image processing to tasks like natural language processing (NLP), audio analysis, and time series prediction. Lastly, the scalability of CNN [55] allows it to be scaled up for large datasets and complex problems. As highlighted with specific references, these features underline the power of CNN as a tool for machine learning practitioners and researchers, enabling innovative and effective problem-solving.

In the architecture of the CNN designed for predicting drought conditions (Figure 2), a series of layers is meticulously integrated, with each layer fulfilling a specific role and providing unique benefits. The Image Input Layer (imageInputLayer) acts as the gateway for input images, ensuring uniformity in data dimensions essential for consistently processing drought imagery, a critical factor in preserving spatial detail. The Convolution Layer (convolution2dLayer) specializes in feature extraction, offering the significant advantage of automatically learning feature representations from the data, removing manual feature extraction. To bolster the stability and accelerate the training process by mitigating internal covariate shifts, the Batch Normalization Layer (batchNormalizationLayer) adjusts and normalizes the inputs for every layer [56]. Internal covariate shift pertains to the alteration in the input data distribution to the following layers caused by frequent weight modifications during the network’s training phase, which can lead to instability in the learning process [27].

Non-linearity, crucial for capturing complex patterns within the data, is introduced by the ReLU Layer, employing the ReLU activation function. To ensure translation invariance and minimize computational burden, the Max Pooling 2D Layer (maxPooling2dLayer) is utilized twice, efficiently reducing the spatial dimensions of the feature maps by dividing the input into rectangular pooling regions and then selecting the maximum value from each region. This process effectively down-samples the input representation, which reduces the number of parameters and computational load while retaining the most critical features detected by the convolutional layers. By doing so, it helps in achieving translation invariance and reduces the risk of overfitting. The Fully Connected Layer (fullyConnectedLayer) serves as a critical link between the network’s depth and its predictive abilities. It consolidates high-level features extracted from earlier layers and facilitates the classification of inputs into predefined categories, thereby greatly enhancing the network’s overall learning capacity. Concluding the architecture, the Softmax Layer (softmaxLayer) transforms the output into a probability distribution, and the Classification Layer (classificationLayer) assigns class labels based on these probabilities, enabling precise drought condition predictions. Due to their combined effectiveness in analyzing complex spatial data, utilizing these layers within the CNN architecture for drought forecasting enables the model to achieve accurate predictions by leveraging their synergistic strengths in a unified framework. It should be noted that CNN modeling in the current study was performed using the MATLAB environment and its Deep Learning Toolbox.

2.6. Flowchart of the Proposed Framework

This study initially extracted Pr and T values from 1979 to 2023 for 199 study points (Figure 3) using code scripting in the GEE environment and utilizing ERA5-Land data. These data were then utilized as inputs for the machine learning model. Subsequently, the CDM values were extracted from the Canada.ca website, which served as the output of the machine learning model representing drought classes. The CNN method was employed for modeling CDM drought classes, and the best model for drought classification was determined. The input combinations are as follows: Long, Lat, Pr(t), Pr(t−1), Pr(t−2), Pr(t−3), Pr(t−12), T(t), T(t−1), T(t−2), T(t−3), and T(t−12). Therefore, there are 12 inputs in total. Longitude and latitude are fixed in all models. Consequently, all input combinations with 1 to 10 inputs (excluding longitude and latitude) result in 1022 different models. Longitude and latitude were also included to introduce future data to the model and determine its related location.

To determine the input combinations from all 1022 different possibilities, the Extreme Learning Machine (ELM) was employed due to its remarkably fast training method [57,58], requiring only a single iteration for modeling. The ELM framework consists of three main matrices: input weights, biases of hidden neurons, and output weights. The input weights and biases are randomly selected, while the output weights are calculated using the Moore–Penrose pseudoinverse of the hidden layer activation matrix. Detailed mathematical formulations of the ELM can be found in recently published documents [59,60]. Using this model, the gaps in areas labeled as “drought not analyzed” were filled to analyze drought conditions across Canada comprehensively.

The CDM database occasionally designates regions as “drought not analyzed” due to several factors. Firstly, some regions in Canada lack sufficient and continuous meteorological and hydrological data, which hinders the accurate and consistent monitoring of drought conditions, leading to these areas being marked as “drought not analyzed.” Secondly, Canada’s vast and diverse landscape, including remote and sparsely populated areas, poses significant challenges for uniform drought monitoring, as the variations in climate across different regions add complexity to the analysis. Lastly, the CDM employs specific methodologies and indices to assess drought conditions, which may not be fully applicable or effective in certain regions, resulting in excluding those areas from the analysis. In the final stage, using the CanESM5 model, Pr and T values were computed for future periods (2024–2100). Using the DL-based model trained in the previous stage, the drought trends for future years were projected for Canada.

Two well-known indices are used to evaluate the performance of deep learning techniques: Classification Accuracy (CA) and Area Under the Curve (AUC). The mathematical formulations for CA and AUC are as follows:

C A = \frac{T P + T N}{T P + T N + F P + F N}

(1)

The components of these indices are defined as follows: True Positives (TP) represent the correctly predicted positive instances, while True Negatives (TN) denote the correctly predicted negative instances. False Positives (FP) are the incorrectly predicted positive instances and False Negatives (FN) refer to the incorrectly predicted negative instances.

The AUC is calculated using the True Positive Rate (TPR) and the False Positive Rate (FPR) at different threshold values. The TPR, also known as sensitivity or recall:

A U C = \int_{0}^{1} T P R (t) d (F P R (t))

(2)

where

T P R = \frac{T P}{T P + F N}

(3)

F P R = \frac{F P}{F P + T N}

(4)

3. Results

3.1. Performance Evaluation of the CNN

Figure 4 reveals the CA and AUC for 1022 models. The latitude and longitude are constant in all these models, along with 1 to 10 additional inputs. The Classification Accuracy values range from a minimum of 0.6092 to a maximum of 0.6465, with a mean accuracy of 0.6316. This indicates that, on average, the models correctly classified approximately 63.16% of the data points. Indeed, on average, the ELM frameworks have performed with moderate precision in forecasting dry spell categories. The quartiles (25th percentile at 0.6273 and 75th percentile at 0.6361) further confirm that most models have accuracies clustered around the mean, indicating a stable performance distribution across the different input combinations tested. The distribution of forecasts (IQR) is relatively low (IQR = Q3 − Q1 = 0.6361 − 0.6273 = 0.0088, less than 1%, 0.88%), indicating that the frameworks show fairly good stability in categorizing CDM. Specifically, the standard deviation is 0.0064, indicating that the models perform consistently around the mean accuracy. This consistency is beneficial as it suggests that the ELM frameworks have a stable performance in forecasting dry spell categories, with most models achieving similar accuracy levels. While appropriate variability is desired to reflect underlying climate uncertainty, the low variability in our models’ performance is advantageous. It indicates that the models are reliable and not prone to significant fluctuations, which is crucial for making consistent and dependable predictions. The interquartile range (IQR) of 0.0088 further supports this, showing that the models’ accuracies are closely clustered around the mean, reflecting a stable and robust forecasting performance.

Nonetheless, various exceptions are underneath the graph, showing instances where frameworks have shown reduced precision. The mean precision of ELM frameworks could differ due to factors such as framework intricacy, information quality, and parameter choice (Figure 4a).

The AUC values exhibit more variation than CA, with a mean of 0.6418 and a standard deviation of 0.0311, ranging from 0.5595 to 0.7392. This higher variance in AUC indicates a broader range of model performances in distinguishing between classes, reflecting variability in the quality of predictions. The median AUC is 0.6410, with an interquartile range from 0.6202 to 0.6620, showing that while some models significantly outperform others, a large portion achieves moderate AUC scores. This variability underscores the complexity of the modeling process and the importance of various factors affecting model outcomes. The main differences between the 1022 models are the use of different input combinations. Further examination revealed outliers, consisting of four models with notably poor performance. These outliers may indicate potential issues with the quality of input variables (i.e., failure to select effective parameters as input to the model) (Figure 4b).

Considering both CA and AUC, it was found that the best ELM-based model (CA = 0.65 and AUC = 0.74) uses longitude and latitude along with T at the current time as well as with two and three lags (i.e., T(t), T(t−2), T(t−3)), and precipitation with one and two lags (Pr(t−1) and Pr(t−2)). This input combination was used for the CDM modeling by CNN. The modeling results by CNN indicate that its CA and AUC are 0.72 and 0.89, respectively, more than 11% and 20% higher than those for the ELM. This implies that the CNN model more efficiently categorizes samples, resulting in fewer misclassifications. Furthermore, regarding AUC, the CNN model once again outperforms the ELM model. With an AUC of 0.8887, the CNN model displays a greater ability to distinguish between different classes, suggesting superior overall predictive performance compared to the AUC of 0.7392 achieved by the ELM model. These results suggest that the CNN model is superior in forecasting drought incidents using Pr and T data, as it attains higher classification accuracy and AUC values than the ELM model. Therefore, this study chooses the CNN model for future drought category estimation.

3.2. Precipitation Analysis Under Climate Change Scenarios

To effectively illustrate the trends in Pr changes for future periods and compare them with observed values, maps were generated for the observational periods of 1983–2023 and future periods of 2024–2040, 2041–2060, 2061–2080, and 2081–2100 under the SSP126, SSP245, SSP370, and SSP585 scenarios. Analyzing the Pr data across Canadian provinces under different scenarios reveals intricate patterns and offers valuable insights into potential climate impacts. Let us thoroughly compare the scenarios with each other and with the observed Pr levels. In the observed data from 1983 to 2023, most provinces exhibit relatively stable Pr levels, with some showing gradual increases over time (Supplementary Material—Figures S1–S3). NL, PE, and NS are notable for experiencing upward trends, indicating potential positive impacts on water resources and agriculture. However, provinces like BC and SK display notable fluctuations, highlighting inherent variability in Pr patterns across Canada.

Analysis of Pr changes under the SSP126 scenario for Canadian provinces reveals diverse trends across different regions. The most substantial increase is observed in NS, peaking at 2.20% during 2024–2040, while the highest decrease is noted in NU, dropping to −7.15% by 2081–2100. These variations reflect the complex interplay of climatic factors and regional characteristics. While some provinces may experience augmented Pr, potentially alleviating drought concerns, others face heightened water scarcity risks due to reduced Pr.

Under the SSP245 scenario, provinces exhibit varying Pr trends, with some showing stability while others exhibit fluctuations. BC experiences relatively stable Pr levels, which are consistent with observations. The percentage changes in Pr under the SSP245 scenario for Canadian provinces reveal notable variations across different intervals. NS experiences the most significant increase, with a 2.21% rise in Pr from 2024 to 2040, suggesting potentially positive implications for water resources and ecosystems. Conversely, NU faces the most substantial decrease, with Pr declining by −6.32% from 2081 to 2100, indicating heightened aridity and potential challenges for various sectors reliant on adequate water supply. Provinces like AB and BC show relatively moderate changes, suggesting a more stable Pr regime than others.

Under the SSP370 scenario, the analysis of percentage changes in Pr for Canadian provinces reveals divergent trends across different intervals. NS exhibits the most significant increase in Pr, with a rise of 2.23% from 1983 to 2023, suggesting potential positive implications for water resources. Conversely, NU experiences the most substantial decrease, with Pr declining by −7.36% from 2041 to 2060, indicating heightened aridity and potential challenges for water availability. Notably, provinces like AB and BC show relatively moderate changes, suggesting a more stable Pr regime than others. These findings underscore the spatial and temporal variability in Pr changes, emphasizing the need for tailored adaptation strategies to address potential impacts on water resources and ecosystems across Canada under the SSP370 scenario.

Under the SSP370 scenario, the analysis of Pr changes for Canadian provinces from 2024 to 2100 reveals notable variations. NS experiences the highest increase in Pr, reaching a maximum of 2.23% by 2024–2040, suggesting potential positive impacts on water availability. In contrast, NU exhibits the most substantial decrease, with Pr dropping to a minimum of −7.36% during 2081–2100, indicating heightened aridity and potential challenges for water resources.

Under the SSP585 scenario, NS demonstrates the most substantial increase, peaking at 2.20% during 2024–2040, while NU experiences the most pronounced decrease, plummeting to −8.23% by 2081–2100 (Supplementary Material—Figures S1–S3).

These variations underscore the diverse impacts of climate change, with some regions witnessing heightened Pr, potentially leading to flooding risks. In contrast, others face intensified aridity, posing challenges for water resource management [61,62]. The findings emphasize the urgency of implementing adaptive measures tailored to the specific climatic challenges faced by each region to mitigate adverse impacts and enhance resilience.

The analysis of precipitation changes for different periods and their percentage changes compared to the observed values under the SSP126 scenario reveals various trends across regions. In the 2024–2040 period, most areas exhibit decreased precipitation. For instance, NS shows a significant decrease of 2.35%, and NB has a reduction of 2.46%. However, regions like ON and SK show slight increases of 0.47% and 2.83%, respectively.

Moving to the 2041–2060 period, most regions observe a noticeable precipitation increase. Prince Edward Island and NS show significant increases of 4.92% and 5.73%, respectively. QC and NB also exhibit substantial increases of 4.53% and 6.62%. This period reflects a general trend toward increased precipitation across the regions compared to the previous period.

The 2061–2080 and 2081–2100 trends continue with some fluctuations. For the 2061–2080 period, SK and AB show notable increases of 2.22% and 4.71%, respectively, while regions like NL and PE experience decreases of 0.83% and 1.68%. By 2081–2100, most regions show a positive trend, with significant increases in MB and NU of 4.86% and 5.78%, respectively. These trends indicate that while some regions may experience fluctuations, there is an overall trend toward increased precipitation in the long term (Figure 5a).

The SSP245 scenario shows distinct trends across different regions. In the 2024–2040 period, most regions experience a decrease in precipitation, similar to the SSP126 scenario. For instance, NS shows a decrease of 2.36%, and NB has a reduction of 2.47%. However, ON and SK show slight increases of 0.49% and 2.80%, respectively, mirroring the slight increases seen in the SSP126 scenario.

In the 2041–2060 period, there was a noticeable increase in precipitation in most regions under the SSP245 scenario, which was similar to the SSP126 scenario. PE and NS show significant increases of 4.90% and 5.72%, respectively. QC and NB also exhibit substantial increases of 4.52% and 6.60%. These increases are comparable to those observed under SSP126, although the exact percentages differ slightly. This period continues to reflect a general trend toward increased precipitation.

For the 2061–2080 and 2081–2100 periods, the trends under SSP245 continue with some fluctuations. In the 2061–2080 period, SK and AB show notable increases of 2.42% and 4.93%, respectively, while regions like NL and PE experience decreases of 0.74% and 1.67%. By 2081–2100, most regions show a positive trend, with significant increases in MB and NU of 5.01% and 6.56%, respectively. These trends indicate a similar overall trend toward increased precipitation in the long term, as seen under SSP126, but with some variations in the magnitude of changes (Figure 5b).

The analysis of precipitation changes for future periods under the SSP370 scenario shows diverse trends across different regions. In the 2024–2040 period, similar to the SSP126 and SSP245 scenarios, most regions experience a decrease in precipitation. NS shows a decrease of 2.37%, and NB has a reduction of 2.47%. ON and SK display slight increases of 0.47% and 2.86%, respectively, consistent with the trends observed in the previous scenarios.

In the 2041–2060 period, an increase in precipitation is observed across most regions under SSP370, similar to SSP126 and SSP245. PE and NS show significant increases of 4.89% and 5.70%, respectively. QC and NB also exhibit substantial increases of 4.55% and 6.60%, respectively. These increases align closely with those seen in the SSP126 and SSP245 scenarios, indicating a general trend toward increased precipitation in the mid-century across various regions.

For the 2061–2080 and 2081–2100 periods, the trends under SSP370 continue to show fluctuations. In the 2061–2080 period, SK and AB show notable increases of 2.55% and 5.12%, respectively, while regions like NL and PE experience decreases of 0.71% and 1.69%. By 2081–2100, most regions exhibit positive trends, with significant increases in MB and NU of 5.17% and 7.61%, respectively. When compared to SSP126 and SSP245, SSP370 shows similar overall trends but with slightly higher variations in some regions, indicating that while the general trend toward increased precipitation remains, the magnitude of these changes can differ based on the specific scenario (Figure 5c).

The SSP585 scenario analysis reveals significant changes in precipitation patterns across different regions for various future periods. For 2024–2040, most regions exhibit decreases in precipitation similar to the SSP126, SSP245, and SSP370 scenarios. NS shows a decrease of 2.37%, and NB experiences a reduction of 2.47%. However, regions like SK and AB see increases of 2.86% and 3.37%, respectively, reflecting a consistent trend across all scenarios where some areas may experience slight increases despite an overall tendency for reductions.

In the 2041–2060 period, there is a notable increase in precipitation across most regions, consistent with trends observed in SSP126, SSP245, and SSP370. For example, PE and NS show increases of 4.92% and 5.73%, respectively, aligning closely with the increments observed in the previous scenarios. QC and NB also exhibit substantial increases of 4.59% and 6.64%, respectively. These increases indicate a general trend toward wetter conditions in the mid-century, a pattern that is consistent across different SSP scenarios.

In the 2061–2080 and 2081–2100 periods, the SSP585 scenario continues to show variations with an overall tendency for increased precipitation. SK and AB display significant increases of 2.77% and 5.29%, respectively, during the 2061–2080 period, while NL and PE show minor reductions. By 2081–2100, most regions will experience positive trends, with MB and NU seeing increases of 5.38% and 8.48%, respectively. Compared to SSP126, SSP245, and SSP370, the SSP585 scenario generally shows higher variability in precipitation changes, indicating that while the overall trend toward increased precipitation remains, the magnitude and specific impacts can differ significantly depending on the scenario (Figure 5d).

The analysis of scenarios SSP126, SSP245, SSP370, and SSP585 reveals significant variations in precipitation patterns across different regions for future periods. Generally, during the 2024–2040 period, a decrease in precipitation is observed in most regions, although some areas like SK and AB experience slight increases. In the 2041–2060 period, a notable increase in precipitation is seen in most areas, especially in PE and NS, which is consistent across all scenarios. In the 2061–2080 and 2081–2100 periods, the overall trend shifts toward increased precipitation, with substantial rises in regions like MB and NU in the SSP585 scenario. Comparing these scenarios indicates that while there is a general trend toward increased precipitation, the specific amounts and impacts of these changes vary significantly depending on the scenario.

3.3. Temperature Analysis under Climate Change Scenarios

Same as Pr, to effectively demonstrate the trend of T changes for future periods and compare them with observed values, maps were generated for the observational periods 1983–2023 and future periods 2024–2040, 2041–2060, 2061–2080, and 2081–2100 under the SSP126, SSP245, SSP370, and SSP585 scenarios (Supplementary Material—Figures S4–S6). In the observed data spanning from 1983 to 2023, most provinces exhibit relatively stable T trends, with some variations over time. Provinces like NL, PE, and NS show upward trends, indicating a gradual T increase. For instance, NL experiences an increase from −2.34 °C to −0.95 °C, representing a notable rise of approximately 1.39 °C over the observed period.

Under the SSP126 scenario, provinces face even more significant T increases, with some regions experiencing substantial warming trends. NL and PE show exceptionally high T rises, with NL increasing from −2.34 °C to 6.49 °C by 2100, indicating a notable rise of approximately 8.83 °C over the observed period. Provinces in northern Canada, such as YT and NT, also experience notable T increases, highlighting the vulnerability of northern regions to climate change.

The SSP370 and SSP585 scenarios further amplify T increases, with provinces experiencing unprecedented warming trends. NL and PE continue to face significant T rises, with NL increasing from −2.34 °C to 8.76 °C by 2100 under SSP370, marking an alarming rise of approximately 11.10 °C over the observed period. Provinces across Canada must prioritize robust adaptation strategies to address the impacts of climate change and ensure the resilience of communities and ecosystems in the face of rising Ts.

Based on the data illustrated in Figure 6, which depicts the deviation in T from observed values, a significant trend emerges, indicating a noteworthy increase in T across all future periods until 2100. There is a substantial rise in T from 2024 onwards, with a pronounced intensification observed in subsequent periods, notably from 2061 to 2080 and 2081 to 2100. This upward trajectory underscores the urgent need for proactive measures to mitigate and adapt to the accelerating impacts of climate change as we progress further into the century.

Notably, northern provinces, especially NU and NT, consistently exhibit the highest observed T increases across all future periods. This increase reaches its maximum for 2081–2100 and scenario SSP585, amounting to 13.47 °C and 16.07 °C, respectively. These regions, characterized by fragile ecosystems and vulnerable communities, face disproportionate impacts of climate change, exacerbating concerns regarding environmental sustainability and social resilience. In contrast, southern provinces such as BC and AB demonstrate relatively lower percentage changes in observed T, indicating a comparatively reduced impact, though not immunity, from climate change effects. This highlights regional disparities in the magnitude of T changes and underscores the need for tailored approaches to address specific regional vulnerabilities.

Among the scenarios considered, SSP585 consistently emerges as the scenario with the most pronounced T increase across all provinces and future periods. Driven by high greenhouse gas emissions, SSP585 underscores the potential consequences of inadequate climate action. It emphasizes the urgent need for ambitious mitigation efforts to curb T rise and its associated impacts. Conversely, scenario SSP126 exhibits relatively smaller T changes, signaling a less extreme trajectory under lower emission scenarios. However, these scenarios still necessitate coordinated global action to transition toward sustainable, low-carbon pathways.

According to Figure 6, the southern parts of Canada are predicted to experience Temperatures (Ts) above 10 °C in almost all scenarios. These conditions will be more common in southern QC, ON, NB, and NS. Similarly, but less severely, the BC, AB, SK, and MB provinces expect an average annual T exceeding 10 °C. This circumstance affects more expansive areas toward the last period and scenarios (Supplementary Material—Figures S4–S6).

A comprehensive analysis of observed T changes underscores the intricate interplay of temporal, regional, and scenario-specific factors in shaping climate change impacts. It emphasizes the critical imperative for multifaceted adaptation and mitigation strategies tailored to regional conditions to address the diverse challenges of climate change. Urgent and coordinated global action is required to navigate the complex challenges of climate change and forge a sustainable future for all.

Figure 6a illustrates the temperature changes across various regions for different periods under the SSP 126 scenario. The data reveal notable increases in temperature from 2024 to 2040 across all regions. For instance, NL will experience a significant rise of 4.32 °C, representing a 320.22% change compared to historical observations. Similarly, PE will increase by 3.11 °C (51.87%), while NS records a 2.81 °C (40.57%) rise. NB follows with a 3.22 °C increase (66.43%). QC undergoes a substantial change of 4.48 °C (179.80%), with ON seeing a 3.84 °C rise (328.98%). MB experiences a 3.83 °C increase (245.51%), SK 3.26 °C (378.77%), AB 3.13 °C (221.89%), and BC 3.08 °C (287.31%). The northern regions, YT and NT, show increases of 4.03 °C (70.98%) and 4.81 °C (58.94%), respectively, while NU registers the highest temperature change of 6.21 °C (45.14%).

Moving forward from 2041 to 2060, the temperature increases become even more pronounced. NL will experience a further rise of 5.48 °C (406.62%), and PE will experience a 4.15 °C (69.21%) increase. NS records a 3.66 °C (52.78%) rise, while NB sees an increase of 4.03 °C (83.13%). QC undergoes a significant temperature rise of 5.62 °C (225.49%), and Ontario (ON) records a 4.54 °C (389.31%) increase. MB experiences a 4.45 °C rise (284.98%), SK 3.59 °C (417.42%), and AB 3.42 °C (242.35%). BC sees a 3.59 °C increase (334.28%). In the northern regions, YT experiences a rise of 4.61 °C (81.30%), NT 5.76 °C (70.53%), and NU, the highest change of 7.34 °C (53.34%) (Figure 6a).

In the SSP245 Scenario, between 2024 and 2040, significant temperature changes are observed across all regions. NL see an increase of 4.68 °C, representing a 347.11% change. PE experiences a rise of 3.58 °C (59.80%), NS records a 3.17 °C increase (45.77%), and NB sees a rise of 3.72 °C (76.72%). QC experiences a substantial change of 4.90 °C (196.60%), with ON seeing a 4.27 °C rise (365.83%). MB experiences a 4.16 °C increase (266.44%), SK 3.55 °C (412.31%), AB 3.38 °C (239.22%), and BC 3.23 °C (301.10%). The northern regions, YT and NT, show increases of 4.20 °C (73.96%) and 5.03 °C (61.56%), respectively, while NU registers a change of 6.51 °C (47.32%).

From 2041 to 2060, temperature changes become even more pronounced. NL’s increase is 5.92 °C (438.51%), and PE will experience a 4.69 °C rise (78.21%). NS records a 4.09 °C increase (58.95%), NB sees a rise of 4.59 °C (94.71%), and QC undergoes a significant rise of 6.10 °C (245.01%). ON sees a 5.03 °C increase (431.02%), MB 4.83 °C (309.49%), SK 3.93 °C (456.66%), AB 3.71 °C (262.86%), and BC 3.79 °C (352.82%). YT experiences a 4.86 °C rise (85.69%), NT 6.06 °C (74.28%), and NU 7.73 °C (56.23%).

For the period from 2061 to 2080, the temperature increases further intensify. NL see an increase of 6.89 °C (510.93%), PE 5.59 °C (93.35%), NS 5.12 °C (73.87%), and NB 5.73 °C (118.11%). QC experiences a significant change of 7.28 °C (292.42%), ON 6.39 °C (547.77%), MB 6.47 °C (414.40%), SK 5.69 °C (660.89%), AB 5.25 °C (372.34%), and BC 4.95 °C (461.52%). YT experiences a rise of 6.40 °C (112.90%), NT 7.94 °C (97.22%), and NU 9.72 °C (70.65%).

Finally, from 2081 to 2100, NL will experience an increase of 7.84 °C (581.15%), PE 6.41 °C (106.91%), NS 5.69 °C (82.13%), and NB 6.30 °C (130.01%). QC experiences a significant change of 8.21 °C (329.47%), ON 6.95 °C (595.39%), MB 6.85 °C (438.87%), SK 5.77 °C (669.93%), AB 5.39 °C (381.71%), and BC 5.40 °C (503.39%). YT experiences a rise of 7.16 °C (126.28%), NT 8.77 °C (107.40%), and NU 10.75 °C (78.18%). Compared to SSP 126, SSP 245 shows consistently higher temperature increases across Canada. Every region experiences this, with NL and PE as specific examples. The difference grows over time, with later periods in SSP 245 showing rises several degrees higher than SSP 126 (Figure 6b).

In the SSP 370 scenario, between 2024 and 2040, notable temperature increases are observed in all regions under the SSP 370 scenario. NL sees a rise of 4.98 °C (368.93%), while PE experiences an increase of 3.89 °C (64.89%). NS records a 3.49 °C increase (50.33%), and NB observes a 4.03 °C rise (83.09%). QC shows a significant change of 5.23 °C (210.12%), and ON experiences a 4.66 °C rise (399.48%). MB records a 4.59 °C increase (294.20%), SK 3.91 °C (454.14%), AB 3.60 °C (255.39%), and BC 3.28 °C (306.17%). Northern regions such as YT and NT see increases of 4.51 °C (79.48%) and 5.55 °C (68.01%), respectively, while NU registers a rise of 6.89 °C (50.05%).

From 2041 to 2060, temperature changes in the SSP 370 scenario become more pronounced. NL sees an increase of 6.31 °C (467.66%), and PE experiences a 5.07 °C rise (84.65%). NS records a 4.47 °C increase (64.50%), NB a 4.99 °C rise (102.82%), and QC a substantial rise of 6.56 °C (263.18%). ON observes a 5.53 °C increase (474.29%), MB 5.38 °C (344.67%), SK 4.39 °C (509.58%), AB 4.02 °C (284.82%), and BC 3.91 °C (364.63%). YT and NT show increases of 5.25 °C (92.53%) and 6.70 °C (82.07%), respectively, while NU registers a change of 8.26 °C (60.05%).

From 2061 to 2080, temperature increases intensify further in the SSP 370 scenario. NL sees a rise of 9.16 °C (679.37%), PE 7.52 °C (125.42%), NS 6.81 °C (98.32%), and NB 7.72 °C (159.27%). QC experiences a significant change of 9.94 °C (399.08%), ON 9.03 °C (773.80%), MB 9.22 °C (590.52%), SK 7.96 °C (924.59%), AB 7.12 °C (504.27%), and BC 6.45 °C (601.42%). YT sees an increase of 8.22 °C (144.87%), NT 10.64 °C (130.36%), and NU 13.08 °C (95.10%).

When comparing the SSP 370 scenario with SSP 245 and SSP 126, it is evident that the SSP 370 scenario presents the most severe temperature increases across all regions and periods. For instance, from 2024 to 2040, NL will experience a 4.98 °C rise in SSP 370 compared to 4.68 °C in SSP 245 and 4.32 °C in SSP 126. Similarly, PE sees a 3.89 °C increase in SSP 370 compared to 3.58 °C in SSP 245 and 3.11 °C in SSP 126. This trend continues across all periods, with SSP 370 consistently showing higher temperature increases.

From 2041 to 2060, the SSP 370 scenario shows more dramatic changes, such as a 6.31 °C (467.66%) increase in NL compared to 5.92 °C (438.51%) in SSP 245 and 5.48 °C (406.62%) in SSP 126. PE also sees higher increases of 5.07 °C (84.65%) in SSP 370 compared to 4.69 °C (78.21%) in SSP 245 and 4.15 °C (69.21%) in SSP 126. This pattern is consistent across other regions, highlighting the severity of temperature changes in the SSP 370 scenario.

By 2061 to 2080, the differences become even more pronounced. NL sees a 9.16 °C (679.37%) rise in SSP 370 compared to 6.89 °C (510.93%) in SSP 245 and 5.18 °C (384.06%) in SSP 126. This trend continues between 2081 and 2100, with SSP 370 showing the highest increases across all regions. For example, NU sees a rise of 13.08 °C (95.10%) in SSP 370 compared to 9.72 °C (70.65%) in SSP 245 and 6.21 °C (45.14%) in SSP 126.

Overall, the SSP 370 scenario indicates the most severe and rapid temperature increases, emphasizing the urgent need for mitigation efforts to prevent the catastrophic impacts of such drastic global temperature changes (Figure 6c).

The SSP585 Scenario between 2024 and 2040, the SSP585 scenario indicates significant temperature increases across all regions. NL sees a rise of 5.22 °C (386.84%), while PE experiences an increase of 4.02 °C (67.15%). NS records a 3.64 °C increase (52.45%), and NB observes a 4.15 °C rise (85.49%). QC shows a significant change of 5.37 °C (215.37%), and ON experiences a 4.73 °C rise (405.25%). MB records a 4.60 °C increase (294.40%), SK 3.86 °C (448.59%), AB 3.61 °C (255.70%), and BC 3.44 °C (320.94%). Northern regions such as YT and NT see increases of 4.52 °C (79.72%) and 5.65 °C (69.18%), respectively, while NU registers a rise of 7.38 °C (53.64%).

From 2041 to 2060, temperature changes in the SSP585 scenario become more pronounced. NL sees an increase of 6.61 °C (489.67%), and PE experiences a 5.26 °C rise (87.71%). NS records a 4.66 °C increase (67.30%), NB a 5.15 °C rise (106.25%), and QC a substantial rise of 6.75 °C (270.97%). ON observes a 5.67 °C increase (485.90%), MB 5.47 °C (350.17%), SK 4.42 °C (513.89%), AB 4.11 °C (291.02%), and BC 4.15 °C (386.38%). YT and NT show increases of 5.36 °C (94.46%) and 6.89 °C (84.43%), respectively, while NU registers a change of 8.83 °C (64.19%).

From 2061 to 2080, temperature increases intensify further in the SSP585 scenario. NL sees a rise of 10.51 °C (779.22%), PE 8.62 °C (143.76%), NS 7.90 °C (114.01%), and NB 8.83 °C (182.04%). QC experiences a significant change of 11.34 °C (455.18%), ON 10.46 °C (896.30%), MB 10.88 °C (696.81%), SK 9.61 °C (1116.20%), AB 8.79 °C (622.54%), and BC 8.11 °C (755.82%). YT sees an increase of 10.14 °C (178.76%), NT 12.69 °C (155.40%), and NU 15.08 °C (109.63%).

Comparing the SSP585 scenario with SSP 370, SSP 245, and SSP 126 shows that it is evident that the SSP585 scenario presents the most severe temperature increases across all regions and periods. For instance, from 2024 to 2040, NL will experience a 5.22 °C rise in SSP585 compared to 4.98 °C in SSP 370, 4.68 °C in SSP 245, and 4.32 °C in SSP 126. Similarly, PE sees a 4.02 °C increase in SSP585 compared to 3.89 °C in SSP 370, 3.58 °C in SSP 245, and 3.11 °C in SSP 126. This trend continues across all periods, with SSP585 consistently showing higher temperature increases.

From 2041 to 2060, the SSP585 scenario shows more dramatic changes, such as a 6.61 °C (489.67%) increase in NL compared to 6.31 °C (467.66%) in SSP 370, 5.92 °C (438.51%) in SSP 245, and 5.48 °C (406.62%) in SSP 126. PE also sees higher increases of 5.26 °C (87.71%) in SSP585 compared to 5.07 °C (84.65%) in SSP 370, 4.69 °C (78.21%) in SSP 245, and 4.15 °C (69.21%) in SSP 126. This pattern is consistent across other regions, highlighting the severity of temperature changes in the SSP585 scenario.

By 2061 to 2080, the differences become even more pronounced. NL sees a 10.51 °C (779.22%) rise in SSP585 compared to 9.16 °C (679.37%) in SSP 370, 6.89 °C (510.93%) in SSP 245, and 5.18 °C (384.06%) in SSP 126. This trend continues between 2081 and 2100, with SSP585 showing the highest increases across all regions. For example, NU will experience a rise of 15.08 °C (109.63%) in SSP585 compared to 13.08 °C (95.10%) in SSP 370, 9.72 °C (70.65%) in SSP 245, and 6.21 °C (45.14%) in SSP 126.

So, the SSP585 scenario indicates the most severe and rapid temperature increases, highlighting the urgent need for mitigation efforts to prevent the catastrophic impacts of these drastic global temperature changes Figure 6d).

The analysis of temperature changes across Canada under the four SSP scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5) reveals significant differences in projected warming. Under the SSP5-8.5 scenario, the most extreme case, all regions experience the highest temperature increases, particularly notable in the latter part of the century (2081–2100). This scenario indicates severe warming, potentially impacting ecosystems, human health, and infrastructure. In contrast, SSP1-2.6, the most optimistic scenario, shows the least temperature rise, suggesting more manageable climate impacts. The intermediate scenarios, SSP2-4.5 and SSP3-7.0, depict moderate warming trends, reflecting a balance between mitigation efforts and continued emissions. These projections underscore the importance of global climate policies in determining the extent of future warming and its associated impacts.

3.4. Canadian Drought Monitor Projections

The graphs in Figure 7, Figure 8, Figure 9 and Figure 10 show the percentage of Canada’s area affected by drought of various degrees (D1 to D4) and monthly Pr and T from January 2003 to December 2100. The horizontal axis represents the time from January 2003 to December 2100, and the vertical axis represents the percentage of Canada affected by drought.

Different colors in the graph show different intensities of drought. The yellow indicates the area percentage under each degree of drought (D1 to D4). The orange indicates the area’s percentage under moderate to exceptional drought (D2-D4). Brown color represents the percentage of areas under extreme to exceptional drought (D3–D4), and red indicates the area under exceptional drought (D4).

Figure 7 shows the dynamics of drought area percentage, Pr levels, and T variations for Canada within the context of the Mahate model under the SSP126 scenario. The dotted red line illustrates a discernible upward trend, indicating a linear increase in the extent of drought-affected regions over time. This trend is mathematically represented by the equation y = 1 × 10⁻⁰⁴x + 7.0543, implying a trajectory toward drought occurrences across Canada. The graph depicting monthly T fluctuations in Canada exhibits a consistent rise, as depicted by the equation y = 0.0002x − 13.625. Conversely, the depiction of monthly Pr changes in Canada reveals a marginal increase, suggesting a near constancy in Pr levels (y = 3 × 10⁻⁰⁵x + 55.82). Despite rising Ts, this minimal alteration in Pr implies a static water input, further complicating the water availability scenario. Under the SSP126 scenario, Canada is poised to confront more frequent and severe drought episodes due to these climatic shifts.

Figure 8 depicts the changes in the percentage of drought-affected areas, Pr, and T for Canada under the SSP245 scenario. The dotted red line represents the linear trend of increasing drought-affected areas, described by the equation y = 0.0002x + 2.3507. This upward trend indicates that droughts in Canada will become more frequent and severe over time. The graph of monthly T changes in Canada shows a gradual increase in T, represented by the equation y = 0.0003x−15.788. Changes in monthly Pr in Canada indicate a slight increase, suggesting that Pr has remained almost constant (y = 4 × 10⁻⁰⁵x + 55.613). This minimal change in Pr, coupled with rising Ts, suggests a constant water input, complicating water availability further. According to the SSP245 scenario, Canada is expected to experience more frequent and severe droughts due to these climatic changes.

Compared with the SSP126 scenario, both scenarios show an observable increase in T and drought. However, there are significant differences. In the SSP245 scenario, the trend of increasing drought is slightly higher (with a slope y = 0.0002x + 2.3507), indicating more severe droughts than SSP126. The increase in T in SSP245, with a slope y = 0.0003x − 15.788, is higher than that in SSP126 (y = 0.0002x − 13.625), indicating a greater rise in T under this scenario. In both scenarios, the changes in Pr are minimal and almost constant. However, in SSP245, the slope is slightly higher (y = 4 × 10⁻⁰⁵x + 55.613), indicating a marginally higher increase in Pr compared to SSP126. So, the SSP245 scenario indicates more critical conditions than SSP126, leading to more severe droughts and higher T increases.

Figure 9 shows the changes in the percentage of drought-affected areas, Pr, and T for Canada under the SSP370 scenario. The red line represents the linear trend of increasing drought-affected areas, described by the equation y = 0.0004x − 4.2277. This trend suggests that droughts in Canada will become significantly more frequent and severe over time. The graph illustrating monthly T changes shows a notable rise, represented by the equation y = 0.0003x − 18.912. The analysis of monthly Pr changes in Canada reveals a slight increase, suggesting that Pr levels have remained relatively constant (y = 4 × 10⁻⁰⁵x + 55.415). Despite the minimal increase in Pr, rising Ts imply a constant water input, complicating water availability and exacerbating the impact of higher Ts on drought conditions. These graphs highlight the intricate relationship between T, Pr, and drought. Under the SSP370 scenario, Canada is expected to face significantly more frequent and severe droughts due to these climatic changes. Comparing SSP370 with SSP126 and SSP245 scenarios reveals notable differences. In the SSP370 scenario, the trend of increasing drought is more pronounced, with a steeper slope (y = 0.0004x − 4.2277), indicating a much higher droughting severity than SSP126 and SSP245. The T increase in SSP370, indicated by the equation (y = 0.0003x − 18.912), is significant and mirrors the trend observed in SSP245 (y = 0.0003x − 15.788), but it is considerably higher than in SSP126 (y = 0.0002x − 13.625). Pr trends remain relatively constant across all three scenarios, but the slight increase in SSP370 (y = 4 × 10⁻⁰⁵x + 55.415) is similar to SSP245 (y = 4 × 10⁻⁰⁵x + 55.613) and slightly higher than SSP126 (y = 3 × 10⁻⁰⁵x + 55.82).

So, the SSP370 scenario suggests the most critical conditions, leading to the most severe droughts and the highest T increases compared to SSP126 and SSP245. This indicates that the SSP370 scenario will have the most drastic impacts on Canada’s climate, making it imperative to consider these projections in future planning and mitigation strategies.

Figure 10 depicts the changes in the percentage of drought-affected areas, Pr, and T for Canada under the SSP585 scenario. The red line represents the linear trend of increasing drought-affected areas, described by the equation y = 0.0005x − 9.6157. This steep upward trend suggests that droughts in Canada will become significantly more frequent and severe over time. The graph of monthly T changes shows a marked increase, represented by the equation y = 0.0004x − 20.958. The analysis of monthly Pr changes in Canada reveals a slight increase, with the equation y = 5 × 10⁻⁰⁵x + 55.249 indicating that Pr levels have remained relatively constant. These graphs underscore the intricate relationship between T, Pr, and drought. Under the SSP585 scenario, Canada is expected to face significantly more frequent and severe droughts due to these climatic changes.

Several significant differences emerge when comparing the SSP585 scenario with SSP126, SSP245, and SSP370. In the SSP585 scenario, the trend of increasing drought is the most pronounced, with the steepest slope (y = 0.0005x − 9.6157), indicating the highest severity of droughts compared to the other scenarios. The T increase in SSP585, indicated by the equation y = 0.0004x − 20.958, is the most significant, surpassing the T increases in SSP370 (y = 0.0003x − 18.912), SSP245 (y = 0.0003x − 15.788), and SSP126 (y = 0.0002x − 13.625). Pr trends remain relatively constant across all four scenarios, but the slight increase in SSP585 (y = 5 × 10⁻⁰⁵x + 55.249) is higher than the increases observed in SSP370 (y = 4 × 10⁻⁰⁵x + 55.415), SSP245 (y = 4 × 10⁻⁰⁵x + 55.613), and SSP126 (y = 3 × 10⁻⁰⁵x + 55.82).

The SSP585 scenario suggests the most critical conditions, leading to the most severe droughts and the highest T increases compared to SSP126, SSP245, and SSP370. This scenario indicates the most drastic impacts on Canada’s climate, making it imperative to incorporate these projections into future planning and mitigation strategies to address the anticipated extreme climate changes.

To better show the changes in precipitation and temperature, Standard Deviation (SD) and Coefficient of Variation (CV) were used under four different SSP scenarios (ssp126, ssp245, ssp370, ssp585) from 2003 to 2100. The SD of precipitation is relatively consistent across the SSPs, with values of 17.0086 for ssp126, 16.9883 for ssp245, 16.9619 for ssp370, and 16.9386 for ssp585. This consistency suggests stable variability in precipitation projections across these different future scenarios. In contrast, the SD of temperature shows a gradual increase, with values of 12.0840 for ssp126, 12.1385 for ssp245, 12.2457 for ssp370, and 12.3347 for ssp585. The CV for precipitation remains relatively stable, with values of 0.2946 for ssp126, 0.2941 for ssp245, 0.2933 for ssp370, and 0.2926 for ssp585. This reflects consistent relative variability in precipitation across different scenarios, indicating that the relative distribution of precipitation remains similar regardless of the SSP. However, the CV for temperature shows significant fluctuations. For SSP126, the CV is −12.1646; for SSP245, it is −20.7236; for SSP370, it drastically deviates to −496.1145; and for SSP585, it is 37.9153. The extreme value for SSP370 suggests potential inconsistencies or anomalies in the temperature projections for this scenario. The negative CV values for SSP126 and SSP245 indicate that temperature variations might be inversely proportional to the mean temperature, while the positive CV for SSP585 suggests direct proportionality.

So, the data highlight that while the variability in precipitation remains relatively stable across different SSP scenarios, temperature variability is more erratic and shows significant differences. The increasing SD of temperature from SSP126 to SSP585 suggests that as future scenarios become more extreme, temperature variations are expected to rise. The CV for precipitation remains stable, reflecting consistent relative variability, whereas the CV for temperature shows substantial variability, particularly under SSP370, indicating potential anomalies or higher uncertainty in temperature projections for that scenario. This analysis underscores the importance of considering both SD and CV to understand the full scope of variability in climate projections.

4. Discussion

4.1. Interpretation of Results

The results of this study indicate a clear trend toward more frequent and severe droughts across Canada, particularly under the SSP585 scenario. This scenario, characterized by high greenhouse gas emissions and minimal climate change mitigation efforts, projects the most significant temperature increases and the most severe drought conditions compared to SSP126, SSP245, and SSP370. These findings align with previous studies that indicated that higher emission scenarios are associated with greater climate extremes, including droughts.

The linear trend of increasing drought-affected areas underscores droughts’ escalating frequency and severity over time. Similarly, the marked increase in temperature highlights the urgent need for effective climate change mitigation strategies. The relatively stable Pr levels across all scenarios suggest that changes in T are likely the primary driver of the increasing drought risk in Canada.

4.2. Policy Implication

Based on the presented research findings, Canada faces significant risks from increasing drought and rising Ts. Immediate and decisive actions are essential to address these challenges and adapt to the changing climate. Figure 11 outlines critical short-term and long-term strategies to combat climate change, with brief explanations below.

Based on analyses of various scenarios (SSP126, SSP245, SSP370, and SSP585), critical policy implications are identified for addressing the increasing frequency and severity of droughts, rising Ts, and relatively stable Pr levels in Canada. Recommendations include prioritizing the reduction in greenhouse gas emissions, exceeding Paris Agreement targets, and transitioning to renewable energy sources like wind, solar, and hydropower. Promoting water conservation methods across agricultural, industrial, and domestic sectors, including low-consumption irrigation systems and drought-resistant crops, is crucial. Additionally, developing integrated water resource management programs that consider anticipated water availability variability and sustainable groundwater and surface water storage is essential. Investing in weather-resistant infrastructure, urban green spaces, and energy-efficient buildings to withstand extreme weather events is imperative. Providing financial and technical support to farmers to adopt innovative farming practices, diversify agriculture, and enhance climate resilience is necessary. Enhancing climate monitoring, research, and data collection for better prediction and response to changing climate conditions is vital. Conducting public awareness campaigns and integrating climate education into school curricula to foster climate awareness is essential. Actively participating in international climate forums, sharing knowledge, and adhering to global agreements like the Paris Agreement are crucial. Creating integrated policy frameworks aligning climate action with economic, social, and environmental goals is necessary. Establishing financial mechanisms such as green bonds, climate funds, and insurance plans to protect against weather-related damage is essential. Implementation of these recommendations can help Canada mitigate the adverse impacts projected in various scenarios and promote a resilient and sustainable future.

4.3. Future Research Directions

Future research should focus on refining climate models to improve drought predictions’ accuracy and better understand the complex interactions between temperature, precipitation, and other climatic factors. Additionally, research should explore the effectiveness of various mitigation and adaptation strategies in reducing the impacts of droughts and enhancing climate resilience. Investigating the socioeconomic impacts of droughts on different regions and communities in Canada can also provide valuable insights for targeted policy interventions. Moreover, exploring and comparing other deep learning models to further enhance the accuracy and reliability of drought forecasting in Canada is suggested.

4.4. Broader Context

The implications of this study extend beyond Canada, highlighting the global need for coordinated climate action. As other regions may face similar or even more severe climate challenges, the insights gained from this research can inform international efforts to combat climate change. By actively participating in global climate forums and adhering to international agreements like the Paris Agreement, Canada can contribute to global efforts to mitigate climate change and promote a sustainable future.

5. Conclusions

This study analyzes the projected impacts of climate change on Canada’s precipitation, temperature, and drought conditions under four Shared Socioeconomic Pathways (SSP126, SSP245, SSP370, and SSP585) based on the 6th IPCC report. The main findings of this research are summarized as follows:

The SSP126 scenario assumes significant emission reductions, resulting in the least severe increases in temperature and drought frequency. The results highlight the benefits of aggressive emission reduction strategies in mitigating climate change impacts.
The SSP245 scenario represents moderate emission reductions, projecting increased temperature and drought severity. The findings emphasize the need for enhanced emissions reduction efforts to prevent moderate climate impacts, suggesting that while some mitigation efforts are effective, they need to be strengthened to avoid more severe outcomes.
The SSP370 scenario, characterized by higher emissions, forecasts more pronounced temperature increases and a significant rise in drought conditions. The implications of this scenario stress that without substantial mitigation efforts, Canada could face strained water resources, reduced agricultural productivity, and an increased frequency and intensity of extreme weather events, illustrating the critical need for robust climate policies to curb emissions.
The SSP585 scenario assumes the highest emissions levels, predicting the most severe temperature increases and widespread droughts. This scenario underscores the catastrophic potential of unmitigated climate change, presenting a dire outlook with profound implications for ecosystems, the economy, and public health. Severe drought conditions under this scenario will likely cause significant disruptions in water access, agricultural failures, and heightened risks of heat-related illnesses and fatalities.
Across all scenarios, stable precipitation levels are offset by higher evaporation rates and decreased soil moisture due to temperature increases, exacerbating drought conditions. This trend indicates that maintaining current precipitation levels alone will not counteract the adverse effects of rising temperatures.

So, the analysis underscores the urgency of implementing effective climate policies to reduce emissions, adapt to climate change, and enhance resilience. Proactive measures are critical to mitigating climate change’s adverse effects and ensuring a sustainable future for Canada.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cli12080119/s1. Figure S1. Precipitation variability across the 10 Canadian provinces and 3 territories, spanning from the observational period (1983–2023) to future projections (2024–2040, 2041–2060, 2061–2080, and 2081–2100) under five distinct scenarios (SSP126, SSP245, SSP370, and SSP585), employing the CanESM5 model within the CMIP6 framework. Figure S2. Precipitation changes relative to observed temperatures across the 10 Canadian provinces and 3 territories for each scenario. Figure S3. Zoning of Projected Average Annual Precipitation Anomalies in Canada (2024–2100) Compared to the Observed Period (1983–2023) using CanESM5 from CMIP6. Figure S4. Temperature variability across the 10 Canadian provinces and 3 territories, spanning from the observational period (1983–2023) to future projections (2024–2040, 2041–2060, 2061–2080, and 2081–2100) under five distinct scenarios (SSP126, SSP245, SSP370, and SSP585), employing the CanESM5 model within the CMIP6 framework. Figure S5. Temperature changes relative to observed temperatures across the 10 Canadian provinces and 3 territories for each scenario. Figure S6. Zoning of Projected Average Annual Temperature Anomalies in Canada (2024–2100) Compared to the Observed Period (1983–2023) using CanESM5 from CMIP6.

Author Contributions

K.S.: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Climate Change Analysis, Writing—original draft, Writing—review and editing, Visualization; A.A.: Conceptualization, Methodology, Validation, Formal analysis, Preparation maps, Writing—review and editing; I.E.: Conceptualization, Methodology, Validation, Formal analysis, Writing—review and editing; H.C.: Data collection, Formal analysis, Writing—review and editing; S.F.: Data collection, Formal analysis, Writing—review and editing; S.J.G.: Conceptualization, Writing—review and editing, Supervising; H.B.: Conceptualization, Writing—review and editing, Supervising. All authors have read and agreed to the published version of the manuscript.

Funding

The corresponding authors acknowledged the financial support provided by the Natural Science and Engineering Research Council of Canada (NSERC) Discover Grant (#RGPIN-2020-04583) and the “Fond de Recherche du Québec- Nature et Technologies,” Québec Government (#B2X—315020).

Data Availability Statement

The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request. For researchers and readers of the article, we suggest accessing the ERA5-Land reanalysis dataset. The data related to the monitoring and reporting of drought in Canada can be downloaded for free in shapefile format from Agriculture Canada’s website at agriculture.canada.ca.

Acknowledgments

The authors express their gratitude to the providers of all the datasets used in this study. We also appreciate the anonymous reviewers and editors for their insightful comments that helped enhance this paper.

Conflicts of Interest

The authors declare no conflicts of interest.

References

Ehteram, M.; Achite, M.; Sheikh Khozani, Z.; Farrokhi, A. Drought prediction using ensemble models. Acta Geophys. 2024, 72, 945–982. [Google Scholar] [CrossRef]
Ha, T.V.; Huth, J.; Bachofer, F.; Kuenzer, C. A review of earth observation-based drought studies in Southeast Asia. Remote Sens. 2022, 14, 3763. [Google Scholar] [CrossRef]
Wang, T.; Tu, X.; Singh, V.P.; Chen, X.; Lin, K.; Zhou, Z. Drought prediction: Insights from the fusion of LSTM and multi-source factors. Sci. Total Environ. 2023, 902, 166361. [Google Scholar] [CrossRef] [PubMed]
Livneh, B.; Hoerling, M.P. The physics of drought in the US central Great Plains. J. Clim. 2016, 29, 6783–6804. [Google Scholar] [CrossRef]
Luo, L.; Apps, D.; Arcand, S.; Xu, H.; Pan, M.; Hoerling, M. Contribution of temperature and precipitation anomalies to the California drought during 2012–2015. Geophys. Res. Lett. 2017, 44, 3184–3192. [Google Scholar] [CrossRef]
Veettil, A.V.; Mishra, A.K. Quantifying thresholds for advancing impact-based drought assessment using classification and regression tree (CART) models. J. Hydrol. 2023, 625, 129966. [Google Scholar] [CrossRef]
Bonsal, B.R.; Wheaton, E.E.; Chipanshi, A.C.; Lin, C.; Sauchyn, D.J.; Wen, L. Drought research in Canada: A review. Atmos.-Ocean 2011, 49, 303–319. [Google Scholar] [CrossRef]
Carrillo, J.; Hernández-Barrera, S.; Expósito, F.J.; Díaz, J.P.; González, A.; Pérez, J.C. The uneven impact of climate change on drought with elevation in the Canary Islands. NPJ Clim. Atmos. Sci. 2023, 6, 31. [Google Scholar] [CrossRef]
Park, S.; Seo, E.; Kang, D.; Im, J.; Lee, M.I. Prediction of drought on pentad scale using remote sensing data and MJO index through random forest over East Asia. Remote Sens. 2018, 10, 1811. [Google Scholar] [CrossRef]
IPCC. Climate Change 2007: The Physical Science Basis: Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change; Cambridge University Press: Cambridge, UK; New York, NY, USA, 2007. [Google Scholar]
Ebtehaj, I.; Bonakdari, H. A comprehensive comparison of the fifth and sixth phases of the coupled model intercomparison project based on the Canadian earth system models in spatio-temporal variability of long-term flood susceptibility using remote sensing and flood frequency analysis. J. Hydrol. 2023, 617, 128851. [Google Scholar] [CrossRef]
Zhou, Z.; Tang, W.; Li, M.; Cao, W.; Yuan, Z. A novel hybrid intelligent SOPDEL model with comprehensive data preprocessing for long-time-series climate prediction. Remote Sens. 2023, 15, 1951. [Google Scholar] [CrossRef]
Luo, N.; Guo, Y.; Chou, J.; Gao, Z. Added value of CMIP6 models over CMIP5 models in simulating the climatological precipitation extremes in China. Int. J. Climatol. 2022, 42, 1148–1164. [Google Scholar] [CrossRef]
Salimi, A.; Ghobrial, T.; Bonakdari, H. Comparison of the Performance of CMIP5 and CMIP6 in the Predic-tion of Rainfall Trends, Case Study Quebec City. Environ. Sci. Proc. 2023, 25, 42. [Google Scholar] [CrossRef]
Sobie, S.R.; Zwiers, F.W.; Curry, C.L. Climate Model Projections for Canada: A Comparison of CMIP5 and CMIP6. Atmos.-Ocean 2021, 59, 269–284. [Google Scholar] [CrossRef]
Lawrimore, J.; Heim, R.R.; Svoboda, M.; Swail, V.; Englehart, P.J. Beginning a new era of drought monitoring across North America. Bull. Am. Meterol. Soc. 2002, 83, 1191–1192. Available online: https://journals.ametsoc.org/downloadpdf/view/journals/bams/83/8/1520-0477-83_8_1191.pdf (accessed on 22 March 2024). [CrossRef]
Mardian, J.; Champagne, C.; Bonsal, B.; Berg, A. A machine learning framework for predicting and understanding the Canadian drought monitor. Water Resour. Res. 2023, 59, e2022WR033847. [Google Scholar] [CrossRef]
Tadesse, T.; Champagne, C.; Wardlow, B.D.; Hadwen, T.A.; Brown, J.F.; Demisse, G.B.; Bayissa, Y.A.; Davidson, A.M. Building the vegetation drought response index for Canada (VegDRI-Canada) to monitor agricultural drought: First results. GIScience Remote Sens. 2017, 54, 230–257. [Google Scholar] [CrossRef]
Mardian, J.; Champagne, C.; Bonsal, B.; Daneshfar, B.; Berg, A. Agricultural Risk Assessments Based on the Canadian Drought Monitor: A Bayesian Neural Network Approach. Agric. For. Meteorol. 2023, 353, 110056, Available at SSRN 4583872. [Google Scholar] [CrossRef]
Prasad, R.; Deo, R.C.; Li, Y.; Maraseni, T. Ensemble committee-based data intelligent approach for generating soil moisture forecasts with multivariate hydro-meteorological predictors. Soil Tillage Res. 2018, 181, 63–81. [Google Scholar] [CrossRef]
Salimi, A.; Noori, A.; Ebtehaj, I.; Ghobrial, T.; Bonakdari, H. Advancing Spatial Drought Forecasts by Integrating an Improved Outlier Robust Extreme Learning Machine with Gridded Data: A Case Study of the Lower Mainland Basin, British Columbia, Canada. Sustainability 2024, 16, 3461. [Google Scholar] [CrossRef]
Haidar, A.; Verma, B. Monthly rainfall forecasting using one-dimensional deep convolutional neural network. IEEE Access 2018, 6, 69053–69063. [Google Scholar] [CrossRef]
Wang, J.H.; Lin, G.F.; Chang, M.J.; Huang, I.H.; Chen, Y.R. Real-time water-level forecasting using dilated causal convolutional neural networks. Water Resour. Manag. 2019, 33, 3759–3780. [Google Scholar] [CrossRef]
Kim, D.Y.; Song, C.M. Developing a discharge estimation model for ungauged watershed using CNN and hydrological image. Water 2020, 12, 3534. [Google Scholar] [CrossRef]
Khosravi, K.; Panahi, M.; Golkarian, A.; Keesstra, S.D.; Saco, P.M.; Bui, D.T.; Lee, S. Convolutional neural network approach for spatial prediction of flood hazard at national scale of Iran. J. Hydrol. 2020, 591, 125552. [Google Scholar] [CrossRef]
Song, C.M. Data construction methodology for convolution neural network based daily runoff prediction and assessment of its applicability. J. Hydrol. 2022, 605, 127324. [Google Scholar] [CrossRef]
Tripathy, K.P.; Mishra, A.K. Deep learning in hydrology and water resources disciplines: Concepts, methods, applications, and research directions. J. Hydrol. 2023, 628, 130458. [Google Scholar] [CrossRef]
Ehteram, M.; Ahmed, A.N.; Sheikh Khozani, Z.; El-Shafie, A. Convolutional neural network-support vector machine model-gaussian process regression: A new machine model for predicting monthly and daily rainfall. Water Resour. Manag. 2023, 37, 3631–3655. [Google Scholar] [CrossRef]
Maity, R.; Khan, M.I.; Sarkar, S.; Dutta, R.; Maity, S.S.; Pal, M.; Chanda, K. Potential of Deep Learning in drought assessment by extracting information from hydrometeorological precursors. J. Water Clim. Chang. 2021, 12, 2774–2796. [Google Scholar] [CrossRef]
Hao, R.; Yan, H.; Chiang, Y.M. Forecasting the Propagation from Meteorological to Hydrological and Agricultural Drought in the Huaihe River Basin with Machine Learning Methods. Remote Sens. 2023, 15, 5524. [Google Scholar] [CrossRef]
Danandeh Mehr, A.; Rikhtehgar Ghiasi, A.; Yaseen, Z.M.; Sorman, A.U.; Abualigah, L. A novel intelligent deep learning predictive model for meteorological drought forecasting. J. Ambient Intell. Humaniz. Comput. 2023, 14, 10441–10455. [Google Scholar] [CrossRef]
Chen, Z.; Wang, G.; Wei, X.; Liu, Y.; Duan, Z.; Hu, Y.; Jiang, H. Basin-Scale Daily Drought Prediction Using Convolutional Neural Networks in Fenhe River Basin, China. Atmosphere 2024, 15, 155. [Google Scholar] [CrossRef]
Liu, W.; Huang, Y.; Wang, H. Effective deep learning seasonal prediction model for summer drought over China. Earth’s Future 2024, 12, e2023EF004409. [Google Scholar] [CrossRef]
Khan, M.I.; Maity, R. Development of a Long-Range Hydrological Drought Prediction Framework Using Deep Learning. Water Resour. Manag. 2024, 38, 1497–1509. [Google Scholar] [CrossRef]
Muñoz-Sabater, J.; Dutra, E.; Agustí-Panareda, A.; Albergel, C.; Arduini, G.; Balsamo, G.; Boussetta, S.; Choulga, M.; Harrigan, S.; Hersbach, H.; et al. ERA5-Land: A state-of-the-art global reanalysis dataset for land applications. Earth Syst. Sci. Data 2021, 13, 4349–4383. [Google Scholar] [CrossRef]
Soltani, K.; Ebtehaj, I.; Amiri, A.; Azari, A.; Gharabaghi, B.; Bonakdari, H. Mapping the spatial and temporal variability of flood susceptibility using remotely sensed normalized difference vegetation index and the forecasted changes in the future. Sci. Total Environ. 2021, 770, 145288. [Google Scholar] [CrossRef] [PubMed]
Jensen, J.R. Introductory Digital Image Processing: A Remote Sensing Perspective, 4th ed.; Prentice Hall Press: Upper Saddle River, NJ, USA, 2015; Available online: https://www.pearson.com/en-us/subject-catalog/p/introductory-digital-image-processing-a-remote-sensing-perspective/P200000006907/9780137551118 (accessed on 22 March 2024).
Intergovernmental Panel on Climate Change. Climate Change 2021–2023: The Sixth Assessment Report; Pörtner, H.O., Roberts, D.C., Tignor, M., Poloczanska, E.S., Mintenbeck, K., Alegría, A., Craig, M.H., Langsdorf, S., Löschke, S., Möller, V., et al., Eds.; Cambridge University Press: Cambridge, UK, 2021–2023. [Google Scholar]
Intergovernmental Panel on Climate Change (IPCC). Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change; Cambridge University Press: Cambridge, UK, 2021. [Google Scholar]
Intergovernmental Panel on Climate Change (IPCC). Climate Change 2022: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change; Cambridge University Press: Cambridge, UK, 2022. [Google Scholar]
Intergovernmental Panel on Climate Change (IPCC). Climate Change 2022: Mitigation of Climate Change. Contribution of Working Group III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC); Cambridge University Press: Cambridge, UK, 2022. [Google Scholar]
Intergovernmental Panel on Climate Change (IPCC). AR6 Synthesis Report: Climate Change 2023. Contribution of Working Groups I, II, and III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change; Cambridge University Press: Cambridge, UK, 2023. [Google Scholar]
Soltani, K.; Azari, A. Forecasting Groundwater Anomaly in the Future Using Satellite Information and Machine Learning. J. Hydrol. 2022, 612, 128052. [Google Scholar] [CrossRef]
Amiri, A.; Soltani, K.; Ebtehaj, I.; Bonakdari, H. A novel machine learning tool for current and future flood susceptibility mapping by integrating remote sensing and geographic information systems. J. Hydrol. 2024, 632, 130936. [Google Scholar] [CrossRef]
Canada, Agriculture and Agri-Food Canada. Canadian Drought Monitor. agriculture.canada.ca. Available online: https://agriculture.canada.ca/en/agricultural-production/weather/canadian-drought-monitor (accessed on 9 May 2024).
Imanian, H.; Mohammadian, A.; Farhangmehr, V.; Payeur, P.; Goodarzi, D.; Hiedra Cobo, J.; Shirkhani, H. A comparative analysis of deep learning models for soil temperature prediction in cold climates. Theor. Appl. Climatol. 2024, 155, 2571–2587. [Google Scholar] [CrossRef]
Hoa, P.V.; Binh, N.A.; Hong, P.V.; An, N.N.; Thao, G.T.P.; Hanh, N.C.; Ngo, P.T.T.; Bui, D.T. One-dimensional deep learning driven geospatial analysis for flash flood susceptibility mapping: A case study in North Central Vietnam. Earth Sci. Inform. 2024, 1–22. [Google Scholar] [CrossRef]
Gyaneshwar, A.; Mishra, A.; Chadha, U.; Raj Vincent, P.D.; Rajinikanth, V.; Pattukandan Ganapathy, G.; Srinivasan, K. A contemporary review on deep learning models for drought prediction. Sustainability 2023, 15, 6160. [Google Scholar] [CrossRef]
Grégoire, G.; Fortin, J.; Ebtehaj, I.; Bonakdari, H. Forecasting pesticide use on golf courses by integration of deep learning and decision tree techniques. Agriculture 2023, 13, 1163. [Google Scholar] [CrossRef]
Zhao, Y.; Xu, Y.; Ye, J.; Zhang, X.; Long, Z. Urban Water Supply Forecasting Based on CNN-LSTM-AM Spatiotemporal Deep Learning Model. IEEE Access 2023, 11, 144204–144212. [Google Scholar] [CrossRef]
Pang, Y.; Yu, J.; Xi, L.; Ge, D.; Zhou, P.; Hou, C.; He, P.; Zhao, L. Remote Sensing Extraction of Lakes on the Tibetan Plateau Based on the Google Earth Engine and Deep Learning. Remote Sens. 2024, 16, 583. [Google Scholar] [CrossRef]
Dixit, C.; Satapathy, S.M. Deep CNN with late fusion for real time multimodal emotion recognition. Expert Syst. Appl. 2024, 240, 122579. [Google Scholar] [CrossRef]
Zamani, M.G.; Nikoo, M.R.; Jahanshahi, S.; Barzegar, R.; Meydani, A. Forecasting water quality variable using deep learning and weighted averaging ensemble models. Environ. Sci. Pollut. Res. 2023, 30, 124316–124340. [Google Scholar] [CrossRef] [PubMed]
Zeng, G.; Ma, Y.; Du, M.; Chen, T.; Lin, L.; Dai, M.; Luo, H.; Hu, L.; Zhou, Q.; Pan, X. Deep convolutional neural networks for aged microplastics identification by Fourier transform infrared spectra classification. Sci. Total Environ. 2024, 913, 169623. [Google Scholar] [CrossRef] [PubMed]
Feng, F.; Ghorbani, H.; Radwan, A.E. Predicting groundwater level using traditional and deep machine learning algorithms. Front. Environ. Sci. 2024, 12, 1291327. [Google Scholar] [CrossRef]
Ioffe, S.; Szegedy, C. Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift. In Proceedings of the 32nd International Conference on Machine Learning, Lille, France, 7–9 July 2015; pp. 448–456. Available online: https://proceedings.mlr.press/v37/ioffe15.pdf (accessed on 22 March 2024).
Ebtehaj, I.; Sattar, A.M.; Bonakdari, H.; Zaji, A.H. Prediction of scour depth around bridge piers using self-adaptive extreme learning machine. J. Hydroinform. 2017, 19, 207–224. [Google Scholar] [CrossRef]
Ebtehaj, I.; Soltani, K.; Amiri, A.; Faramarzi, M.; Madramootoo, C.A.; Bonakdari, H. Prognostication of shortwave radiation using an improved No-Tuned fast machine learning. Sustainability 2021, 13, 8009. [Google Scholar] [CrossRef]
Bonakdari, H.; Ebtehaj, I.; Ladouceur, J.D. Machine Learning in Earth, Environmental and Planetary Sciences: Theoretical and Practical Applications; Elsevier: Amsterdam, The Netherlands, 2023; ISBN 9780443152856. Available online: https://books.google.com/books?hl=en&lr=&id=dmilEAAAQBAJ&oi=fnd&pg=PP1&dq=60.%09Bonakdari,+H.,+Ebtehaj,+I.,+%26+Ladouceur,+J.+D.+(2023).+Machine+Learning+in+Earth,+Environmental+and+Planetary+Sciences:+Theoretical+and+Practical+Applications.+Elsevier.+ISBN:+9780443152856&ots=XUXJLaY_h0&sig=II-r41AtdFHMchVyVgyYUAJh3Q0#v=onepage&q&f=false (accessed on 22 March 2024).
Ebtehaj, I.; Bonakdari, H.; Safari, M.J.S.; Gharabaghi, B.; Zaji, A.H.; Madavar, H.R.; Khozani, Z.S.; Es-haghi, M.S.; Shishegaran, A.; Mehr, A.D. Combination of sensitivity and uncertainty analyses for sediment transport modeling in sewer pipes. Int. J. Sediment Res. 2020, 35, 157–170. [Google Scholar] [CrossRef]
Bonakdari, H.; Zaji, A.H.; Soltani, K.; Gharabaghi, B. Improving the accuracy of a remotely-sensed flood warning system using a multi-objective pre-processing method for signal defects detection and elimination. C. R. Geosci. 2020, 352, 73–86. [Google Scholar] [CrossRef]
Soltani, K.; Amiri, A.; Zeynoddin, M.; Ebtehaj, I.; Gharabaghi, B.; Bonakdari, H. Forecasting monthly fluctuations of lake surface areas using remote sensing techniques and novel machine learning methods. Theor. Appl. Climatol. 2021, 143, 713–735. [Google Scholar] [CrossRef]

Figure 1. The relief map of the study area. The black dots show the distribution of sample points across Canada.

Figure 2. The Schematic of the CNN’s structure.

Figure 3. Research flowchart.

Figure 4. (a) Classification Accuracy for 1022 ELM models. (b) Area Under the Curve (AUC) for 1022 ELM models.

Figure 5. Zoning of Projected Average Annual Precipitation Anomalies in Canada (2024–2100) Compared to the Observed Period (1983–2023) using CanESM5 from CMIP6.

Figure 6. Zoning of Projected Average Annual Temperature Anomalies in Canada (2024–2100) Compared to the Observed Period (1983–2023) using CanESM5 from CMIP6.

Figure 7. Canada’s Percentage of Drought-Affected Area, Pr, and T under the SSP126 Scenario: A Comparative Analysis from 2003–2023 to 2024–2100, Examining Historical Trends and Future Projections Using CanESM5 within the CMIP6 Framework.

Figure 8. Canada’s Percentage of Drought-Affected Area, Pr, and T under the SSP245 Scenario: A Comparative Analysis from 2003–2023 to 2024–2100, Examining Historical Trends and Future Projections Using CanESM5 within the CMIP6 Framework.

Figure 9. Canada’s Percentage of Drought-Affected Area, Pr, and T under the SSP370 Scenario: A Comparative Analysis from 2003–2023 to 2024–2100, Examining Historical Trends and Future Projections Using CanESM5 within the CMIP6 Framework.

Figure 10. Canada’s Percentage of Drought-Affected Area, Pr, and T under the SSP585 Scenario: A Comparative Analysis from 2003–2023 to 2024–2100, Examining Historical Trends and Future Projections Using CanESM5 within the CMIP6 Framework.

Figure 11. Policy Implications for Canada in Addressing Climate Change and Drought Conditions.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

Share and Cite

MDPI and ACS Style

Soltani, K.; Amiri, A.; Ebtehaj, I.; Cheshmehghasabani, H.; Fazeli, S.; Gumiere, S.J.; Bonakdari, H. Advanced Forecasting of Drought Zones in Canada Using Deep Learning and CMIP6 Projections. Climate 2024, 12, 119. https://doi.org/10.3390/cli12080119

AMA Style

Soltani K, Amiri A, Ebtehaj I, Cheshmehghasabani H, Fazeli S, Gumiere SJ, Bonakdari H. Advanced Forecasting of Drought Zones in Canada Using Deep Learning and CMIP6 Projections. Climate. 2024; 12(8):119. https://doi.org/10.3390/cli12080119

Chicago/Turabian Style

Soltani, Keyvan, Afshin Amiri, Isa Ebtehaj, Hanieh Cheshmehghasabani, Sina Fazeli, Silvio José Gumiere, and Hossein Bonakdari. 2024. "Advanced Forecasting of Drought Zones in Canada Using Deep Learning and CMIP6 Projections" Climate 12, no. 8: 119. https://doi.org/10.3390/cli12080119

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Menu