I was born and raised in Ecuador. Weather and climate shape our lives in this country. For example, our energy supply depends on sufficient rainfall for hydroelectric power generation. I remember when I was a kid, there were continuous power outages. Unfortunately, Ecuador is not resilient. As I write this article, the power is out again. Paradoxically, El Niño Southern Oscillation causes flooding every year. I love hiking, but it made me very sad to see how our glaciers are receding.
Ten years ago, I decided to study for a PhD in meteorology. Climate change and its effects bothered me. It is a difficult challenge facing humanity in this century. Significant advances have been made in the scientific understanding of this issue. But we still need more action.
When I started my PhD, very few researchers were using artificial intelligence (AI) techniques. Today, there is a consensus that we can harness the potential of AI to make a difference. Especially in climate change mitigation and adaptation.
ML, especially computer vision (CV), allows us to make sense of the large amounts of data available. This power allows us to take action. Uncovering hidden patterns in visual data (such as satellite data) is a key challenge in tackling climate change.
This article introduces the relationship between CV and climate change. This is the first article in a series on this topic. This article has five sections on him. First, let me give you an overview. Next, this article defines some basic concepts related to CV. We then explore the potential of CV to tackle climate change with a case study. The article then discusses challenges and future directions. Finally, I will provide a summary.
understand computer vision
CV uses computational techniques to learn patterns from images. Earth Observation (EO) primarily relies on satellite imagery. Therefore, CV is a suitable tool for climate change analysis. Understanding climate patterns from images requires several techniques. The most important ones include classification, object detection, and segmentation.
Classification: I need to classify a (single) image based on a predefined class (single label). Fire detection and burned area mapping uses image classification techniques from satellite imagery. These images provide spectral features associated with burnt vegetation. Using these unique patterns, researchers can track the effects of wildfires.
Object detection: Involves locating objects within a region of interest. This technique is used to track hurricanes and cyclones. Detecting cloud patterns can help reduce impacts on coastal areas.
Image segmentation: Assign a class to each pixel in the image. This technique helps identify areas and their boundaries. Segmentation is also called “semantic segmentation.” Each region (target class) receives a label, so its definition contains “semantics.” For example, this technology is used to track the retreat of glaciers. By segmenting satellite images from glaciers, we can track changes in glaciers. For example, we monitor the extent, area, and volume of glaciers over time.
In this section, we have provided some examples of CVs in action to tackle climate change. In the next section, we analyze them as case studies.
Case study 1: Wildfire detection
Climate change has several effects on wildfires. For example, extreme events are more likely to occur. It also extends the duration of the fire season. Likewise, the severity of fires will worsen. Therefore, it is essential to invest resources in innovative solutions to prevent catastrophic wildfires.
This type of research relies on image analysis for early detection of wildfires. In general, ML techniques have proven effective in predicting these events.
However, advanced AI deep learning algorithms provide the best results. An example of these advanced algorithms is neural networks (NNs). NN is an ML technique inspired by human cognition. This technique relies on one or more convolutional layers to detect features.
Convolutional neural networks (CNNs) are popular in geoscience applications. CNN shows the greatest potential to improve the accuracy of fire detection. Several models use this algorithm, including VGGNet, AlexNet, and GoogleNet. These models have improved accuracy for CV tasks.
Fire detection using CV algorithms requires image segmentation. However, before segmenting the data, preprocessing is required. For example, reduce noise, normalize values, and resize. The analysis then labels the pixels that represent fire. This distinguishes it from other image information.
Case study 2: Cyclone tracking
Climate change will increase the frequency and intensity of cyclones. In this case, large amounts of data are not processed by real-time applications. For example, data from models, satellites, radar, and ground-based weather stations. We show that CV can efficiently process these data. It also reduced the bias and errors associated with human intervention.
For example, numerical weather forecasting models use only 3% to 7% of the data. In this case, the observations are from the Geostationary Operational Environmental Satellite (GOES). Even less data is used in the data assimilation process. The CNN model selects the most relevant observations from this vast amount of images. These observations refer to the region of interest (ROI) where the cyclone is active (or will soon be active).
Identifying this ROI is a segmentation task. There are several models used in the geosciences to approach this problem. Still, U-Net CNN is one of the most popular choices. Model design is relevant for medical segmentation tasks. But it has also been proven to help solve weather problems.
Case study 3: Tracking glacier retreat
Glaciers are thermometers of climate change. The effects of climate change on glaciers are visible (recession of contours). Therefore, they symbolize the consequences of climate change and change. In addition to the visual impact, glacier retreat has other effects as well. For example, the negative impact on the sustainability of water resources. Destabilization of hydropower generation. Affects drinking water quality. Decrease in agricultural production. upset the ecological balance. On a global scale, even rising sea levels threaten coastal areas.
The process of monitoring glaciers used to be time-consuming. Interpreting satellite images requires experts to digitize and analyze them. CV helps automate this process. Additionally, computer vision can be used to make processes more efficient. For example, you can incorporate more data into your modeling. CNN models such as GlacierNet harness the power of deep learning to track glaciers.
There are several techniques for detecting glacier boundaries. For example, segmentation, object detection, edge detection, etc. CV can perform more complex tasks. Comparing images of glaciers over time is one example. Similarly, it determines the speed at which the glacier moves and its thickness. These are powerful tools for tracking glacier dynamics. These processes can extract valuable information for adaptation purposes.
Challenges and future direction
Using CV to tackle climate change presents special challenges. We might need a whole book to discuss each one. However, our purpose here is modest. I will put it on the table for reference.
- Data complexity: The need to use many data sources and the associated complexity. Examples include satellite and aerial imagery, LIDAR data, and ground sensors. Data fusion is an evolving technology that attempts to address this difficult problem.
- Model interpretability: The current challenge is the development of a hybrid model. It means harmonizing statistical data-driven models with physical models. Incorporating knowledge about the climate system increases the interpretability of CV algorithms. Therefore, these models are better at fitting complex functions. But it also needs to provide an understanding of the underlying causal relationships.
- Labeled sample: Availability of high quality labeled samples. These samples should be specific to the EO problem for training the CV model. Their generation is a time-consuming and costly task. Addressing this challenge is an active area of research.
- ethics: The challenge is to incorporate ethical considerations into AI development. Privacy, fairness, and accountability play important roles in ensuring trust with stakeholders. Considering environmental justice is also a sound strategy from a climate change perspective.
summary
CV is a powerful tool to tackle climate change. From detecting wildfires to tracking cyclone formation and glacier retreat. CV is transforming the way climate impacts are monitored, forecasted and predicted. The study of these effects relies on CV techniques. For example, classification, object detection, segmentation, etc. Finally, several challenges arise in the relationship between CV and climate change. For example, managing multiple data sources. Enhance the interpretability of machine learning models. Generate high-quality labeled samples for training CV models. and incorporate ethical considerations when designing AI systems. The following article provides a guide to collecting and curation of image datasets. Especially those related to climate change.
References
- Kumler-Bonfanti, C., Stewart, J., Hall, D., and Govett, M. (2020). Tropical and extratropical cyclone detection using deep learning. Journal of Applied Meteorology and Climatology, 59(12), 1971-1985.
- Maslov, K. A., Persello, C., Shellenberger, T., and Stein, A. (2024). Towards global glacier mapping using deep learning and open earth observation data. arXiv preprint arXiv:2401.15113.
- Mumgiakumas, SS, Samatas, GG, Papakostas, Georgia (2021). Computer vision for UAV fire detection—from software to hardware. future internet, 13(8), 200.
- Rolnick, D., Donti, P. L., Kaack, L. H., Kochanski, K., Lacoste, A., Sankaran, K., … & Bengio, Y. (2022). Tackling climate change with machine learning. ACM Computing Research (CSUR), 55(2), 1–96.
- Tuia, D., Schindler, K., Demir, B., Camps-Valls, G., Zhu, XX, Kochupilai, M., … & Schneider, R. (2023). Artificial Intelligence Advancing Earth Observation: A Perspective. arXiv preprint arXiv:2305.08413.
