Abstract
Land degradation and desertification are considered substantial issues in ecological and social research; hence the need for a quantitative, reliable, and repeatable methodology to evaluate desertification processes is urgent. This chapter aims to review the advances and limitations of existing work seeking to predict desertification through the use of automated, data-driven methods (i.e. machine learning). Using the CRISP-DM framework, existing research was classified into classic (supervised) ML models using field data, classic ML models using remote sensing data, and deep-learning models using remote sensing data. However, more research is needed to incorporate feedback effects and human intervention, as well as to make the distinction between desertification risk and desertification in ML models. Finally, the chapter suggests that existing research should be collated and formed into a “desertification warning system,” addressing the need to harmonize data, better understand desertification, and aim for the greater inclusion of local end-stakeholders.
TopPromises And Challenges Relating To Machine Learning Techniques To Predict Areas At Risk Of Desertification: A State-Of-The-Art Review
In 1994, the United Nations Convention to Combat Desertification (UNCCD) set out to define desertification as the process of land degradation in dry sub-humid, semiarid, and arid areas as the result of various factors, including improper human activities as well as climatic variation (Ma & Zhao, 1994). “Drylands,” i.e. the umbrella term used herein, are defined as ‘areas other than polar and sub-polar regions, in which the ratio of annual precipitation to potential evapotranspiration falls within the range from 0.05 to 0.65’ (Ma & Zhao, 1994, as cited in McSweeney, 2019). Amongst other effects, land degradation can cause the decline of vegetation, water resources, soil quality, and wildlife, as well as the deterioration of economic productivity (McSweeney, 2019). Desertification has occurred throughout history, and 40% of the earth’s terrestrial surface is now classified as drylands (Nunez, 2019). However, according to the United Nations, the pace of encroachment has recently increased exponentially by up to 30 to 35 times its historical rate. In total, 75 billion tons of fertile soil is lost every year due to land degradation, and 12 million hectares of land due to drought and desertification (UNCCD, n.d.). A further 3.97 billion hectares, which is equal to 75.1% of all drylands, is at least moderately at risk of desertification (Baartman et al., 2007). It is generally understood that both natural (i.e. biophysical traits, climate), as well as human influence, are drivers of this issue (Baartman et al., 2007). However, population growth (Yaojie et al., 2019) and climate change, resulting in an increase in average temperatures and drought frequency, the continuation of dry conditions, as well as precipitation variability (Buckland et al., 2019), are likely to magnify these effects. The results include food production problems, a lack of clean water, and the shifting of habitats, thereby sharply decreasing the ability of the land to nourish surrounding populations of people and animals (Nunez, 2019).
In order to effectively counteract desertification processes, it is vital to monitor the situation continuously to prevent aggravation and possibly irreversibility once climatic conditions and/or disturbances caused by human intervention reach a certain threshold. Early warning monitoring can have a significant impact when developing sustainable environmental restoration and management strategies (Liang et al., 2021). A major challenge in developing action plans against desertification remains the gathering of baseline data through the mapping and inventory of linked processes. The dynamic nature of desertification requires mapping at regular intervals with high accuracy (Meng et al., 2021). Knowledge on the current state of land degradation, or the magnitude of its potential globally relevant threat, remains mostly fragmentary or incomplete; for some areas, there is even a complete lack of awareness (Pinet et al., 2006).
The complexity involved in quantifying land degradation mainly arises due to the interdisciplinary forces driving its processes, including ecological, meteorological, biophysical, and social components, each of which plays out differently on a regional level. Moreover, there is particular concern in relation to understanding the interactions between individual environmental forces, causing instability and land degradation (Buckland et al., 2019). In addition, desertification is a slowly emerging phenomenon, which also renders its detection harder (McSweeney, 2019). Given this, it is important to develop a further understanding of the geographic extent of desertification in order to support prevention planning on a local, regional, and global scale (Verstraete et al., 2009).
Key Terms in this Chapter
Supervised Learning: Machine learning methods using labeled datasets designed to train algorithms into classifying data accurately.
NDVI: Normalised Difference Vegetation Index: Commonly used indicator which applies vegetation cover information to conduct trend analysis of vegetation index data (Higginbottom and Symeonakis, 2004 AU10: The in-text citation "Higginbottom and Symeonakis, 2004" is not in the reference list. Please correct the citation, add the reference to the list, or delete the citation. ).
Machine Learning Techniques: Machine learning is part of the field of artificial intelligence (AI) which is used to derive patterns and regularities from a big set of data, including those hidden amongst the dataset variables (Wu et al., 2008 AU9: The in-text citation "Wu et al., 2008" is not in the reference list. Please correct the citation, add the reference to the list, or delete the citation. ).
Unsupervised Learning: Machine learning algorithms used to cluster and analyze unlabeled data, whereby hidden patterns in the data can be discovered without the need for human intervention. Used for clustering, determination of association rules as well as dimensionality reduction of a dataset.
Desertification: The process of land degradation in dry sub-humid, semiarid and arid areas as the result of various factors, including improper human activities as well as climatic variation (UNCCD, 1994 AU8: The in-text citation "UNCCD, 1994" is not in the reference list. Please correct the citation, add the reference to the list, or delete the citation. ).
CRISP-DM (Cross-Industry Standard Process for Data Mining) Framework: Reference point for data mining projects. It is an industry- and technology-independent, end-to-end framework with the aim of rendering large data mining projects more reliable, manageable, repeatable, less costly and also faster ( Wirth and Hipp, 2000 ).
Data Mining Techniques: The application of specific algorithms in order to extract patterns from a data set.