Researchers at IIT Bombay discover the role of environmental resources, genes and mating in species in the development of new species in the same area, challenging the traditional view that new species can develop only in distinct geographies.

Here’s how machine learning could help in choosing the best policy for forest management

Read time: 4 mins
Himachal Pradesh
28 May 2019
Here’s how machine learning could help in choosing the best policy for forest management

Forests, with their rich biodiversity, play a critical role not only in the ecology of a region but also in social systems that involve humans. In India, different forest management policies are in place to protect and promote the sustainable use of forest resources. Community-based forest management, where local communities govern the use of these resources to improve their livelihoods and monitor vegetation growth to maintain healthy ecosystems, is one such policy. How well do such management policies work? A recent study by researchers from the University of Illinois, USA, has used artificial intelligence to evaluate the forest management policies in Himachal Pradesh, India.

The researchers of the study, published in the journal Environmental Research Letters, have analysed the impacts of two forest management policies followed in the Kangra district of Himachal Pradesh. The first policy, called cooperative forest management, involves the formation of a federation of forest cooperatives with financial and technical support provided by the state government. The second policy, called joint forest management, has communities involved in forest regeneration and protection. A critical difference between the two is that the latter does not allocate forest property rights to communities, or the right to derive revenue from forest resources claimed by the state.

The study, for the first time, uses algorithms based on machine learning to evaluate the impacts of the two policies. Machine learning is the application of algorithms and statistical models to make predictions and decisions by training data. “To date, machine learning approaches have seen little applications, especially in relation to evaluating forest and natural resource policy”, say the researchers.

The application of machine learning algorithms in formulating forest management-related policies is currently limited. This could be due to insufficient availability of data, the complexity of socio-ecological contexts such as on forest management policies, the high number of villagers, farmers, land holdings, the area under plantations, and the restricted ability of machine learning to process ‘big data’ and detailed data that can form forest policies.

In this study, the researchers started with a hypothesis that joint forest management policies will be less conducive to vegetation growth in the long run, as compared to cooperative forest management. This difference could be because current institutional land rights and land tenure, under cooperative forest management, favour vegetation growth. The researchers measured the vegetation growth in forests in the Kangra region of the Himalayas, which were managed according to either of the policies, to test their hypothesis.

The researchers analysed 28 socio-ecological variables and associated indicators that indirectly determine the vegetation growth in the region. The variables included the dependence of the communities on forests, governance, tree cover, forest area, interactions between the communities and forest resources, and outcomes of the policies on vegetation growth. The indicators varied from the number of households, tree cover, grass acreage, temperature, forest fires, and several others.

In the next step, the researchers used the Causal Tree and Forests algorithm—a machine learning-based technique that shows relationships between events and causes that lead to them. Developed by researchers at Stanford University, it unpacks the contexts in which the policies perform well as opposed to the traditional evaluation methods that failed to understand the performance of different policies in the region. The machine-learning based algorithm helped identify the factors that led to a long-term improvement in vegetation growth in the area.

The results showed that neither joint forest management nor cooperative forest management affects vegetation growth in the long-term. Cooperative forest management led to a positive impact on vegetation growth for at least 5-9 years after intervention in Kangra, whereas the social and ecological contexts favouring forest growth were better in joint forest management than in cooperative forest management.

Interestingly, the study found that temperature was an important variable in shaping the impact of forest management policies. Forests, where the temperature was higher than 18oC, had more vegetation growth when managed jointly. However, in colder areas where grazing is universal, joint forest management policies led to the loss of vegetation.

“Our results suggest that future community-based plantation programs in the Indian Himalaya would do well to provide alternative grazing options to forest communities and facilitate secure rights over forests to foster positive long-term vegetation growth trajectories”, say the researchers. The study shows how forestry policies, which support strong, local institutions and tenure, could drive positive outcomes.