Recursive intelligent geographic modeling based on the "data-knowledge-model" tripartite collaboration

Geographic modeling, which might be one of the most complex modeling domains due to the essential characteristics of spatial heterogeneity, currently is within a sweeping intelligentization trend of using the state-of-the-art data-driven AI, specifically, large language models (LLMs). A new position paper, recently published in Journal of Geo-information Science, advocates the potential of recursive intelligent geographic modeling based on the "data-knowledge-model" tripartite collaboration, which highlights the fundamental role of domain modeling knowledge driving geocomputation with geospatial data to achieve successful geographic model workflow adaptive to application context.

 

The lead author Cheng-Zhi Qin, a professor of geographical information science (GIS) at Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, expended his research interests from digital terrain analysis and application modeling to intelligent geographic modeling since about 10 years ago.

 

Why intelligent geographic modeling matters?

“Geographic modeling is to appropriately couple diverse geographic models and their specific algorithm implementations to be an effective and executable workflow model for a specific unsolved application modeling problem. It is highly valuable and demanded in practice,” Cheng-Zhi says. Meanwhile, no single model or workflow could fit even one individual type of geographic modeling problems but always with diverse application contexts (e.g., application target, study area characteristics, data and other resource availability) in reality. “However, traditional geographic modeling is designed to be with an execution-oriented way. Such a way to modeling still lays heavy burden on users, especially those non-expert users, the majority of users in increasing wider real applications." So intelligent geographic modeling, which aims at providing accurate geographic modeling solution in a way being not only automated but also adaptive to application context, is highly demanded in practice and a hot research direction in current GIS.

 

In this position paper, the author team not only reaffirmed the necessity of intelligent geographic modeling, but also advocated that “intelligent geographic modeling should be achieved through a so-called recursive geographic modeling way”. What’s the unique point within “recursive intelligent geographic modeling” called in this paper?

“Actually I borrowed the ‘recursive’ word from computer programming science and added it in front of ‘intelligent geographic modeling’, just during the last round of manuscript revision, after I repeatedly thought how to well clarify the unique point in our intelligent geographic modeling way”, Cheng-Zhi says. The recursive intelligent geographic modeling way originates from the user’s modeling target, which may be formalized as the initial “elementary modeling question”. Then such geographic modeling way heuristically reasons backward to settle down current elemental modeling question and then update new elemental modeling questions in a typical recursive manner, so to automatically construct an appropriate geographic workflow model according to the application context of user’s modeling problem. Thus, the disadvantages in traditional geographic modeling can be solved (see below image)[CQ1] .

 

With such a basic idea, in this position paper the author team introduced a series of intelligent geographic modeling methods which were developed by authors in the past decade. As presented in the paper abstract, Cheng-Zhi says, “each of the proposed intelligent geographic modeling methods aims at solving a specific type of ‘elementary modeling questions’ during intelligent geographic modeling. The elementary questions include such as under current application context, 1) how to determine appropriate model algorithm or its parameter value, 2) how to select appropriate covariate set as input for a model without predetermined count of input (e.g., a soil mapping model without predetermined environmental covariates as the input), 3) how to determine the structure of a model with multiple modules coupled closely (e.g., a watershed system model coupling diverse process simulation modules), and 4) how to determine proper spatial extent of input data of a geographic model with an assigned area of user’s interest.[CQ2]  The key to solving different elemental questions is the effective adoption of geographic modeling knowledge, especially the application-context knowledge. Due to the characteristics of application-context knowledge typically being unsystematic, empirical, and implicit, we developed the case formalization and case-based reasoning strategies to use application-context knowledge within the proposed methods. Based on the proposed recursive intelligent geographic modeling way as well as the correspondingly proposed methods, an application schema of intelligent geographic modeling and computing based on domain modeling knowledge (especially the case-based application-context knowledge) and ‘data-knowledge-model’ tripartite collaboration is proposed.”

 

How about LLMs-driven geographic modeling? An opponent or a collaborator for “recursive intelligent geographic modeling”?

“The recently prevailing LLMs are explored on their potential for intelligent geographic modeling. Both LLMs-driven geographic modeling or our recursive intelligent geographic modeling are started from user’s modeling goal, but with different methodologic ways or driven forces: ours are essentially domain modeling knowledge driven, while LLMs are data-driven,” Cheng-Zhi says. “Current LLMs-driven geographic modeling faces the inevitable hallucination issue, even if the chain-of-thought way was considered for using LLMs. The retrieval-augmented generation has been highlighted for LLMs-driven geographic modeling. This confirms the key role of domain modeling knowledge similarly as what our study advocated. So, the relationship between LLMs-driven geographic modeling and the recursive intelligent geographic modeling based on the ‘data-knowledge-model’ tripartite collaboration should be collaborator, instead of opponent.” How to make good use of each strength, as well as other future works, were also discussed in this position paper.

 

See the article:

Recursive intelligent geographic modeling based on the "data-knowledge-model" tripartite collaboration

https://www.sciengine.com/JGIS/doi/10.12082/dqxxkx.2025.240706（If you want to see the English version of the full text, please click on the 科大讯飞翻译(iFLYTEK Translation)  in the article page.）

This work was supported by National Key Research and Development Program of China, grants 2021YFB3900904.

---

 

---

---