Preliminary exploration of GIS agent application based on DeepSeek

Written by

Audrey Miles

Updated on:July-14th-2025

With the rapid development of generative AI technology, many industries and fields are trying to build intelligent applications based on large models. DeepSeek R1 is an innovative large model that has become popular recently, and its exploration and application are in full swing.

The unique advantages of DeepSee k R1 have also brought many positive impacts to the research of intelligent agents. For example, its deep thinking ability can greatly reduce the dependence on prompt words and optimize the development and adjustment process of intelligent agents. At the same time, its open source characteristics and low reasoning costs have lowered the development threshold and can be efficiently practiced in various application scenarios, thereby promoting the popularization of intelligent applications.

A Preliminary Study on GIS Intelligent Agent

In view of the many advantages of DeepSeek R1, we conducted preliminary exploration of it in GeoScene Pro and tried to create a GIS intelligent application - GeoScene Pro Intelligent Assistant.

The purpose of GeoScene Pro Smart Assistant is to "extend" Pro intelligently, provide users with intelligent questions and answers to help the software, and assist users to use Pro more conveniently. This practice also explored the ability of Agent to call GeoScene Pro tools, allowing users to complete complex GIS mapping and analysis tasks through simple language interaction. Pro Smart Assistant has the following "intelligent" features:

Natural language understanding and perception

Transforming the "cold" software interface interaction into a "human-like" dialogue mode, we only need to tell the Pro Assistant the task like we speak and write in daily life, and it will be able to understand and parse the semantics and perceive the task we assign to it.

Enhanced search generation

Although DeepSeek R1 has a strong reserve and generation capability of general knowledge, it still lacks the "cognition" level when it comes to a professional software in a professional field. Therefore, by adding a lot of additional knowledge to DeepSeek R1, the Pro intelligent assistant provides enhanced retrieval and generation capabilities, so that the Pro assistant can answer questions about the use of GeoScene Pro more richly and accurately. For example, we can ask about the latest developments of GeoScene Pro, the use of specific tools, and how to choose tools in a certain business scenario.

Input: What new features of GeoScene 4.1 are worth paying attention to in terms of 3D, imaging, GeoAI, knowledge graph, etc.?

This question is relatively simple. In the deep thinking session of DeepSeek R1, the assistant retrieves key information from the knowledge base and guides itself to generate clear and key content.

Input: Study the correlation between various influencing factors such as temperature, precipitation, altitude, soil composition and farmland production potential. What GP tools are available in Pro? No need to recommend data preprocessing tools, directly recommend available regression models, and consider the impact of spatial scale.

This question is more complicated. In the deep thinking session, we can see that the intelligent assistant understands the instructions very clearly, focuses on recommending the regression model, and notices the need to consider the spatial scale.

Tool call

In GeoScene Pro intelligent assistant, we try to perform traditional mapping, AI-assisted mapping, and advanced spatial analysis functions to greatly improve work efficiency. DeepSeek R1 has excellent semantic understanding capabilities. We input task instructions composed of natural language, which contain the necessary parameters required by the tool. It can decide which tool to call through reasoning planning, extract parameters from the instructions and convert them into structured parameter descriptions, and then execute the tool and return the results and text responses after execution.

Input: Load the Chinese city layer and symbolize the data in graded colors. Select the City Innovation Index_2021 field with a level of 5.

The Pro Assistant intelligently and accurately interprets the instructions, finds the "Chinese Cities" layer from the current workspace and loads it to the map, correctly calls the graded color symbolization tool, and correctly extracts the Urban Innovation Index_2021 field, 5 levels and other key parameters and assigns them to the tool for execution.

Input: Use E:\firefly map.jpg to run AI color matching tool

The Pro Assistant calls the tool correctly and automatically matches symbol colors based on the specified picture style.

Input: There are two sets of earthquake hotspot analysis results from different years. How to conduct an accurate comparative analysis?

The Pro Assistant not only correctly suggests the tools to be used, but also, based on its understanding of the principles of hotspot analysis, provides suggestions on strategies for setting parameters during input data and use.

Input: Read the "Earthquake Frequency_2004_2013_Half Degree Grid_Optimized Hotspot Analysis_3857" layer and the "Earthquake Frequency_2014_2023_Half Degree Grid_Optimized Hotspot Analysis_3857" layer in the default database and run the Hotspot Analysis Comparison tool

Summarize

As DeepSeek and other large models are continuously explored and applied in the GIS field, intelligent tools are gradually becoming the industry standard. GIS software and other types of tool software will no longer be cold tools, but smart partners that can understand business logic and actively provide help, allowing every GISer to easily master the complex data world and move forward steadily in the intelligent era.