Abstract: Nighttime remote sensing data has unique advantages for estimating metropolitan area populations.Its strong correlation with human activity makes it a popular tool in sociology and human dynamics research.This paper presents a nighttime population estimation model based on multi-tree fusion to predict the population in the Chengdu area with high precision using nighttime remote sensing data.The study uses nighttime remote sensing and population data from various regions in Chengdu from 2000 to 2020 to construct long-term nighttime remote sensing data.XGBoost,random forest,and decision tree models are used as base models for preliminary predictions.These predictions are then integrated and fused using a linear regression meta-model to form a multi-tree fusion nighttime population estimation model.This model is then used to predict the population of various Chengdu regions in 2021 and 2023.The results are compared with the actual values to validate the model.Overall,the model's prediction accuracy is excellent.The average prediction accuracy for 2021 is 99.54%,with an average absolute error of 3437 people.The average accuracy for 2022 is 98.87%,with an average error of 9156 people,and the average accuracy for 2023 is 98.86%,with an average error of 9832 people.Districts such as Xinjin district (99.85% in 2021)and Jinniu district (99.93% in 2023)achieved an accuracy level exceeding 99.5%.However,only a few districts,such as Longquanyi district (96.55% in 2023)and Wuhou district (97.58% in 2022),experienced slight accuracy fluctuations due to rapid urban functional changes.The study confirms that the multi-tree fusion model effectively captures the correlation between nighttime remote sensing data and population changes.This provides reliable data support for Chengdu's urban planning and resource allocation.

Key words: long duration, nighttime remote sensing, multi-tree fusion, human activity