Preliminary Study On Multi-scale And Data-assimilated Simulation Of Land Use/land Cover Change In A Mountainous Area

Land use/land cover change(LUCC)is the core and hot topic of global environmental change research.The large-scale expansion of hot cultivation leads to the drastic development of LUCC in the border mountains of southwest China,which makes it a hot area for LUCC research.LUCC simulation can deeply explore LUCC process,driving mechanism and predict future land use/land cover spatial pattern.LUCC process and its driving factors have multi-scale characteristics,and the optimal scale varies with the research area.Therefore,it is very important to carry out multi-scale comparative study of LUCC simulation in a specific area.However,LUCC simulation often uses fixed parameters,resulting in constant error accumulation.Therefore,the data assimilation algorithm is introduced into the LUCC simulation,and the model structure and parameters are calibrated by combining with the observation data,and the model simulation trajectory is modified,which is of great significance for improving the simulation accuracy of LUCC model in mountainous counties.In this paper,Mengla County of Yunnan Province,a county in the southwest border mountain region,was selected as the study site.Landsat TM in 1989 and 2019 Landsat OLI image were classified using the random forest classification method.Based on the spatial conformity assessment of 1989,1994,1999,2004,2009,2014 and 2019classification results,we explored the land use/land cover change rule from 1989 to2019 in this region.The land use/land cover status in 2014 was simulated at 60 m,90 m,120 m,150 m,180 m,210 m and 240 m grid scales using LCM,FLUS,CA-Markov and CLUE-S models,taking the land use/land cover data classified from Landsat TM images acquired in 1994 and 2004 and natural and socio-economic data between 1994and 2014 as inputs.The simulation results were assessed using the land use/land cover classification data produced from the Landsat OLI images in 2014 as a reference to search for an optimal grid scale for land use/land cover simulation in a mountainous county.Then the land use transition probability during three different periods(1989-2004,1999-2009 and 2004-2009)were used to establish the LCM model to simulate land use/land cover in 2019 using 15 a,10 a and 5 a intervals respectively.And the simulation results were assessed using the land use/land cover classification data produced from the Landsat OLI images in 2019 as a reference to search for an optimal time scale for land use/land cover simulation.Based on the optimized grid scale and time scale,the particle filter data assimilation method was coupled to the LUCC model,and the model structure and parameters were optimized to simulate the spatial distribution pattern of plantation in 2024 based on 2019.The results show that:(1)From 1989 to 2019,the land use/land cover in this county are mainly forest and plantation,and LUCC mainly shows that forest is transformed into plantation.From 1989 to 2014,the expansion of plantation led to drastic deforestation.From 2014 to 2019,the expansion of the plantation slowed down,and some areas showed a decrease.(2)All the four models can simulate LUCC in this county well,and the K_standard is 0.67⁰.84.Among them,the LCM model has the highest overall simulation accuracy,and the K_standard is 0.77⁰.84.Different models have different simulation accuracy on different land use/land cover types.LCM has the highest accuracy in simulating forest,plantation and farmland,while FLUS has the highest accuracy in simulating built-up land.(3)120 m is the best grid scale of LUCC simulation.With the increase of the grid scale,the K_standard and K_no of the four models showed a trend of rising and then falling,and reached the maximum at the grid scale of120 m(K_standard were all greater than 0.70).(4)5 a is the best timely scale of LUCC simulation.With the increase of time scale,the simulation results of K_standard,K_no,K_location,K_quantity and TFOM gradually decreases,and the 5 a maximum time scales(K_standard,K_no,K_location and K_quantity were 0.82,0.88,0.89,0.92 and 90.08%).(5)The neighborhood factor has the greatest influence on the change of plantation.The optimal neighborhood window size is about 500 m,and the posterior probability of the percentage of plantation pixels in the neighborhood window is 0.6¹.0.(6)The Kappa of the CA model coupled with data assimilation(PF-CA)is 0.68,higher than that of tranditional CA.The RMSE of the simulation results of PF-CA is lower than that of CA.The difference of RMSE in the two scenarios decreased with the increase of simulation time,and reached the maximum in 2004,indicating that the data assimilation had a good effect on improving the simulation accuracy of LUCC in a short period of time.Compared with the validation block,the difference between RMSE in scenario 2 and scenario 1 is larger in the calibration block,indicating that the role of data assimilation is better in the calibration block with observation data.