In the PLUM, the reconstruction of prehistoric human land use areas and distributions based on archaeological and environmental data is conducted quantitatively using three sub-models (land use requirement, residential area distribution, and land use allocation sub-models). Detailed information about the model has been presented by Yu et al. (2012). The framework of the model is shown in Figure S1.

In the first step, the total residential and cultivated area in the study area is calculated using the land use requirement sub-model. The estimation assumes that human survival relied entirely on cultivated plants because it is still difficult to quantitatively assess the amounts of natural plant and animal resources in prehistoric human food sources using the available archaeological records. The equation for calculating the total human land use (A_l) is as follows:

where A_r and A_p are the total residential area and the average residential area per person inside the archaeological sites, F_p is the food requirement per person, Y is the yield per unit area, and T_f and T_c are the intervals of fallow and tillage during one cultivation cycle. These variables can be obtained from published archaeological research.

In the second step, the potential distribution of humans is predicted using the residential area distribution sub-model. A weighted overlay method is adopted to make the prediction according to the relationship between the distributions of the archaeological sites and various environmental variables. Two types of weights (class and spot weights) are set for all of the grids in the study area and are used to represent the degrees to which human activities depend on the environmental variables and the different ranges of these variables, respectively. The class weights are set based on the differences in the cumulative frequency distributions of the environmental variables between grids in the study area and known archaeological sites. A higher weight corresponds to a greater difference. The spot weights are set based on the percentage of archaeological sites distributed in the different sub-ranges of each specific environmental variable, and the weights are positively correlated with the percentages. The class and spot weights for a specific environmental variable are multiplied to obtain the total weight, and then, the total weighted grid layers in the study area for the different environmental variables are summed and standardized to the range of 0-100% to create a single layer, which shows the potential distribution of human occupation across the study area from low to high.

In the final step, the land use area of the archaeological sites (estimated using land use requirement sub-model) is allocated to the appropriate grids in the study area using the land use allocation sub-model. Because the distance humans can walk in 1 day is limited, the allocation process starts with the assumption that humans always selected the grids around the archaeological site within a threshold distance, and the grids with higher standardized total weight values according to the residential area distribution sub-model were selected first. The process ends when the total land use areas of the archaeological sites are all distributed to suitable grids.

A total of 8161 archaeological sites in the Liaohe River Valley were used as the inputs of land use requirement sub-model. The information for these sites was obtained from the published Chinese Cultural Relics atlases for Liaoning province, Jilin province, Hebei province, and the Inner Mongolia autonomous region (National Heritage Board, 1993, 2003, 2009, 2013). The information is presented in Table S1. These archaeological sites were documented during the second national cultural survey undertaken in the 1980s, and the related information (e.g., the name, location, area, culture type, culture depth, and archaeological remains) for the sites was attached in the atlases.

About 97% of the archaeological sites belong to specific cultural types (Table 1), which occurred in a period of less than 2000 years according to their absolute ages, and all of these sites were assumed to exist continuously during the corresponding entire cultural periods (Tables 1 and S1). As for 3% of the sites under unclear Neolithic culture types, they were randomly attributed to different 1000-year time windows during 8-4 ka BP based on the number of archaeological sites under clear cultural types in each millennium (Table S1). This attribution technique is acceptable because the percentages of the sites with clear cultural types are overwhelming, and they can be used to determine the pattern of the temporal changes in the sites. The specific culture types and time periods that have been assigned to the archaeological sites are presented in Table S1.

The residential areas of 88% of the archaeological sites were documented. For each cultural type, the median value of the area of the known sites was used for the sites of the same type without documented values. As some of the known sites that existed during different cultural periods only had one documented area value, these sites were excluded during the calculation of the residential areas of the unknown sites.

Several socioeconomic parameters were adopted in the land use requirement and land allocation sub-models, and these data were mainly based on the results presented in the published literature. Table 2 presents the corresponding values for the Liaohe River Valley.

The average value of the residential area per person and its range were set according to a study on typical excavated archaeological sites in the Yellow River Valley in northern China (Wang, 2011).

The average food need per person and its range were calculated according to the minimum annual nutritional requirements of an individual, and the value of 300 (200-400) kg/year estimated by Araus et al. (2003) was taken as a constant during the study period since human physiognomy and physiology changed insignificantly during the Holocene (Wu, 1995).

The average value and range of the yield per unit area for millet cultivation were estimated based on archaeological research, historical documents, and modern slash and burn agriculture observations (Song, 1982; Wu, 1985; Gao and Min, 2008; Zheng et al., 2009).

The time periods of the fallow and tillage intervals before 5 ka BP were obtained from studies of typical archaeological sites that belonged to different types of fallow vegetation (forest, bushes, and short fallows) (Wang, 1997), while the corresponding value for 3-2 ka was obtained from studies on the Han Dynasty and modern slash and burn agriculture (Guo, 1994; Yang et al., 2018). These values were further used to estimate the corresponding value during 4-3 ka BP through linear interpolation.

As cultivation usually occurred within a certain distance from the archaeological sites, the threshold value of the distance (10 km) was set based on the maximum time (2 h) that a human can potentially spend on walking to cultivated fields in 1 day (Zheng et al., 2008).

Modern spatial environmental data (altitude, soil type, and water system) throughout the Liaohe River Valley were needed for the residential area distribution sub-model. The grid layer of the altitude (90 m × 90 m horizontal resolution and 1 m vertical resolution) was obtained from the website of the Shuttle Radar Topography Mission (http://srtm.csi. cgiar.org/). The vector layers of the soil type (1:1,000,000) and water system (1:4,000,000) were obtained from the National Earth System Science Data Center (http://www. geodata.cn/). The grid layer of the slope was extracted from the altitude data using ArcGIS 10.2, a geographic information system (GIS) software. The grid layers of the horizontal and vertical distances to the water system were calculated based on the data for the altitude and water system using ArcGIS. All of the layers of the environmental variables mentioned above were finally converted or resampled to grid data with the same resolution of 0.01°×0.01° under the uniform projection WGS_1984.

Fossil pollen data are widely accepted geological records for conducting reliable quantitative paleoclimate reconstruction. In consideration of the data quality of the records (e.g., number of dates, duration of the record, and sampling resolution), two pollen records from Daihai (Xiao et al., 2004) and Sihailongwan (Stebich et al., 2015) in the western and eastern parts of the Liaohe River Valley were selected from the published literature and were used to reflect the climate changes in the western and eastern parts of the valley, respectively. More than five independent radiocarbon dates have been reported for each of these two pollen records. The durations of all of the records exceed the entire Holocene, and the samples younger than 1 ka BP used for the anomaly calculation are also included in the records. The sampling resolutions of the records are less than 60 years, and each sample contains >200 pollen counts.

The modern pollen dataset utilized in this study includes 1863 surface pollen samples from Li et al. (2009) and the East Asian Pollen Database (Zheng et al., 2014). This dataset was used to establish the transfer function during the climate reconstruction. These surface pollen samples are distributed across China, covering climate conditions from cold temperate zone to tropical zone. Such wide climatic and ecological ranges span the changes in the climate in the Liaohe River Valley during the Holocene. Modern spatial climate data of the whole valley were from a 30-second gridded monthly mean climate dataset (1960-1990) (https://www.worldclim.org/data/v1.4/worldclim14.html).

The climatic factors in the pollen records for Daihai and Sihailongwan have been quantitatively reconstructed in previous studies using different methods (Xu et al., 2010; Stebich et al., 2015). In this study, to reduce the uncertainty of the comparison between the records based on different reconstruction methods, the modern analogy technique (MAT), which is implemented in the R package Rioja (Juggins, 2020), was applied to both records to reconstruct their annual mean temperature and annual precipitation (ANNT and ANNP). In this study, the MAT was based on the scores of the plant functional type (PFT, i.e., plant group characterized by common phenological and climate constraints) because the application of PFTs rather than pollen taxa can decrease the amount of poor analogue cases and reduce non-climatic influences (Prentice et al., 1992; Peyron et al., 1998; Davis et al., 2003).

The reconstruction process for each pollen record was as follows. First, all of the radiocarbon dates in the pollen records were recalibrated based on the IntCal20 calibration curve and were interpolated to sample levels using the R package clam (Blaauw, 2010; Reimer et al., 2020). Then, the fossil pollen samples with different ages were matched with modern pollen datasets using the dissimilarity of the biome score compositions according to the squared chord distance (Overpeck et al., 1985), and the risk of non-analogy bias was decreased by using a dissimilarity threshold value (T=0.06) for the non-analogy/analogy definition. Finally, the climatic conditions under which the sample in the pollen record was deposited were reconstructed according to the dissimilarity-weighted mean climate of the seven closest modern analogues (Simpson, 2012).

All of the significant climate reconstructions in the records from the Daihai and Sihailongwan sites were converted to anomalies through comparison with the mean value of their latest 1000 years of reconstructions. The reconstructed anomalies of the samples were further linearly interpolated to 200-year bins, which were finally averaged into different 1000-year intervals (Marsicek et al., 2018). To visualize the spatial changes in the climate during the past millennia, the anomalies of the Daihai and Sihailongwan records in different 1000-year intervals were overlaid onto the layers of the modern climatic variables (temperature and precipitation) in the western and eastern parts of the Liaohe River Valley, respectively.

The number of archaeological sites in the Liaohe River Valley increased from 321 to 3481 between 8 and 2 ka BP (Figures 2 and S2), and the total residential areas of the sites increased from 14.6 km² during 8-7 ka BP to 70.6 km² during 3-2 ka BP. In addition, the human population increased from 3.4×10⁴ individuals (average of 2.5×10⁴-4.2×10⁴) during 8-7 ka BP to 75.8×10⁴ individuals (average of 56.5×10⁴-97.3×10⁴) during 3-2 ka BP, indicating that the total population increased by a factor of more than 20. The changes in the above three variables during 8-2 ka BP were further divided into three stages. First, the site numbers, residential areas, and population size in the valley all slowly increased during 8-5 ka BP. Then, they synchronously decreased significantly during 5-4 ka BP. Finally, these variables all increased significantly after 4 ka BP.

Regarding the average human activity intensity at the different archaeological sites in the valley, the average residential area decreased overall with time, while the average human population remained relatively stable, within the range of 100-200 individuals. Because the population size of each archaeological site was estimated according to its total residential area and the corresponding value per person, the synchronous decreases in these two variables during 8-2 ka BP caused the small variations in the average population.

Spatially, the changes in the distribution of the archaeological sites with time can also be divided into three stages (Figure S2). Before 5 ka BP, most of the archaeological sites were distributed in the western part of the Liaohe River Valley, and the distribution of the sites experienced an expansion process from higher altitude to different altitudes. During 5-4 ka BP, the range of the distribution of the archaeological sites contracted significantly to small areas in the western part of the valley. After 4 ka BP, the spatial distribution of the archaeological sites expanded again, and the density of the sites increased more significantly in the southern part of the valley. There was an obvious boundary near 42.5°N, and the densities of the site distribution on the two sides of this boundary were obviously different.

The temporal changes in the reconstructed total human land use area in the Liaohe River Valley during 8-2 ka BP exhibited characteristics similar to the changes in the archaeological sites and population in the valley (Figure 2). First, it gradually increased from 3053 km² to 5411 km² to 5531 km² between 8 and 5 ka BP; then, it sharply decreased to 398 km² during 5-4 ka BP, and finally, it rebounded to values as high as 9428 km² and 4556 km² during 4-2 ka BP. Overall, the variation range of the human land use was smaller than those of the number of sites and population size during 8-2 ka BP.

The reconstructed average land use area among the different archaeological sites continuously decreased from 8 to 2 ka BP. Before 5 ka BP, <50% of the archaeological sites had land use areas <1 km² during each millennium, while >7% of the sites had land use areas >10 km². In contrast, >77% of the archaeological sites had land use areas <1 km² during 4-2 ka BP, and <2% of the sites had land use areas >10 km². These differences may be partly due to an increase in land use efficiency.

Regarding the spatial changes in the land use from 8 to 2 ka BP (Figure 3), the reconstruction indicates that the areas with higher percentages of land use were mainly distributed around the archaeological sites in the northwestern part of the valley during 8-6 ka BP. The land use expanded across the entire western part of the valley during 6-5 ka BP. However, the distribution of the land use contracted significantly, which is consistent with the changes in the archaeological sites during 5-4 ka BP. After 4 ka BP, the land use was widely distributed in the southern part of the valley, and the areas with high land use intensities more frequently appeared in the areas around 42°N from west to east. Overall, the significant differences in the land use distribution in the east-west direction were disrupted.

Currently, the human land use within the Liaohe River Valley accounts for about 39% of the total area (https://www.geodata.cn/), while based on the reconstruction conducted using the PLUM, it only accounted for 3% of the area from 8 to 2 ka BP, which indicates that prehistoric human activity intensity was low before 2 ka BP. However, after 4 ka BP, the spatial pattern of the land use distribution became similar to the current pattern (https://www. geodata.cn/). It should be noted that our population and human land use reconstructions based on the discovered archaeological sites in the Liaohe River Valley only reveal the lower limits of the actual human activity during the prehistoric period because fluvial erosion, human disturbances, and other taphonomic factors also affect the preservation of archaeological sites.

The history database of the global environment version 3.2 (HYDE3.2) and the Global Land Use and Environment Lab (GLUE) database have been widely utilized to quantitatively reconstruct the cultivated land use during the Holocene (Lemmen, 2009; Goldewijk et al., 2017). For HYDE3.2, the prehistoric land use area was calculated via extrapolations of the population and land use per capita during the historic period, while the land use distribution was reconstructed based on the modern distributions of the population and environmental variables (Goldewijk et al., 2017). The same principle for land use reconstruction was adopted for GLUE; however, for the GLUE dataset, the prehistoric population was estimated according to model simulations (Lemmen, 2009). From the temporal perspective, the reconstructions conducted using the PLUM (our study), HYDE3.2 (Goldewijk et al., 2017), and GLUE (Lemmen, 2009) have similar levels of land use size during 8-2 ka BP; however, the variations among the three reconstructions during this period were different (Figure 4a). An overall increase was revealed by the reconstructions of the PLUM and HYDE3.2, but the land use area was much lower in the reconstruction of HYDE3.2. Additionally, according to the HYDE3.2 reconstruction, the land use growth was persistent, while the PLUM reconstruction indicates that it temporarily decreased during 5-4 ka BP. Unlike the two reconstructions above, the GLUE reconstruction indicates that the land use area sharply increased during 5-4 ka BP, while it remained stable in the other periods (Figure 4a).

Regarding the spatial changes, the distribution of the land use in the HYDE3.2 reconstruction indicates a continuous dependence on low elevation and sloping land, and the land use percentage was always higher in the plain areas in the middle and eastern parts of the Liaohe River Valley, which is consistent with the pattern during the recent period (Figure S3). The spatial resolution of the GLUE reconstruction is low, which indicates that there were no significant differences in the land use intensity throughout the valley (Figure S3). In contrast, our reconstruction using the PLUM indicates that the human land use areas were predominantly located in the relatively high-altitude areas in the western part of the valley in the earlier stages of the study period, while a pattern of north-south differentiation occurred during the later stage of the study period.

The adoption of direct prehistoric human activity evidences (archaeological sites) as control points in the PLUM improved the accuracy of the quantitative land use reconstruction because the properties of the archaeological sites (number, size, and distribution) can represent the differences in the population and land use efficiency on local scales, as well as human migration on large spatial scales. Compared with extrapolations of historical and modern human activity information to the prehistoric period conducted in previous studies, the temporal fluctuations and spatial variations in our land use reconstruction conducted using the PLUM are more reliable.

Our reconstruction results are also supported by previous studies of prehistoric human activity in and around the Liaohe River Valley based on various archaeological and geological records. For example, the analysis of starch residues revealed that the proportion of domesticated millet continuously increased in the western part of the Liaohe River Valley during the Neolithic period, indicating strengthening of agricultural activities (Ma et al., 2016). A geochemical study of an archeological site in the Liaohe River Valley revealed that human agricultural practices potentially caused the higher C₄ biomass in the cultural layers during the process of millet domestication (Wang et al., 2020). Different pollen records in and around the Liaohe River Valley have also revealed that disturbances of vegetation caused by early human activities began as early as 6000 years ago and intensified since 4 ka BP at local and regional scales, while the high correlation between the microfossil charcoal influx and the occurrence of crop remains indicates that human burning was an important means of vegetation clearance and agriculture development (Li et al., 2006; Yang et al., 2020; Niu et al., 2024). Spatial differences in the human activity intensity have also been revealed through synthesis research on pollen records (Ren, 2000; Li et al., 2020), and the results of these studies indicate the expansion of agriculture from the western part of the valley to the eastern part. Compared with the above semi-quantitative studies, the series of spatially continuous maps of human land use reconstructed using the PLUM in this study is favorable for better quantitative evaluation of the changes in human activity from the spatiotemporal perspective.

In our study, the residential and cultivated human land use areas were reconstructed using the PLUM; however, the cultivated land use accounted for over 99% of the total human land use in the Liaohe River Valley during 8-2 ka BP. Because different crop types have suitable temperature and precipitation ranges for growth (Zhou et al., 2020), the distribution and size of the cultivated land use areas in the valley were directly related to changes in the climatic condition.

According to crop seed remains from archaeological sites (Ma et al., 2016), millet was the main crop cultivated in the Liaohe River Valley during the prehistoric period, and the optimum climatic conditions for its growth are 500-750 mm of precipitation and more than 60 days with temperatures >16°C in 1 year (Zhou et al., 2020). Regarding the modern temperature and precipitation distributions, almost the entire Liaohe River Valley falls within the optimum temperature range for millet cultivation (Figure 5a), while only the southern and central-eastern parts of the valley fall within the optimum precipitation range (Figure 5b). Therefore, the variations in temperature and precipitation during 8-2 ka BP would have affected the distribution of the millet cultivation areas. Paleoclimate reconstructions based on pollen records in and around the valley have revealed that it was warmer and wetter during 8-2 ka BP than in modern times (Xu et al., 2010; Stebich et al., 2015), and thus the temperature was not the restricting factor for regional millet cultivation. However, the land use distribution was potentially sensitive to changes in precipitation in the valley. This deduction is confirmed by the comparisons of the time series and spatial evolutions of the reconstructed land use, temperature, and precipitation between 8 and 2 ka BP in different parts of the Liaohe River Valley (Figures 4b-4e and 5c-5f).

From the temporal perspective, synchronous changes in human land use and precipitation occurred in both the western and eastern parts of the valley during 8-2 ka BP, while the variations in temperature were insignificant in both parts of the valley. In addition, the decreases of land use occurred in both parts of the valley during 5-4 ka BP; however, the more obvious decrease in the land use area in the western part of the valley than in the eastern part of the valley also corresponded to the different levels of precipitation decline in the two parts of the valley (Figure 4).

From the spatial perspective, the comparisons of the reconstructed spatial distributions of the climatic factors and land use areas during the different millennia further revealed that the ranges of human activity corresponded well with the optimum range of precipitation for millet cultivation during 8-3 ka BP. Before 5 ka BP, the optimum range of precipitation for millet cultivation mainly occurred in the western part of the valley, and this part of the valley also had a high human land use intensity (Figure 5c). After 5 ka BP, the land use distribution patterns changed obviously around 5-4 ka BP, and the core regions of land use shifted from the western part of the valley to the southern part of the valley (Figure 3). This shift was directly related to the retreat of the optimum precipitation range in the northwestern part of the valley, and the northern boundary of the area with a high percentage of land use coincided with the area with the optimum precipitation range for millet cultivation in the western part of the valley during 5-3 ka BP (Figures 5d and 5e). In the eastern part of the valley, the human land use widely and quickly expanded after 4 ka BP, and the precipitation conditions were higher than the optimal range for millet cultivation in this region; therefore, the regional land use expansion was potentially accelerated by the population pressure from the western valley at this time. In addition, the further aridification in the western part of the valley after 3 ka BP did not significantly affect the range of the land use distribution, which may have been due to some of the non-climatic factors, such as the transition between different subsistence strategies and the development of agricultural technology (Su and Chen, 2008; Jia et al., 2016b; Zheng et al., 2021).

The strong correlations between climate change and prehistoric human activity in the Holocene have been revealed in many previous studies conducted in and around the Liaohe River Valley (Hu et al., 2002; Jia et al., 2017; Guo et al., 2018; Zhao et al., 2019; Zheng et al., 2021; Wang et al., 2024). In the western part of the Liaohe River Valley, the significant decrease in the human activity intensity during 5-4 ka BP has always been attributed to the change in the climate to colder and drier conditions in accordance with more frequent wind and sand activity (Hu et al., 2002; Guo et al., 2018). In the eastern part of the valley, the responses of the regional hydrological conditions to climate changes have been concluded to be a direct driver of the human activity distribution (Zhao et al., 2019). In addition to climatic factors, previous studies have also clarified the roles that non-climatic factors played in the evolution of the human activity in this region. For example, the unbalanced distribution of the human activity between the northern and southern parts of the Liaohe River Valley during 8-2 ka BP (especially after 4 ka BP) was also related to the migration of the population from the Yellow River Valley (Hu et al., 2002; Leipe et al., 2019). In addition, the pattern of human activity was sometimes determined by the subsistence strategies of the society, which were further affected by the different environmental conditions (e.g., the landforms and water system) (Jia et al., 2016a; Wang et al., 2016; Zhao et al., 2019). Compared with previous studies, our study provides not only comparisons of the temporal curves for different climate and human activity variables from the qualitative perspective but also a series of maps that illustrate the spatial reactions of prehistoric human activity to climate change from the quantitative perspective. Furthermore, we clarified that precipitation was the key climatic factor affecting the changes in the human activities in the valley.