Optimizing coordination and trade-offs between food security and biodiversity conservation goals
Data files
Aug 28, 2025 version files 96.77 MB
Abstract
Balancing food security and biodiversity conservation—two often conflicting objectives—is essential for achieving global goals (e.g., SDG 2 and 15; GBF Targets 1, 3, and 10). While previous studies have explored global or national-level trade-offs, there is a lack of spatially explicit, scenario-based planning frameworks at regional scales to reconcile cropland expansion and biodiversity conservation. The study develops a multi-objective spatial planning framework to assess how future cropland expansion may be optimized to reduce biodiversity impacts while ensuring food security in the northwestern dry geo-eco region of China. Using a random forest model trained with environmental, socio-economic and trend variables to project cropland expansion from 2020 to 2030 and identify areas of spatial conflict with biodiversity priority regions. Results reveal intense conflicts in ecologically sensitive areas such as the Altai and Tianshan Mountains. Under a food-security-first scenario, expanding 300,000 km² of cropland would result in 167,978 km² of conflict areas and a 12.20% habitat loss rate. In contrast, a biodiversity-priority scenario achieves only 199,782 km² of cropland expansion, reducing habitat loss to 2.39%. A trade-off coordination scenario offers an optimized balance, enabling 300,000 km² of cropland expansion while protecting 30% of biodiversity priority areas and limiting habitat loss to 3.52%. This study highlights a novel framework for integrating food security and biodiversity conservation, offering spatially explicit strategies to support region-specific sustainable land-use planning.
Dataset DOI: 10.5061/dryad.02v6wwqgn
Description of the data and file structure
This dataset was generated as part of the study "Optimizing Coordination and Trade-offs between Food Security and Biodiversity Conservation Goals" (submitted to Ecography).
The data include two raster layers:
A probability map of future cropland expansion in the northwestern dry geo-eco region of China for the period 2020–2030, produced using a Random Forest (RF) model trained with environmental, socio-economic, and trend variables.
A biodiversity conservation priority map derived from spatial prioritization analysis based on area of habitat (AOH).
These datasets were used to identify spatial conflicts between predicted cropland expansion and biodiversity priority areas, and to develop land-use planning scenarios under food security and biodiversity conservation objectives.
Files and variables
File: Probability_map_of_cropland_expansion_in_the_study_area_for_2020-2030.tif
Description: Raster layer representing the modeled probability (range: 0–1) of cropland expansion for each 1 km × 1 km grid cell within the study area for the period 2020–2030.
Value: Probability of cropland expansion (unitless; 0 = no likelihood, 1 = maximum likelihood).
Missing values: None.
File: Biodiversity_conservation_priority_map_of_the_study_area.tif
Description: Raster layer showing biodiversity conservation priority scores derived from a spatial prioritization analysis. Higher scores indicate higher priority for conservation action.
Value: Conservation priority index (unitless; range: 0–1, with 1 being highest priority).
Missing values: None.
Code/software
Viewing software: Any GIS software capable of reading GeoTIFF format, such as ArcGIS Pro.
Modeling software:
- R (v4.3.1) with packages: ranger, raster, sp, rgdal, biovars, dismo.
- ArcGIS Pro (v3.0+) for spatial processing (Euclidean Distance, slope extraction, raster resampling).
- Zonation (v4.0) for biodiversity prioritization.
Workflow:
1.Model training:
- Time-lagged Random Forest (RF) modeling: predictor variables from an earlier period (e.g., 2000–2005) used to model expansion in the subsequent period (e.g., 2005–2015).
- Training sample balancing: tested ratios of expansion to non-expansion grids (1:1, 1:2, …, all available points).
- RF implemented with 500 trees using the ranger package in R.
2.Model evaluation:
- Predictions for 2010–2020 compared with observed expansion.
- Evaluated using Overall Accuracy, Precision, Recall, F1 Score, Brier Score, AUC-ROC, and AUC-PRC.
- Best-performing RF model selected based on combined performance metrics.
3.Probability mapping:
Best RF model applied with 2015–2020 predictor variables to project cropland expansion probability (0–1) for 2020–2030 across the Northwestern dry geo-eco region of China.
4.Biodiversity prioritization:
- Species Area of Habitat (AOH) maps from Lumbierres et al. (2022) were processed to 1 km resolution.
- Zonation software applied with additive benefit function (ABF) and species weights based on IUCN Red List categories (CR: 8, EN: 6, VU: 4, NT/DD: 2, LC: 1).
- Output: biodiversity conservation priority index map (0–1).
5.Conflict analysis:
Spatial overlay of cropland expansion probability map and biodiversity priority map to identify potential conflict zones under different land-use planning scenarios.
Access information
Data was derived from the following sources:
- LULC: Nine major land cover classes (cropland, forest, shrub, grassland, water, snow/ice, barren, impervious, and wetland) were obtained from the China Land Cover Dataset (CLCD, https://zenodo.org/records/12779975).
- Climate data: We obtained 1 km resolution datasets for China covering temperature, precipitation, potential evapotranspiration, relative humidity, and average wind speed from the National Earth System Science Data Center (http://www.geodata.cn/).
- Topography: Elevation data were obtained from WorldClim (https://www.worldclim.org/).
- Soil: We used the Harmonized World Soil Database (HWSD v1.2) at 1 km resolution from the FAO (https://www.fao.org/soils-portal/data-hub/soil-maps-and-databases/harmonized-world-soil-database-v12/en/).
- Population: Population data at 1 km resolution for China were obtained from WorldPop (https://hub.worldpop.org/).
- Roads: Road network data were sourced from OpenStreetMap (https://www.openstreetmap.org/), and Euclidean Distance analysis was used to calculate distance to roads.
