When we observe large-scale environmental changes from satellite imagery—such as a desert shrinking, a forest thinning, or a landscape growing greener—we are looking at the net outcome of multiple overlapping forces. Is a grassland recovering because it rained more this decade, or because local governments enforced sustainable grazing policies?
To answer this question, researchers rely on a powerful framework known as Residual Trend Analysis (RESTREND). The core intuition behind RESTREND is simple subtraction: if we can accurately model how an ecosystem should behave based solely on natural climate variations, then any systematic deviation from that prediction must be caused by non-climatic factors. By analyzing the long-term trend of these leftovers (the residuals), we can isolate and quantify the human footprint on the landscape.
Step 1: Building the Climate Baseline
The first step in residual analysis is to establish a statistical relationship between environmental drivers (like temperature, precipitation, or solar radiation) and an ecosystem metric of interest, such as the Normalized Difference Vegetation Index (NDVI), Leaf Area Index (LAI), or Net Primary Productivity (NPP).
Multiple Linear Regression (MLR) is frequently used to construct this baseline. For instance, Wanqiang Han et al. applied linear regression to link vegetation loss directly to temperature and precipitation anomalies in Assessment of Vegetation Drought Loss and Recovery in Central Asia Considering a Comprehensive Vegetation Index.
Depending on the ecosystem, the relevant climate variables change:
- Terrestrial Vegetation: Baozhang Chen et al. used MLR to partition the effects of temperature, vapor pressure deficit, and soil wetness on global leaf area trends in Inhibitive Effects of Recent Exceeding Air Temperature Optima of Vegetation Productivity and Increasing Water Limitation on Photosynthesis Reversed Global Greening.
- Marine Ecosystems: The same mathematical intuition applies to oceans. Xinyu Zheng et al. leveraged MLR to isolate how marine environmental factors like sea surface temperature, wind speed, and sea ice concentration drive changes in phytoplankton productivity in Diversity mechanisms of long-term changes in phytoplankton biomass and productivity in the Southern Ocean: comparison in the Amundsen and Cosmonaut Seas.
Step 2: Extracting the Residuals
Once the climate model is trained, it can calculate a "predicted" or "potential" value for each time step or geographic pixel. This prediction represents what the environment would look like if climate were the only variable changing.
The residual is then calculated as:
$$\text{Residual} = \text{Observed Value} - \text{Predicted Value}$$
If the observed vegetation matches the climate prediction perfectly, the residual is zero. A positive residual means the vegetation is healthier or denser than the weather would suggest; a negative residual means it is underperforming.
Step 3: Analyzing the Trend for Attribution
A single residual value only tells us the prediction error for one moment in time. To isolate human impact, researchers calculate the trend of these residuals over a multi-decade span.
Because the climate's influence has already been mathematically subtracted out, any persistent upward or downward trajectory in the residuals represents a non-climatic driver. Performing this calculation pixel-by-pixel across satellite images yields a spatial map of human impact:
- Isolating Restoration and Mining: Aman Fang et al. used pixel-based residual trends to separate natural precipitation variations from the positive impacts of human-driven ecological restoration in industrial zones in Tempo-Spatial Variation of Vegetation Coverage and Influencing Factors of Large-Scale Mining Areas in Eastern Inner Mongolia, China.
- Differentiating Agriculture from Climate: In Trend in Satellite-Observed Vegetation Cover and Its Drivers in the Gannan Plateau, Upper Reaches of the Yellow River, from 2000 to 2020, Zhengyang Zhang et al. isolated the specific regions where human agricultural practices overrode local aridity trends.
- Mapping Urbanization: Siya Chen et al. applied this to the changing landscape of Beijing in The Spatio-Temporal Variation of Vegetation and Its Driving Factors during the Recent 20 Years in Beijing, demonstrating how positive residual trends tracked urban greening policies while negative residual trends perfectly captured rapid suburban expansion.
Quantifying Relative Contributions
Beyond merely identifying whether humans or climate caused a change, the framework can calculate exactly how much each driver contributed. By comparing the slope of the climate-predicted trend against the slope of the residual trend, the total observed change can be broken down into exact percentages.
For example, Jiali Xie et al. used multiple regression residual analysis in The Latest Desertification Process and Its Driving Force in Alxa League from 2000 to 2020 to disentangle the drivers of desertification reversal, mapping out exactly where policy interventions succeeded relative to shifting weather patterns. To handle complex, non-linear interactions across borders, Xueqin Zhang et al. introduced a first-difference multivariate regression approach in Precipitation and Anthropogenic Activities Jointly Green the China-Mongolia-Russia Economic Corridor to isolate the relative contributions of climate change from non-climatic factors.
This quantification is equally vital for diagnosing environmental degradation. Fei Wang 0120 et al. utilized residual analysis in Temporal and Spatial Change in Vegetation and Its Interaction with Climate Change in Argentina from 1982 to 2015 to prove that while climate shifts played a minor role, human activities like deforestation and agricultural expansion were the primary drivers of vegetation loss across the Pampas and Patagonia. Similarly, Xingming Yuan and Bing Guo mapped the spatial mosaic of the Qinghai-Tibet Plateau in The Responses of Vegetation NPP Dynamics to the Influences of Climate-Human Factors on Qinghai-Tibet Plateau from 2000 to 2020, showing exactly where human activities had a beneficial versus detrimental impact on potential Net Primary Productivity.
Alternative Baselines: Simulation vs. Regression
While statistical regression is the most common way to build a baseline, the same residual logic can be applied using process-based simulation models. Instead of fitting a line to historical data, researchers simulate the physical ecosystem under a controlled scenario.
In Attribution and Risk Assessment of Wind Erosion in the Aral Sea Regions Using Multi-Source Remote Sensing and RWEQ on GEE, Jianli Ding et al. simulated wind erosion under fixed land-use conditions. By keeping land use frozen in the model, they isolated the pure contribution of climate shifts; any real-world deviation from this simulation was attributed to human activity.
Taking a similar comparative approach, Zhong'en Niu et al. studied the greening of the Chinese Loess Plateau in Climate Change and CO2 Fertilization Have Played Important Roles in the Recent Decadal Vegetation Greening Trend on the Chinese Loess Plateau. They compared satellite-observed LAI against a process model simulation that included all environmental changes, including CO2 fertilization. The residual difference between the real-world observations and the environmental simulation isolated the specific impact of human land-use management, such as large-scale reforestation programs.
Ultimately, whether using simple linear models or complex physical simulations, subtracting what nature explains leaves behind a clear signal of human impact. This makes residual trend analysis an indispensable tool for verifying the success of conservation policies and managing ecosystems sustainably, as emphasized by Panpan Chen et al. in Vegetation Dynamic Assessment by NDVI and Field Observations for Sustainability of China’s Wulagai River Basin.
Key Foundational Work
Vegetation Dynamic Assessment by NDVI and Field Observations for Sustainability of China’s Wulagai River Basin (2021) — contributed 1 concept elements to this direction.
Tempo-Spatial Variation of Vegetation Coverage and Influencing Factors of Large-Scale Mining Areas in Eastern Inner Mongolia, China (2019) — contributed 1 concept elements to this direction.