Regional Evapotranspiration Estimation and Partitioning Model Based on Energy Balance: A Case Study of the Tibetan Plateau
Description
This research presents a novel regional two-source evapotranspiration (ET) model designed to simulate and partition total evapotranspiration into soil evaporation (E) and vegetation transpiration (Tr) over the Qinghai-Tibetan Plateau (QTP) with high precision. Developed by extending the energy balance framework of Wang and Yamanaka (2014) from point to kilometer scale, the model's key innovation is its ability to operate without relying on land surface temperature—a parameter often challenging to obtain—enabling daily simulations at a 1 km resolution from 2003 to 2018. The model is grounded in the physical energy balance principles of both the vegetation canopy and the soil layer. It simultaneously solves the energy balance equations for these two surfaces through an iterative Newton-Raphson method, optimizing leaf and soil surface temperatures to derive the corresponding latent heat fluxes for Tr and E. Critical model components include the calculation of canopy solar radiation transmittance based on leaf area index and clumping index, as well as the dynamic parameterization of canopy and soil surface resistances, which are informed by variables such as soil moisture and incoming solar radiation. Inputs to the model integrate multi-source data, including meteorological drivers, vegetation parameters from satellite-derived LAI and land cover products, and soil properties from reanalysis and observational datasets. This comprehensive data integration supports robust simulation across the plateau's diverse ecosystems. Validation against flux tower observations at nine sites and comparison with existing regional ET products confirm the model's strong performance and reliability. It effectively captures seasonal and spatial ET dynamics and demonstrates superior skill in partitioning the total flux into its soil and vegetation components. By combining physical rigor, operational feasibility without land surface temperature, and high spatiotemporal resolution, this model provides a valuable tool for advancing water and energy cycle studies across the heterogeneous landscapes of the Tibetan Plateau. The datasets used for the model simulation are detailed in the Data Availability Statement. The corresponding code, written in Python, is located in the file SPAC.py.
Files
Steps to reproduce
The datasets used in this study were generated through a combination of multi-source remote sensing products, land surface reanalysis, in situ observations, and model-derived parameters. The meteorological forcing data were obtained from the high-resolution China Regional Surface Meteorological Elements Driving Dataset, which integrates ground station measurements, satellite retrievals, and reanalysis outputs using dynamical downscaling and spatial interpolation techniques. Vegetation parameters, including Leaf Area Index (LAI), were derived from the GIMMS LAI4g dataset, which applies a radiative transfer-based algorithm to AVHRR satellite observations with temporal smoothing and gap-filling procedures. Land cover classification was based on the MODIS MCD12Q1 product, generated using a supervised decision tree algorithm applied to Terra and Aqua MODIS reflectance data. Soil temperature data were sourced from the ERA5-LAND reanalysis, which assimilates observational data into the HTESSEL land surface model. Soil moisture data came from the China Soil Moisture Dataset, produced by merging satellite microwave retrievals with ground measurements using a machine learning-based blending approach. Soil hydraulic parameters (e.g., saturated water content) were estimated using the Rosetta3 pedotransfer model, which predicts soil properties based on texture class and bulk density using a hierarchical neural network approach. Site-level validation data were collected from eddy covariance flux towers within the ChinaFLUX, HiWATER, and Tibetan Observation and Research Platform networks. These stations follow standardized protocols for instrument calibration, data collection, and post-processing, including spike detection, turbulence stationarity tests, and energy balance correction. The evapotranspiration products used for regional comparison (SEBS, ETMonitor, PML-V2, MOD16-STM) were generated using distinct physical or empirical algorithms applied to similar remote sensing and meteorological inputs. The core model was implemented in Python (version 3.8) and relied on standard scientific libraries including NumPy, SciPy, and pandas for numerical computations and data handling. The model workflow involved daily sequential execution over each 1 km grid cell, with iterative solution of energy balance equations using the Newton-Raphson method. All input datasets were spatially resampled to a common 1 km resolution WGS84 grid and temporally aggregated or interpolated to daily time steps to ensure consistency. This integrated data generation and processing framework ensures reproducibility and provides a transparent pathway for reconstructing the model simulations.
Institutions
- Beijing Normal University