DrylANNd: twenty years of monthly, 0.05° gross primary production and evapotranspiration estimates for global drylands

Matthew Paul Dannenberg; Mallory Barnes; Joel Biederman; Miriam Johnston; Susan Meerdink; Sophie Ruehr; Russell Scott; William Kolby Smith; A Park Williams

doi:10.1088/2752-664X/ae8f4f

Back

DrylANNd: twenty years of monthly, 0.05° gross primary production and evapotranspiration estimates for global drylands

Journal article

Open access

Peer reviewed

DrylANNd: twenty years of monthly, 0.05° gross primary production and evapotranspiration estimates for global drylands

Matthew Paul Dannenberg, Mallory Barnes, Joel Biederman, Miriam Johnston, Susan Meerdink, Sophie Ruehr, Russell Scott, William Kolby Smith and A Park Williams

Environmental research. Ecology, Vol.5(3), 034501

09/08/2026

DOI: 10.1088/2752-664X/ae8f4f

Appears in UI Libraries Support Open Access

Files and links (1)

url

https://doi.org/10.1088/2752-664X/ae8f4fView

Published (Version of record) Open Access

Abstract

Drylands provide ecosystem services to one-third of humanity, are home to many rare and endemic species, and are major drivers of trends and interannual variability (IAV) in land carbon uptake. However, many satellite-based carbon and water flux products systematically underestimate the variability of dryland fluxes. Here, we present two decades (2003-2022) of monthly, 0.05°-resolution gross primary production (GPP) and evapotranspiration (ET) estimates for global drylands using a new-and-improved version of our Dryland Artificial Neural Network (DrylANNd) approach driven by surface reflectance and land surface temperature from the Moderate Resolution Imaging Spectroradiometer (MODIS) and soil moisture/temperature from the Soil Moisture Active-Passive (SMAP) sensor. Neural networks were calibrated using 33 eddy covariance sites in arid to subhumid regions of North America, southern Europe, and Australia. To optimize DrylANNd’s ability to capture IAV, we first decomposed monthly predictor (MODIS and SMAP) and response (GPP and ET) variables at each site into mean monthly fluxes and monthly anomalies from those means. We then fit two sets of DrylANNd models: one predicting mean monthly fluxes (capturing the seasonal and spatial variability of GPP and ET) and one predicting monthly flux anomalies (capturing the IAV in GPP and ET). Finally, we summed the predicted mean monthly flux and monthly flux anomaly to give the actual monthly flux. This approach forces one set of neural networks to train specifically on the IAV, rather than primarily fitting to spatial and seasonal patterns. DrylANNd captures ~60% of the monthly flux variance and 50-60% of the IAV, a substantial improvement over many other satellite-based estimates, which only capture ~30% of IAV. The improved ability of DrylANNd to capture the IAV of GPP and ET will improve future dryland-specific research on carbon and water fluxes, including their sensitivities to extreme events and global change.

UIOWA OA Agreement

Details

Title: Subtitle: DrylANNd: twenty years of monthly, 0.05° gross primary production and evapotranspiration estimates for global drylands
Creators: Matthew Paul Dannenberg - University of Iowa
Mallory Barnes - Indiana University Bloomington
Joel Biederman - Agricultural Research Service
Miriam Johnston - University of Maryland Center for Environmental Science
Susan Meerdink - University of Iowa
Sophie Ruehr - Carnegie Institution for Science
Russell Scott - Agricultural Research Service
William Kolby Smith - University of Arizona
A Park Williams - University of California, Los Angeles
Resource Type: Journal article
Publication Details: Environmental research. Ecology, Vol.5(3), 034501
DOI: 10.1088/2752-664X/ae8f4f
ISSN: 2752-664X
eISSN: 2752-664X
Publisher: IOP Publishing
Language: English
Date published: 09/08/2026
Academic Unit: School of Earth, Environment, and Sustainability
Record Identifier: 9985214919402771

Metrics

6 Record Views