Gridded Datasets | Dr. Mauricio Zambrano-Bigiarini

Article on the evaluation of gridded soil moisture products published in HESS

Sat, 04 Apr 2026 00:00:00 +0000

On January 12th, 2026, published our article entitled . This study investigates how spatial patterns, temporal trends, and record length in hourly precipitation data affect annual maximum intensities estimated with stationary and non-stationary models across a climatically and topographically diverse region.

Motivation

Soil moisture is a key variable controlling how water moves through landscapes, supports vegetation, and interacts with the atmosphere. It plays a central role in drought monitoring, ecosystem management, and hydrological modelling. In many regions—particularly natural or remote ecosystems—direct soil moisture measurements are scarce. As a result, scientists and practitioners often rely on large-scale datasets derived from satellites or land surface models. This study evaluates how accurately these datasets represent soil moisture dynamics across Chile’s wide range of climates, from arid northern zones to humid southern forests.

What was the novelty?

The study assessed four widely used soil moisture datasets, ERA5, ERA5-Land, SMAP-L4, and GLDAS-Noah, against detailed field observations collected every three hours from the Kimün-Ko monitoring network. The monitoring sites span ten near-natural ecosystems along Chile’s hydroclimatic gradient.

Study area: (a) catchment location (CAMELS-CL; Alvarez-Garreton et al., 2018); (b) elevation (SRTMv4.1; Jarvis et al., 2008); (c) land cover classification (CLDynamicLandCover.V2; Galleguillos et al., 2024); (d) soil properties (CLSoilMaps; Dinamarca et al., 2023); and (e) aridity index (AI=P/PET) 1970–2000 (Global-AI-PET-v3; Zomer et al., 2022).

Locations of in situ TEROS 10 and TEROS 12 sensors. (a) Example of TEROS 10 and TEROS 12 sensors installed across various land cover types; (b) northern arid sites in the Petorca (PRB) and Mapocho (MRB) river basins; and (c) southern humid sites in the Cauquenes (CRB) and Trancura (TRB) river basins. Red triangles indicate the locations of in situ SM monitoring sites. Grid cell boundaries of each gridded SM product are shown for ERA5 (green), ERA5-Land (purple), SPL4SMAU (blue), and GLDAS-Noah (lightblue).

In addition to standard statistical indicators, the researchers applied an event-based diagnostic method that examines how soil moisture responds to individual rainfall events. This approach evaluates both the magnitude of the response and how quickly the soil becomes wetter after rainfall.

What we found

The evaluation revealed consistent patterns with direct implications for environmental monitoring and modelling:

ERA5 and ERA5-Land showed the most reliable overall performance. These datasets reproduced seasonal soil moisture dynamics reasonably well across most regions, particularly in the wetter southern ecosystems.
Deeper soil layers were simulated more accurately than surface layers. Root-zone soil moisture changes more slowly and is less sensitive to short-term fluctuations, making it easier for large-scale models to represent.
Arid regions remain difficult to simulate. In northern ecosystems, all datasets struggled to reproduce the first rainfall response after long dry periods, typically overestimating how much and how quickly soil moisture increased.
Performance varies by product and location. Some datasets performed well under specific conditions—for example, one showed relatively strong skill for surface soil moisture in selected arid sites—while others systematically underestimated soil moisture in wetter environments.

Schematic summary of the main conclusions.

Why the new diagnostic approach is important

A key contribution of the study is the demonstration that traditional performance metrics can overlook important timing and response errors. A dataset may appear accurate when evaluated over long periods but still fail to capture the rapid changes that occur during individual storms. Event-based diagnostics provide a clearer understanding of how models represent real hydrological processes, especially during extreme or short-lived events.

Why this matters for practice and decision-making

The findings provide practical guidance for selecting soil moisture datasets in regions where field measurements are limited. In particular, identifying the most reliable products supports better drought monitoring, improved hydrological simulations, and more informed ecosystem and water resource management.

Our study also highlights the importance of evaluating not only average performance, but also the dynamic response of soils to rainfall—an aspect that becomes increasingly critical under changing climate conditions.

The full article can be found here: .

Infographic summary, created by Google NotebookLM (23-Apr-2026)

Article on gridded IDF curves published in HESS

Mon, 12 Jan 2026 00:00:00 +0000

Motivation

Intensity–Duration–Frequency (IDF) curves are essential for designing infrastructure that must safely manage extreme rainfall, including urban drainage systems, culverts, and flood protection works. Traditionally, these curves depend on long-term observations from rain gauges. In many parts of Chile, however, such records are sparse, unevenly distributed, or too short to support robust design. This study evaluates whether modern gridded precipitation datasets can provide reliable alternatives for estimating rainfall extremes across Chile’s diverse climatic and topographic regions.

What is new in this study

The study analysed -for the first time in Chile- data from 161 quality-controlled hourly rain gauges together with five widely used gridded precipitation products:

IMERG (versions v06B and v07B)
ERA5
ERA5-Land
CMORPH-CDR

Then, a new systematic evaluation framework was developed to:

Correct systematic biases in gridded precipitation estimates using local observations.
Detect long-term changes in extreme precipitation intensity.
Compare conventional (stationary) and trend-aware (non-stationary) statistical models for estimating design storms with return periods from 2 to 100 years and durations from 1 to 72 hours.
Assess how the length of the precipitation record influences the reliability of design estimates

Flowchart summarising the methodology used in this study

What we found

Several findings are directly relevant for engineering practice and hydrological planning:

Extreme precipitation does not mirror average precipitation patterns. The most intense short-duration storms occur in central–southern Chile, even though total annual precipitation increases farther south.
Mountains experience substantially higher extremes. For longer storm durations, the Andes show markedly higher intensities than nearby lowland areas, indicating that design values derived from valley stations may underestimate risk in mountainous terrain.
Recent decades show declining extremes in central Chile. This pattern is consistent with the prolonged regional drought and reduced frequency of winter storm systems.
Traditional statistical assumptions remain adequate for design. Differences between stationary and non-stationary models were generally small, suggesting that standard engineering approaches remain appropriate in most applications.
Shorter records can still provide reliable estimates. In many cases, 20 years of data produced results comparable to those obtained from 40-year records, which is operationally important in data-limited regions.

Why this study is important for infrastructure and risk management

The results demonstrate that carefully evaluated gridded precipitation datasets can extend reliable rainfall design information to areas without rain gauges. This capability is particularly relevant in Chile, where steep topography and strong climatic gradients create large spatial variability in extreme rainfall.

To facilitate practical use, the authors implemented these findings in an operational web platform that provides location-specific IDF curves for continental Chile: . This tool enables engineers, planners, and public agencies to access consistent design rainfall estimates, supporting safer infrastructure development and more resilient water management under changing climatic conditions. We hope this tool might be incorporated in future design manuals in Chile.

Main screen of the web platform

The full article can be found here: .

Infographic summary, created by Google NotebookLM (23-Apr-2026)

XXVII Congreso Chileno de Hidráulica: Course 'Using R for spatio-temporal data analysis: application to daily data CR2Met v2.5'

Tue, 21 Oct 2025 00:00:00 +0000

R Course: Spatiotemporal Data Analysis

From October 20th to 25th, the was held at the Faculty of Engineering of Concepción (FI UdeC). The congress was convened by the Chilean Society of Hydraulic Engineering (SOCHID) and organized by the Department of Civil Engineering of the University of Concepción. The event consisted of courses, scientific presentations, and lectures featuring the participation of engineers, academics, and students.

On October 21th, I taught the course Using R for spatio-temporal data analysis: application to daily data CR2Met v2.5, which was attended by undergraduate and graduate students, as well as public and private sector professionals.

Dr. Zambrano-Bigiarini at Foro UdeC

UFRO team at Foro UdeC

UFRO team with Dr. Oscar Link (UdeC)

RFmerge

Fri, 22 May 2020 00:00:00 +0000

R package.

Description

is an R package (currently not on CRAN, but working with the terra package on Github) designed to generate more reliable environmental datasets by combining information from gridded datasets and ground-based observations. It implements the Random Forest Merging Procedure (RF-MEP) (Baez-Villanueva et al., 2020), a machine-learning approach developed to improve the spatial and temporal representation of environmental variables—particularly precipitation—by leveraging the complementary strengths of different data sources.

The package addresses a persistent challenge in hydrology and Earth system sciences: no single dataset provides a complete and unbiased representation of environmental conditions. Rain gauges offer accurate point measurements but limited spatial coverage, while satellite products provide broad spatial information that may contain systematic errors. By integrating these sources within a unified statistical framework, produces merged datasets that better capture variability, reduce bias, and enhance the reliability of environmental analyses, especially in data-scarce regions.

Built with operational applications in mind, provides a transparent and reproducible workflow for dataset merging that can be adapted to a wide range of variables beyond precipitation, including temperature, soil moisture, or other gridded datasets. It is particularly well suited for researchers and practitioners who require spatially consistent datasets to support hydrological modelling, climate analysis, and water resources assessment.

Grounded in peer-reviewed research and real-world applications, offers a technically robust and methodologically sound foundation for transforming heterogeneous environmental observations into coherent, analysis-ready datasets.

Reference

Baez-Villanueva, O.M.; Zambrano-Bigiarini, M.; Beck, H.; McNamara, I.; Ribbe, L.; Nauditt, A.; Birkel, C.; Verbist, K.; Giraldo-Osorio, J.D.; Thinh, N.X. (2020). , Remote Sensing of Environment, 239, 111610. doi:10.1016/j.rse.2019.111606.