📄 earth_science

Can Satellite and Reanalysis Estimates Capture the Intra-Annual Structure of Localised Temperature Trends? Novel Diagnostic Metrics and Station Evidence from a Data-Sparse Tropical Region, with Implications for Climate Adaptation

This study evaluates the ability of satellite and reanalysis estimates (ERA5 and CHIRTS) to capture the intra-annual structure of localised temperature trends in Ghana by introducing novel diagnostic metrics (SS var and SPA), revealing significant biases in seasonal timing and warming intensity that necessitate trend-preserving bias correction before these datasets can be reliably used for climate adaptation planning in data-sparse tropical regions.

Original authors: John Bagiliko, David Stern, Danny Parsons, Denis Ndanguza

Published 2026-08-19

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: John Bagiliko, David Stern, Danny Parsons, Denis Ndanguza

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Climate change is not a single, uniform event that happens everywhere at the same speed. While the planet as a whole is warming, the experience of that heat varies wildly from one valley to the next, and from one season to the next. For farmers, doctors, and city planners, the most critical details are often the local ones: exactly when the hottest days arrive, how fast the nights are getting warmer, and whether the difference between day and night temperatures is shrinking. In many parts of the world, especially across the tropics, ground-based weather stations that record these daily details are rare or have stopped working. To fill these gaps, scientists increasingly rely on satellite images and computer models that estimate temperatures over vast areas. These digital estimates are powerful tools, but they have rarely been tested to see if they can capture the subtle, shifting patterns of warming that happen within a single year. If these tools smooth out the local peaks and valleys of heat, they could lead to adaptation plans that miss the mark entirely.

A team of researchers set out to test whether these digital estimates can truly see the local rhythm of warming. They focused on Ghana, a country in West Africa where the landscape shifts from dry savannah in the north to lush forests and a humid coast in the south. The team gathered daily records of the highest and lowest temperatures from twenty-two weather stations across the country, covering the years from 1983 to 2021. They cleaned this data rigorously to remove errors and then compared it against two popular digital products: one is a high-resolution computer model of the atmosphere, and the other is a satellite-based estimate that blends satellite data with model outputs. The goal was not just to see if the digital tools got the average temperature right, but to see if they could reproduce the specific timing and intensity of warming that happens in different months.

The researchers found that the reality on the ground is far more complex than the digital maps suggest. In Ghana, the warming is highly localised. Even within the same climatic zone, one town might be warming at a rate of 0.1 degrees Celsius per decade while another warms at 0.5 degrees. The timing of this heat is also specific. The fastest warming for daytime highs occurs during the dry season, particularly in December and January, while the nights are warming up even faster than the days. This creates a shrinking gap between day and night temperatures, a pattern that is most pronounced along the coast. This rapid warming of the nights is a significant concern for agriculture, as crops like maize need cool nights to recover from the heat of the day.

When the team compared these detailed ground observations with the digital estimates, the results revealed a mixed and often misleading picture. The computer model tended to overestimate how fast the daytime temperatures were rising in many places, while the satellite-based product consistently underestimated the warming. Both tools struggled to capture the speed at which nighttime temperatures were climbing, often missing the rapid warming entirely. Perhaps most importantly, both digital products tended to smooth out the local details. They failed to show the sharp differences in warming rates between stations that are close to each other, and they often got the timing wrong, suggesting that warming happened in different months than it actually did on the ground.

The study also introduced two new ways to measure these errors. One method checks if the digital tools capture the full intensity of the seasonal swings in warming, while the other checks if they identify the correct months when the heat is strongest. Using these tools, the researchers found that while the digital products could sometimes guess the general direction of the trend, they frequently missed the seasonal structure. For nighttime temperatures, the digital tools often failed to align with the actual timing of the warming, suggesting that the months driving the annual heat increase were being misidentified. This is a critical failure for adaptation planning, because a strategy designed for heat in March will not work if the real danger arrives in December.

The implications of these findings are clear for anyone trying to prepare for a changing climate. The digital estimates, while useful for filling in large gaps where no data exists, cannot be trusted to tell the whole story of local warming without correction. They tend to flatten the unique, jagged edges of local climate change, hiding the specific risks that communities face. The researchers recommend that before these digital tools are used to design policies for agriculture or public health, they must be adjusted to preserve the true trends and seasonal patterns observed on the ground. Without this step, adaptation plans risk being built on a smoothed-out version of reality, potentially leaving vulnerable populations unprepared for the specific timing and intensity of the heat they will actually experience.

Technical Summary

Problem Statement
Satellite and reanalysis estimates (SREs) are increasingly utilized as substitutes or complements for scarce station data in data-sparse tropical regions. While previous validations have demonstrated that SREs perform well in capturing daily, seasonal, and annual temperature averages, there is limited systematic attention regarding their ability to reproduce the intra-annual structure of long-term temperature trends. Specifically, it remains unclear whether these products accurately capture the correct months, intensity, and timing of warming. This limitation is critical for climate adaptation planning, where the seasonal timing and intensity of warming dictate the design of sector-specific responses in agriculture and human health. To date, no equivalent analysis examining these specific trend characteristics has been conducted for West Africa.

Methodology
The study utilizes daily maximum ( $T_{max}$ ) and minimum ( $T_{min}$ ) temperature observations from 22 synoptic stations in Ghana (1983–2021), serving as a representative case for data-sparse tropical regions. The station data underwent rigorous quality control and homogenization using four distinct tests (SNHT, Pettitt's Test, Buishand's BLRT and BUT) with AgERA5 as a reference series. Trends were estimated using the Modified Mann-Kendall test and Sen's slope estimator.

The study evaluates two SREs: ERA5 (reanalysis) and CHIRTS (satellite-based). To assess performance beyond simple mean bias, the authors introduce two novel diagnostic metrics:

Trend Variability Skill Score ( $SS_{var}$ ): Adapted from the Ratio of Standard Deviations, this metric quantifies "smoothing bias" by comparing the spread and intensity of intra-annual trends in gridded data against station observations. It measures whether the product captures the full intensity of localised trend variability or flattens unique station-level signatures.
Seasonal Phase Alignment (SPA): Adapted from the Pearson Correlation Coefficient, this metric measures "temporal phase shift" by correlating observed and gridded monthly trend residuals. It determines whether the product correctly identifies the specific months driving the annual trend.

Statistical significance for SPA was assessed against a critical value ( $|SPA| \ge 0.576$ ) for $N=12$ monthly observations. Stations with very low observed intra-annual trend variability (below the first quartile) were excluded from $SS_{var}$ and SPA conclusions to avoid hypersensitivity to minor discrepancies.

Key Results

Observed Trends: Temperature trends in Ghana are highly localised, with warming rates for $T_{max}$ ranging from 0.1–0.3°C decade $^{-1}$ and $T_{min}$ from 0.1–0.5°C decade $^{-1}$ . The strongest warming occurs during the dry season (December–February). A distinct diurnal asymmetry was observed where $T_{min}$ warmed faster than $T_{max}$ , leading to widespread significant declines in the Diurnal Temperature Range (DTR) of 0.1–0.2°C decade $^{-1}$ , particularly in the Coastal zone.
SRE Performance (Bias): The two products exhibited opposing biases for $T_{max}$ : ERA5 overestimated trends at 16 of 22 stations, while CHIRTS underestimated them at 20 of 22. Both products consistently underestimated $T_{min}$ warming rates and the magnitude of DTR declines.
SRE Performance (Smoothing and Phase):
- Smoothing Bias ( $SS_{var}$ ): Both products frequently smoothed local seasonal signals, particularly for $T_{min}$ , failing to capture the full intensity of intra-annual trend variability at stations with high observed variability (e.g., Sunyani).
- Phase Alignment (SPA): For $T_{max}$ , both products demonstrated a relative ability to track the seasonal phase of warming (significant at 56% for ERA5 and 81% for CHIRTS among stations with sufficient variability). However, phase alignment for $T_{min}$ and DTR was markedly weaker, with many stations showing weak or negative correlations, indicating that the products often misidentify the months driving nocturnal and diurnal-range warming.
DTR Specifics: Neither product successfully replicated the direction or significance of DTR trends outside the Coastal zone. CHIRTS failed to capture any significant decreasing coastal DTR trends, while ERA5 captured only one.

Significance and Claims
The paper claims that the introduced $SS_{var}$ and SPA metrics offer transferable tools for evaluating SREs in other data-sparse tropical regions. The study concludes that while SREs are essential for filling data gaps, they cannot be assumed universally superior for local-scale trend analysis due to systematic biases in magnitude, smoothing, and phase alignment.

The authors assert that these findings have direct implications for climate adaptation. The observed localisation of trends and diurnal asymmetry suggests that regional averaging may mask critical signals relevant to sectors like agriculture (e.g., maize yield variability driven by minimum temperatures) and public health. Consequently, the paper recommends that SREs undergo trend-preserving bias correction (such as Quantile Delta Mapping or ISIMIP3 BASD) before being utilized in adaptation planning for temperature-sensitive sectors. The study positions its diagnostic framework as a low-cost means for auditing SRE products to ensure they faithfully represent both the direction and intensity of local warming before adoption for climate services.

More like this