论文信息
论文解读
AI 辅助整理- 方法
- CHIRPS降雨、Landsat NDVI/LST、趋势检验、熵权法与随机森林。
- 发现
- 降雨与温度无显著单调趋势,年际变率主导;干旱与极端降雨风险指数空间差异明显。
- 机会点
- 可引入跨传感器协调NDVI与作物物候数据,构建农业干旱—洪涝复合风险预警模型。
论文摘要
引言 理解水文气候变率对于热带河流流域的有效水资源管理至关重要。本研究评估了2000—2024年跨境塔诺河流域(Tano River Basin)的降雨、极端降水、地表温度(land surface temperature, LST)、植被响应,以及干旱状况与极端降雨灾害的分离组分。方法 本研究整合了CHIRPS降雨数据、基于Landsat的NDVI和LST、Hamed–Rao修正Mann–Kendall检验与Sen斜率、Spearman秩相关、熵权灾害组分以及随机森林敏感性诊断,并结合GPM IMERG和GMet验证。结果 重新计算的ETCCDI连续干旱日数(consecutive dry days, CDD)平均为31.93天(范围24.85—44.68天)。降雨(p = 0.726)、Rx1day(p = 0.107)、Rx5day(p = 0.315)、CDD(p = 0.216)或LST(p = 0.129)均不存在显著的单调趋势。由于Landsat ETM+/OLI NDVI序列未进行跨传感器协调,因此未报告2000—2024年合并NDVI趋势。保留传感器时段诊断仅用于表明2000—2012年(p = 0.057)和2013—2024年(p = 1.000)均不存在显著的时段内趋势。干旱状况指数(Drought Condition Index, DCI)范围为0.138—0.661,极端降雨潜在指数(Extreme Rainfall Potential Index, ERPI)范围为0.005—1.000。留一年随机森林诊断得出DCI的R² = 0.597,ERPI的R² = 0.783;这些值具有描述性,因为目标变量由预测变量构建。月度CHIRPS–GPM验证得出R² = 0.881、r = 0.939、RMSE = 27.96 mm、MAE = 18.56 mm。讨论 结果强调了年际水文气候变率;未协调的NDVI记录不用于推断合并的长期绿化趋势。
全文
Frontiers in Remote SensingTYPE Original Research
PUBLISHED 17 November 2026
DOI 10.3389/frsen.2026.1972697
Remote sensing and machine
learning-based assessment of
hydro-climatic variability and
drought–flood hazard in the
transboundary Tano River Basin,
West Africa
Francis Quayson1,2*, Xiaoli Ding1,2 and George Owusu Amoah3
1Department of Land Surveying and Geospatial Science, The Hong Kong Polytechnic University, Hong
Kong, SAR, China, 2Research Institute for Land and Space, The Hong Kong Polytechnic University, Hong
Kong, SAR, China, 3Department of Geography and Regional Planning, University of Cape Coast, Cape
Coast, Ghana
Introduction: Understanding hydro-climatic variability is essential for effective
water-resource management in tropical river basins. This study assesses rainfall,
precipitation extremes, land surface temperature (LST), vegetation response, and
separate drought-condition and extreme-rainfall hazard components in the
transboundary Tano River Basin over 2000–2024.
Methods: CHIRPS rainfall, Landsat-derived NDVI and LST, Hamed–Rao modified
Mann–Kendall tests with Sen’s slope, Spearman rank correlations, entropy-
weighted hazard components, and Random Forest sensitivity diagnostics were
integrated with GPM IMERG and GMet validation.
Results: Recomputed ETCCDI consecutive dry days (CDD) average 31.93 days
(range 24.85–44.68 days). No significant monotonic trend occurs in rainfall (p =
0.726), Rx1day (p = 0.107), Rx5day (p = 0.315), CDD (p = 0.216), or LST (p = 0.129).
Because the Landsat ETM+/OLI NDVI series was not cross-sensor harmonized, no
pooled 2000–2024 NDVI trend is reported. Sensor-era diagnostics are retained
only to show that neither 2000–2012 (p = 0.057) nor 2013–2024 (p = 1.000)
exhibits a significant within-era trend. The Drought Condition Index (DCI) ranges
from 0.138 to 0.661 and the Extreme Rainfall Potential Index (ERPI) from 0.005 to
1.000. Leave-one-year-out Random Forest diagnostics yield R2 = 0.597 for DCI
and 0.783 for ERPI; these values are descriptive because the targets are
constructed from the predictors. Monthly CHIRPS–GPM validation yields R2 =
0.881, r = 0.939, RMSE = 27.96 mm, and MAE = 18.56 mm.
Discussion: Results emphasize interannual hydro-climatic variability; the non-
harmonized NDVI record is not used to infer a pooled long-term greening trend.
KEYWORDS
chirps, drought–flood hazard, extreme rainfall, hydro-climatic variability, land surface
temperature, NDVI, random forest
OPEN ACCESS
EDITED BY
Xiaotong Zhang,
Beijing Normal University, China
REVIEWED BY
Fatima Zohra Tebbi,
University of Batna 2, Algeria
Gopal Chowdhury,
University of Delhi, India
*CORRESPONDENCE
Francis Quayson,
francis.quayson@connect.polyu.hk
RECEIVED 19 August 2026
REVISED 17 October 2026
ACCEPTED 18 October 2026
PUBLISHED 17 November 2026
CITATION
Quayson F, Ding X and Amoah GO (2026)
Remote sensing and machine learning-
based assessment of hydro-climatic
variability and drought–flood hazard in
the transboundary Tano River Basin,
West Africa.
Front. Remote Sens. 7:1972697.
doi: 10.3389/frsen.2026.1972697
COPYRIGHT
© 2026 Quayson, Ding and Amoah. This is
an open-access article distributed under
the terms of the Creative Commons
Attribution License (CC BY). The use,
distribution or reproduction in other
forums is permitted, provided the original
author(s) and the copyright owner(s) are
credited and that the original publication
in this journal is cited, in accordance with
accepted academic practice. No use,
distribution or reproduction is permitted
which does not comply with these terms.
Frontiers in Remote Sensing frontiersin.org01
1 Introduction
Hydro-climatic variability has become a defining feature of
contemporary environmental change, with significant
implications for water resources, ecosystem stability, and socio-
economic resilience, particularly in climate-sensitive regions such as
sub-Saharan Africa (Intergovernmental Panel on Climate Change,
2021; World Bank Group, 2021). Increasing global temperatures
have intensified the hydrological cycle, leading to greater
atmospheric moisture capacity and consequently more extreme
precipitation events, alongside prolonged dry periods (Trenberth,
2011; Donat et al., 2016; Quayson et al., 2026b). These shifts have
resulted in a growing occurrence of compound hydro-climatic
extremes, where droughts and floods interact within the same
system, amplifying environmental risks and management
challenges (Zscheischler et al., 2020; Ridder et al., 2020).
West Africa represents a region of heightened vulnerability due
to its strong climatic gradients and high rainfall variability, largely
governed by the seasonal migration of the Intertropical
Convergence Zone (Nicholson, 2013; Amoah et al., 2026). While
historical records highlight prolonged drought conditions in the
Sahel during the late 20th century, recent decades have been
characterized by a partial recovery in rainfall accompanied by
increased variability and intensification of extreme events
(Nicholson et al., 2018; Panthou et al., 2018; Taylor et al., 2017;
Quayson et al., 2026a). Emerging evidence suggests that rainfall in
West Africa is becoming more erratic, with increasing frequency of
high-intensity rainfall events and intermittent dry spells, even in
regions where annual totals remain relatively stable (Bichet and
Diedhiou, 2018; Panthou et al., 2018). These dynamics significantly
influence hydrological processes such as runoff generation, soil
moisture availability, and evapotranspiration, thereby shaping
both flood and drought hazards (Trenberth, 2011; Donat
et al., 2016).
In Ghana, hydro-climatic variability is expressed through
spatially heterogeneous rainfall patterns and increasing
temperature trends, with important consequences for
agriculture, water supply, and ecosystem functioning (World
Bank Group, 2021; Sultan and Gaetani, 2016). The southern
regions, including the Tano River Basin, experience a bimodal
rainfall regime, yet recent studies indicate shifts in rainfall onset,
shortening of growing seasons, and rising land surface
temperatures (Panthou et al., 2018). These changes are further
exacerbated by land use and land cover transformations,
including deforestation and agricultural expansion, which alter
surface energy balance and hydrological responses (Serdeczny
et al., 2017; Sultan and Gaetani, 2016). As a result, the basin is
increasingly exposed to both water scarcity and flood hazards,
necessitating integrated assessment approaches.
Despite the growing body of literature on hydro-climatic
variability in West Africa, most existing studies adopt fragmented
approaches, focusing on individual components such as rainfall
trends, temperature changes, or vegetation dynamics in isolation
(Panthou et al., 2018; Quayson et al., 2026a). While these studies
provide valuable insights, they often fail to capture the complex
interactions between atmospheric processes, land surface
conditions, and hydrological responses. In particular, the
integration of rainfall extremes with land surface indicators such
as vegetation and temperature remains limited, despite their critical
role in modulating hydro-climatic processes. Furthermore, the
application of advanced analytical techniques, including machine
learning, to quantify the relative importance of hydro-climatic
drivers is still emerging in regional studies (Reichstein et al., 2019).
Recent advances in Earth Observation (EO) technologies
provide a powerful means of addressing these limitations by
enabling consistent, long-term monitoring of hydro-climatic
variables at multiple scales (Gorelick et al., 2017). Satellite-
derived datasets such as the Climate Hazards Group InfraRed
Precipitation with Station data (CHIRPS) have demonstrated
strong performance in representing rainfall variability across
Africa (Funk et al., 2015; Beck et al., 2019; Amoah et al., 2026;
Quayson et al., 2026a), while Landsat-based products allow for
the derivation of vegetation indices and land surface
temperature, offering insights into ecosystem dynamics and
surface energy balance (Ermida et al., 2020; Roy et al., 2014).
In addition, standardized extreme climate indices developed by
the Expert Team on Climate Change Detection and Indices
(ETCCDI), including maximum 1-day rainfall (Rx1day),
maximum 5-day rainfall (Rx5day), and consecutive dry days
(CDD), provide robust measures of rainfall intensity,
accumulation, and persistence (Zhang et al., 2011; Alexander
et al., 2006).
However, several key gaps remain. First, rainfall extremes,
vegetation, and thermal indicators are rarely evaluated together
within a consistent basin-scale Earth-observation framework.
Second, drought-related and extreme-rainfall conditions are often
collapsed into a single score even though they represent distinct
hydro-climatic processes. Third, satellite-rainfall validation is
frequently reported at an aggregation scale different from that of
the extreme indices, which can encourage overinterpretation of
daily-event skill (Satgé et al., 2020; Dinku et al., 2018; Huffman
et al., 2020).
This study addresses these gaps by developing a multi-
dimensional framework for hydro-climatic variability and
hazard in the Tano River Basin over 2000–2024. Rainfall,
Rx1day, Rx5day, CDD, NDVI, and LST are first evaluated
individually. Drought-related conditions are then summarized
with an entropy-weighted Drought Condition Index (DCI), while
rainfall and precipitation extremes are summarized separately
with an Extreme Rainfall Potential Index (ERPI). These indices
are descriptive hazard components, not disaster-risk measures,
because exposure and vulnerability are not represented. Random
Forest models are retained only as sensitivity diagnostics of the
two constructed indices; their feature importances are not
independent evidence of causal or physical drivers. Multi-
source validation combines GPM IMERG and ground
observations to assess CHIRPS rainfall estimates at the
evaluated aggregation scales.
The contribution of this study lies in integrating multiple hydro-
climatic indicators within one Earth-observation framework while
retaining the conceptual distinction between drought-related and
extreme-rainfall hazard components. The analysis evaluates
temporal trends, spatial co-variability, multi-source rainfall
validation, weighting sensitivity, and descriptive machine-learning
reconstruction without interpreting constructed-index feature
importance as causal attribution.
Frontiers in Remote Sensing frontiersin.org02
Quayson et al. 10.3389/frsen.2026.1972697
2 Methods
2.1 Study area
The Tano River Basin is a transboundary basin spanning
southwestern Ghana and eastern Côte d’Ivoire (Figure 1). The
spatial analyses use the full basin boundary rather than only the
Ghanaian portion. The basin lies within the humid tropical climatic
zone of West Africa and is influenced by the seasonal migration of
the Intertropical Convergence Zone (ITCZ), which governs rainfall
seasonality and spatial distribution (Nicholson, 2013). The climate is
characterized by a bimodal rainfall regime, with major and minor
rainy seasons typically occurring during April–July and September–
October, respectively; mean annual rainfall is approximately 1200–
1800 mm and temperatures are generally 26 °C–28 °C (IPCC, 2021).
The basin is ecologically significant, hosting dense forest
vegetation and supporting agricultural production, water supply
systems, and biodiversity conservation. However, increasing
anthropogenic pressures such as deforestation, agricultural
expansion, and settlement growth have altered land surface
characteristics, influencing energy balance and hydrological
processes. These factors, combined with climatic variability, make
the basin highly vulnerable to both drought and flood hazards,
thereby justifying its selection for integrated hydro-climatic analysis
(World Bank Group, 2021). Multi-source satellite and ground-based
datasets were used to capture hydro-climatic variability over the
period 2000–2024. Rainfall data were obtained from the Climate
Hazards Group InfraRed Precipitation with Station data (CHIRPS),
which provides quasi-global precipitation estimates at a spatial
resolution of approximately 0.05° (~5 km). CHIRPS integrates
satellite infrared observations with in-situ station data, making it
particularly suitable for regions with sparse gauge networks such as
West Africa (Funk et al., 2015; Beck et al., 2019).
Land-surface variables were derived from USGS Landsat
Collection 2 Tier 1 Level 2 products from Landsat 5 TM, Landsat
7 ETM+, Landsat 8 OLI/TIRS, and Landsat 9 OLI-2/TIRS-2. Surface
reflectance was scaled using the Collection 2 factor 0.0000275 and
offset −0.2; surface temperature was scaled using 0.00341802 and
offset 149.0 K and then converted to °C. Annual NDVI and LST were
calculated from cloud- and saturation-masked observations to
represent vegetation condition and surface thermal dynamics
(Roy et al., 2014; Ermida et al., 2020).
For validation purposes, rainfall data from the Global
Precipitation Measurement (GPM) Integrated Multi-satellite
Retrievals (IMERG) product were used, alongside ground-based
observations obtained from the Ghana Meteorological Agency
(GMet). The integration of satellite and station data provides a
robust framework for evaluating precipitation accuracy and
reliability (Huffman et al., 2020; Satgé et al., 2020; Dinku et al.,
2018). The datasets used in this study, including their sources, spatial
resolutions, temporal coverage, and respective applications, are
summarized in Table 1.
2.2 Data preprocessing
All datasets were processed within the Google Earth Engine
(GEE) platform, which enables efficient handling of large geospatial
datasets and reproducible analysis workflows (Gorelick et al., 2017).
The preprocessing stage involved spatial clipping of all datasets to
the Tano River Basin boundary, temporal filtering to extract the
FIGURE 1
Location of the Tano River Basin in West Africa, showing its position within Ghana and its transboundary extent into Côte d’Ivoire. The map illustrates
the basin boundary, major river network, elevation, and key settlements, providing the geographical and environmental context for hydro-
climatic analysis.
Frontiers in Remote Sensing frontiersin.org03
Quayson et al. 10.3389/frsen.2026.1972697
study period (2000–2024), and standardization of spatial
resolution (Figure 2).
Cloud masking was applied using the Collection 2 QA_
PIXEL flags for fill, dilated cloud, cirrus, cloud, cloud
shadow, and snow, together with QA_RADSAT to exclude
saturated pixels. For each year, the available Landsat 5/7/8/
9 observations were merged and annual median NDVI and
LST composites were generated at 30 m. Scene composition
changes through the record, including the introduction of
Landsat 8 OLI in 2013 and Landsat 9 in 2021. Because the
ETM+/OLI NDVI record was not cross-sensor harmonized with
an overlap-based calibration, the pooled 2000–2024 NDVI series
was not subjected to long-term trend inference. Sensor-era
analyses (2000–2012 and 2013–2024) are reported only as
diagnostics of within-era behavior. Rainfall was aggregated
from daily CHIRPS data to annual totals, while daily data
were retained for Rx1day, Rx5day, and CDD derivation.
2.3 Rainfall analysis
Annual rainfall was computed by summing daily precipitation
values for each year (Equation 1):
Rannual �
n
i�1
Ri (1)
where Ri represents daily rainfall (mm) and n is the number of days
in a given year. This approach captures total precipitation input and
provides a basis for assessing interannual variability (Funk et al.,
2015).
2.4 Extreme rainfall indices
(ETCCDI framework)
Extreme rainfall characteristics were quantified using indices
defined by the Expert Team on Climate Change Detection and
Indices (ETCCDI), which are widely used for assessing climate
extremes (Zhang et al., 2011; Alexander et al., 2006).
The maximum 1-day rainfall (Rx1day) was computed as
(Equation 2):
Rx1day� max Ri( ) (2)
The maximum 5-day rainfall (Rx5day) was calculated as
(Equation 3):
Rx5day� max
j+4
i�j
Ri⎛⎝ ⎞⎠ (3)
Consecutive dry days (CDD) were recomputed from the daily
CHIRPS series using the ETCCDI definition: for each year, CDD is
the maximum run length of consecutive days with rainfall <1 mm.
This corrects the earlier implementation that had counted all dry
days in a year rather than the longest uninterrupted dry spell
(Equation 4).
CDD� max di | Ri < 1 mm( ) (4)
FIGURE 2
Methodological framework for hydro-climatic hazard assessment in the Tano River Basin, integrating multi-source data, preprocessing, indicator
derivation, statistical and machine-learning analyses, validation, and final outputs. DCI and ERPI are treated as separate hydro-climatic hazard
components; the workflow does not estimate disaster risk because exposure and vulnerability are not included.
Frontiers in Remote Sensing frontiersin.org04
Quayson et al. 10.3389/frsen.2026.1972697
These indices provide complementary measures of rainfall
intensity, accumulation, and persistence, which are critical for
assessing both flood and drought hazards (Donat et al., 2016;
Zhang et al., 2011; Alexander et al., 2006).
2.5 Vegetation analysis (NDVI)
Vegetation condition was assessed using the Normalized
Difference Vegetation Index (NDVI), which is a widely used
indicator of vegetation health and productivity (Equation 5):
NDVI� NIR − REDNIR+ RED (5)
where and represent reflectance values in the near-infrared and red
spectral bands, respectively. NDVI values range from -1 to +1, with
higher values indicating denser and healthier vegetation (Tucker,
1979). NIRRED
2.6 Land surface temperature (LST)
Land Surface Temperature was derived from Landsat thermal
bands using radiometric calibration and conversion to temperature
units (Equation 6):
LST� DN × 0.00341802+ 149.0( ) − 273.15 (6)
where represents the digital number of the thermal band. The
resulting temperature is converted from Kelvin to Celsius. LST
provides insights into surface energy balance and thermal
conditions, which influence evapotranspiration and moisture
availability (Ermida et al., 2020). LST was derived using sensor-
specific calibration parameters following USGS guidelines DN
2.7 Trend analysis
Temporal trends in rainfall, Rx1day, Rx5day, CDD, and LST were
evaluated using ordinary least-squares (OLS) regression and the
Hamed–Rao variance-corrected modified Mann–Kendall test with
Sen’s slope. The modified test adjusts the Mann–Kendall variance
for significant serial correlation in the detrended ranked series, while
Sen’s slope provides a robust estimate of monotonic rate of change.
The original sample size (n = 25), effective sample size (n*), and the
ratio n/n* are reported in Table 2 to make the serial-correlation
adjustment explicit. For these five tested series, no significant
detrended rank-autocorrelation lags were retained, giving n* =
25 and n/n* = 1.000. OLS was retained for descriptive comparison.
Statistical significance was evaluated at α = 0.05; directional language is
used only where the trend test supports it (Equation 7).
Y� aX+ b (7)
For each tested annual series, the OLS slope represents the fitted
linear rate of change per year. Hamed–Rao modified Mann–Kendall
Z, Kendall’s tau, two-sided p-value, Sen’s slope, and n/n* are reported
in Table 2. Non-significant results are interpreted as interannual
variability rather than evidence of persistent directional change.
NDVI is excluded from pooled 2000–2024 trend inference
because the ETM+/OLI series is not cross-sensor harmonized.
Separate 2000–2012 and 2013–2024 NDVI trend diagnostics are
provided only to characterize within-era behavior and are not
combined into a single long-term greening estimate.
2.8 Correlation analysis
Relationships among annual hydro-climatic variables were
quantified using Spearman rank correlation (ρ). The rank-based
statistic was selected because the annual record contains only
25 observations and the variables are not assumed to follow a
common normal distribution. Two-sided p-values were
computed for each pair, and correlations are interpreted as
statistical co-variation rather than causation (Equation 8).
ρ� corr rank X( ), rank Y( )( ) (8)
rσ where ρ is Spearman’s rank correlation coefficient. Statistical
significance was evaluated using two-sided p-values at α = 0.05
(Wilks, 2019).
2.9 Drought and extreme-rainfall hazard
components
To avoid conflating drought and flood-related processes, two
normalized hazard components were recomputed separately. The
Drought Condition Index (DCI) combines rainfall deficit, CDD, LST,
and NDVI deficit; the Extreme Rainfall Potential Index (ERPI) combines
rainfall, Rx1day, and Rx5day. All components were min–max
normalized to [0,1]. Entropy weights were estimated from the 2000–
2024 annual series: DCI weights were 0.2614 for rainfall deficit,
0.3238 for CDD, 0.1598 for LST, and 0.2550 for NDVI deficit; ERPI
weights were 0.3028 for rainfall, 0.3980 for Rx1day, and 0.2992 for
Rx5day. Equal-weight alternatives were retained as a sensitivity check
(Equation 9).
DCI� 0.2614R def+ 0.3238CDD n+ 0.1598LST n
- 0.2550 1 − NDVI n( ); ERPI
� 0.3028R n+ 0.3980Rx1day n+ 0.2992Rx5day n (9)
Higher DCI values indicate stronger drought-related conditions,
whereas higher ERPI values indicate greater extreme-rainfall
potential. The two indices are interpreted separately and jointly
as descriptive hydro-climatic hazard conditions; they do not
represent disaster risk because exposure and vulnerability are not
included. Equal-weight sensitivity produced rank correlations of
0.972 for DCI and 0.996 for ERPI relative to entropy weighting, with
100% agreement among the top 3 years for both indices.
2.10 Machine learning (Random Forest
model)
Separate Random Forest regressions (Breiman, 2001) were
recomputed as descriptive sensitivity diagnostics for DCI and
ERPI. The DCI model used rainfall, CDD, LST, and NDVI as
predictors; the ERPI model used rainfall, Rx1day, and Rx5day.
Each model used 300 trees, bootstrap sampling, max_features =
1.0, min_samples_leaf = 1, and random_state = 42. With only
25 annual observations, generalization was assessed using leave-
one-year-out cross-validation (LOOCV), so every year served once
as the held-out observation.
Frontiers in Remote Sensing frontiersin.org05
Quayson et al. 10.3389/frsen.2026.1972697
Feature importance was obtained from the fitted Random Forest
impurity-based importance values and is reported only as a
decomposition of each mathematically constructed index. LOOCV
performance was summarized using R2, RMSE, and MAE. Because
the targets are constructed from the same variables used as
predictors, neither feature importance nor cross-validated
reconstruction is interpreted as independent evidence of physical
causation or as validation against observed impacts (Equation 10).
FIj � 1TTt�1Ij,t (10)
This design directly addresses the circularity of the original
single-index Random Forest analysis: the machine-learning results
are retained only to quantify how the components of DCI and ERPI
reproduce variation in their respective constructed indices.
2.11 Validation of rainfall data
The performance of CHIRPS rainfall data was evaluated using
statistical metrics against GPM IMERG and GMet observations.
Coefficient of determination (Equation 11):
R2 � 1 − yi − ˆyi 2
yi − y
ˉ
2 (11)
Root Mean Square Error (Equation 12):
RMSE�
�����������1
n yi − ˆyi 2
(12)
Mean Absolute Error (Equation 13):
MAE� 1n| yi − ˆyi | (13)
Bias (Equation 14):
Bias� 1n ˆyi − yi (14)
These metrics evaluate agreement in the temporal scale at which
the comparison was performed. The monthly/aggregated CHIRPS
comparisons with GPM IMERG and GMet do not constitute
validation of daily Rx1day or Rx5day extremes; this distinction is
maintained in the interpretation.
2.12 Methodological framework
The methodological framework integrates satellite data
acquisition, preprocessing, variable derivation, statistical
analysis, machine learning, and validation into a unified
workflow. This approach supports reproducible assessment of
hydro-climatic variability and hazard in the Tano River Basin
(Gorelick et al., 2017).
3 Results
3.1 Descriptive characteristics of hydro-
climatic variables
The statistical summary of hydro-climatic variables across
the Tano River Basin for 2000–2024 is presented in Table 3.
Annual rainfall averaged 1441.96 mm (SD = 174.74 mm),
ranging from 1144.48 to 1766.45 mm Rx1day averaged
59.66 mm and reached 100.28 mm, while Rx5day averaged
111.80 mm and reached 164.64 mm. These values indicate
substantial interannual variability and episodic high-
intensity rainfall events.
Recomputed ETCCDI consecutive dry days (CDD) averaged
31.93 days, with annual values ranging from 24.85 to 44.68 days.
This run-length metric is physically consistent with the
basin’s bimodal rainfall regime and replaces the erroneous
annual dry-day count reported in the original submission. In
the final per-image-NDVI annual series, NDVI averaged 0.671
(range 0.523–0.747), while LST averaged 29.49 °C (range 26.38 °C–
32.07 °C).
3.2 Temporal trends in hydro-climatic
variables
The temporal evolution of hydro-climatic variables is presented
in Figure 3 and the recomputed Hamed–Rao modified Mann–
Kendall statistics in Table 2. No significant trend occurs in
annual rainfall (Sen’s slope = +1.944 mm yr-1, p = 0.726), Rx1day
(−0.438 mm yr-1, p = 0.107), Rx5day (−0.430 mm yr-1, p = 0.315),
CDD (+0.141 days yr-1, p = 0.216), or LST (+0.0447 °C yr-1, p =
0.129). For each of these five series, n = 25, n* = 25, and n/n* = 1.000.
The pooled NDVI trend is not tested because the ETM+/OLI record
is not cross-sensor harmonized.
Accordingly, individual wet, dry, hot, cool, or high/low-
NDVI years are treated as episodic variability rather than
evidence of persistent long-term change unless supported by
an appropriate trend test. The recomputed CDD series shows
year-to-year dry-spell variability without a significant
monotonic change, and the earlier claim of statistically
significant LST warming is withdrawn.
TABLE 1 Summary of datasets used in the study, including source, spatial resolution, temporal coverage, and application.
Dataset Source Resolution Temporal coverage Purpose
CHIRPS rainfall Climate hazards group ~5 km 2000–2024 Rainfall analysis and extremes
Landsat C2 L2 USGS 30 m 2000–2024 NDVI and LST
GPM IMERG NASA ~10 km 2000–2024 Satellite validation
GMet data Ghana meteorological agency Station 2000–2024 Ground validation
Frontiers in Remote Sensing frontiersin.org06
Quayson et al. 10.3389/frsen.2026.1972697
3.3 Rainfall variability and extreme
rainfall dynamics
The interannual variability of rainfall and extreme rainfall
indices is illustrated in Figure 4a, while CDD dynamics are shown
in Figure 4b. Peak rainfall years such as 2007 and 2023 coincide
with elevated extreme-rainfall magnitudes: in 2007 R × 1 day
reached 100.28 mm and Rx5day 164.64 mm, while 2023 recorded
71.44 mm and 142.38 mm, respectively. Relatively low-rainfall
years such as 2013 and 2015 had lower extreme-rainfall
magnitudes. Recomputed CDD varies from 24.85 to 44.68 days,
with the maximum occurring in 2024. These observations
characterize interannual variability and are not interpreted as
significant long-term trends.
3.4 Vegetation–temperature interactions
and eco-climatic response
The coupled dynamics of vegetation condition and land surface
temperature are presented in Figure 5. Across 2000–2024, NDVI
ranges from 0.523 to 0.747 (mean = 0.671), while LST ranges from
26.38 °C to 32.07 °C (mean = 29.49 °C). The annual series show
appreciable interannual variability.
Because the Landsat ETM+/OLI NDVI record was not cross-
sensor harmonized, a pooled 2000–2024 NDVI trend is not reported.
As a diagnostic, the 2000–2012 sensor-era series has Sen’s slope =
+0.00521 yr-1 (p = 0.057), while the 2013–2024 series has Sen’s slope
≈ −0.00003 yr-1 (p = 1.000); neither within-era trend is statistically
significant. Segmented regression detects neither a significant
2013 level shift (p = 0.632) nor a significant post-2013 slope change
(p = 0.114). These diagnostics do not substitute for formal cross-sensor
harmonization and therefore are not interpreted as evidence of
persistent basin-wide ecological greening. LST shows no significant
monotonic trend (p = 0.129).
Spearman correlation between NDVI and LST is weakly positive
(ρ = 0.278, p = 0.179), indicating no statistically significant
monotonic association between annual basin-mean greenness and
surface temperature. Vegetation variability therefore cannot be
attributed to temperature alone and may also reflect rainfall,
phenology, land-cover change, management, and sensor/sampling
effects.
3.5 Interrelationships among hydro-
climatic variables
The recomputed Spearman correlation matrix (Figure 6a) shows
moderate positive co-variation between rainfall and Rx1day (ρ =
0.524, p = 0.007) and between rainfall and Rx5day (ρ = 0.467, p =
0.019). Rx1day and Rx5day are also positively associated (ρ = 0.551,
p = 0.004). Other pairwise relationships are weaker and are not
statistically significant at α = 0.05.
Rainfall has a weak negative association with CDD (ρ = -0.211),
while rainfall–NDVI (ρ = -0.008) and rainfall–LST (ρ = -0.233) are
negligible to weak. CDD is weakly related to NDVI (ρ = 0.102) and
LST (ρ = -0.070). These recomputed relationships replace the
stronger pre-revision correlations and do not support
deterministic statements that annual vegetation or dry-spell
variability is controlled by annual rainfall alone.
TABLE 3 Hamed–Rao modified Mann–Kendall trend analysis of hydro-climatic variables. Here n/n* is the ratio of the original sample size to the effective
sample size after serial-correlation adjustment; n/n* = 1.000 indicates no retained variance inflation. Pooled NDVI trend inference is omitted because the
ETM+/OLI series was not cross-sensor harmonized.
Variable n/
n*
Hamed–Rao modified
MK Z
Kendall τ p-value Sen slope yr-1 Interpretation
Rainfall 1.000 0.350 0.053 0.726 +1.944 mm No significant trend
Rx1day 1.000 −1.611 −0.233 0.107 −0.438 mm No significant trend
Rx5day 1.000 −1.004 −0.147 0.315 −0.430 mm No significant trend
CDD 1.000 1.238 0.180 0.216 +0.141 days No significant trend
NDVI — — — — — Pooled trend not tested; ETM+/OLI not
harmonized
LST 1.000 1.518 0.220 0.129 +0.04470 °C No significant trend
TABLE 2 Descriptive statistics of hydro-climatic variables in the Tano River Basin (2000–2024).
Variable Count Mean Std dev Min 25% Median 75% Max
Rainfall (mm) 25 1441.96 174.74 1144.48 1339.90 1388.44 1564.35 1766.45
Rx1day (mm) 25 59.66 12.79 42.44 51.65 56.21 62.98 100.28
Rx5day (mm) 25 111.80 18.04 83.57 101.64 107.47 115.93 164.64
CDD (days) 25 31.93 4.58 24.85 29.53 30.59 35.06 44.68
NDVI 25 0.671 0.047 0.523 0.655 0.684 0.695 0.747
LST (°C) 25 29.49 1.26 26.38 29.11 29.38 30.29 32.07
Frontiers in Remote Sensing frontiersin.org07
Quayson et al. 10.3389/frsen.2026.1972697
The recomputed Random Forest sensitivity diagnostics
(Figures 6b,c) are reported separately for the two hazard
components. For DCI, impurity-based importance is rainfall
deficit 0.594, CDD 0.293, NDVI deficit 0.086, and LST 0.028;
LOOCV gives R2 = 0.597, RMSE = 0.082, and MAE = 0.065. For
ERPI, importance is rainfall 0.407, Rx1day 0.422, and Rx5day
0.171; LOOCV gives R2 = 0.783, RMSE = 0.094, and MAE =
0.059. These values describe reconstruction of the constructed
indices and are not interpreted as independent physical-driver
evidence.
FIGURE 4
Hydro-climatic variability in the Tano River Basin: (a) rainfall, Rx1day, and Rx5day dynamics, and (b) consecutive dry days (CDD).
FIGURE 3
Annual hydro-climatic variability in the Tano River Basin (2000–2024): (a) rainfall, (b) Rx1day, (c) Rx5day, (d) ETCCDI consecutive dry days (CDD), (e)
NDVI, and (f) LST. Dashed Sen-slope lines are shown only for rainfall, Rx1day, Rx5day, CDD, and LST. The 2013 marker in the NDVI panel identifies the
ETM+/OLI transition; no pooled NDVI trend is fitted or tested because the series is not cross-sensor harmonized.
Frontiers in Remote Sensing frontiersin.org08
Quayson et al. 10.3389/frsen.2026.1972697
3.6 Drought and extreme-rainfall hazard
dynamics
The recomputed DCI and ERPI time series are shown in
Figures 7a,b. DCI ranges from 0.138 to 0.661 (mean = 0.425).
The highest DCI occurs in 2024 (0.661), followed by 2016 (0.648)
and 2004 (0.624), whereas the lowest occurs in 2023 (0.138).
ERPI ranges from 0.005 to 1.000 (mean = 0.368), with the highest
values in 2007 (1.000), 2023 (0.700), and 2002 (0.620), and the
lowest in 2013 (0.005). Keeping the components separate shows
that years characterized by stronger drought-related conditions
are not necessarily the years with the greatest extreme-
rainfall potential.
Figures 7c,d compares held-out LOOCV predictions with the
observed constructed indices. Cross-validated performance is R2 =
0.597 for DCI and R2 = 0.783 for ERPI, with RMSE values of
0.082 and 0.094, respectively. These statistics quantify descriptive
reconstruction of the indices rather than independent predictive
validation, because each target is mathematically derived from its
predictors.
3.7 Validation of satellite rainfall data
The reliability of CHIRPS rainfall estimates was rigorously
evaluated through comparison with both satellite-based (GPM
IMERG) and ground-based (GMet station) datasets, as presented
in Figures 8, 9. This multi-source validation framework provides a
comprehensive assessment of the accuracy, consistency, and
applicability of satellite-derived precipitation data for hydro-
climatic analysis in the Tano River Basin.
The revised monthly CHIRPS–GPM IMERG comparison uses
288 paired months and shows strong positive co-variation (R2 =
0.881; Pearson r = 0.939). This supports CHIRPS for representing
basin-scale monthly rainfall variability at the evaluated
aggregation, while leaving daily extreme-event performance
unresolved.
FIGURE 6
(a) Recomputed Spearman rank-correlation matrix among rainfall, Rx1day, Rx5day, CDD, NDVI, and LST. (b,c) recomputed Random Forest feature
importance for DCI and ERPI, respectively. Importance values are descriptive decompositions of constructed indices, not independent driver attribution.
FIGURE 5
Temporal interaction between NDVI and LST in the Tano River Basin from 2000 to 2024. The 2013 vertical marker identifies the ETM+/OLI transition.
The annual NDVI–LST association is weak and non-significant (Spearman ρ = 0.278, p = 0.179).
Frontiers in Remote Sensing frontiersin.org09
Quayson et al. 10.3389/frsen.2026.1972697
The corresponding monthly error metrics are RMSE =
27.96 mm, MAE = 18.56 mm and mean bias = +6.66 mm under
the reported comparison convention. These errors are explicitly
interpreted at monthly scale. They do not constitute direct
validation of daily Rx1day, Rx5day or CDD, and no negative
precipitation observations are used in the analysis.
Seasonal validation against GMet station data (Figure 9a)
further confirms the reliability of CHIRPS in capturing the
characteristic bimodal rainfall regime of the Tano River Basin.
The primary rainfall peak occurs in June (approximately
225 mm), while a secondary peak is observed between
September and October (approximately 190–210 mm). CHIRPS
successfully reproduces this seasonal pattern with high fidelity,
demonstrating its capability to capture both the timing and
magnitude of rainfall events. Minor overestimations during
peak months, particularly in June and October, are observed
and can be attributed to the spatial averaging inherent in
satellite-based measurements compared to localized point
observations from rain gauge stations. The CHIRPS–GMet
scatterplot comparison (Figure 9b) further supports this
observation, showing a strong correlation (R2 ≈ 0.845) between
satellite-derived and ground-based rainfall estimates. While this
relationship confirms the overall consistency between the
FIGURE 7
(a,b) Recomputed temporal evolution of the drought condition index (DCI) and extreme rainfall potential index (ERPI) for 2000–2024. (c,d) Leave-
one-year-out observed-versus-predicted diagnostics for DCI and ERPI. Cross-validated fit is interpreted only as descriptive reconstruction of the
constructed indices.
FIGURE 8
Monthly CHIRPS–GPM IMERG validation at the evaluated
aggregation scale (288 paired months): R2 = 0.881, Pearson r = 0.939,
RMSE = 27.96 mm, MAE = 18.56 mm and mean bias = +6.66 mm. The
monthly comparison does not directly validate daily Rx1day,
Rx5day or CDD extremes.
Frontiers in Remote Sensing frontiersin.org10
Quayson et al. 10.3389/frsen.2026.1972697
datasets, slight deviations are evident, particularly at lower rainfall
values, where CHIRPS tends to marginally overestimate
precipitation. These discrepancies are likely influenced by
localized rainfall variability, uneven spatial distribution of
meteorological stations, and inherent uncertainties in satellite
retrieval algorithms.
Overall, the validation results support the use of CHIRPS for
basin-scale rainfall variability at the evaluated temporal aggregation.
They should not be interpreted as direct validation of daily extreme
indices (Rx1day and Rx5day), because the validation comparison is
not performed at the daily-extreme scale. Differences in spatial
support between gridded satellite products and point-based GMet
observations also contribute to uncertainty.
3.8 Spatial patterns of rainfall and
extreme rainfall
The spatial distribution of rainfall and extreme rainfall
indices across the Tano River Basin is presented in Figure 10
for four benchmark years (2001, 2010, 2020 and 2023), selected
to span the beginning, middle and recent part of the record and
to illustrate contrasting hydro-climatic conditions rather than
to imply a formal event sample. The results reveal a pronounced
and persistent south–north hydro-climatic gradient that defines
rainfall distribution and associated flood hazard within the
basin. Annual rainfall patterns (Figures 10a–d ) show
consistently higher rainfall concentrations in the southern
and southwestern regions, with values ranging between
approximately 1700 and 1800 mm. In contrast, the northern
portions of the basin record lower rainfall amounts, typically
between 1100 and 1300 mm. This spatial gradient remains
stable across all selected years, indicating the dominant
influence of large-scale atmospheric processes, particularly
moisture transport from the Gulf of Guinea and the
latitudinal positioning of the Intertropical Convergence
Zone (ITCZ).
The spatial distribution of extreme rainfall further shows
localized variability superimposed on the broader rainfall
gradient. Rx1day and Rx5day hotspots vary among the
selected benchmark years, including locally elevated values
in the southern basin. Because the temporal trend tests for
Rx1day and Rx5day are non-significant, the benchmark maps
are interpreted as examples of spatial and interannual contrast
rather than evidence of progressive intensification
through time.
3.9 Spatial patterns of dryness and
vegetation
The spatial distribution of dryness and vegetation condition is
presented in Figure 11 using the fully recomputed analytical
rasters for 2001, 2010, 2020, and 2023. CDD represents the
maximum annual run of consecutive days with CHIRPS
rainfall <1 mm, not total annual dry-day counts. The maps
therefore show the spatial structure of annual dry-spell
persistence for the selected benchmark years and are
interpreted as spatial/interannual contrasts rather than a
temporal trend.
NDVI maps (Figures 11e–h), recomputed from quality-
screened Landsat Collection 2 Level 2 observations, show
heterogeneous vegetation conditions across the basin and
among benchmark years. White areas in the early-year panels
denote pixels without valid observations after quality screening;
no spatial interpolation or artificial gap filling was applied. Spatial
correspondence with rainfall and CDD is treated descriptively
because the annual Spearman correlations are weak. Because the
ETM+/OLI NDVI record is not cross-sensor harmonized, the
benchmark maps are not used to infer a pooled long-term
greening trend.
3.10 Spatial patterns of land surface
temperature
The spatial distribution of LST is shown in Figure 12 using
quality-screened Landsat Collection 2 Level 2 annual median
composites for 2001, 2010, 2020, and 2023. White areas in the
early-year panels denote pixels without valid observations after
quality screening; no spatial interpolation or artificial gap filling
was applied. The maps reveal spatial and interannual thermal
variability across the basin. Because the recomputed 2000–
FIGURE 9
Comparison of CHIRPS rainfall estimates with ghana meteorological agency (GMet) observations: (a) scatterplot showing the relationship between
CHIRPS and GMet rainfall estimates; and (b) monthly rainfall distributions showing the seasonal variation in both datasets.
Frontiers in Remote Sensing frontiersin.org11
Quayson et al. 10.3389/frsen.2026.1972697
2024 temporal test shows no significant basin-wide LST trend, the
benchmark maps are not interpreted as evidence of progressive
long-term warming.
4 Discussion
4.1 Hydro-climatic variability and
rainfall dynamics
As shown in Figures 3, 4, rainfall in the Tano River Basin is
characterized by pronounced interannual variability, yet lacks
a statistically significant long-term trend. This pattern
indicates that rainfall dynamics are primarily governed by
short-term climatic oscillations and large-scale atmospheric
controls rather than persistent directional change. Such
variability is consistent with tropical precipitation systems
influenced by the seasonal migration of the Intertropical
Convergence Zone (ITCZ) and regional ocean–atmosphere
interactions (Nicholson, 2013 ; Sultan and Gaetani, 2016 ;
IPCC, 2021 ).
Rx1day and Rx5day also show no statistically significant
monotonic trend over 2000–2024 (Mann–Kendall p =
0.107 and 0.315, respectively). Their year-to-year fluctuations
therefore indicate episodic variability in rainfall intensity and
accumulation rather than a demonstrated directional shift. This
distinction is important because individual high-impact rainfall
years can occur in a series that lacks a significant long-term
trend.
4.2 Land surface temperature and land-
surface processes
The recomputed LST series does not support a statistically
significant long-term warming trend over 2000–2024 (Hamed–
Rao modified Mann–Kendall Sen’s slope = +0.0447 °C yr-1, p =
0.129). We therefore withdraw the original claim of significant
thermal intensification. LST remains hydro-climatically relevant
because surface temperature modulates evapotranspiration, soil-
moisture demand and land–atmosphere exchanges, but the
present annual record supports interannual thermal variability
rather than a robust directional trend.
FIGURE 10
Spatial distribution of rainfall and extreme rainfall indices in the Tano River Basin for selected years (2001, 2010, 2020, and 2023): (a–d) annual rainfall,
(e–h) maximum 1-day rainfall (Rx1day), and (i–l) maximum 5-day rainfall (Rx5day).
Frontiers in Remote Sensing frontiersin.org12
Quayson et al. 10.3389/frsen.2026.1972697
4.3 Vegetation resilience under climate
variability
The Landsat NDVI record spans the ETM+/OLI sensor
transition and was not cross-sensor harmonized with an overlap-
based calibration. Consequently, this revision does not report or
interpret a pooled 2000–2024 NDVI trend. Within-era diagnostics
show a positive but non-significant 2000–2012 Sen slope
(+0.00521 yr-1, p = 0.057) and essentially no 2013–2024 trend
(−0.00003 yr-1, p = 1.000). Segmented regression likewise detects
neither a significant 2013 level shift nor a significant post-2013 slope
change. These results support a conservative interpretation of
interannual vegetation variability and do not establish persistent
ecological greening.
4.4 Interpretation of the hazard
components and random forest diagnostics
The recomputed Spearman analysis identifies statistically
significant positive co-variation among the rainfall and
extreme-rainfall measures, particularly rainfall–Rx1day,
rainfall–Rx5day, and Rx1day–Rx5day, while associations
involving CDD, NDVI, and LST are weaker. The Random
Forest analysis is now separated into DCI and ERPI
diagnostics. Because each target is constructed from its
predictors, feature importance is circular with respect to causal
inference; it describes the internal statistical composition of the
indices rather than independent hydro-climatic drivers.
4.5 Spatial hydro-climatic differentiation
and management implications
Spatial patterns in Figures 10–12 show persistent geographic
heterogeneity across the transboundary basin, including wetter
sectors toward the south and southwest and local contrasts in dry-
spell persistence, vegetation, and LST. These spatial patterns
should not be conflated with temporal trends: the 2000–
2024 tests are non-significant for rainfall, Rx1day, Rx5day,
CDD, and LST. From a management perspective, relatively
drier sectors warrant attention to seasonal water availability,
agricultural drought preparedness, and storage planning,
whereas wetter sectors require rainfall monitoring, drainage
maintenance, flood-warning systems, and floodplain
management. Coordinated Ghana–Côte d’Ivoire hydro-climatic
monitoring and data exchange would improve basin-wide
preparedness.
4.6 Reliability of satellite-based rainfall data
At the evaluated monthly aggregation scale, the revised
CHIRPS–GPM comparison (288 paired months) gives R2 =
0.881, Pearson r = 0.939, RMSE = 27.96 mm, MAE = 18.56 mm
and mean bias = +6.66 mm for the reported comparison
convention. These statistics indicate strong monthly co-
variation but do not validate daily Rx1day, Rx5day or CDD.
GMet comparison is likewise interpreted at its evaluated
aggregation and sampling support. Negative values
FIGURE 11
Spatial patterns of (a–d) ETCCDI consecutive dry days (CDD) and (e–h) normalized difference vegetation index (NDVI) in the Tano River Basin for
2001, 2010, 2020, and 2023. CDD represents the maximum annual run of consecutive days with CHIRPS rainfall <1 mm. NDVI was derived from quality-
screened Landsat Collection 2 Level 2 observations. White areas denote pixels without valid observations after quality screening; no spatial interpolation or
artificial gap filling was applied.
Frontiers in Remote Sensing frontiersin.org13
Quayson et al. 10.3389/frsen.2026.1972697
implied by a fitted regression-line extension are not
physical precipitation observations and are not used in the
analysis.
4.7 Limitations and future research
Several limitations should be acknowledged. First, satellite
products differ in spatial resolution, sampling, and algorithmic
uncertainty. Second, annual aggregation can mask sub-seasonal
and event-scale flood processes; the present results therefore
characterize interannual hydro-climatic hazard rather than
individual flood events. Third, although CDD is recomputed
as the ETCCDI longest dry-spell run, daily extreme indices are
not independently validated at daily gauge scale. Fourth, DCI
and ERPI are constructed hazard components, and their
Random Forest feature importances cannot establish
independent drivers or causality. Fifth, Landsat sensor
composition and scene availability vary through the record.
Because the NDVI record was not cross-sensor harmonized,
pooled NDVI trend inference was removed; the reported sensor-
era analyses are diagnostic only. A future overlap-based ETM+/
OLI harmonization would be required for a defensible pooled
long-term NDVI trend. Finally, monthly rainfall validation
cannot be generalized directly to daily extremes. Future work
should incorporate event-scale hydrology, exposure and
vulnerability, sensor-harmonized land-surface analyses, daily
gauge validation, and independent impact data for predictive
or causal modelling.
5 Conclusion
This study assesses hydro-climatic variability and separate
drought-condition and extreme-rainfall hazard components in
the transboundary Tano River Basin using multi-source
Earth observation and statistical analysis. Recomputed
Hamed–Rao modified Mann–Kendall/Sen’s-slope tests show
no significant monotonic trend in rainfall, Rx1day, Rx5day,
CDD, or LST over 2000–2024. For all five tested series, n = 25,
n* = 25, and n/n* = 1.000. Because the Landsat ETM+/OLI
NDVI record was not cross-sensor harmonized, the pooled
NDVI trend is not reported. Neither the 2000–2012 nor 2013–
2024 within-era NDVI diagnostic is statistically significant.
DCI and ERPI are interpreted as hazard components rather
than disaster risk because exposure and vulnerability are
not included.
The recomputed Random Forest diagnostics are reported
separately for DCI and ERPI. DCI reconstruction gives LOOCV
R2 = 0.597, with rainfall deficit and CDD receiving the largest
within-index importance values; ERPI gives LOOCV R2 = 0.783,
with Rx1day and rainfall receiving the largest values. Because the
targets are constructed from these predictors, the results are
descriptive index decompositions rather than independent driver
attribution. Spatial analysis nevertheless demonstrates substantial
hydro-climatic heterogeneity across the transboundary basin,
supporting geographically differentiated monitoring and
adaptation.
The revised CHIRPS–GPM comparison shows strong
monthly co-variation (R2 = 0.881; r = 0.939; 288 paired
months), with RMSE = 27.96 mm and MAE = 18.56 mm. This
evidence supports use of CHIRPS for basin-scale monthly
variability but does not directly validate daily extreme indices;
spatial-resolution and station-versus-grid differences
remain important.
Overall, the study demonstrates the value of combining
rainfall, thermal, and vegetation observations while
maintaining conceptual separation between drought-related
and extreme-rainfall hazards. The revision corrects CDD,
reports the Hamed–Rao effective-sample-size diagnostic (n/n*),
removes pooled NDVI trend inference in the absence of ETM+/
OLI harmonization, recomputes the correlations, DCI/ERPI
indices, Random Forest diagnostics, and affected figures from
the corrected inputs, withdraws unsupported LST-warming and
causal-driver claims, and separates monthly rainfall validation
from daily-extreme validation. Future work should incorporate
event-scale hydrology, exposure and vulnerability, sensor-
harmonized land-surface analyses, daily gauge validation, and
independent impact data.
FIGURE 12
Spatial distribution of annual median land surface temperature (LST) in the Tano River Basin for 2001, 2010, 2020, and 2023, derived from quality-
screened landsat collection 2 Level 2 surface-temperature observations. White areas denote pixels without valid observations after quality screening; no
spatial interpolation or artificial gap filling was applied. The maps illustrate spatial and interannual thermal variability and are not interpreted as evidence of
progressive warming because the 2000–2024 basin-scale LST trend is non-significant.
Frontiers in Remote Sensing frontiersin.org14
Quayson et al. 10.3389/frsen.2026.1972697
Data availability statement
The datasets supporting the findings of this study are available in
the Zenodo repository (Quayson et al., 2026) at https://doi.org/10.
5281/zenodo.23138334.
Author contributions
FQ: Conceptualization, Data curation, Formal Analysis,
Investigation, Methodology, Project administration, Software,
Validation, Visualization, Writing – original draft, Writing – review
and editing. XD: Conceptualization, Data curation, Formal Analysis,
Funding acquisition, Investigation, Methodology, Project
administration, Resources, Software, Supervision, Validation,
Visualization, Writing – original draft, Writing – review and
editing. GA: Conceptualization, Data curation, Formal Analysis,
Investigation, Methodology, Software, Validation, Visualization,
Writing – original draft, Writing – review and editing.
Funding
The author(s) declared that financial support was received
for this work and/or its publication. This research was supported
by the Research Grants Council (RGC) of the Hong Kong Special
Administrative Region (PolyU 152318/22E and 152344/23E),
the National Natural Science Foundation of China
(42330717), the Innovation and Technology Commission
(ITC) through K-BBY1 – Smart Railway Technology, and the
Guangdong–Hong Kong Joint Laboratory for Marine
Infrastructure.
Conflict of interest
The author(s) declared that this work was conducted in the
absence of any commercial or financial relationships that could be
construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that generative AI was not used in the
creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this
article has been generated by Frontiers with the support of artificial
intelligence and reasonable efforts have been made to ensure
accuracy, including review by the authors wherever possible. If
you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors
and do not necessarily represent those of their affiliated
organizations, or those of the publisher, the editors and the
reviewers. Any product that may be evaluated in this article, or
claim that may be made by its manufacturer, is not guaranteed or
endorsed by the publisher.
References
Alexander, L. V., Zhang, X., Peterson, T. C., Caesar, J., Gleason, B., Klein Tank, A. M. G.,
et al. (2006). Global observed changes in daily climate extremes of temperature and
precipitation. J. Geophys. Res. Atmos. 111, D05109. doi: 10.1029/2005JD006290
Amoah, G. O., Olawoyin, R., and Quayson, F. (2026). Hydro-climatic ecosystem risk
mapping and spatial propagation analysis in the upper niger basin, West Africa. Environ.
Challenges 24, 101567. doi: 10.1016/J.ENVC.2026.101567
Beck, H. E., Wood, E. F., Pan, M., Fisher, C. K., Miralles, D. G., Van Dijk, A. I. J. M., et al.
(2019). MSWEP V2 global 3-hourly precipitation: methodology and quantitative
assessment. Bulletin of the American Meteorological Society 100, 473–500. doi: 10.
1175/BAMS-D-17-0138.1
Bichet, A., and Diedhiou, A. (2018). West African Sahel has become wetter during the
last 30 years, but dry spells are shorter and more frequent. Clim. Res. 75 (2), 155–162.
doi: 10.3354/cr01502
Breiman, L. (2001). Random forests. Mach. Learn. 45 (1), 5–32. doi: 10.1023/A:
1010933404324
Dinku, T., Funk, C., Peterson, P., Maidment, R., Tadesse, T., Gadain, H., et al. (2018).
Validation of the CHIRPS satellite rainfall estimates over eastern Africa. Q. J. R.
Meteorological Soc. 144 (S1), 292–312. doi: 10.1002/qj.3244
Donat, M. G., Lowry, A. L., Alexander, L. V., O’Gorman, P. A., and Maher, N. (2016).
More extreme precipitation in the world’s dry and wet regions. Nat. Clim. Change 6,
508–513. doi: 10.1038/nclimate2941
Ermida, S. L., Soares, P., Mantas, V., Göttsche, F. M., and Trigo, I. F. (2020). Google
Earth engine open-source code for land surface temperature estimation from the landsat
series. Remote Sens. 12 (9), 1471. doi: 10.3390/rs12091471
Funk, C., Peterson, P., Landsfeld, M., Pedreros, D., Verdin, J., Shukla, S., et al.
(2015). The climate hazards infrared precipitation with stations—A new
environmental record for monitoring extremes. Sci. Data 2, 150066. doi: 10.
1038/sdata.2015.66
Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., and Moore, R. (2017).
Google Earth engine: planetary-scale geospatial analysis for everyone. Remote Sens.
Environ. 202, 18–27. doi: 10.1016/j.rse.2017.06.031
Huffman, G. J., Stocker, E. F., Bolvin, D. T., Nelkin, E. J., and Tan, J. (2020). GPM
IMERG final precipitation data. NASA Precip. Process. Syst. doi: 10.5067/GPM/IMERG/
3B-MONTH/06
Intergovernmental Panel on Climate Change (IPCC) (2021). “Climate change 2021:
the physical science basis,”. Cambridge University Press. doi: 10.1017/
9781009157896
Nicholson, S. E. (2013). The west African Sahel: a review of recent studies on rainfall
regime and variability. ISRN Meteorol. 2013, 1–32. doi: 10.1155/2013/453521
Nicholson, S. E., Funk, C., and Fink, A. H. (2018). Rainfall over the African continent
from the 19th through the 21st century. Glob. Planet. Change 165, 114–127. doi: 10.1016/
j.gloplacha.2017.12.014
Panthou, G., Vischel, T., and Lebel, T. (2018). Recent trends in the regime of
extreme rainfall in the Sahel. Environ. Res. Lett. 13 (6), 064013. doi: 10.1088/1748-
9326/aac334
Quayson, F., Ding, X., and Amoah, G. O. (2026). Remote sensing and machine learning-
based assessment of hydro-climatic variability and drought–flood hazard in the
transboundary tano river basin, West Africa dataset v3. Zenodo. doi: 10.5281/
zenodo.23138334
Quayson, F., Ding, X., and Dadson, I. Y. (2026a). Satellite-based assessment of hydro-
climatic extremes and integrated risk in the niger river basin. Environ. Challenges 24,
101584. doi: 10.1016/j.envc.2026.101584
Quayson, F., Ding, X., and Dadson, I. Y. (2026b). Spatiotemporal evaluation and
uncertainty assessment of multi-source evapotranspiration products across Hong
Kong: implications for mai Po wetland water-balance monitoring. Water 18 (15),
1897. doi: 10.3390/w18151897
Reichstein, M., Camps-Valls, G., Stevens, B., Jung, M., Denzler, J., Carvalhais, N., et al.
(2019). Deep learning and process understanding for data-driven Earth system science.
Nature 566, 195–204. doi: 10.1038/s41586-019-0912-1
Ridder, N. N., Pitman, A. J., Westra, S., Ukkola, A. M., Do, H. X., Bador, M., et al. (2020).
Global hotspots for the occurrence of compound events. Nat. Commun. 11, 5956. doi:
10.1038/s41467-020-19639-3
Frontiers in Remote Sensing frontiersin.org15
Quayson et al. 10.3389/frsen.2026.1972697
Roy, D. P., Wulder, M., Loveland, T., C.E., W., Allen, R., Anderson, M., et al. (2014).
Landsat-8: science and product vision for terrestrial global change research. Remote
Sens. Environ. 145, 154–172. doi: 10.1016/j.rse.2014.02.001
Satgé, F., Defrance, D., Sultan, B., Bonnet, M.-P., Seyler, F., Rouché, N., et al. (2020).
Evaluation of 23 gridded precipitation datasets across West Africa. J. Hydrology 581,
124412. doi: 10.1016/j.jhydrol.2019.124412
Serdeczny, O., Adams, S., Baarsch, F., Coumou, D., Robinson, A., Hare, W., et al.
(2017). Climate change impacts in Sub-Saharan Africa: from physical changes to
their social repercussions. Reg. Environ. Change 17, 1585–1600. doi: 10.1007/s10113-
015-0910-2
Sultan, B., and Gaetani, M. (2016). Agriculture in West Africa in the twenty-first century:
climate change and impacts scenarios. Environ. Res. Lett. 11, 034008. doi: 10.1088/1748-
9326/11/3/034008
Taylor, C. M., Belušić, D., Guichard, F., Parker, D. J., Vischel, T., Bock, O., et al. (2017).
Frequency of extreme Sahelian storms tripled since 1982. Nature 544, 475–478. doi: 10.
1038/nature22069
Trenberth, K. E. (2011). Changes in precipitation with climate change. Clim. Res. 47,
123–138. doi: 10.3354/cr00953
Tucker, C. J. (1979). Red and photographic infrared linear combinations for monitoring
vegetation. Remote Sens. Environ. 8 (2), 127–150. doi: 10.1016/0034-4257(79)90013-0
Wilks, D. S. (2019). “Statistical methods in the atmospheric sciences,”. 4th ed. Academic
Press. doi: 10.1016/C2017-0-03921-6
World Bank Group (2021). “Climate Risk Country Profile: Ghana. Washington, DC,
USA: World Bank Group. Available online at: https://climateknowledgeportal.
worldbank.org/sites/default/files/country-profiles/15857WB_Ghana Country%
20Profile-WEB.pdf.
Zhang, X., Alexander, L., Hegerl, G. C., Jones, P., Tank, A. K., Peterson, T. C., et al.
(2011). Indices for monitoring changes in extremes based on daily temperature and
precipitation data. WIREs Clim. Change 2, 851–870. doi: 10.1002/wcc.147
Zscheischler, J., Westra, S., Van Den Hurk, B. J. J. M., Seneviratne, S. I., Ward, P. J.,
Pitman, A., et al. (2020). A typology of compound weather and climate events. Nat. Rev.
Earth and Environ. 1, 333–347. doi: 10.1038/s43017-020-0060-z
Frontiers in Remote Sensing frontiersin.org16
Quayson et al. 10.3389/frsen.2026.1972697
论文原文与出版信息
Frontiers in Remote Sensing