Chronic scarcity of continuous in situ sampling severely hampers trophic monitoring in tropical reservoirs, restricting the application of data-intensive models in regulatory environments. Therefore, the aim of this study was to develop a cross-validation framework integrating a Mamdani Fuzzy Inference System (FIS) with Sentinel-2 Normalized Difference Chlorophyll Index (NDCI) retrievals via Google Earth Engine (GEE) to overcome these limitations. The methods involved parameterizing the FIS with subtropical thresholds using a 16-year limnological dataset from the Billings Reservoir (Brazil). This system was then cross-validated against satellite-derived NDCI distributions using non-parametric statistical tests. Subsequently, it was applied to the Funil Reservoir to evaluate spatial transferability based exclusively on NDCI mapping, without concurrent in situ validation. The results demonstrated a statistically significant validation at the Billings Reservoir (Kruskal-Wallis, <InlineEquation ID="IEq1"> <EquationSource Format="TEX">$p < 0.05$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>p</mi> <mo><</mo> <mn>0.05</mn> </mrow> </math></EquationSource> </InlineEquation>), confirming that NDCI medians vary systematically across fuzzy-classified trophic states. Additionally, Levene’s test revealed homogeneous variance for total phosphorus (TP) (<InlineEquation ID="IEq2"> <EquationSource Format="TEX">$W = 3.09$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>W</mi> <mo>=</mo> <mn>3.09</mn> </mrow> </math></EquationSource> </InlineEquation>, <InlineEquation ID="IEq3"> <EquationSource Format="TEX">$p = 0.051$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>p</mi> <mo>=</mo> <mn>0.051</mn> </mrow> </math></EquationSource> </InlineEquation>) contrasting with heteroscedastic chlorophyll-a (Chl-a) (<InlineEquation ID="IEq4"> <EquationSource Format="TEX">$p < 0.05$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>p</mi> <mo><</mo> <mn>0.05</mn> </mrow> </math></EquationSource> </InlineEquation>), characterizing the distinct dispersion regimes of eutrophication drivers and biological responses. Furthermore, the NDCI-based application to the Funil Reservoir yielded synoptic maps that indicated potential algal bloom expansion at resolutions unattainable by conventional monitoring. In conclusion, this approach establishes remote sensing-based surveillance via NDCI, initially calibrated by fuzzy logic, as a valuable screening tool for eutrophication monitoring in data-scarce aquatic systems.

Addressing field data scarcity in algal bloom surveillance: integrating fuzzy inference and orbital remote sensing in Brazilian reservoirs

期刊论文
出版模式：

开放获取
发表日期：

2026年7月30日

查看全文

Ebenézer de Paula Carvalho,
Nilo Antonio de Souza Sampaio,
Hugo Pimentel Tavares,
Carin von Mühlen

摘要

Chronic scarcity of continuous in situ sampling severely hampers trophic monitoring in tropical reservoirs, restricting the application of data-intensive models in regulatory environments. Therefore, the aim of this study was to develop a cross-validation framework integrating a Mamdani Fuzzy Inference System (FIS) with Sentinel-2 Normalized Difference Chlorophyll Index (NDCI) retrievals via Google Earth Engine (GEE) to overcome these limitations. The methods involved parameterizing the FIS with subtropical thresholds using a 16-year limnological dataset from the Billings Reservoir (Brazil). This system was then cross-validated against satellite-derived NDCI distributions using non-parametric statistical tests. Subsequently, it was applied to the Funil Reservoir to evaluate spatial transferability based exclusively on NDCI mapping, without concurrent in situ validation. The results demonstrated a statistically significant validation at the Billings Reservoir (Kruskal-Wallis, $p < 0.05$ $p < 0.05$ ), confirming that NDCI medians vary systematically across fuzzy-classified trophic states. Additionally, Levene’s test revealed homogeneous variance for total phosphorus (TP) ( $W = 3.09$ $W = 3.09$ , $p = 0.051$ $p = 0.051$ ) contrasting with heteroscedastic chlorophyll-a (Chl-a) ( $p < 0.05$ $p < 0.05$ ), characterizing the distinct dispersion regimes of eutrophication drivers and biological responses. Furthermore, the NDCI-based application to the Funil Reservoir yielded synoptic maps that indicated potential algal bloom expansion at resolutions unattainable by conventional monitoring. In conclusion, this approach establishes remote sensing-based surveillance via NDCI, initially calibrated by fuzzy logic, as a valuable screening tool for eutrophication monitoring in data-scarce aquatic systems.