An Interpretable, WFD‐Aligned Fuzzy Inference Framework for Phytoplankton‐Based Ecological Status Assessment Across Three Hydroecologically Contrasting Mediterranean Lakes (Menderes Basin, Türkiye)

ABSTRACT

Three lakes in the Menderes Basin of western Türkiye carry sharply different water regimes: brackish, lagoonal Bafa, hypertrophic in phosphorus but not in algal biomass; turbid, alkaline, endorheic Yayla; and spring‐fed karstic Işıklı, close to reference condition. No basin‐wide assessment of their phytoplankton and water chemistry has been published. We assessed 16 station‐seasons with a three‐layer fuzzy inference system aligned to the Water Framework Directive. It scores the biological and the physicochemical component separately, then combines them by the Directive's one‐out‐all‐out rule, which keeps the limiting element visible. Ordination of 97 taxa separated the three lakes along a gradient set by conductivity, temperature, phosphorus, suspended solids and transparency, the constrained axes carrying 59.4% of species inertia. With only three lakes, that separation is descriptive and cannot be tested. The phytoplankton index MPTI tracked every major stressor (phosphorus ρ = 0.976, suspended solids ρ = 0.859), whereas the alternative Phyto‐Tr index responded to phosphorus alone. Cross‐validated agreement with the MPTI‐derived reference classification was 81.2% (13/16; Cohen's κ = 0.751). That reference is itself derived from MPTI, so the figure is an internal consistency check, not external validation. The independent anchor is MPTI's agreement with Carlson's Trophic State Index (ρ = 0.879). Membership functions anchored to regional Directive thresholds matched, and did not exceed, four data‐driven recalibration schemes (13/16 against 14/16). We keep them because they express status against a fixed reference, not rank within the sample. A limiting element is attributed only where the two components lie further apart than the system can resolve, which holds at 10 of the 16 stations. The six that do not resolve are at near‐reference Işıklı and at the dry‐season Yayla stations, where both components agree. The attribution is co‐monotone with the trophic gradient and reads as transparent bookkeeping, not an emergent signature. Within‐lake clustering is strong, so the 16 observations amount to roughly three to six independent lake‐level units, and the framework is offered as a proof of concept for a single basin.