The study of phytoplankton through satellite remote sensing requires accurate estimates of the abundance and physiological status of autotrophic communities (^{Behrenfeld et al., 2008}). However, complex optical conditions or time-varying patterns of autotrophic activity can greatly bias retrievals of chlorophyll-a (Chl-a) biomass using default algorithms (^{Vilas et al., 2011}). In this way, conventional ocean color products are not fully operational in coastal and/or interior water (so called Case II waters), where substances from terrestrial origin change optical properties and introduce errors in Chl-a estimates (^{Gitelson et al., 2011}).

Following high levels of organic matter (^{Iriarte et al., 2007}; ^{Silva & Astorga, 2010}) and the significant input of freshwater runoff (^{León-Muñoz et al., 2013}; ^{Iriarte et al., 2016}) to the Inner Sea of Chiloé (ISC), we consider this region as Case II waters. The ISC (41- 45°S, Fig. 1) is an extensive coastal region with a convoluted coastal topography marked by numerous fjords, inlets and islands and receives large freshwater discharges of fluvial and/or glacial origin bearing important loads of suspended organic and inorganic material (^{Dávila et al., 2002}; ^{Calvete & Sobarzo, 2011}; ^{Pantoja et al., 2011}). One of the main features of envi ronmental variability in the region is the marked seasonal cycle in Chl-a concentration and the associated changes in patterns of photosynthetic activity (^{Iriarte et al., 2007}). These patterns have been difficult to quantify using satellite data at different spatial and temporal scales, a shortcoming that is further hindered by persistent cloud cover. Recent work in the region has helped to assess the relationships between oceanographic forcing and biological response in the ISC (^{Tello & Rodríguez-Benito, 2009}; ^{Lara et al., 2010}, ²⁰¹⁶). These studies have successfully used satellite data only as a proxy of autotrophic activity (biomass and primary production) observed in situ, underlining the potential of satellite-borne sensors in the region and providing support for our results showing that it is possible to identify cyclical patterns (e.g., interannual variability).

Figure 1 Study area off southern Chile showing the average climatology (2003-2012) of normalized Fluorescence Line Height (nFLH) for a) winter (JJA: June, July, August), and b) spring (SON: September, October, November). The position of in situ sampling stations of surface chlorophyll-a are shown as red dots. The black arrows denote the main axis of Desertores Islands. CI: Chiloé Island, RF: Reloncaví Fjord. 

Errors in satellite Chl-a algorithm retrievals in Case II waters (e.g., MODIS-OC3; OC4 for SeaWIFS; ^{O’Reilly et al., 1998}) do not provide accurate Chl-a estimates in coastal environments (e.g., ^{Gitelson et al., 2007}) and hamper our understanding of the dynamics of bio-optical properties (e.g., ^{Van Der Woerd & Pasterkamp, 2008}; ^{Ruddick et al., 2014}). To improve the noise:error ratios associated with Chl-a retrievals, a new satellite product was developed following the SeaWiFS mission to better understand space-time patterns of Chl-a biomass and other phytoplankton properties. This product, called normalized Fluorescence Line Height (nFLH) is not affected by CDOM (colored dissolved organic matter) in the same way as Chl-a biomass estimates (^{Siegel et al., 2005}; ^{Szeto et al., 2011}). In this way, the nFLH product stands as an improved indicator of physiological variability or phytoplankton biomass (^{Behrenfeld et al., 2009}). In coastal waters, the precision of default Chl-a algorithms is deprecated due to the high concentration of CDOM and non-algal particles. In this cases (and inner waters), the use of sun-induced fluorescence tend to be a better option to estimate phytoplankton biomass (^{Huot et al., 2013}).

Located at the northern tip of Chilean Patagonia, the ISC receives freshwater inflow from fluvial and/or glacial origin (^{Calvete & Sobarzo, 2011}) and terrigenous sediment supply (^{Silva et al., 2011}). The West Wind Drift (WWD) flow impinges on the continent near ~42°S and drives a strong oceanic-atmospheric coupling in the region (^{Garreaud et al., 2013}) that is manifested as a pronounced horizontal density gradient along ISC waters (^{Calvete & Sobarzo, 2011}). Retrospective satellite Chl-a validation studies in the ISC are limited by the lack of matching in situ measurements for validation (e.g., ^{Tello & Rodríguez-Benito, 2009}; ^{Lara et al., 2010}, ²⁰¹⁶). Here, to evaluate the performance of the standard MODIS Chl-a algorithm (MODIS-OC3), we address these limitations using satellite high-resolution Chl-a and nFLH estimates, together with a compilation of historic in situ Chl-a, and evaluate the nFLH algorithm as an estimator of photosynthetic biomass in the ISC ecosystem.

Oceanographic research cruises were conducted between 2003-2012 during austral winter (June, July and August, JJA, Fig. 1a) and spring (September, October and November, SON, Fig. 1b) as part of the CIMAR-FIORDOS program. Each seasonal cruise extended over the entire ISC region. Water samples were collected to measure Chl-a by filtering 250-500 mL of surface seawater on to glass fiber filters (0.7 μm size). Filters were immediately frozen (-20°C) until later fluorometric analysis (Turner Design TD-700), using acetone (90% v/v) for the pigment extraction according to ^{Parsons et al. (1984)}.

To characterize autotrophic biomass using remote sensing we used 10 years (2003-2012) of daily MODIS-OC3 high resolution (1 km) data. Satellite ocean color data were processed using NASA's software SeaDAS (SeaWIFs Data Analysis Systems) version 6.4, and following the recommendation for files with low processing level (Level-1A), and containing the information at its maximum spatial resolution (e.g., ^{Saldías et al., 2012}). To reduce the influence of additional light absorbing components, other than chlorophyll, we retrieved matching normalized Fluorescence Line Height data (^{Behrenfeld et al., 2009}) derived using normalized water leaving radiance as described in ^{Huot et al. (2013)}. Finally, to reduce the noise associated with clouds and land edges and potential low accuracy of a single pixel, a 3 × 3 pixels two-dimensional median filter was used so that each in situ measurement was centered in a nine pixel box (e.g., ^{Bailey & Werdell, 2006}). As the distribution of phytoplankton pigment biomass exhibits a log-normal distribution (^{Bricaud et al., 2002}), we used a log₁₀-transformed data for statistical analyses.

We obtained the highest fraction of matching MODIS satellite retrievals, and in situ observations, by averaging over a spatial window of 3×3 pixels compared to a “nearest-neighbor” approach. The low accuracy of a single pixel (^{Patt, 2002}), and the increase of useful MODIS observations, when using a 3×3 window (^{Bailey & Werdell, 2006}), enhances this approach for increasing/optimizing the number of match-ups. A greater number of match-ups were found during the spring months (137) than in winter (74) due to high cloud cover (Fig. 1). Histograms illustrating the frequency distribution of co-located log₁₀-transformed in situ Chl-a, MODIS-OC3, and nFLH observations during the two seasons, are shown in Figure 2. During austral winter, MODIS-OC3 overestimated in situ Chl-a biomass at all Chl-a values, while in spring months MODIS-OC3 matched well in situ Chl-a (MODIS-OC3 ≤ 1.0 mg m⁻³) towards MODIS-OC3 high values (>1.0 mg m⁻³) (Figs. 2a-2b). Seasonal histograms of log₁₀-transformed nFLH and in situ Chl-a are similar during winter months, while in austral spring (high phytoplankton activity and biomass) nFLH underestimates the center of the distribution, with a better performance towards high values (Figs. 2c-2d).

Figure 2 Histograms of match-ups data between Chl-a (black lines) and MODIS-OC3 in a) winter (JJA: June, July, August) and b) spring (SON: September, October, November). Histograms of match-ups data between Chl-a (black lines) an nFLH in c) winter and d) spring. Note that the histogram of Chl-a is repeated between seasonal panels for better comparison. 

A linear regression of in situ Chl-a versus MODIS-OC3 explained a low, non-significant percentage of variance during austral winter (R² = 0.06, P = 0.37, F = 0.80). During austral spring, in situ Chl-a showed a higher and significant relationship with MODIS-OC3, but accounted for a low fraction of the variance (R² = 0.2, P = 0.03, F = 5.33). The strength of the linear relationship (Fig. 3) may be influenced by the poor performance of the MODIS-OC3 algorithm in coastal waters where suspended particles bias estimates at the wavelengths used to estimate chlorophyll with this algorithm (^{Zhang et al., 2006}).

Figure 3 Scatterplots between in situ Chl-a and MODIS-OC3 data for the matching dates. Observations for winter and spring are with and black circles, respectively. Regression statistics are presented in the text. 

Supporting the interpretation that fluorescence provides a better estimate of autotrophic biomass despite the impact of physiology (Beherenfeld et al., 2009; ^{McKibben et al., 2012}) a linear regression of in situ Chl-a versus nFLH explained a larger and significant fraction of variance during austral winter (R² = 0.54, P = <0.01, F = 10.68; Fig. 4). However, the relationship was not significant during spring months (R² = 0.04, P = 0.42, F = 0.67). The largest mismatches between satellite observations and in situ Chl-a are associated with areas of freshwater input, mainly in the region located north of the Desertores Islands. Several rivers (e.g., Puelo River, Reñihue River) provide a large amount of allochthonous substances (^{Silva et al., 2011}). These materials of terrigenous origin may induce errors in the quantification of autotrophic biomass in the ISC during particularly seasons of high stream flow, particularly spring (^{Calvete & Sobarzo, 2011}; ^{León-Muñoz et al., 2013}).

Figure 4 Scatterplots between in situ Chl-a and nFLH data for the matching dates. Observations for winter and spring are with and black circles, respectively. Regression statistics are presented in the text. 

Our work corresponds to an initial investigation in the study of ecological processes of autotrophic communities through analysis of high-resolution satellite data. Due to the high cloud cover and the limited in situ information, it is important to consider the validation and calibration of satellite products that provide information of the spatiotemporal variability of autotrophic communities in an optically complex ecosystem. Besides, this information is relevant to the assessment of changes in ocean color, e.g., due to phytoplankton blooms anomalies (^{Hu et al., 2005}). Our work, as ^{Hu et al. (2005)} and ^{Behrenfeld et al. (2009)}, among others, highlights the importance of the nFLH from satellite to avoid interference from other components (e.g., CDOM, shallow bottom) because this product is centered at 667, 678 and 748 nm, which increases its correlation with measurements of in situ Chl-a (e.g., ^{Hu et al., 2005}; R² > 0.91, Florida, USA). Future research should be directed toward the correction of algorithms MODIS- OC3 and nFLH for a better understanding of ecological processes operating on the variability of phytoplankton biomass in the ISC. Long-term time series of ocean color imagery will also provide a future assessment of the influence of climate variability on local oceanographic conditions (e.g., ^{Saldías et al., 2016}).