Assesment of water elevation measurement quality from multi nadir altimetry missions over a 'small' river: the Garonne River case

For two decades, nadir altimetry mission measurements have been used to derive water elevation over rivers. It has been proven to be a powerful tool to estimate water elevation over big rivers (> 500 m), providing useful complementary data to in-situ gage networks. More recently, nadir altimetry data have been used successfully for some smaller rivers. It is therefore needed to assess potential errors from these measurements for these kind of river, using a well gaged basin.That's why we have investigated water elevations estimates from three altimetry missions (ENVISAT, Jason-2 and SARAL) over the Garonne River (South West of France). River width at studied river goes from 160 to 230 m.By comparison to nearby in-situ gages (IG), it has been shown that measurements from ENVISAT and Jason-2 virtual stations (VS) 100 km upstream of the estuary have errors between 20 to 70 cm for water anomaly, whereas most of the bias comes from river slope between IG and VS. 160 km upstream, the few usable ENVISAT VS have errors spanning from 80 to 160 cm. SARAL/AltiKa however, does not provide any water elevation information: there are rather no data in the record or measurements have huge bias with no correlation with water level variations. This is mainly due to high reliefs surrounding the river valley (up to 100 m difference between the valley and the top of the hills over few 10 km). SARAL/AlitKa, compared to previous instruments, has higher pulse bandwidth which results in a smaller range detection window of 30 m. It allows a higher vertical accuracy, but causes loss of data or to be 'locked' on the top of the hill even when it flies over the river valley, when there are important soil elevation variation over 'short' distance. This fact could also be observed for some Jason-2 and ENVISAT VS but to a much lower extent.Besides, water elevations time series has high frequencies due to local and upstream precipitation events. Therefore, it is more difficult to discriminate only from the VS measurements themselves if time series are physically coherent, contrarily to big rivers for which an annual cycle helps to identify usable VS. This is especially true for ENVISAT data, which has 35 days revisit period, which is not well suited for following dynamic of such kind of rivers.