On June 25, 2013, at 1645 central daylight time, an Avery Glastar GS-1 experimental amateur-built airplane, N103A, impacted water while landing on Shoal Lake, Grand Rapids, Minnesota. The airline transport pilot, who was the sole occupant, sustained serious injuries. The airplane sustained substantial damage to the fuselage. The airplane was registered to Enviro Resources, LLC, and operated by the pilot under the provisions of 14 Code of Federal Regulations Part 91 as a personal flight. Visual meteorological conditions prevailed for the flight, and the flight was not operated on a flight plan. The local flight departed Grand Rapids/Itasca County Airport pond, Grand Rapids, Minnesota, at 1605. According to the pilot, prior to takeoff from the airport pond, the pilot performed a few high-speed taxi and on-the-step maneuvers with no anomalies noted. During the short flight to the intended landing area, the pilot heard two to three aural "autopilot disconnect" voice messages. Because the pilot believed he did not engage the autopilot during the flight, he was uncertain why he was hearing the autopilot disconnect messages and was not overly concerned. The pilot established the airplane on the visual approach to the lake and was prepared to execute a touch and go landing. After the floats contacted the water, the airplane's right wing immediately lifted up, and the left wing tip contacted the water. Subsequently, the airplane nosed over. The pilot exited the airplane and was rescued by local authorities. The pilot, an ex-military and airline pilot, accumulated approximately 40 total hours in the accident airplane. The pilot reported the weather at the time of the accident as ideal for flying; sky clear and steady 5 knot wind. According to the pilot, a Dynon SkyView (model number: SV-D1000) avionics system, which included autopilot, was installed on the airplane in May 2013, at a facility in Brainerd, Minnesota. After the installation, a technician performed a flight test and noted no issues with the autopilot. The pilot picked up the airplane from Brainerd and flew to Grand Rapids for the installation of float landing gear. The accident flight was the first flight since the floats were installed. The airplane was recovered from the lake and showed structural and system damage due to water exposure. The Dynon SkyView system was removed by the pilot and delivered to the NTSB investigator-in-charge for further examination. On September 19, 2013, the Dynon SkyView unit was examined at Dynon's facilities in Washington under the supervision of a NTSB investigator. The unit was damaged by corrosion due to water exposure; however, non-volatile memory data was extracted from the unit. The data indicated that there were 9 autopilot engagements that synced the heading/track and altitude bugs during the accident flight. The data showed that something was asserting the disconnect switch or the input to both autopilot servos for a long enough period to engage the autopilot. The behavior was sporadic and occurred for various amounts of time. The autopilot would engage for a varying amount of time, then disengage when the disconnect input was again asserted. The data did not include the accident sequence; however, it ended near the time of the accident. The data for the disconnect button ended with the disconnect button being asserted for a long enough time to engage the autopilot. The data did not show the de-assertion of the disconnect input. The avionics manufacturer suggested two possibilities related to inadvertent autopilot engagement. First, it was suggested that a loose wire may have been in the system; however, the wiring of the airplane was not examined due to recovery and water damage. A resistor, which is recommended by the manufacturer, may be installed to detect a break in the circuitry and alert the pilot. The airplane's avionics system did not have the resistor installed. Second, it was possible the pilot may have inadvertently triggered the autopilot engagements. The pilot did not recall inadvertently selecting the disconnect switch on the control stick. According to the Dynon User Guide, the autopilot servos are designed so the pilot can always overpower them and move the control stick manually with minimal force, even when the servo is operating at its highest torque output capability.